Health and environment analysis and indicators for decision-making = Analyse et indicateurs santé et environnement pour la prise de décisions [full issue]

Vol. 48, No. 2, 1995 World Health STATISTICS Quarterly Rapport trimestriel de STATISTIQUES sanitaires mondiales Health and environment analysis and indicators for decision-making Analyse et indicateurs sante et environnement pour la prise de decision World Health Organization Organisation mondiale de la Sante Geneve The World health statistics quarterly replaces (since 1978) the monthly World health statistics report(publ1shed since 1967) and its forerunner the Epidemwlogical and vital statistics report (published since 194 7). It deals with the detailed analysis of selected health topics of current interest. Starting with Vol. 4 1 ( 1988). the Quarterly contains articles in either French or English with a summary in both languages. Annual subscription Sw. fr. 110.- Price per copy Sw. fr. 31 - Material from the Quarterly may be reproduced providing due acknowledgement is made. Symbols used in tables Preliminary, approximate or estimated data Data not available NII or magnitude negligible Category not applicable T Total M Male F Female N Absolute numbers The World Health Organization welcomes requests for permission to reproduce or translate its publications. 1n part or n full Applications and enquiries should be addressed to the Office of Publications. World Health Organ12at1on. Geneva. Switzerland. wrich will be glad to provide the latest information on any changes made to the text. plans for new editions. and reprints and translations already available. © World Health Organization 1995 Publications of 1he World Health Organ12at1on en1oy copyright protection 1n accor- dance with the prov1s1ons of Protocol 2 of the Universal Copyright Convention. All rights reserved. The designations employed and the presentation of the material 1n this publ1cat1on do not imply the expression of any opinion whatsoever on the part of the Secretariat of the World Health Organ12at1on concerning the legal status of any country, territory. city or area or of its autorities. or concerning the del1m1tat1on of its front,ers or boundaries. The mention of specific companies or of certain manufacturers' products does not imply that they are endorsed or recommended by the World Health Organization 1n preference to others of a similar nature that are not mentioned. Errors and omissions excepted, the names of proprietary products are d1st1ngu1shed by initial capital letters Le Rapport trimestriel de statJstiques sanitaires mond1ales remplace (depu1s 1978) le Rapport de statlstiques sanita1res mond1ales (publie depuis 1967) et son precurseur leRapportep1dem1ologiqueetdemogra- phique (publie depuis 1947) II presente des analyses detail lees sur des sujets specifiques d'inter~t courant. A compter du Vol. 41 ( 1988). le Trimestnel presente des articles orig1naux en franr;ais ou en angla1s. accompagnes d'un resume dans les deux langues. Prix de l'abonnement annuel Fr. s. 110.- Le numero Fr s. 31 .- La reproduction d'extraits du fomestriel est autorisee. sous reserve d'indication de la source. Explication des signes Donnee preliminaire. approx1mat1ve ou estimative Donnee non disponible Zero ou quantile negligeable Categorie non applicable T Total M Mascul1n F Fem1nin N Nombres absolus L'Organisat1on mond1ale de la Sante est touiours heureuse de recevo1r des deman- des d'autonsat1on de reprodu1re ou de tradu1re ses publications. en partte ou 1ntegralement Les demandes a cet effet et les demandes de rense1gnements do1vent etre adressees au Bureau des Pubhcat1ons, Organisation mondlale de la Sante. Geneve, Suisse. qu1 se fera un plaisir de fourn1r les rense1gnements les plus recants sur les changements apportes au texte, les nouvelles 6dit1ons prevues et les reimpress1ons et traductions deja disponibles © Organisation mondlafe de la Sante 1995 Les publications de l'Organisat1on mondiale de la Santa benef1c1ent de la protection prevue par les dispositions du Protocole No 2 de la Convention universelle pour la Protection du Droit d'Auteur Taus dro1ts reserves Les appellations employees dans cette publication et la presentation des donnees qui y f1gurent n'1mpl1quent de la part du Secretariat de l'Organisat1on mondiale de la Sante aucune prise de position quant au statut 1undique des pays, temtoires, v1lles ou zones. ou de leurs auton::es. ni quant au trace de leurs front1E:res ou hm1tes. La mention de f1rmes et de produ1ts commerc1aux n'1mpl1que pas que ces f1rmes et produits commerc1aux sont agrees au recommandes par l'Organisat1on mond1ale de la Sante de preference a d'autres. Sauf erreurou om1ss1on, une ma1uscule m1t1ale ind1que qu'il s'agit d'un nom depose. IX ISSN 0043-8510 Printed in Switzerland 95/10597 -Atar SA. Geneva- 5200 World Health Statistics Quarterly Rapport trimestriel de statistiques sanitaires mondiales Vol. 48, N° 2, 1995 Health and environment analysis and indicators for decision-making Contents Foreword Office of Global and Integrated Environmental Health, World Health Organization 70 Health and environment analysis for decision making Carlos Corvalan & Tord Kjellstrom 71 Linkage failures in ecological studies Markku Nurminen 78 The use of geographical information systems in studies on environment and health David J. Briggs & Paul Elliot 85 Health and environment in Sao Paulo, Brazil: methods of data linkage and questions of policy Carolyn Stephens, Marco Akerman & Paulo Borlima Maia 95 Health and environment analysis for decision-making (HEADLAMP): field study in Accra, Ghana Jacob Songsore & Greg Goldstein 108 The effect of outdoor air pollution on mortality risk: an ecological study from Santiago, Chile Manuel Salinas & Jeanette Vega 118 Assessment of the impact of ambient air pollutants on health in Helsinki, Finland Antti Ponka Environmental health indicators and sanitation-related disease in developing countries: limitations to the use of routine data sources 126 Peter J. Ko/sky & Ursula J. Blumenthal 132 An epidemiological perspective on environmental health indicators Harris Pastides 140 Framework for the development of environmental health indicators Tord Kjellstr6m & Carlos Corvalan 144 Developing indicators for environment and health John T. Wills & David J. Briggs 155 Decision-making in environmental health Eugene Schwartz & Carlos Corvalan 164 Wld hlth statist. quart., 48 (1995) Analyse et indicateurs sante et environnement pour la prise de decisions Somma ire Avant-propos [anglais seulement] Bureau de /'Hygiene du Milieu integree et mondiale, Organisation mondiale de la sante 70 Analyse sante et environnement pour la prise de decision [resume] Carlos Corvalan & Tord Kjellstrom 76 Echec d'appariement des etudes ecologiques [resume] Markku Nurminen 83 Utilisation des systemes d'information geographique dans le cadre d'etudes sur l'environnement et la sante [resume] David J. Briggs & Paul Elliott 94 Sante et environnement a Sao Paulo, Bresil: methodes d'appariement des donnees et politique adaptee [resume] Carolyn Stephens, Marco Akerman & Paulo Borlina Maia 106 Analyse sante et environnement pour la prise de decisions (HEADLAMP): etude de terrain a Accra, Ghana [resume] Jacob Songsore & Greg Goldstein 117 Incidence de la pollution de l'air ambiant sur le risque de mortalite. Etude ecologique menee a Santiago, Chili [resume] Manuel Salinas & Jeanette Vega 125 Evaluation de l'impact des polluants de l'air ambiant sur la sante a Helsinki, Finlande [resume] Antti Ponka 131 lndicateurs de la salubrite de l'environnement et maladies liees a !'hygiene du milieu dans les pays en developpement: obstacles a !'utilisation des donnees recueillies systematiquement [resume] Peter J. Ko/sky & Ursula J. Blumenthal 138 lndicateurs de la salubrite de l'environnement: perspectives epidemiologiques [resume] Harris Pastides 142 Schema pour la mise au point d'indicateurs de la salubrite de l'environnement [resume] Tord Kjellstr6m & Carlos Corvalan 153 Mise au point d'indicateurs environnement et sante [resume] John T. Wiiis & David J. Briggs 162 Prise de decision en matiere d'hygiene de l'environnement [resume] Eugene Schwartz & Carlos Corvalan 169 Fors word HEADLAMP (the Health and Environment Analysis for Decision-Making Project) is a joint col- laborative project of the United Nations Environ- ment Programme (UNEP), the United States Envi- ronmental Protection Agency (USEPA) and the World Health Organization (WHO), with addi- tional funding provided by the German govern- ment. The project began towards the end of 1993 with a feasibility study describing the epidemiological methods which could be used with routinely-col- lected health and environmental data, at the local level, to estimate the health impact of environmen- tal contamination. Field studies were later carried out in Accra (Ghana) and Sao Paulo (Brazil) to examine data availability and quality as well as the feasibility of linking health and environment data. An important milestone in the project was a con- sultation with international experts held in Geneva in August 1994. The consultation reviewed a report entitled Epidemiowgic methods for linking health and environment data for decision-making and provided advice on the future direction of the project. Among the recommendations of the consulta- tion was the publication of selected papers present- ed during that meeting. The result is this special issue of the World health statistics quarterly. For 1995- 96 our focus will be the development and testing of environmental health indicators in further field studies. Instructional materials and workshops on HEADLAMP methods and the identification, col- 70 lection and use of environmental health indicators are part of the project's ongoing activities. Numerous people have contributed comments and ideas on this project. We are greatly indebted to Dr J. Stober of WHO, who has provided continu- ous advice in the many stages and aspects of this project. We are grateful to Drs B. Nussbaum and N.P. Ross and Ms M. Conomos ofUSEPA who have provided advice and encouragement at every stage. We are also indebted to Dr H. Gopalan and Mr A. Dahl ofUNEP for their continued support, and for their comments on the reports. Among the many reviewers of the first HEAD- LAMP document on which some of these papers are based, we wish to especially thank Dr G. Gold- stein of WHO/HQ; Drs M. Krzyzanowski and C. Dora of the European Centre for Environment and Health (ECEH/WHO); Drs T. McMichael, C. Stephens and T. Fletcher of the London School of Hygiene and Tropical Medicine; Dr D. Briggs of Nene College, Northampton; and Dr K. Katsouyanni from the University of Athens. Finally we wish to thank the authors of the papers in this issue and the other participants of the HEADLAMP consultation who have made this work possible. Office of Gwbal and Integrated Environmental Health, World Health Organization, Geneva, Switzer/,and. Rapp. trimest. statist. sanit. mond., 48 (1995) Health and environment analysis for decision making Carlos CorvaJana & T ord Kjellstromb Introduction The aim of the Health and Environment Analysis for Decision-making Project (HEADLAMP) is to provide valid and useful information on the local and national health impacts of environmental haz- ards to decision-makers, environmental health pro- fessionals and the community. HEADLAMP com- bines methodologies in environmental epidemiol- ogy, human exposure assessment and other health and environmental sciences to produce and anal- yse data and present the resulting information so that it can be understood, interpreted and acted upon by those responsible for environmental health protection. The information created via lo- cal and national HEADLAMP applications will help monitor progress towards sustainable devel- opment as recommended in Agenda 21 (1). The important tools in HEADLAMP are health and environment data linkage methods, environmen- tal health indicators used to quantify and monitor the local situation, and the interpretation and deci- sion-making process. This article gives an overview of the HEAD- LAMP project and its background, while other re- ports in this issue of the World health statistics quarter- ly present further details on HEADLAMP methods, field studies examples, the development and use of environmental health indicators and the decision making-process, based on materials discussed at the first expert meeting on HEADLAMP held in Geneva, in August 1994. The health and environment situation Air, water, soil and food contamination - whether short-term at high levels or long-term at lower lev- els - have been associated with increases in morbid- ity and mortality. However, because levels of gener- al environmental pollution fluctuate, and various extraneous determinants exist, precise measures of the association between pollution levels and health outcomes are rare. Exposure to environmental pol- lution is usually involuntary and people are often ignorant of its possible effects and therefore may exert little control over it. a Scientist, Office of Global and Integrated Environmental Health, World Health Organization, Geneva. b Director, Office of Global and Integrated Environmental Health, World Health Organization, Geneva. Wld hlth statist. quart., 48 (1995) Biological and chemical agents in the environ- ment contribute to millions of premature deaths and hundreds of millions of disabilities every year (2). Certain environmental hazards affect very large populations. In many large cities millions of people are exposed to severe outdoor air pollu- tion. In addition, indoor air pollution is a wide- spread and serious problem particularly in devel- oping countries. The disease burden of indoor and outdoor air pollution is such that hundreds of mil- lions of people are affected by respiratory diseases (2). Although these estimates are very approximate - since the methodology is poorly developed and the data quality uncertain - they highlight the need to control harmful exposure. Health and environment analysis for decision- making - and data linkage where relevant - may be particularly important for those less-developed countries in which environmental pollution issues have traditionally taken second place to economic development issues. In many of these countries environmental pollution is increasing and popula- tions are growing, particularly in urban centres. Since governments are beginning to appreciate the links between economic growth and environ- mental protection, an estimate of health impact attributable to environmental pollution may allow decision-makers to prevent irreversible and costly health and environmental damage. Various epidemiological studies have been un- dertaken, mostly in developed countries, to link specific forms of environmental pollution and health-outcome data. The methods used, however, are not always easily applicable to other settings, especially if high quality data are unavailable. A serious limitation in conducting these studies con- cerns the measurement of exposure in individuals. Routine monitoring provides average exposures for geographical regions, and all persons within the relevant area are assumed to have exposures that are close to the average exposure. Lack of individual exposure data, and the limitations of routinely collected health and environment data, means that the most feasible linkage is based on the "ecological" method, for which the statistical unit of observation is a population rather than an individual (3). The World Health Organization (WHO) has developed environmental quality guidelines for dif- ferent pollutants in air, water, food, soil and the workplace (4-12). These guidelines are based on 71 epidemiological and toxicological studies and indi- cate the maximum environmental levels or the lev- els of human exposure considered acceptable in order to protect human health. However, some persons may still experience adverse health effects at levels that fall below the maximum recommend- ed levels. Moreover, in many areas of the world these levels are frequently exceeded, in some places by as much as several times the guideline levels, and actions to reduce human exposure may be difficult or very costly. In these cases, health and environ- ment data analysis provides a tool for obtaining valid estimates of the health impact of pollution, which can be used to set priorities for action. In many countries significant quantities of data on environmental quality are routinely collected via the Global Environment Monitoring System (GEMS), as well as by local authorities.c,d,e The environmental levels are compared with guideline values or standards for maximum recommended levels, but seldom is the data used to quantify any health impacts. Many countries routinely collect health outcome data in the form of morbidity and mortality statistics, but seldom are these analysed in terms of their causation by environmental or other factors. One of HEADLAMP's aims is to ap- ply epidemiological methods to link the two types of data in order to better estimate the health out- comes of environmental pollution, with a view to reducing exposure whenever possible. The link- age is typically based on routinely-collected data, and on an ongoing process to observe trends over time. Linkages must also include adequate con- trol for extraneous determinants and account for changes in the underlying population structure. Given that data are often observed over time, the linkage must also account for artefactual changes that are result from changes in disease classifica- tion or in exposure monitoring methods. The HEADLAMP methods could also play a significant role in Environmental Health Impact Assessment (EHIA), which aims at predicting the health im- pacts of development projects which may intro- duce new forms of pollution or increase existing pollution levels. Specific environmental health problems Broadly speaking, environmental contamination may occur in the general environment, the occupa- tional environment, or an individual's personal c World Health Organization. GEMS/Water 1990-2000: The challenge ahead. Geneva, WHO, 1991 (WHO/PEP/91.2). d World Health Organization. GEMS/Food. joint UNEP/FAO/ WHO food contamination monitoring programme. Geneva, WHO, 1990 (WHO/EHE/FOS/90.2). e World Health Organization. Global environmental monitoring system: A global programme JM urban air quality monitoring and assessment. Geneva, WHO, 1993 (unpublished GEMS/ Air document). 72 or domestic environment. Human exposure to pol- lution, in any of these environments, may occur via air, water, food or soil contamination. Air pollution is a general term that describes the admixture of harmful substances with the air we breathe. The most well-documented of these substances (and those usually monitored on a rou- tine basis) include sulphur dioxide (S02), nitro- gen oxides (NOx, including NO and N02), carbon monoxide (CO), ozone (03), lead (Pb), and total suspended particles (TSP, also known as sus- pended particulate matter, or SPM of which the respirable particles are of most concern, e.g., par- ticulates of up to 10 µm in size, or PM10). The major source of these pollutants is combustion: that of fossil fuels for energy generation, industrial processes and transportation; and that of solid fuels, such as coal and wood, for domestic purpos- es. The combination and concentration of outdoor air pollutants varies from city to city, according to the quantity of fossil fuels used, and according to other environmental factors, such as the geograph- ical and meteorological features of the given area. Air pollution is different from other forms of pollu- tion in that once the pollutants are in the air, exposure cannot be easily avoided. If high levels of air pollution occur in a city one can expect that a large proportion of the population will be ex- posed. However, this will vary according to the concentration of pollutants in time and space, to the proportion of time people spend outdoors, and to the ability of the individual pollutants to enter the indoor environment. Indoor air pollution is in some situations con- sidered to be more serious than outdoor air pol- lution. Pollutants tend to become trapped in- doors, resulting in higher concentrations. In addi- tion, most people spend a much larger propor- tion of their life indoors than outdoors. Indoor air pollution is a serious problem in some devel- oping countries, and occurs independently of outdoor air pollution. In many rural areas, for example, where ambient air pollution is low, the use of biomass fuel in unventilated houses causes pollution concentration levels which are much higher than in the most polluted cities. This form of pollution is likely to affect women and children more severely because they spend the longest time indoors. The problems associated with in- door air pollution in this context have been de- scribed in detail by Chen et al. (13). In this report, the authors identify the main source of pollution to be solid-fuel-fired cooking and heating stoves. The authors conclude that the evidence argues strongly that this source of indoor air pollution is a risk factor for chronic lung disease in adults, where women are most adversely affected. Coal smoke may also be a risk factor for cancer in women. The health effects in children are of great concern. Combustion-related pollutants are a risk Rapp. trimest. statist. sanit. mond., 48 (1995) factor for acute respiratory disease in young chil- dren, which is one of the main causes of infant and childhood morbidity and mortality in devel- oping countries.£ The levels of suspended particulates in the most polluted indoor environments in developing coun- tries may be several times higher than the daily averages measured in cities with severe air pollu- tion problems. However, the main difference be- tween outdoor and indoor air pollution is that indoor pollution has a larger range, with periods of no exposure. Outdoor air pollution tends to be relatively more constant (for example, over the period of one day), and even though the level of outdoor pollution may vary greatly, it is hardly ever absent (an ubiquitous exposure). Water pollution is a pressing problem in many areas of the world, irrespective of the level of devel- opment. Most drinking water is obtained from groundwater or surface water and can be contami- nated by the presence of physical, chemical and biological agents. Biological pollution is often of greatest concern, particularly in less developed countries, and in rural areas. Diarrhoeal disease due to faecal pollution of water is a widespread problem and a major cause of infant deaths (14). Chemical pollution in water includes nitrates and nitrites, pesticides, volatile organic compounds, and heavy metals such as arsenic and lead, and to a lesser extent, mercury, cadmium and other metals. In local "hot-spot" areas these have caused im- portant outbreaks of poisoning, for example, Minamata disease and Itai-Itai disease (15,16). Environmental pollution can also be transmit- ted through food and soil, often involving contam- ination of foodstuff with chemicals, specifically pes- ticides, or by biological agents. In addition, food contamination also includes the accumulation of pollutants in the food chain, as for example some metals become concentrated in the bodies of fish. Soil, in tum, may be also chemically contaminated with pesticides, industrial waste products and heavy metals such as lead or cadmium. Another environmental health hazard of con- cern is ionizing radiation which can be emitted from nuclear power station accidents ( e.g., Cher- nobyl) or from natural sources, such as in the form of radon gas creating indoor air pollution. The health effect of main concern after general envi- ronmental exposure is cancer, but a number of other effects are being investigated after the Cher- nobyl accident.g f World Health Organization.lndoorairpollution.frombi.omass fuel. Reporlof a WHOamsultation. Geneva, WHO, 1992 (WHO/PEP I 92.3A). K World Health Organization. International programme on the health effects of the Chernobyl accident (IPHECA). Reporl of the management committee meeting, Geneva 16-17 March 1994. Geneva, WHO, 1994 (WHO/EOS/94.24). Wld hlth statist. quart., 48 (1995) The environmental health hazard pathway The environmental health hazard pathway is de- scribed in Fig. 1. Traditional hazards (such as hu- man faeces in densely populated areas) or modern hazards (such as air pollution from cars) result in environmental emissions. Once in the environ- ment, pollutants may be transmitted via air, water, food or soil, thus entering the human body by inhalation, ingestion or dermal absorption. The amount of any given pollutant that is absorbed is often termed the dose, and may be dependent on the duration and intensity of the exposure. Target organ dose refers specifically to the amount that reaches the human organ where the relevant ef- fects can occur. The first effects may be sub-clinical changes, which in tum may be followed by disease and in some cases even death. Brief examples are given below. Air Contaminated air may enter the human body by inhalation of air pollutants, but may also be ab- sorbed through dermal contact. The most com- mon health effects are associated with the respira- tory system, particularly in more sensitive persons, such as children and the elderly. For example, particulate matter and sulphur dioxide, two very common air pollutants, may cause bronchocon- striction, chronic bronchitis or chronic obstructive lung disease. Water Contaminated water is usually absorbed by the hu- man body by ingestion, but some contaminants may also be absorbed by inhalation or via dermal contact. Depending on the type of contamination, different vital organs may be targeted by the differ- ent contaminants. For example, contamination with volatile organic compounds may affect the liver or the kidneys, causing hepatitis or kidney failure. Food and soil Consider, for example, lead contamination. Lead in food or soil is absorbed from the gastrointestinal tract (up to 50% may be absorbed in children compared to 10% in adults). Almost all organ sys- tems can be potential targets for lead, including effects on haem biosynthesis, the nervous system and on blood pressure. Measuring exposure accurately and precisely is of great importance when seeking to establish ex- act associations with health outcomes. It is often impossible to measure exactly how much of a par- ticular exposure reaches the target human organ. Biological monitoring techniques can provide good estimates of dose but they are not always practical - or available - when assessing exposure to environmental pollutants. In some cases, exposure 73 Fig. 1 The environmental health hazard pathway: conceptual framework at the individual level Chaine des risques pour la sante lies a l'environnement: schema theorique au niveau individuel Traditlonal hazards - Risques traditionnels Human activities - Activites humaines Natural phenomena - Phenomenes naturals Mortality- Mortalite Modern hazards - Rlsques nouveaux Development activities - Activites liees au developpement levels around an individual can be measured by using personal monitors. However, it is often the case that measurements are based on samples which provide averages (e.g., of geographical areas), not on actual individual exposure. There- fore, the existence of a measurable amount of con- centration of a pollutant, even when higher than recommended levels, is not always sufficient infor- mation to infer health effects. The joint use of health and environmental data, plus data linkage where feasible, is essential for the management of known environmental health problems, such as those described above. 74 Developing new analysis and interpretation tools Two important criteria must be considered when developing new methods for linking health and environment data, or assessing and adapting exist- ing ones. On the one hand the methods must be simple, inexpensive to implement and operable with available data, thus allowing rapid assessment. On the other, they must be unbiased and produce results that agree with those obtained from more detailed studies, for which the statistical precision can be quantified. Accuracy is of the greatest im- portance, since credible results are indispensable for promoting appropriate action. If the methods are overly complex, requiring extensive resources and large amounts of additional data collection, few of the less-developed countries will be able to apply them. HEADLAMP methods are generally based on routinely collected data and, where relevant, on data collected from specifically designed rapid sur- veys. Therefore, even with some control for extra- neous determinants, these methods should not be seen as substitutes for epidemiological linkages performed at an individual level. As in many other environmental epidemiology projects, the long- term challenge is to develop new forms of study design and data analysis techniques for environ- mental epidemiology. Nevertheless, much can be achieved for deci- sion-making purposes through careful application of existing methods. In addition, researchers in countries where no studies have been performed should be encouraged and supported in undertak- ing epidemiological studies that would help to shed light on the effects of specific forms of envi- ronmental pollution in their particular settings. If detailed information on the dose-response re- lationship of pollutants in different settings around the world were available, techniques of risk analysis could be used to estimate the impact of exposure on different populations. This implies knowledge about exposure, estimates of the popu- lation exposed and of the health effects associated with the exposure in the form of a dose-response function. At present, this is only possible by extrap- olating from available study results done in one country ( often developed) to other countries ( of- ten less developed), or by using the limited num- ber of studies available from developing countries. In addition, this can be done only with respect to pollutants for which well researched exposure-re- sponse relationships have been established. One of the limitations of this approach is that background exposure levels and the distribution of extraneous determinants may differ between populations. A more serious concern is the lack of certainty re- garding the assumed association between environ- mental pollution levels and actual exposure in in- dividuals. In addition there is a significant time lag Rapp. trimest. statist. sanit. mond., 48 (1995) between exposure and health effects for many pol- lutants. This means that the health outcomes ob- served at present may be due to exposures which occurred many years or even decades earlier. Nevertheless, in spite of these limitations, risk anal- ysis may be the only tool available for estimating the health outcomes of environmental pollution in areas where health monitoring is not undertaken, or for which data quality is poor, or for obtaining crude estimates of health impacts in very large population groups. Establishing environmental health indicators To monitor progress in environmental health management and to quantify the health impacts, at both local and national levels, it is important to establish an appropriate set of environmental health indicators based on health and environ- ment monitoring data, and data linkage analysis. The key aspect of an indicator is the transition from "data" to "information". In this context en- vironmental health indicators can be understood as synthesized information regarding known envi- ronment-related diseases or contaminants with known adverse health effects. Once identified, these indicators can be used to establish improved and more cost-effective environmental monitoring and management programmes. Data linkage methods and development of envi- ronmental health indicators can be very useful tools for policy making and management. Reduc- tion of exposure requires investment by people and authorities, and given the shortage of resourc- es for essential development activities in virtually all countries, this investment can only occur if sound information is available to support it. Data linkage and the development of indicators can pro- vide decision makers with tools for monitoring en- vironmental health problems and assessing the ef- fect of their policies. The HEADLAMP process The project has 3 defining characteristics which differentiate it from ad hoe epidemiological studies. 1. HEADIAMP is based on already known and scientifically established relationships between environmental exposure and health effects. Based on these relationships it is possible to define environmental health indicators, which within the context of HEADIAMP are chosen for their potential value in the decision-making process. Research to establish new relationships between exposure and health effects is a related but separate activity. 2. The environmental health indicators used in HEADIAMP are usually based on routinely- collected data. This is cost-effective and encour- Wld hlth statist. quart., 48 (1995) ages improved use of these data and provides guidance for more useful and valid future data collection. To measure the relevant environ- mental health indicators it may also be neces- sary to collect new data. In these situations HEADIAMP will encourage the use of appro- priate, low-cost techniques. 3. The ultimate aim of HEADIAMP and its envi- ronmental health indicators is to obtain infor- mation on which to base preventive action against environmental health problems. HEADIAMP is intended to be an ongoing activity focusing on information needs at local and national levels. On-going assessment will indicate environmental health trends, and en- able policy-makers and managers to assess the value and performance of their policies over time. National and local capacity-building is, therefore, also an integral part of the HEAD- IAMP approach. Based on these characteristics, a framework for implementing HEADIAMP in the field has been developed and is summarized in Fig. 2. Application of HEADIAMP methods is motivated by concern regarding specific environmental conditions and their potential adverse impact on human health. In practice, application of the HEADIAMP process follows 3 stages, reflecting the 3 characteristics de- scribed above. The first stage of the process is the definition and validation of the problem. The known links between a defined environmental factor and its associated health outcomes provide the starting point. These links will already have been estab- lished in previous research and in the literature. Basic information requirements are identified at this stage. At the second stage (application/assessment and quantification environmental health indica- tors), detailed specifications for data are identified based on the specific setting, taking into account that "ideal" data will not always be available. Rou- tinely-collected data, and when necessary, specifi- cally designed rapid survey data, are then analysed to obtain information on environmental health effects or conditions. The variables produced through this process are the environmental health indicators. Depending on the problem and/ or fea- sibility of obtaining all the relevant data, environ- mental health indicators will be derived from: health data (e.g., morbidity rates attributable to definable environmental factors); environmental data (e.g., pollution levels with human health implications); or data linkage (e.g., time-series analyses). At the third stage (policy formulation/imple- mentation), policy action is taken based on the levels and trends of the environmental health indi- cators. Repeated assessments may be undertaken at appropriate intervals in order to monitor changes 75 Fig. 2 The HEADLAMP Process - Processus HEADLAMP Environmental factors - Facteurs li6s A l'environnement + Relallonshlp - Relation t Health factors - Facteurs sanitaires WH09535t Specification of environmental health data or information requirements - D6termination des donn6es et renseignements n6cessaires sur la salubrit6 de l'environnement Monitoring - Surveillance continue Confounder data - Donn6es sur les facteurs de confusion Demographic data - Donn6es d6mographiques Health - Sant6 Environment - Environnement Definition/validation - D6finition/validation Surveys - Enqultes Application/assessment and quantification or ascertainment of environmental health indicators - Application/6valuation et mesure ou v6rification des indicateurs de la salubrit6 de l'environnement Policy formulation/ implementation - Elaboration/mise en oeuvre de la politique in health and/or environmental status and to as- certain if any particular trend has been established. Thus, repeated assessment would contribute to monitoring the effects of policy implementation, provide support for changes in policy, and convey environmental health information to the public and other stakeholders. A decision to cease moni- toring activities may be taken once pre-set targets have been met on a sustained basis. The application of HEADLAMP's tools for environmental health management strives to- wards the prevention of environmentally related disease and the promotion of a healthy environ- ment. This is consistent with UNCED's Agenda 21 efforts for sustainable development as well as the ideals in the Alma-At.a declaration of "Health for All". Agenda 21 recognizes that both insufficient and inappropriate development can result in se- vere environmental health problems. Thus, while development cannot occur without a healthy population, such development should in turn not create additional environmental health problems (1). "Health for All" ideals of equity in health are also closely linked to environmentally related health problems, where it is clearly recognized that some sectors of the population are adversely affected by their living environment, and by poor access to health services. The implementation of HEADl.AMP activities at the local level aim at providing a contribution to these two processes already in motion. If effective environmental health decision-making and actions can be sus- tained and multiplied in many local situations, they will have a significant impact at the national and global levels. 76 Summary This article gives an overview of the HEADLAMP project and its background. The project aims at bringing valid and useful information on the local and national health impacts of environmental hazards to those responsible for environmental health protection. Its main tools are health and environment data linkage methods, environ- mental health indicators used to quantify and monitor the local situation, and the interpretation and decision- making process. The project has 3 main characteristics which differenti- ate it from ad hoe epidemiological studies (i) it is based on scientifically established relationships between envi- ronmental exposures and health effects; (ii) it uses routinely-collected data, or where necessary, new data collected using low-cost techniques; and (iii) it aims at providing information on which to base preventive ac- tion against environmental health problems. Based on these characteristics, a framework is proposed for the application of HEADLAMP for managing specific envi- ronmental health problems. Resume Analyse sante et environnement pour la prise de decision Cetarticle presente le projet HEADLAMP (analyse sante et environnement pour la prise de decisions) et le contexte dans lequel ii se place. Ce projet est destine a fournir aux responsables de !'hygiene de l'environne- ment des informations exactes et utiles sur les conse- quences sanitaires. aux niveaux local et national, des risques lies a l'environnement aux responsables de !'hygiene de l'environnement. II fait appel principale- Rapp. trimest. statist. sanit. mond., 48 (1995) ment aux methodes de raccordement des donnees sur la sante et sur l'environnement, aux indicateurs de la salubrite de l'environnement utilises pour quantifier et surveiller les problemes locaux, ainsi qu'au processus d'interpretation de prise de decisions. Le projet presente trois caracteristiques principales qui le differencient des etudes epidemiologiques ponctuel- les : i) ii se fonde sur des relations scientifiquement averees entre des risques lies a l'environnement et des effets sur la sante; ii) ii utilise des donnees recueillies de maniere systematique, et, si necessaire, d'autres don- nees rassemblees a l'aide de techniques peu coOteu- ses; iii) ii fournit des informations qui serviront de base a la prevention des problemes de salubrite de l'environ- nement. Le schema etabli a partir de ces trois points pourra servir de cadre a !'application du projet pour gerer les problemes concrets de salubrite de l'environ- nement. References/References 1. Agenda 21. The Unitf.d Nations Programme of Action from Rio. New York, United Nations, 1993. 2. Our Pio.net Our Health. Report of the WHO Commission on Health andEnvironment. Geneva, World Health Organization, 1992. 3. Beaglehole, R. etal. Basic epidemiology. Geneva, World Health Organization, 1993. 4. Air quality guidelines forEurape. WHO European series No. 23, Copenhagen, World Health Organization, 1987. 5. Guidelines for drinking-water quality. Vol. 1, R.erommendations. Geneva, World Health Organization, 1993. 6. Summary of acceptanus: worldwide and regional Codex standards. Codex Alimentarius part 1, rev.4. Food and Agriculture Organization/World Health Organization, 1989. 7. WHO Technical Report Series, No. 647, 1980 (R.erommended health-based limits in occupational exposure to heavy metals). OMS Serie de Rapports techniques N° 647, 1980 (Exposition aux metaux lourds: limites recommandees d'exposition profes- sionnelle a visee sanitaire). Wld hlth statist. quart., 48 (1995) 8. WHO Technical Report Series, N° 664, 1981 (R.erommended health-based limits in occupational exposure to sel.ectf.d inganic solvents). OMS Serie de Rapports techniques N° 664, 1981 (Exposition a certains solvants inganiques: limites recommandees d 'exposition professionnelle a visee sanitaire). 9. WHO Technical Report Series N° 677, 1982 (R.erommended health-based limits in occupational exposure to pesticides). OMS Serie de Rapports techniques No. 677, 1982 (Exposition aux pesticides: limites recommandees d 'exposition professionnelle a visee sanitaire). 10. WHO Technical Report Series N° 684, 1983 (Recommended health-based occupational exposure limits for sel.ected vegetabl.e dusts). OMS Serie de Rapports techniques N° 684, 1983 (Exposition a certaines poussieres vegetales: limites rerommandees d 'exposition professionnell.e a visee sanitaire). 11. WHO Technical Report Series N° 707, 1984 (Recommended health-based occupational exposure limits for respiratory irritants). OMS Serie de Rapports techniques No. 707, 1984 (Exposition aux substances irritantes pour l.es voies respiratoires: limites rerommandees d 'exposition professionnelle a visee sanitaire). 12. WHO Technical Report Series N° 734, 1986 (R.erommended health-based limits in occupational exposure to sel.ected mineral dusts). OMS Serie de Rapports techniques N° 734, 1986 (Exposition a certaines poussieres miniral.es (silice, charbon): limites recom- mandies d 'exposition professionnelle a visee sanitaire). 13. Chen, B.H. et al. Indoor air pollution in developing countries. Worldhealthstatisticsquarterly, 48(3) :127-38 ( 1990). Chen, B.H. et al. Pollution de !'air a l'interieur des habitations dans les pays en developpernent [resume]. Rapport trimestriel de statistiques sanitaires mondial.es, 43 ( 3) : 136 (1990). 14. Martinez,J. etal. Diarrhoeal diseases. In:Jarnison D.T. etal., Disease control priorities in develnping countries, Oxford University Press, 1993. 15. World Health Organization. Environmental health criteria No. 1, Mercury. Geneva, WHO, 1976. Organisation mondiale de la Sante. Criteres d 'hygiene de l'environnement N' 1, Mercure. Geneve, OMS, 1976. 16. World Health Organization. Environmental health criteria No. 134, Cadmium - environmental aspects. Geneva, WHO, 1992. [anglais seulernent]. 77 Linkage failures in ecological studiesa Markku Nurminenb The objective of this paper is to give a brief review of ecological studies which deal with aggregate data for groups of people rather than for individu- als. The types of ecological study can be either explorative ( disease mapping), multi-group ( dis- ease-exposure correlation or regression) or time- trend (time-series) studies. Ecological studies are the main epidemiological method used in the Health and Environment Analysis for Decision- making (HEADLAMP) project, described in detail in the article by Corvalan and ~ellstrom in this issue (1). The main advantage of the ecological approach is that it permits the study of very large popula- tions, and studies are often relatively easy to con- duct using existing databases in a fairly short peri- od of time. On the other hand, ecological studies are subject to unique biases not present in individ- ual-level studies. The various sources of biases in ecological data derive from linkage failures; that is, an ecological study does not link individual disease events to individual exposure or covariate data. In addition to sampling error, ecological estimates are prone to measurement error, sensitive to mis- specification of the regression model form, and the conditions of confounding differ from those of individual-level studies. The primary strategy for bias prevention in ecological studies, as in epidemi- ological studies in general, must be the design of a valid study. If bias persists, there are limited statisti- cal methods available for bias reduction in the data analysis phase. Finally, this paper recalls method- ological recommendations proposed by Morgen- stern and Greenland for valid ecological study de- sign and data analysis (2,3). Basic characteristics of ecological studies Objectives Ecological studies describe and analyse correla- tions between variates measured in populations in groups or regions rather than in individuals. The a This article is part of the methodology report titled Epidemiowgic methods j<ff linking health and environment monitoring dataj<ffdecision making on the World Health Organization's Health and Environment Analysis for Decision-making (HEADLAMP) Project. It was prepared for the HEADLAMP project during the author's visits to the WHO Office of Global and Integrated Environmental Health, Geneva, in 1994. b Department of Epidemiology and Biostatistics, Finnish Institute of Occupational Health, Helsinki (Finland). 78 variates concerned are aggregated observations of individuals, for example, rates and averages mea- sured in groups of people. The ecological method is a research technique used in observational studies to detect and recog- nize patterns of disease across space and time, and to relate the rates of disease frequency to environ- mental, behavioural, and constitutional factors. The ecological design is also useful in epidemiol- ogy for evaluating intervention on risk factors for various diseases, for example, the effect of a low- cholesterol diet on the rate of ischaemic heart disease. Ecological analyses thus use aggregate or grouped data, rather than individual level data, as the basic sampling unit of analysis. The grouping variate is usually a geographical region, although other factors such as ethnicity, socio-economic class, time period, etc., could also be used for grouping. Ecological analyses of exposure-disease relationships can be biased by model mis-specifica- tion, confounding, nonadditivity of exposure and covariate effects (effect modification) and non- comparable standardization. Ecological correla- tions and rate estimates can be more sensitive to these sources of bias than individual-level esti- mates, because ecological estimates are based on extrapolations from unobserved individual-level data. This article has assembled recent results which describe various methodological issues concerning ecological studies in environmental health re- search, with emphasis on biases in ecological esti- mates when compared to their counterparts in in- dividual-level designs. Some strategies described here may be useful for minimizing bias in ecologi- cal estimates. The description draws heavily on the important works of Brenner, Greenland, Morgen- stern and Robins (3, 4, 5, 6, 7, 8, 9). Types of ecological design An exploratory disease mapping can detect geo- graphical disease clusters without any direct incor- poration of exposure information. An example is a cancer atlas in which the spatial variations in mor- tality that exist within and between countries for some cancers may form the point of departure for further studies to uncover possible determinants of risk. With the availability of exposure information, an epidemiologist can study its association with Rapp. trimest. statist. sanit. mond., 48 (1995) disease outcome in a single population at a given point in time. Alternatively one can compare the correlations or, preferably, the slopes ofregression lines in two or more populations. In either design, the data accrue in a relatively short time span but there are no multiple measurements over an ex- tended time period. In time-trend ecological studies, a single popu- lation may be followed up with regard to its changes in exposure over time and the respective changes in the rates of disease over the same period of time. For example, the association between the amount of fluoridation in the drink- ing water and the corresponding incidence rates of caries in the teeth was studied in a population of one town in Eastern Finland for several years (before the community experiment was stopped in fear of possible adverse health effects). An example of a 2-group comparison design with time trend data is the community health pro- motion program (towards healthier eating habits, among other changes) conducted in the province of North Karelia in Eastern Finland, with the prov- ince of Kuopio, in the same region, selected as a control ( 10). The incidence of coronary heart dis- ease declined in the intervention population over time, but a similar favourable trend was also ob- served in the neighbouring, control population. Thus the relative risk was by and large unaffected by the program. This study's result emphasizes the general epidemiological principle that inclu- sion of an unexposed control population is neces- sary for estimating the possible effects of exposure. In the multi-group comparison design, data on exposure to an agent and the health outcome are collected on a group basis for several regions. A study of this type is the time-series data from the Nordic countries showing the change in the mor- tality rate from cervical cancer following the initia- tion of Pap-smear screening ( 11). Because the rate of cervical cancer was declining before the intro- duction of the programmes, it is again uncertain whether screening indeed had any favourable ef- fect on cancer mortality. Another example of this kind is an ecological study on the chlorination of drinking water and lung cancer incidence in Nor- way (12). This study was a 2-level, geographical investigation in 19 Norwegian counties and 97 mu- nicipalities, in which a slightly increased incidence of cancer of the colon and rectum was found. Advantages of ecological studies The main advantage of the ecological approach is that it permits the study of very large populations ( e.g., entire countries). An example is an analysis of the relation between the national coronary heart disease mortality rates and fatty acid balance ( the estimated polyunsaturated/ saturated fat ra- tio) in the diet of populations of approximately 20 countries (13). This study used WHO/FAO Wld hlth statist. quart., 48 (1995) data (14) and regularly published mortality statis- tics from the participating countries. Ecological studies remain common because they are relatively easy to conduct using existing data bases in a relatively short period of time. Thus a judiciously implemented ecological approach can serve as a cost-effective alternative for screen- ing or monitoring many diseases and environmen- tal conditions across geographical areas. Because it is feasible to study large populations ecologically, relatively small increases in risk can be detected. For example, the association of air pollu- tion and daily mortality was studied in Birming- ham, Alabama (United States of America) (popula- tion 884 OOO) with time series data from 1985 through 1988 (15). A small significant association (relative risk= 1.1) was found for a 100-µg increase in respirable particles. Ecological studies sometimes cover popula- tions more markedly divergent in their exposures than those that can be readily obtained in studies of individuals. The ecological approach may also be useful for the investigation of clusters of dis- ease in relatively small geographical areas. In ad- dition, description of variations in disease fre- quency among or within geographical regions can provide suggestions for more analytical epidemio- logical research. Disadvantages of ecological studies The ecological design provides no information at all on the joint distribution of the exposure and disease variates at the individual level. Thus there is no way of knowing from the ecological data which individuals experiencing the health outcome have in fact been exposed to the environmental risk factor. Inferences of individual-level exposure-disease relationships from ecological data are justified only under exceptional conditions. Therefore, deriving individual-level relationships from ecological data should be viewed as a particularly tentative and exploratory process, which may yield very mislead- ing results. Routinely-registered health event data (e.g., hos- pital discharges) may not suit the purposes of the ecological research in question, for instance, be- cause of an unusable classification system of diseas- es. For less severe health events such as acute asth- matic attacks there may be no records available at all. It may also be difficult to define population denomi- nators ( e.g., the catchment population of the hospi- tal) corresponding to health event numerators. Extra effort may be needed to create health and environmental data sets with comparable popula- tion subgroups. Health data are usually available for administrative units such as municipal health districts, municipalities, or provinces, whereas en- vironmental pollutants and other exposures tran- 79 scend their boundaries. For example, air quality data are measured in Helsinki (Finland) by district municipal authorities at several automatic moni- toring stations which are located irregularly all over the city. The air pollutant levels in various parts of the city are assessed by mathematical mod- els to provide estimates of exposure for popula- tions in the ecological units of analysis. As an illustration of the problem of obtaining inadequate exposure data, consider an ecological study to detect an association between the birth rate of infants with neural-tube defects in the Aus- tralian state of New South Wales and Australia's use of trichlorophenoxyacetic (2,4,5-T) acid dur- ing the previous year (16) (Fig. 1). The only Austra- lian data accessible to the investigators for record- linkage were relatively complete epidemiological information on annual rates of birth malforma- tions in New South Wales in 1965-1976 and herbi- cide usage for the whole of Australia in the same period. Data on the use of 2,4,5-T in each state were not available. Estimates of the actual number of persons in New South Wales occupationally or environmentally exposed to this herbicide were evidently too difficult to obtain. The proportion of the population classified as living in rural areas in this Australian state was presumably approximately the same as in the whole of Australia (i.e., 15%), and a linear correlation (r = 0.65) between the rate for neural-tube defects in New South Wales with the previous year's use of the herbicide in Aus- tralia was found. However, as the authors pointed out, the crudeness of the exposure data and the record-linkage nature of the analysis cannot be Flg.1 Ecological relationship between the rate of neural-tube defects (NTD) per 1 OOO births and the utilization of trichloro- phenoxyacetic (2,4,s-n acid (in equivalent tonnes), Australia. Correlation entre le taux d'anomalies du tube neural (NTD) pour 1 OOO naissances et !'utilisation de l'acide trichloro- pMnoxyac6tique (2,4,s-n (en tonnes), Australie :a .. 3 c .,.:JI .c.!!! !: 2! i! ~~ !.8. rl i!! .; i ,:,C 2 "'.! a-= _:::, I! 't:I :::,u, !~ I o~ "'0 i iJi 't:I § {!!. 50 200 350 500 Utilization of 2,4,5-Tacid-Utillsatlon de l'acide 2,4,5-T Note: normal regression line and smoothed regression curve fitted to the time-series data of Field & Kerr (Ref. 16). The circles represent yearly observations between 1965 and 1976. -Oroite de regression normale et courbe de regression lissee appliquee aux series de donnees chronologiques de Field et Kerr (Ref. 16). Les cercles correspondent au nombre annuel de NTD observes entre 1965 et 1976. 80 taken as direct evidence of any causal association involving 2,4,5-T herbicide. Ecological studies are subject to unique biases not present in individual-level studies. Therefore the demand for methodological rigor is great. When biases cannot be ruled out or quantified with available ecological data, further exploration will require individual-level studies. Biases in ecological vs. individual-level studies The ecological fallacy Unlike an individual-level study, an ecological study does not link individual disease events to individual exposure or covariate data, nor does it link individual exposure and covariate data to one another. The special biases of the ecological studies spring from these linkage failures. The aggregation bias or cross-level bias refers to the incorrect estimates of exposure effect that re- sult from the analysis of data aggregated across study groups. Because the groups are typically het- erogeneously exposed, cross-level bias is a more complex issue than that of a simple confounding by group ( specification bias). In addition to the sources of bias inherent in individual-level studies, ecological estimates of ef- fect can be biased by confounding by group and effect modification by group. Covariates responsi- ble for ecological bias may not even be confound- ers or effect modifiers at the individual level (4). Because the geographical units on which the ecological sampling is based are divisible, ecologi- cal analyses are often done at several levels that may not give identical results. Therefore, it is essen- tial to check the stability of the results for their interpretation. Sampling error One problem that has not been sufficiently dis- cussed in the literature on ecological studies is that the ecological estimates of exposure are based on sample surveys and so are subject to sampling er- ror. If the sampling error is not negligible, expo- sure variates (ecological regressors) have standard errors, which will bias the regression coefficients ( 17). If estimates of the standard errors are avail- able from surveys, these may be incorporated to correct for the bias. For example, the smoking prevalence estimates obtained in a survey in Ger- many were based on samples always exceeding 2 OOO and generally exceeding 5 OOO (from popula- tions ranging in size from 371 OOO to 2 760 OOO): for prevalence estimates ranging from 12% to 26%, the standard errors were very small (always under 0.8 and mostly under 0.5%) (5). Thus, in this example there was no need to correct the regres- sion coefficient (and its variance) by using an errors-in-variables model ( 17). Rapp. trimest. statist. sanit. mond., 48 (1995) Measurement error The proneness of ecological estimates to measure- ment error can be a far more important source of uncertainty than the sampling error in the basic variates. Apart from basic demographic variates (such as sex, age, and vital status), most measure- ments used in ecological analyses contain errors. Measurement error has different effects for ecolog- ical and individual-level studies. Sample design in an ecological study is more complex than that in classical epidemiological ( e.g., cohort) studies. This is because the samples used to estimate the distribution of the disease, exposure and covariate distributions for an ecological study often are n_ot necessarily the same. Therefore, the measurement errors that arise from this structure of an ecological study have to be considered separately for the ex- posure, disease outcome and covariates. One fortunate result regarding ecological esti- mates is that non-differential mis-classification of a binary covariate does not reduce the ability to con- trol confounding by the covariate. Naturally, if, in addition to the misclassification, there are sam- pling errors, control of the confounder may be compromised by these errors (6). Ecological studies in epidemiology typically deal with cause-specific mortality (and morbidity) rates rather than total mortality. Therefore, mis- classification of disease outcome can be a source of severe bias. This bias can be considerably greater than the sampling variability of the disease out- come (a dependent variate in the regression). Im- perfect disease specificity (i.e., false positive rate) induces no bias in the risk difference estimate, but this estimate is biased towards the null value by imperfect sensitivity. With disease misclassification ( due to either imperfect sensitivity or specificity) the linear regression estimate of risk ratio is biased toward the null (5). The most demanding sampling problem in eco- logical studies is adequate measurement of the po- tential confounders. If existing databases are used, then obviously one is limited by the extent of those information sources. The use of routinely collected health and environmental data will by necessity restrict confounder control possibilities to those covariates that have been measured. These variates usually do not include all the relevant covariates for the relationships being studied. Most covariates used in ecological regressions are either surrogates or rather crude measures of the true confounders. The problem is com- pounded by the need for measurement of the with- in-region multivariate (joint) distribution of the confounders; univariate distributions of the covari- ates or a confounder score may not suffice to achieve full control of confounding. For example, an ecological study was carried out in the Czech Republic to test the hypothesis that atmospheric pollution levels affect infant mor- Wld hlth statist. quart., 48 (1995) tality risk (18). The study permitted only limited control of confounding. The effects of smoking, indoor pollution from heating or cooking, and family size were not adequately controlled by the adjustments made using the socioeconomic vari- ables for which data were available. These com- prised mean income, mean savings, mean number of persons per car, proportions of total births out- side marriage, and legal abortions per 100 live births. Thus, as acknowledged by the investigators, an unknown amount of residual confounding was likely. One fortunate result regarding ecological esti- mates is that nondifferential misclassification of a binary confounder does not cause bias. Naturally, if sampling errors are added to the misclassifica- tion, control of the confounder may be compro- mised by these errors (7). Mis-specification of model form An important issue in the study of ecological data concerns the sensitivity of these analyses to model specification. Ecological summary rate ratios can be very sensitive to the chosen model form ( 11). In contrast, individual-level effects summarized in terms of rate ratios appear insensitive to the choice of model structure (8). Recall the linear correlation between herbicide use and incidence rate of neural-tube defects found in New South Wales (16). The 12-year time- series data were depicted as a scatterplot with a line drawn to indicate the average trend. The investiga- tors were keen to observe that the figure suggests that linear correlation disappeared in 1975 and 1976 (Fig. 1). Thus there is reason to check this observation. The fitted regression line was estim- ated to be: R = l.76+0.00115T, where R is rate of neural-tube defects in New South Wales per 1000 births, and T is usage of trichlorophenoxyacetic acid (2,4,5-T acid in equivalent tonnes) in Austra- lia in the previous year. A nonparametric, smooth regression curve ( 19) displayed an initial steady rise in the incidence rate which seemed to level off despite the notable increase in herbicide use in the last two years. Another analysis confirmed that a quadratic term for exposure effect was significant with a negative coefficient in the polynomial re- gression. Thus the simple linear model would seem to be a misrepresentation of this ecological time-series. The bias of an exposure-disease relation may depend on other aspects of the nature of the data. For example, even if the ecological regression were nearly linear, there might be covariates whose ef- fects were nonlinear in the individual-level regres- sion (9). While most individual-level studies employ some type of exponential or log-linear (multiplica- tive) model for the rates, the ecological studies typically use a linear (additive) model. Despite the 81 frequent use of multiple linear-regression models in the analysis of ecological data, these models are not well suited to modelling disease rates and do not provide a good fit for most cancer and cardio- vascular data sets. The reason for this inadequacy is that the effects of most ecological regressor variates are likely to be nonlinear and nonadditive over the studied range. For example, most rates vary so much with age that the rate ratio is almost always more nearly constant over age spans than rate dif- ference. Confounding Failure to identify, measure, or control important covariates of the exposure-disease relation is known as confounding. This problem is shared by cohort and case-referent study designs as well as all types of ecological designs, but is more serious in ecological studies than in individual-level studies. This is because ecological bias can be produced by other factors, such as effect modifiers acting inde- pendently of the confounders or tangling with their effects. Thus, the conditions which would prevent con- founding in ecological studies are logically inde- pendent of the conditions that guarantee no con- founding in individual-level studies. If the latter conditions are mistakenly applied in ecological studies, it can lead to omission of important covari- ates from the analysis. As Greenland (3) reminds us, a covariate may be ignorable at the individual level but not at the ecological level, or vice versa. There will be no ecological association of expo- sure distributions with disease outcome rates across groups if there are no exposure effects on either disease risk or the distributions of other risk factors by group. This condition obtains even if within each group exposure levels are associated with the other risk factors (9). Let us recall the study in the Czech Republic that found pollutant-specific associations between neo- natal mortality and measured atmospheric pollu- tion concentrations ( 18). The number of geograph- ical districts included in the study was rather large ( there were 46). To analyse relatively homogeneous categories of exposure, the districts were grouped in successive quintiles of each air pollutant However, because within-district information on the key cova- riates such as smoking, indoor air pollution and family circumstances was missing, there is no guar- antee of no confounding in this ecological study. Moreover, it is the essential feature of ecological data that the available information is average over the margins of a multivariate distribution of expo- sures and covariates. Thus, it is not even possible to estimate the direction and magnitude of the ecolog- ical bias. Despite these shortcomings, there were other considerations which made the study findings plausible (i.e., the consistency and magnitude of the adjusted relative risk estimates). 82 Noncomparable standardization There is a need for standardization in ecological studies for variates whose distribution is not con- stant across population. For example, published disease rates are invariably age-standardized, whereas published exposure rates are seldom so standardized. Since age is often associated with duration of exposure, regression of the published disease rates on the published exposure rates is biased, even if the precision of the rates used in the regression is high. Greenland (3) gives an ex- ample, based on actual data, of a study of the relationship of smoking to lung cancer that may incorporate this bias. The United States death rates for lung cancer were age-standardized, whereas smoking information was based on crude prevalence. Methodological recommendations The following recommendations have been pro- posed for design and data analysis in ecological studies (2,3). In the design of an ecological study select areas with populations that: are homogeneously exposed (i.e., minimize within-area exposure variations); represent different extremes of exposure dis- tribution (i.e., maximize between-area expo- sure variations) ; - are comparable with respect to covariate distri- butions; and use the smallest possible sampling units for eco- logical analysis. In the analysis of ecological data: - use weighted regression, instead of correlation, with weights proportional to the amount of in- formation contained in each group; - include in the regression model all variates that are thought to be related to the grouping pro- cess; examine multiple regression models with dif- ferent and flexible structural forms beyond the standard linear form - such as exponential and product-term models; consider the ecological implications of differ- ent individual-level model form specifications; - conduct an influence analysis by examining the effect of deleting from the analysis various areas with unusual outcome, exposure or covariate combinations; conduct a sensitivity analysis of ecological esti- mates to misclassification; take into account latency and induction periods separating causes and effects (e.g., consider the relevant exposures); consider the effect of migration on exposure estimates from geographical regions; and Rapp. trimest. statist. sanit. mond., 48 (1995) complement ecological analysis with thorough consideration of biases unique to such an analy- sis, and to biases common to all epidemiologi- cal studies. Finally, a general suggestion: whenever feasible supplement approximate aggregate data with accu- rate data at the individual level in a hybrid epide- miological analysis (20). Summary Ecological studies require a methodological theory dis- tinct from that used in individual-level epidemiological studies. This article discusses the special problems that need to be considered when planning ecological stud- ies or using the results of such studies. Ecological studies are much more sensitive to bias from model mis-specification than are results from individual- level studies. For example, deviations from linearity in the underlying individual-level regressions can lead to inability to control for confounding in ecological studies, even if no misclassification is present. Conditions for confounding differ in individual-level and ecological analyses. For ecological analyses of means, for example, a covariate will not be a confounder if its mean value in a study region is not associated with either (i) the mean exposure level across regions, or (ii) the mean outcome (disease rate) across regions. On the other hand, effect modification across areas can induce ecological bias even when the number of areas is very large and there is no confounding. In contrast to individual-level studies, independent and nondifferential misclassification of a dichotomous expo- sure usually leads to bias away from the null hypothesis in aggregate data studies. Failure to standardize dis- ease, exposure and covariate data for other confound- ers (not included in the regression model) can lead to bias. It should be borne in mind that there is no method available to identify or measure ecological bias. While this conclusion may sound like a general criticism of ecological studies, it is not. It does, however, serve as a reminder of the problems that need to be considered when one designs, analyses, or critically evaluates ecological studies. Resume Echec d'appariement des eludes eco/ogiques Les etudes ecologiques requierent une methodologie distincte de celle utilisee pour les etudes epidemiologi- ques au niveau individuel, comme cela a ete souligne par Greenland (3). II est done important de prendre en compte les considerations suivantes avant de realiser des etudes ecologiques ou d'en utiliser les resultats finaux. Wld hlth statist. quart., 48 (1995) Les resultats des etudes ecologiques sont bien plus sensibles au biais resultant des erreurs de modelisation que ceux des etudes realisees au niveau individuel. Par exemple, les ecarts a la linearite du modele ajuste au niveau individuel peuvent empecher de controler les variables de confusion dans une etude ecologique, meme en !'absence d'erreur de classification. Les facteurs de confusion different entre les etudes au niveau individuel et les analyses ecologiques. Pour les analyses ecologiques de moyennes, par exemple, une covariable ne sera pas un facteur de confusion si sa valeur moyenne dans une zone donnee n'est pas fonc- tion soit i) du niveau d'exposition moyen pour !'ensemble des zones, soit ii) du taux moyen de maladie pour !'ensemble des zones. Par contre, un effet modif- icateur selon les zones etudiees peut induire un biais meme lorsque les zones en question sont tres nom- breuses et qu'il n'existe pas d'effet de confusion. Contrairement a ce qui se passe pour les etudes au niveau individuel, une erreur de classificaiton non diffe- rentielle et independante concernant une exposition dichotomique biaise generalement les resultats vers l'hypothese alternative dans les etudes de correlation. Le fait de ne pas standardiser entre elles les donnees concernant la maladie, !'exposition et les covariables pour d'autres facteurs de confusion non pris en compte dans le modele de regression peut entrainer un biais. Entin, ii ne taut pasoublier qu'il n'existe aucune methode pour identifier ou mesurer les biais dans les etudes ecologiques. Cette remarque de conclusion ne doit pas etre consideree comme une «condamnation» des etu- des ecologiques, mais souligne !'importance de pren- dre en compte ces problemes lors de la conception, de !'analyse ou de !'evaluation critique des etudes ecologi- ques. References/References 1. Corvalan, C. &: Kjellstrom, T. Health and environment analysis for decision making. World health statistics quarterly, 48(2): 71-77 (1995). Corvalan, C. &: de Kjellstrom, T. Analyse sante et environnement pour la prise de decision [resume]. &fr port trimestriel de statistiques sanitaires mondiaks, 48 (2): 76 (1995). 2. Morgenstern, H. Uses of ecologic analysis in epidemiologic research. American journal of public health, 72: 1336-1344 (1982). 3. Greenland, S. Divergent biases in ecologic and individual- level studies. Statistics in medicine, 11: 1209-1223 (1992). 4. Greenland, S. &: Morgenstern, H. Ecological bias, confounding, and effect modification. International journal of epidemiowgy, 18: 269-274 (1989). 5. Greenland, S. &: Brenner, H. Correcting for non-differential misclassification in ecologic analyses. Applied statistics, 42: 117-126 (1993). 6. Brenner, H. et al. Effect of nondifferential exposure misclassification in ecologic studies. American journal of epidemiowgy, 135: 456-458 ( 1992). 7. Brenner, H. et al. The effects of nondifferential confounder misclassification in ecologic studies. Epidemio/,ogy, 3: 456-459 (1992). 83 8. Maldonado, G. & Greenland, S. Interpreting model coefficients when the true model form is unknown. Epidemiology, 4: 310-318 (1993). 9. Greenland, S. & Robins, J. Ecologic studies: biases, misconceptions, and counterexamples. American journal of epidemiology, 139: 747-759 (1994). 10. Salonen, J.T. et al. Changes in morbidity and mortality during comprehensive community programme to control cardiovascular diseases during 1972-7 in North Karelia. British medicaljourna~ 2: 1178-1183 (1979). 11. Hakama, M. Trends in the incidence of ceIVical cancer in the Nordic Countries. In: Magnus, K. (ed), Trends in cancer incidenr.e. Washington, D.C., Hemisphere Publishing, 1982, pp. 279-292. 12. Flaten, T.P. Chlorination of drinking water and cancer incidence in Norway. International journal of epidemiology, 21: 6-15 (1992). 84 13. Shaper,A.G. & Marr,J.W. Dietary recommendations for the community towards the postponement of coronary heart disease. British =dicaljourna~ 1: 867-871 (1977). 14. Staernler,J. et al. In: J. H. de Haas et al. (eds). Ischo.emic heart disease. London, London University Press, 1970, p. 84. 15. Schwartz,J. Air pollution and daily mortality in Birmingham, Alabama. American journal of epidemiology, 137: 1136-1147 (1993). 16. Field, B. & Kerr, C. Herbicide use and incidence ofneural- tube defects. (Letter). Lanr.eti: 1341-1342 (1979). 17. Fuller, W.A. Measur~terrormodels,NewYork, Wiley.1987. 18. Bobak, M. & Leon, D.A. Air pollution and infant mortality in the Czech Republic, 1986-88. Lant.et, 340: 1010-1014 (1992). 19. Cleveland, W.S. Robust locally weighted regression and smoothing scatterplots. Journal of the American statistical association 74: 829-836 (1979). 20. Prentice, R.L. & Sheppard, L. Aggregate data studies of disease risk factors. Bi~trics, in press. Rapp. trimest. statist. sanit. mond., 48 (1995) The use of geographical information systems in studies on environment and health David J. Briggsa & Paul Ellioffb Introduction The study of the environment in relation to health is, by its very nature, a spatial problem. Environ- mental exposure varies geographically in response to variations in environmental conditions; health outcomes and associated levels of health needs and support vary as a consequence. Many of the ques- tions facing environmental epidemiologists and policy-makers are thus inherently geographical, and spatial analysis and mapping are vital compo- nents of their work. In research terms, they provide an important step in describing problems and in formulating and testing hypotheses about possible links between environment and health. In policy terms, they are a potentially valuable means of directing policy to areas and problems of greatest need, and of monitoring policy performance and effects. Spatial analysis and mapping in environmental health have a long history. It is now traditional to trace their origin back at least as far as John Snow's seminal study of cholera in London (1). Until recently, they could only be carried out man- ually, or using relatively simple mapping packages. Over the last ten years, however, the capability for spatial data analysis has been revolutionised by the development of geographical information systems (GIS). These have not only made mapping and many spatial analytical techniques much easier, but have also stimulated a wide range of new re- search into spatial operations and concepts which have greatly advanced our understanding of how to analyse and interpret spatial phenomena. GIS functionality GIS may be defined as systems for the manipula- tion and presentation of georeferenced (i.e., spa- tial) data. As such, they are able to perform a range of functions, including: • data capture: the acquisition of the data in digi- tal form, normally by manual encoding, digitis- ing or scanning; • data cleaning: the checking, correction and ed- iting of data to remove errors during data cap- ture or transfer; a Professor, The Nene Centre for Research, Nene College, Northampton. b Doctor, Small Area Health Statistics Unit, London School of Hygiene and Tropical Medicine. Wld hlth statist. quart., 48 (1995) • data integration: a series of operations (e.g. generalization, projection conversion, registra- tion) which convert the data to a consistent geographical structure; • data storage: holding the data in a data base in a form suitable for easy retrieval and analysis; • data search and retrieval: the identification and recall of data on the basis of specific conditions defined by the user; • spatial analysis: the geographical manipulation and transformation of the data (e.g., buffering, point-in-polygon analysis, interpolation, map overlay); • statistical analysis: the computation of descrip- tive statistics for either single or multiple cover- ages; • display: the generation of maps or other output (graphs, tables, etc.) either on screen or as hard copy. In broad terms, these functions can be seen as a logical series of operations which describe the data stream from original source to final map ( or other) output. The full data stream is clearly long and complex (Box 1). The early stages in the process - data capture, cleaning and integration - can be especially time-consuming. Typically, they may account for 70% or more of the total time and resources needed to set up and implement a GIS for any specific application. On the other hand, once the data have been integrated into the GIS, it is possible to perform a wide range of spatial analyses on them, at relatively low cost. Some of the more commonly used functions available for these purposes are summarized in Box 2. Multiple use of data is only possible, however, if the data held in the GIS meet the needs of different users and different applications. To en- sure this requires careful consideration of the basic data needs of any GIS, and of the ways in which the data will be collected, integrated and stored. GIS applications in environmental health The wide-ranging functionality of GIS, outlined above, clearly makes them potentially powerful tools for the analysis oflinks between environment and health. Three particular areas of application can be defined: 85 Box 1 Steps in the application of GIS 1. Data capture 2. Data cleaning 3. Integration 4. Data analysis 5. Display Box 2 Acquisition of the data in digital form, normally by manual encoding, digitising, scanning or electronic transfer from pre-existing data bases. Checking, conversion, reformatting, correction and editing of data to remove gaps in the data and to eliminate or resolve errors either contained in the original data or introduced during data capture. Important operations include checking that all poly- gons are closed and that digitising errors (e.g., spikes, loops, false lines, unlabelled polygons or duplicate labels) are removed. The registration of the spatial data to a consistent geographic base. Important operations include: Edge-matching - joining coverages for adjacent map sheets to ensure that they are seamless; Generalization- reducing the scale of more detailed (larger scale) coverages to match the scale of the base map. This may involve quantitative generalization (weeding) by removal of excess data points, or logical generalization, by selected removal of specific classes of feature (e.g., by eliminating minor roads); Projection conversion - the conversion of the coverages to the same projection system (mainly important when mapping large areas); Registration - the matching of boundaries and other features on different coverages (e.g., to ensure that administrative boundaries match up with the rivers which they follow, or that settlement boundaries do not extend into the sea). This is often partially a manual process, conducted by selected removal and editing of map features; and Tiling - subdivision of the coverages into smaller spatial blocks (sheets or tiles) to facilitate data searching and retrieval (normally important only for coverages of large areas). Processing of the spatial and attribute data to extract information. Common operations include: Data search and retrieval - selected retrieval of information on the basis of defined criteria (e.g., by location, proximity, size, value) Spatial analysis - modelling and analysis of spatial patterns and relationships (see Box 2). The production and presentation of output, either to screen, hard copy or electronic media. Output may be in the form of tables, graphs or maps. Spatial analysis techniques in GIS for environmental health applications 86 Technique Point in polygon Line in polygon Buffering Interpolation Proximal estimation Smoothing Overlay Description Identifies intersection between point features and the areas (polygons) in which they lie ldentifes intersection between line features and the areas (polygons) in which they lie Construction of zones (buffers) of specified width around points, lines or areas Estimates conditions at unsampled locations Analysing conditions at a point based upon conditions in a specified neighbourhood Construction of a smoothed (generalized) surface Combination of one map coverage with another Example of application To identify all "cases" within a specified exposure zone To identify line sources (e.g., roads) passing through a specified area To define areas of exposure around emission sources (e.g., chimneys, power lines) Mapping of pollution surfaces Estimating pollution levels on the basis of surrounding land use Mapping of generalized exposure surfaces Combination of pollution and population density maps to assess exposed populations Rapp. trimest. statist. sanit. mond., 48 (1995) Table 1 Estimates of mean and 98th-percentile N02 levels in Huddersfield using different interpolation methods Tableau 1 Estimation des niveaux moyens et au 9ae percentile de N02 a Huddersfield, selon differentes methodes d'interpolation Method Methode Arithmetic averaging - Moyenne arithmetique Voronoi tessellation - Decoupage en polygone de Thiessen Contouring - Courbe de niveau Kriging mapping environmental risk; mapping health outcome; and linking environmental risk and health out- come. Risk assessment Risk assessment in relation to environment and health is essentially an attempt to estimate the level of exposure to specified pollutants, either for indi- viduals or for particular population groups. Direct measurements of exposure are rare, and usually available only for a relatively small sample of peo- ple. Instead, levels of exposure commonly have to be estimated by indirect means. Two main ap- proaches are available: spatial interpolation from measured data on ambient pollution levels; or modelling on the basis of data on emission levels and sources. Spatial interpolation Data on pollution are typically in the form of point measurements, derived either from routine moni- toring stations or from purposely designed surveys. Air pollution, for example, may be measured at permanently-established monitoring sites or at temporary sites set up for a specific project. Stream water quality is often monitored by collecting sam- ples from selected sites along the stream network. Data on soil and geological properties may be col- lected from specific sample points as part of a field survey, or from permanent monitoring sites. The problem with pollution mapping - as with other forms of risk assessment - is thus essentially one of spatial interpolation: how to estimate pollu- tion levels at unsampled sites. It is a problem to which GIS can make a major contribution. Most GIS provide a range of interpolation techniques which can be used as a basis for mapping. Some of the main techniques are summarized in Box 3. These different methods of spatial interpola- tion clearly help to make GIS a powerful tool for risk mapping. Nevertheless, the range of methods available presents the user with an important ques- Wld hlth statist. quart., 48 (1995) System Systeme SPANS ARC/INFO SPANS ARC/INFO Estimated concentrations (µg/m3) Estimation des concentrations (µg/m3) Mean 98th percentile Moyenne 98• percentile 30.1 55.7 27.6 56.2 30.2 46.2 27.9 46.1 29.1 40.1 tion of choice. This is not a trivial decision, since the method of interpolation may have marked ef- fects on the modelled outcome. Tab/,e 1, for exam- ple, shows estimates of the mean and 98th-percen- tile concentration of nitrogen dioxide in Hudders- field. This is based upon a total of 79 sample points, within an area of about 300 km2. The table compares 4 different methods of interpolation: arithmetic averaging (i.e., simple averaging of the data for the measured points), tessellation (area weighting) in borh SPANS and ARC/INFO, con- touring (using a linear interpolation routine in SPANS) and kriging (using a circular semivario- gram in ARC/INFO). As can be seen, considerable differences occur in the measured statistics, most notably in the 98th-percentile value. The performance of the different interpolation methods depends upon a number of factors in- cluding the nature of the underlying spatial varia- tion in the phenomenon under consideration and the sample density and distribution. A number of comparative studies have been carried out without clear consensus (2-4). In general, however, there are reasons to favour local methods of interpola- tion (such as kriging) over global methods (such as trend surface analysis) because the former are more sensitive to local variations in the data and thus do not produce as much smoothing of the modelled surface. Kriging and thin-plate spline techniques also provide error estimates for the modelled surface. In recent years, kriging has become a preferred technique in many situations. Oliver and Webster (5) provide a useful review in relation to GIS appli- cations. As Box 3 shows, kriging comprises a suite of techniques, each adapted to different applications. All, however, are based on the principle that spatial variation in any phenomenon can be divided into three main components: systematic trend ( or drift); random but spatially-correlated variation (such that near points are more similar to each other than distant points); and random spatially- uncorrelated variation (noise). The procedure in- volves calculating the semivariance, summarizing 87 Box 3 Spatial interpolation techniques available in GIS Method Trend surface analysis Voronoi tessellation Contouring Potential surface mapping Kriging Type Global Proximal Local Local Local Description An extension of the regression model; fits a generalized surface (linear or polynomial) through data points using least square methods Creates "minimum distance" polygons around each data point (locus), within which the values of the data point are applied. Uses Triangulated Irregular Networks (TIN) algorithms to construct isolines through a network of measurement points using either linear or polynomial functions. A moving-window approach, which examines all selected points within the window to define an average for the surface at the centroid. The user is able to control the radius of influence of the window, the decay rate of the surface away from the centroid and the maximum number of points to be used in making each window. A suite of local weighted averaging techniques which estimate conditions at unsampled locations on the basis of a spatial semiovariogram. Methods include: a) Punctual kriging - estimates values at specific points b) Block kriging- estimates values for areas (blocks) c) Universal kriging - estimates values at unsampled locations in the presence of regional trend (drift) d) Disjunctive kriging-estimates the likelihood of the value at each unsampled point exceeding a specified threshold e) Co-kriging - estimates conditions at unsampled loca- tions on the basis of more than one predictor variable the relationship between the distance between each pair of points ( or lag) and the difference in values of the measured phenomenon. The semi- variance is defined by the equation: Different models may be used to describe the semivariance quantitatively. Two groups of models are usually recognised: bounded models ( charac- terized by a finite a priori variance - the sill) and unbounded models (lacking a pre-defined sill). Circular, spherical and exponential models are the most common bounded forms used; power func- tions are the most widely used unbounded models. In either case, the choice of model may be ex- tremely important, since it may significantly affect interpolation estimates. y(h) = I I (2m(h)) :E {z(x)- z(xi + h) }2 where y(h) = the semivariance m(h) = the number of paired observations sepa- rated by a distance h z(x) = the value of property z at location i. This relationship is then used to predict condi- tions at any unsampled point. Plotting the semivariance typically produces a curve (the semivariogram), the shape of which in- dicates the nature of spatial variation involved. Often the semivariogram is convex in form, the semivariance initially rising from some point above the origin, reaching a threshold (the sill) and then levelling off. The rising curve defines the spatially correlated variation in the data. The point of inter- section on the y axis is known as the nugget vari- ance and defines the "noise". A concave-upward semivariogram may indicate trend ( drift) in the data. This normally needs to be removed prior to further analysis, for example using trend surface methods. The semivariogram is then recomputed using the detrended residuals. 88 Model fitting, however, poses particular prob- lems. Often it is done by eye, though this is clearly subject to considerable error and observer varia- tion. Alternatively, least squares techniques may be used. Because different numbers of data pairs may contribute to the semivariance at any lag (in- ter-point distance), however, these methods are not strictly appropriate. Ideally, weighted least squares techniques should be used instead (6). Unfortunately, the model-fitting methods current- ly used in existing GIS are not clearly specified and are not necessarily statistically optimal. This undoubtedly weakens the utility of the kriging techniques presently offered by GIS. For greater rigour, it seems necessary to carry out kriging ex- ternally, then import the results into the GIS for mapping. Rapp. trimest. statist. sanit. mond., 48 (1995) Pollution modelling Monitored data on pollution are far from ubiqui- tous. National networks are sparse, focused on a few key pollutants, and often spatially clustered. As a result, they do not provide an ideal basis for pollution mapping. Purpose-designed surveys are likely to be time-consuming and expensive and provide only relatively short-term data, and thus are also often inappropriate. In the absence of sufficient monitored data, therefore, pollution sur- faces must be calculated on the basis of data on emission levels or sources, using relevant model- ling techniques. In recent years, a wide range of pollution mod- els have been constructed, for both point and non- point sources. Examples include the suite of BREEZE models developed for the United States of America's Environmental Protection Agency (7), the CAR model developed by the Dutch Envi- ronmental Ministry (8) and the point-source mod- els for dispersal of radionuclides developed by the United Kingdom's National Radiological Protec- tion Board (9). These are generally based on Gaus- sian plume dispersion equations and take account of emission source, meteorological and terrain ef- fects. In general, they provide good estimates of pollution concentrations under simple and con- trolled conditions ( e.g., relatively stable weather conditions, regular emission rate, simple terrain and inert pollutants). In more complex condi- tions, however, the appropriateness of the models will be reduced. Problems also arise with the use of dispersion models because of their data demands. As yet, few attempts have been made to link these models into GIS, although it is expected that Version 7 of ARC/INFO will include a point dis- persion model. Loose coupling of existing pollu- tion models with GIS is, however, possible both to facilitate collection and input of geographic data required by the models, and to facilitate and ex- tend mapping. Kim et al., for example, used PC- ARC/INFO in association with SLAMM (Source Loading and Management Model) to assess water pollution in Beaver Dam City, Wisconsin ( 10). Briggs and Collins have also developed a GIS-based system for mapping air pollution from road traffic (11). This takes output from the CALINE line source model and uses the GRID facilities in ARC/ INFO to generate pollution surfaces for road net- works. Use of dispersion models to assess pollution surfaces has a number of advantages. In particular, it does not rely on measured concentrations and can thus be extended to areas and pollutants for which no monitored data are available. It also al- lows knowledge about dispersion processes to be used in modelling the pollution surface, and can take account of local factors which influence these processes (e.g., terrain, weather). On the other hand, the approach suffers from 3 major limita- Wld hlth statist. quart., 48 (1995) tions. The first is its often-heavy data demands. Most models require information not only on the location of the emission sources, but also emission rates and conditions (e.g., emission temperature, gas flow rate), local topography and climate. These data are often unavailable or of doubtful quality, especially where a large number of sources exist. The second constraint is that dispersion models only operate successfully relatively close to the emission source, where Gaussian dispersion pro- cesses can be assumed. This is a limitation especial- ly in relation to line-source or low-level point source emissions. The CAR model, for example, provides estimates only up to 35 metres from the roadway; the CALINE models which form part of the BREEZE suite give reliable predictions only up to 150-200 metres from the highway. The models cannot therefore be used for estimating variations in background concentrations. The third limitation is that the models are, for the most part, designed to work under relatively simple conditions. As the number of sources in- creases, as local terrain and land use become more complex, and as the weather conditions become more variable, their validity declines dramatically. This means that the models are often oflimited use in complex urban environments. Many models also provide estimates only for inert pollutants or for gases and vapours for which knowledge of the atmospheric chemistry is reasonably well-estab- lished (e.g., N02). Without measured data to vali- date the results, it can therefore be dangerous to use the models outside their design conditions. Health mapping Mapping of health outcome is an important step in many health studies, providing a useful description of the data and a valuable basis for further analysis. The display and mapping facilities available in GIS are clearly valuable in this context. Individual data may be presented as point maps, for example, and choropleth maps (e.g., of standardized mortality ratios or SMRs) may be produced by allocating each case to its relevant administrative or other area using point-in-polygon procedures. Several major limitations of health mapping need to be acknowledged. Firstly, mapping of indi- vidual (point) data (i.e., cases) is of limited use in the absence of appropriate data on population, and in any event raises important issues of confi- dentiality. Secondly, crude health maps (e.g., of SMRs) may be dominated by random variation, especially for rare diseases in small areas, or over short time periods. Thirdly, no account is taken of the specific area's underlying social and demo- graphic characteristics (e.g., smoking rates, unem- ployment), which may show marked geographical variation, and may be powerful predictors of dis- ease (12). Nonetheless, health maps are often re- garded as a valuable means of generating hypo the- 89 ses, although, as Barker has observed, "Maps of disease impel speculation about aetiology, but only rarely has such speculation by itself led directly to the discovery of causes." ( 13) Searching tor point clusters In the case of individual (i.e., point) data, spatial variation is likely to be expressed in a number of ways. They may be more or less clustered, and they may be distributed systematically or randomly. Analysis of point data on health outcome therefore commonly involves the search for clusters, and test- ing for spatial structure in the points ( e.g., case/ control locations) as well as searching for putative "clusters" of disease. The search for clusters in point health data has proved to be especially controversial. It is an ap- proach which has been motivated to a large extent by concerns about elevated incidence of disease around industrial installations (e.g., nuclear power stations and processing plants). It is an approach, however, which presents a number offundamental technical and statistical problems. These lie primari- ly in the statistical difficulty of detecting true excess- es, and in the complexity of searching large areas, and large numbers of cases, for possible clusters. Despite the theoretical problems which to date remain unresolved (14), GIS have been used to scan health data through a range of moving-win- dow, buffering and point-in-polygon techniques. This approach was developed and applied by Openshaw et al. using what was called a Geograph- ical Analysis Machine (GAM) which systematically constructed buffer zones around a fixed lattice of points in the study area ( 15). If the number of observed cases exceeded an expected number then a circle was drawn. Following repeated scan- ning with circles of different radii, the results were mapped, and locations found in a large number of overlapping circles were identified. The method attracted considerable criticism, not least because it involves double-counting of individual cases and because of the difficulty of analysing the resulting maps ( 16, 17). Subsequent developments of this approach resolved some of these problems, but failed fully to satisfy its critics ( 18). In the meantime, various other methods have been suggested. Besag & Newell, for example, pro- pose a statistically more robust method, in which the cumulative number of cases is counted with increasing distance from any point of interest, until a required "cluster" size is reached, and an associ- ated p-value is calculated (16). There are, however, problems of multiple inference, and false-positive clusters are bound to be detected. Map smoothing Similar issues occur with area data. The basic ques- tions are whether and to what extent the data show 90 evidence of heterogeneity of risk across areas and spatial structure, be it in the form of local trend or clustering. One recently developed approach to these questions is provided by "map smoothing". This uses Bayesian statistical methods to determine whether there is any underlying structure (i.e., ex- tra-Poisson variability) in the spatial distributions (see, for example, Mollie & Richardson (19)). It is based on the principle that spatial structure in the data will be seen as a tendency for nearby areas to be more similar to each other (in terms of health outcome) than more distant areas. Rates of disease are thus estimated in terms of the adjacency or proximity of the map units. To date, these techniques have not been fully integrated into GIS. Nevertheless, because they are based on the assessment of proximity or adjacency of the survey units, they can clearly benefit from the use ofGIS. Thus Briggs et al. used map smooth- ing techniques in combination with GIS to analyse variations in infant mortality in the Huddersfield area (20). Based on preliminary data, the results indicated a small but statistically significant varia- tion in rates in the area. As part of the SA VIAH study, similar methods are being used to analyse patterns of asthma and wheeze in schoolchildren in urban areas, prior to examining possible links with air pollution.c Assessing links between environment and health Raised incidence of events around emission sources Methods of studying environmental and health data depend to a large extent upon the availability and quality of data, especially measures of expo- sure. Where maps of pollution are available (which is not always the case), these can be combined with the health outcome data, using either overlay or point-in-polygon procedures (see below). Where pollution or exposure has not been mapped, alter- native, proxy indicators of risk may need to be used, such as distance from the emission source. This type of application is exemplified by the work of the Small Area Health Statistics Unit, at the London School of Hygiene and Tropical Medicine (21, 22). Concern about the possible existence of raised levels of cancers and childhood leukaemias around nuclear installations has also motivated a large body of research in this area (23). Commonly, the approach adopted has been to assume that people living near the point source are at greater risk than those living further away. A circle is thus constructed around the point source, and stan- c Elliott, P. et al. Analysing small area variations in air quality and health - the SA VIAH study ( submitted to Journal of epidemiology and community health). Rapp. trimest. statist. sanit. mond., 48 (1995) dardized rates of health outcome within this circle compared to national rates ( or those for other control areas). While much of this research has been conduct- ed without recourse to GIS, it is clear that GIS provide a more flexible and interactive method for circle construction through their buffering capa- bilities. They also provide the opportunity to im- prove on this type of analysis. Circles are inevitably crude indicators of exposure or risk, and without an understanding of the dispersion processes and pathways involved it is easy to use circles of an inappropriate size. GIS clearly allow the modelling of more complex search areas, taking account of dispersion patterns and other effects, where data permit. They also provide the opportunity to extend this type of analysis easily to non-point sources. A growing body of evidence exists, for example, to suggest that levels of respiratory disease and aller- gies are elevated in populations living close to roadways (24,25), while the incidence of leukae- mias and some other cancers has been associated with residence near high voltage power lines (26). Using buffering techniques along roads or power lines may thus help to define possible search areas for raised incidence. The use of buffering techniques in this manner simply divides the population at large into an "ex- posed" and "non-exposed" group. Incidence or prevalence of the disease of concern may then be compared between these groups. At a somewhat more sophisticated level, however, risk may be as- sumed to vary with distance from the pollution source. In this case, tests are required to determine whether the distribution of health outcome away from the source shows a significant distance- related effect. One approach to this problem, us- ing point data, has been developed by Diggle (27), who compared the distribution of cases around the point source with that of controls using "kernel estimation" techniques. These convert the point distributions of cases and controls into a two- dimensional density surface. Rowlingson et al. de- scribe a module called RAISA, designed to per- form this analysis in ARC/INFO (28). Another method, that uses non-parametric isotonic regres- sion to model "non-increasing" risk with distance, has been described by Stone for grouped ( ecologi- cal) data (29). Assumptions of distance decay of exposure and health effect away from an emission source are likely to be appropriate for low-level emission sources, but may be less applicable, to high level emissions (e.g., tall stacks), where significant shadow zones tend to occur close to the source. In these cases, one would ideally model exposure more rigorously, for example using dispersion models ( as outlined above), provided that the nec- essary input data are available. Wld hlth statist. quart., 48 (1995) Comparing exposure and health outcome Analysing possible links between pollution and health solely on the basis of distance from emission source is clearly an uncertain process. In general, stronger inferences can be drawn when a map of pollution levels is available. In these circumstances, the pollution map may be used as an indicator of exposure, and compared to health outcome. At its simplest, this may be achieved by overlay- ing the health and pollution maps. From this, the level of pollution can be compared with the health outcome ( e.g., SMR), using an appropriate statisti- cal model. A number of problems are nevertheless encountered by this approach, especially when ap- plied to aggregated health data. One major difficulty is the mismatch which normally occurs between the spatial structure of the two data sets (i.e., health and pollution). While dispersion modelling or spatial interpola- tion produce data attached to "natural" areas, aggregated health data are often based on pre- defined administrative regions. Overlay of the two coverages produces new map units, which are the result of the combination of the two maps. To assess rates of health outcome for each of these new areas requires assumptions about the distri- bution of cases within each of the original map- ping units. The simplest approach is to assume that the population is distributed evenly within each administrative area. The health rates for the new map units can thus be computed as the area- weighted total of all contributing administrative areas. This, however, may be highly unrealistic, especially where the administrative units are large or where the population is located in nucleated settlements. In these circumstances, exogenous information may be used to model the distribu- tion of the health outcome within the original map units. A useful indicator in this context may be the distribution of population density (e.g., based on small census tracts). Where this is not available, the area of built-up land may be used as a surrogate (if this is available at a higher resolu- tion than the administrative areas - for example from remote-sensing imagery). Postal codes may also be used to reallocate the population (and thus health outcome) across the original map units. The problem, of course, is that unless these methods of allocation are reliable, considerable misclassification may occur which will impair sub- sequent analysis. Further difficulties are encoun- tered in the statistical analysis, although again this can be conveniently approached through the framework of Bayesian statistics (30). Where data on health outcome are available at the individual level, analysis can be carried out using point-in-polygon methods. Pollution levels, attributable to both "cases" and "controls" may be assessed by dropping their respective point loca- tions ( e.g., home address or postal code or place of 91 work) onto the pollution map. Cases and controls may then be compared using standard statistical techniques. This approach clearly avoids the difficulties in- herent in area-aggregate analysis. On the other hand, it suffers from a different problem: that of over-specificity of the exposure data. People are exposed not only at their place of residence ( or other point location to which they are attributed), but typically obtain their overall exposure over a relatively wide area across which they range over a period of time. Point pollution levels therefore do not necessarily provide a good indicator of total exposure. Instead, relationships between individu- al exposure and health outcome are much more complex, with contributions from many different, overlapping and changing sources. Individuals may thus be members of exposed groups for part of the time ( and in some locations) and unexposed groups at other times (and in other places). In addition, the reliable attribution of exposure to a point location may be beyond the scope of the modelling procedure. A potentially valuable development in this re- spect is the research into what have become known as "fuzzy relational models" (31). More generally, there is a need to develop methods for reliably assessing actual exposure, taking account of the "residence times" of individuals in different parts of the area. This involves mapping the time- weighted distribution of the target groups. Com- bining this surface with that of the pollution con- centration would thus permit the estimation of the integrated exposure. Data on which to assess resi- dence-time surfaces are not easily available. For detailed studies at the individual scale, activity dia- ries may provide suitable data. For group estimates, it may be possible to use information on travel patterns (e.g., from census data). The spatial mod- elling facilities available in GIS clearly provide the potential to undertake such studies, but as yet no examples are known. Overlay of health and pollution data - either in point or area form - also faces a major difficulty, namely, confounding. This relates to the circum- stance in which measured covariances in health and pollution do not necessarily represent direct causal links but may be influenced by their mutu- al association with a third, independent variable. Variations in mortality in an area may, for exam- ple, appear to relate to differences in levels of air pollution, not because of any direct link between the two, but because both are a reflection of gen- eral living conditions ( e.g., housing conditions, poverty). The problem of unmeasured confound- ing is particularly acute in ecological ("areal") studies. In many studies of environment and health, socio-economic and lifestyle factors provide the main confounders (12). Data are therefore re- 92 quired on these, in a form compatible with those on the health and environmental factors of inter- est. These data may be derived from a range of different sources, including census statistics, household surveys and hospital or other records. To convert these to a consistent spatial form inev- itably poses all the problems outlined above in relation to matching environment and health data. Again, however, it is clear that the facilities available in GIS for spatial data transformation and integration provide useful tools in this re- spect. Discussion and conclusion As the foregoing discussion has illustrated, GIS have much to offer in attempts to link environment and health. They provide powerful systems for the collection and integration of spatial data on envi- ronment and health. They provide a basis for sub- sequent statistical calculation and analysis. They provide a means for visualizing and displaying these data in map form. As costs of purchase come down, and as more digital data sets become avail- able, GIS are becoming more accessible. The use of GIS for environment-health linkage is therefore likely to increase. The use of GIS for environment and health applications nevertheless carries with it serious dangers, for GIS are influential and persuasive in- struments. It is all too easy to use them to promote false conclusions, especially through the naive or simplistic use of maps and mapping techniques. This highlights a number of important issues. Data quality Great care needs to be taken in terms of the quality of the data used in GIS. Once in the form of maps, the quality of the data can be easily overlooked. Many data used for GIS applications are derived from different sources, and as such contain differ- ent levels and types of error, and may be intrinsical- ly non-comparable. Linking these data sets to pro- vide spatial coverage, or overlaying them to derive new information, may generate complex and un- seen errors. Data quality control is thus of the utmost importance. It must consider not only the spatial properties of the data (e.g., are points or lines located correctly?) but also - and often more importantly - the quality of the attached attributes (e.g., are the values being mapped accurate and unbiased?). Sadly, data quality control is often inhibited by the poor documentation attached to many data sets. In many cases, it is difficult to obtain indepen- dent reference data against which to judge the quality of the data we use or the results we obtain. Manual data checking can be extremely onerous. GIS methods themselves, however, are often useful in data checking. Mapping each variable separately Rapp. trimest. statist. sanit. mond., 48 (1995) can provide a useful means of identifying gaps or errors in the surfaces. Calculation of spatial statis- tics (e.g., point densities, averages for moving win- dows across the surface) can similarly indicate er- rors or discrepancies. Point-in-polygon searches can be a useful means of checking locational accu- racy of data points (e.g., addresses, postal codes): as most users of the United Kingdom's postal code system know, some codes are located in the sea! Scale and resolution Issues of scale and resolution are of special impor- tance. Environmental and health data are often gathered for different spatial structures, at different levels of spatial resolution. Combining them within a GIS may seem relatively easy, but may make little sense. Clearly, all spatial analyses undertaken in a GIS must be carried out at an appropriate and consistent scale, although defining an appropriate and consistent scale is not easy. Environmental phe- nomena often vary over different scales depending upon the context. Atmospheric pollution in urban areas, for example, typically shows high levels of variation even within an individual street, whereas in rural areas nitrogen dioxide pollution surfaces are typically relatively smooth. Similarly, scales of exposure mayvaryaccording to the pollutant under consideration. Exposure may occur as an extreme ( or repeated) event at a point, or as long-term exposure to relatively low concentrations over a wide area. Unless the data available match the spa- tial scale at which these variations occur, then any pollution surfaces defined, or any assessments of exposure, are likely to be of limited use. GIS data are often described as being scale-free. In principle, this is true. Through the generaliza- tion features available in GIS, for example, large- scale (i.e., detailed) maps can be reduced to a smaller, less detailed scale. Conversely, the zoom facilities available in GIS allow maps easily to be enlarged for display either on screen or in hard copy. Generalization, however, involves loss of de- tail in the data; normally the data must be reduced in volume (i.e., weeded) during generalization to remove redundant data and prevent crowding and clogging of the output. Magnification of the data, in contrast, can increase the size at which maps are displayed, but cannot add information to the origi- nal set. It is thus impossible to improve upon the resolution of the data beyond that already stored in the data base. The ultimate limits to map resolu- tion are therefore determined by the scale and resolution of the source data. A rule of thumb is therefore useful for data acquisition in GIS: namely, to obtain the data at the largest scale and highest resolution practicable. Some redundancy may occur as a result, but this can be reduced by producing generalized versions Wld hlth statist. quart., 48 (1995) of the data for routine application (keeping the original, more detailed coverages in back-up form). The cost of this redundancy, however, is likely to be considerably less than that of having to recapture more detailed data at a later date, when it is found that the original coverages are too coarse and generalized. Beyond that, a second "rule" may be invoked. GIS data should only be used for applications for which their scale and resolution are appropriate. Using data at too large a scale may simply be incon- venient (and unnecessarily costly); using data at too small a scale ( or, more precisely, at insufficient resolution) adds uncertainty and error to the anal- yses, and may generate false conclusions. Users therefore need to understand the spatial limita- tions of the data available in their GIS. Spatial bias Many environmental data used for health applica- tions derive from point measurements at monitor- ing or survey stations. As has been seen, to relate these to health outcome often requires a process of spatial interpolation. GIS provide a number of methods for spatial interpolation of point data sets. It is important to appreciate, however, that these cannot improve upon the quality of the original coverage: if the original sample points are unrepresentative of the study area as a whole, then the pollution surface generated by interpola- tion is likely to be unreliable. Unfortunately, many of the environmental variations of interest to health occur at a spatial scale considerably beyond those inherent in the available data. In other words the spatial density of most environ- mental monitoring networks is too sparse to dis- play the true degree of variation in the phenom- ena concerned. Many pollution monitoring net- works are also biased towards certain types of site (e.g., highly polluted industrial or street-level sites). Interpolation is therefore likely to generate biased and unrealistic surfaces. Caution is again needed, therefore, to avoid overly naive use of GIS capabilities. Indeed, cau- tion must be the watchword when using GIS to link environment and health. They are powerful and potentially valuable tools, but they need to be used wisely and carefully. Their proper use requires close collaboration not only between geographers and epidemiologists, but also with environmental scientists, spatial statisticians and experts in com- puting and data base techniques. Summary Geographical information systems (G IS) provide a pow- erful technology for the spatial analysis of environmental and health data. Major areas of application include the assessment and mapping of environmental exposure, 93 mapping of health outcome, and the analysis of spatial relationships between environment and health. The use of GIS nevertheless brings with it many potential prob- lems and pitfalls. This article reviews some of the recent applications in relation to studies of environment and health, and examines some of the research issues involved. Resume Utilisation des systemes d'information geographique dans le cadre d'etudes sur l'environnement et la sante Les systemes d'information geographique fournissent un outil performant pour analyser dans l'espace les donnees sur l'environnement et la sante. Leurs princi- paux domaines d'application sont !'evaluation des ris- ques lies a l'environnement, a ces dangers et leurs consequences sanitaires, et !'analyse des relations dans l'espace entre l'environnement et la sante. Toute- fois, !'utilisation de ces systemes peut soulever de nombreux problemes et induire en erreur. Cet article decrit quelques-unes de leurs applications recentes dans le cadre d'etudes sur l'environnement et la sante, et examine certaines questions connexes de re- cherche. References/References 1. Snow,J. On the mode of communication of cholera. 2nd edition. London, 1855. 2. Dubrule, 0. Comparing splines and kriging. Computers and geoscienus, 10: 327-338 ( 1984). 3. Laslett, G. M. et al. Comparison of several spatial prediction methods for soil pH.Journal of soil scienr.e, 38: 325-70 ( 1987). 4. Abbass, T. et al. A comparison of surface fitting algorithms for geophysical data. Terra nova, 2: 467-75 (1990). 5. Oliver, M.A. &: Webster, R. Kriging: a method of interpolation for geographical information systems. Inter- national journal of geographical information systems, 4: 313-332 (1990). 6. McBratney,A.B. &: Webster, R.Choosing functions for semi- variograms of soil properties and fitting them to sampling estimates.Journal of Soil Science, 37: 617-639 (1986). 7. Trinity Consultants Inc. BREEZE software user's manual. Dallas, Trinity Consultants Inc., 1992. 8. Eerens, H. et al. The CAR model: the Dutch method to determine city street air quality. Atmospheric environment, 278(4): 389-399 (1993). 9. National Radiological Protection Board.A model/or short and medium range dispersion of radionudi.des rekased to the atmosphere. First &port of a Worlcing Group on Atmospheric Dispersion. NRPB- R91. Harwell, National Radiological Protection Board, 1979. 10. Kim, K. et al. Urban non-point source pollution assessment using a geographical information system. Journal of environ- mental management, 39: 157-170 (1993). 11. Briggs, DJ.&: Collins, S. Mapping exposure to road traffic pollution using GIS. In: M. Baranowski (ed.), GIS in ecol.ogical studies and environmental management. Warsaw, Global Resource Information Database, GRID-Warsaw, 1994, pp. 31-38. 94 12. Jolley, DJ. etal. Socio-economic confounding. In: P. Elliott et al. (eds.), Geographical and environmental epidemiol.ogy: methods for small-area studies. Oxford, Oxford University Press, (1992), pp. 115-124. 13. Barker, D.J.P. Geographical variations in disease in Britain. British medical journal, 283, 398-400 (1981). 14. Elliott, P. etal. Spatial statistical methods in environmental epidemiology: a critique. Statistical Methods in Medical Research (in press). 15. Openshaw, S. et al. A mark 1 Geographical Analysis Machine for the automated analysis of point data sets. International journal of geographical information systems, 1: 335-358 ( 1987). 16. Besag, J. &: Newell, J. The detection of clusters in rare diseases.Journal of the Ruyal Statistical Society, A 154: 143-155 (1991). 17. Urquhart,J. etal. Exploring small area methods. In Elliott, P. (ed.), Methodol.ogy of enquiries into disease clustering. Small Area Health Statistics Unit, London School of Hygiene and Tropical Medicine, London, 1989, pp. 41-49. 18. Openshaw, S. et al. Building a prototype Geographical Correlates Exploration Machine. International journal of geographical information systems, 4:(3), 297-311 (1990). 19. Mollie, A. &: Richardson, S. Empirical Bayes estimates of cancer mortality rates using spatial models. Statistics in medicine, 10: 95-112 (1991). 20. Briggs, DJ. et al. Infant mortality in HuddersfieUDistrictHealth Authority, 1982-1991. &port of a Study on behalf o/West Yorlcshire Health Authority. Huddersfield: Institute of Environmental and Policy Analysis, University of Huddersfield and Environmental Epidemiology Unit, London School of Hygiene and Tropical Medicine, 1993. 21. Elliott, P. et. al. Incidence of cancer of the larynx and lung near incinerators of waste solvents and oils in Great Britain. Lanr.et, 339: 854-858 (1992). 22. Elliott, P. et. al. The Small Area Health Statistics Unit: a national facility for investigating health around point sources of environmental pollution in the United Kingdom. Journal of epidemiol.ogy and community health, 46: 345-349 (1992). 23. Thomas, R. W. spatial epidemiol.ogy. Pion Press, London, 1990. 24. Nitta, H. etal. Respiratory health associated with exposure to automobile exhaust I. Results of cross-sectional study in 1979, 1982 and 1983. Archives of environmental health, 48: 53- 58 (1993). 25. Wjst, M., et al. Road traffic and adverse effects on respiratory health in children. British medical journa~ 307: 596-600 (1993). 26. Aldrich, T. &: Easterly, C. Electromagnetic fields and public health. Environmental health perspectives, 75: 159-171 ( 1987). 27. Diggle, PJ. A point process modelling approach to raised incidence of a rare phenomenon in the vicinity of a pre- specified point. Journal of the Ruyal Statistical Society, Al53: 349-362 (1990). 28. Rowlingson, B.S. et al. Statistical spatial analysis in a geographical information systems frameworlc. North West Regional Research Laboratory Research report No. 23. Lancaster University, Lancaster, no date. 29. Stone, R.A. Investigations of excess environmental risks around a putative source: statistical problems and a proposed test. Statistics in medicine, 7: 649-660 (1988). 30. Clayton, D. &: Kalder, J. Empirical Bayes estimates of age- standardised relative risk for use in disease mapping. Biometrics, 43: 671-681 ( 1987). 31. Wang, F. et al. Fuzzy information representation and processing in conventional GIS software: database design and application. International journal of geographical infor- mation systems, 4: 261-283 (1990). Rapp. trimest. statist. sanit. mond., 48 (1995) Health and environment in Sao Paulo, Brazil: methods of data linkage and questions of policy Carolyn Stephensa, Marco Akermanb & Paulo Borlina Maiac Introduction This article is based on a case study prepared at the request of the Health & Environment Analysis for Decision-making (HEADLAMP) project of the Of- fice of Global and Integrated Environmental Health of the World Health Organization (WHO). The main purpose of the HEADLAMP initiative is to obtain information on which to base preventive action against environmental health problems. To achieve this aim, HEADLAMP promotes the devel- opment of methods to link environment and health data collected routinely within countries. The project intends to identify core environmental health indicators for use in environmental health management at local and national levels, and to encourage their use to direct policy. Primarily, this article will address the develop- ment of data linkage methods, and will discuss briefly, based on existing experience with data link- age in Sao Paulo, the potential for routine environ- mental health monitoring and management in a major developing-country industrial centre. At what could be called the macro level, we will look briefly, at an environmental hazard which may have broad health impacts for the population of the city: air pollution. At what could be called the micro level, we look at a complex of environ- mental hazards which affect the population of Sao Paulo differentially depending on household and neighbourhood circumstances. At this level, we re- view the use of routinely collected data on socio- environmental conditions (water consumption, population density, sanitation, income and educa- tion standards) and their linkage to health data. For our analysis of data linkage methods, in- cluding issues of data quality and linkage potential, we draw heavily on recent research in Sao Paulo by Stephens et al. (l).d First, we shall give some background to the case study of Sao Paulo. Sources of population, health, socio-environmental and air pollution data are identified. Data are assessed on the basis of their a Lecturer, Department of Public Health and Policy, London School of Hygiene & Tropical Medicine, London, United Kingdom. b Research Fellow, CEDEC, Sao Paulo, Brazil. c Project Analyst, Funda,;:ao SEADE, Diretoria de Estudos Populacionais, Sao Paulo, Brazil. d Funded by the Environment Policy Department of the Overseas Development Administration, United Kingdom. Wld hlth statist. quart., 48 (1995) availability (frequency of measurement, access and geographical reference) and their quality (com- pleteness and accuracy). Current monitoring and linkage activities are presented, supplemented by ad hoe studies which have used the same type of data to demonstrate health links (2). Finally, the potential for Health and Environmental Data Link- age Analysis and Monitoring Projects in the Sao Paulo Metropolitan Area is discussed briefly. Background information: Sao Paulo Metropolitan Area (SPMA) The Sao Paulo Metropolitan Area (SPMA) is situ- ated in southeastem Brazil and comprises 39 municipalities in a territory of 8 051 km2. Map 1 shows maps of the SPMA and Sao Paulo City (SPC). Approximately 5 OOO km2 of the SPMA is urban- ized. Sao Paulo City, the main municipality of the metropolitan area and capital of the Brazilian state of the same name, occupies 1 577 km2. The SPMA contains 12% of the Brazilian popu- lation. In 1991, 15 416 416 people lived in the SPMA, with 9 626 894 of them living in Sao Paulo City (SPC). According to this most recent census, the annual growth rate of the region has slowed from 4.5% (3.7% for SPC) between 1970-1980 to 1.9% (1.2% for SPC) from 1980-1990 (3). Socio-economic environmental and health situation The SPMA accounts for 18% of Brazilian gross domestic product (or US$ 425 billion), 31 % of Brazil's industrial domestic product and 25% of the industrial labour force. Today, the region is the largest industrial centre in Latin America. Despite its economic stature, there are consid- erable inequalities in income distribution in the metropolitan area: the richest 10% of the popula- tion earned 30% of the total income, and the poor- est 50% earned only 25% of the total income in 1990. The unequal distribution of the economic benefits of urban growth in Sao Paulo has had a marked impact on the distribution of household environmental benefits resulting in "an inequi- table burden of negative environmental impacts which affect the poor" (4). Sao Paulo's growth and industrialization have also been marked by serious environmental problems. The deteriorating quality of environ- mental resources including air, water and land have been mentioned by various authors (4-6). At 95 Map 1 Sao Paulo City (SPC) and Sao Paulo Metropolitan Area (SPMA), Brazil Carte 1 Ville de Sao Paulo et agglomeration de Sao Paulo, Bresil the macro-level, industrial air and water pollution have been tackled with some success. However, there are still serious problems with air pollution produced by motor vehicles and water pollution due to inadequate treatment of sewerage. Solid waste collection has also been a matter for concern, because the annual increase in quantity has not been followed by a corresponding increase in waste treatment sites (2). Erosion, flooding, soil contamination, mudslides, and noise pollution are other environmental prob- lems in the region (2, 4). To set the linkage of environment and health data in Sao Paulo in its broader health context, we shall describe the aggregate epidemiological pro- file of the city. In the City of Sao Paulo in 1992, only 4% of all registered deaths were due to infectious and parasitic diseases, whereas 33% of all deaths were due to diseases of the circulatory system, 12% to respiratory problems, and 14% to external causes. Other cause groups accounted for the re- maining 37% of all registered deaths (Funda~ao SEADE - internal files). In general, Sao Paulo is a city of considerable industrial and economic strength, with emerging problems of the macro environment including air pollution. In socio-economic terms, the city is char- acterized by large disparities in wealth between the rich and the poor and this has repercussions for social and environmental conditions at the micro- level. Bearing in mind the city's overall context, we will now explore monitoring of the environ- 96 ment and health and the use of data linkage in Sao Paulo. Data linkage The type of linkage possible using existing data is dependent on the range and quality of informa- tion available and accessible. There are, broadly, 3 types of data required for the linkage of environ- ment and health data: numerator data (i.e., data on health events - ideally clinically-diagnosed mor- tality or morbidity); denominator data (i.e., data on the population at risk); and lastly, the exposure data (i.e., data on the environment to which the population is exposed). In order to explore the possibilities and practi- calities of data linkage we will review information in Sao Paulo under the general themes of: sources of data; quality of data; and monitoring and data linkage. The discussion focuses on data linkage at the ecological level, that is analysis based on grouped data (rather than individual data). Sources of data and availability Urban areas of both developed and developing countries are often characterized by their multi- plicity of agencies, and if they are regional or na- tional capitals, by the presence of academic institu- tions. Consequently, there is often no shortage of information collected on health-related conditions and on the urban environment. Sao Paulo is no exception. Rapp. trimest. statist. sanit. mond., 48 (1995) Denominator data (population) Census data are normally the main source of popu- lation estimates and can be used to provide ap- proximate information for denominators in rates of mortality and morbidity. One advantage of cen- sus data is that they are based on total counts of the population and are regularly undertaken ( allowing time trends to be analysed). Disadvantages occur if the census occurred many years prior to a current analysis. Mobility of the population can be difficult to estimate accurately and may differ widely be- tween sub-groups. Brazil's population census (most recent in 1991), is carried out every 10 years, and compiles the following information: resident and/ or present population (in other words, de Jure and/ or de facto population), distribution by age, sex, mari- tal status, religion, race, urban/rural condition, education, place of birth, relation to the head of the household, etc. (7).e Population enumeration is based on the small- est geographical fraction ( census units) but results are published by districts or municipalities. The SPMA is formed by 139 districts with 96 of them in SPC. Data on census units are available upon re- quest. Access to the data is obtained by published reports or by terminal and disks. There is no charge to obtain general population data, but IBGEf has set fees for special tabulations. The 1991 Census has recently made available data on age structure of the population. Ad hoe surveys have also been used as sources for population estimates. PCV (Pesquisa de Con- dir;oes de Vida - survey of living conditions) (5)g and Pesquisa Origem Destino ( origin - destination survey) h are two such surveys that can provide population data, and are used particularly for local planning in inter-censual periods. Numerator: health data on mortality and/or morbidity Sources of routine numerator data on mortality and morbidity can be diverse. Routine numerator data can be obtained primarily through vital events information, infectious disease notification and e Basic data are collected for the total population, but more specific information, for example, fertility, mortality, migration, labour force, is collected only for a sample, which, until 1980, represented 25% of the population. In the 1991 Census this sample varied between 10 and 20%, depending on the size of the city. f IBGE (Brazilian Institute of Geography and Statistics) located in Rio de Janeiro is a federal government unit in charge of Census operations. g The Pesquisa de Condi~iies de Vida collected data on family income, labour market, housing, access to health care, education and property for 5 500 households in the SPMA. h The Pesquisa Ori.gem Destina was carried out by the SPMA in 1987. This research resulted from a survey of 26 OOO households and used the population projected by Funda<;ao SEADE for 1990 as its expansion factor. Wld hlth statist. quart., 48 (1995) facility-based information. Here we will discuss only vital events and facility-based information. Vital registration of events (mortality). Annual analyses of mortality data in Sao Paulo are made by Fundai;:ao SEADE, a division of the Sao Paulo State Planning Secretariat. Death certificates are com- piled on a monthly basis from civil registers ( cart6- rios) and deaths are coded by place of residence of the deceased} Data on age, sex, place of occur- rence, main and associated causes of death are also available. Data on yearly mortality is obtained from SEADE upon request or through their annual pub- lications. Mortality data can also be acquired from the Ministry of Health, but their data are always produced after SEADE's data. where registration of death events is a condition for burial in con- trolled cemeteries, and all cemeteries are con- trolled by government, then registration of events is more comprehensive. Daily enumeration of deaths is also compiled through the Sao Paulo Municipal Secretariat of Health (SEMPLA), where PROAIM, a municipal project to improve death certification, obtains death certificates sent by city funerariasj 24 hours after the occurrence of every death. PROAIM data are based on 56 districts and subdistricts in SPC. Access to these data can be obtained through SEMPLA's computer terminals. Morbidity data. Morbidity data are produced rou- tinely in public and private health facilities with an official compilation mechanism embracing public and private hospital admissions. Data on outpatient care are officially compiled only for public facilities. A recent household survey (the PCV) showed that only 40% of the families living in the SPMA are exclusive users of the public health system (publicly-owned facilities plus pri- vately-owned facilities financed by public money). The other 60% have individual private health in- surance, are enroled in company plans or use private facilities on an ad hoe basis. This official data base thus does not contain outpatient data on 60% of the population, making routine mor- bidity data incomplete in terms of the popula- tion's health experience.k i SEADE codes mortality data for all municipalities in the SPMA. In SPC place ofresidence is disaggregated into 56 districts and subdistricts. There are plans to adjust SPC coding to the 96 new districts. j Special shops that organize funerals, make and sell coffins. They are private in the majority of Brazilian cities. In Sao Paulo City for historical reasons they belong to the public sector. k The "Hospital data base" (SIH-SUS) gathers data on admissions to the public-sector hospitals and the "Ambulatory data base" (SIA-SUS) registers data from appointments made with public-sector outpatient facilities. Both systems were set up by the Ministry of Health, but State Secretaries have their own data bases. In Sao Paulo, the State Health Secretary runs a very well structured information system (CIS). 97 Health surveys have been carried out to obtain data on morbidity. In the 1980s, IBGE undertook 2 surveys (8). Cancer registers are also available. The population based register was established in 1969, but there has been discontinuity in its work and the latest data are from 1978. There are also hospital registers in the main cancer units in SPC. Access to this type of data can be obtained through ad hoe reports and publications. Exposure data There is a diverse range of sources for environmen- tal exposure data - at the micro-level, data are ob- tainable through service providers, and intermit- tently from the census or other large-scale house- hold surveys. Macro-environmental problems are often monitored through agencies set up for the purpose. Service-based data. Most cities maintain records of environmental services such as water supplies, sani- tation, electricity connections, sewerage connec- tions and solid waste disposal. These data serve to give a broad idea of the environmental conditions faced by the population in their homes or neigh- bourhoods. At the very least, agencies dealing with environmental services will have information on the more affiuent areas of town where services are regularly provided. Agencies which provide infor- mation on the supply of an amenity to an area may not routinely collect data on aspects of quality, quantity or regularity of that supply. Census or large-scale household survey data. The main advantage of exposure data culled from the census is that information is often detailed and the measurements apply to individual households. Since one of its aims is to represent the whole population, the census may also be the most com- prehensive and detailed available source of micro- level environmental exposure information. De- pending on the degree of disaggregation of data required, an advantage of using exposure data from the census is that the geographical break- down of the data is compatible with the breakdown of census-based denominator data (thus enhanc- ing the reliability of the linkage of the two types of data). A disadvantage of census-based exposure (and population) data may be that geographical disaggregation of information is not compatible with the disaggregation of data on health events such as mortality ( often reported by administra- tive/ political districts). Spot-monitoring systems. Some cities now employ monitoring systems which monitor an environ- mental hazard at certain geographical locations. Such monitoring usually documents a macro-envi- ronmental problem, for example air or water qual- ity. If carried out consistently over time this moni- toring can gauge trends in conditions. A limitation 98 of such spot monitoring may be that a small num- ber of stations or points of monitoring may not be able to represent accurately the overall profile of conditions in a large area. In Sao Paulo, both micro-level and macro level environmental hazards are monitored. At the macro-level, air pollution is monitored by an agen- cy set up for the task. Micro-level environmental health conditions are monitored routinely by ser- vice agencies. Macro-level exposure data: air pollution. CETESB ( the State of Sao Paulo Environmental Agency) is the main source of routinely-collected air pollution data in the SPMA (9). There are 22 automatic and 7 manual stations collecting air pollution data in SPMA. 60% of the automatic and 86% of the man- ual stations are in SPC. Because air quality is only monitored close to the stations, the results are not representative of the whole district. The majority of stations are placed at busy roads in the town's centre. CETESB provides data on air pollution each day to the media plus a forecast of pollutant disper- sion for the next 24 hours. Basing calculations on the Pollutant Standard Index (PSI) created by the Environmental Protection Agency (EPA) of the United States of America, CETESB computes an index which expresses the quality of Sao Pau- lo' s air within 6 categories: "good", "regular", "inadequate", "bad", "worse" and "critical". CETESB also adopts criteria for controlling daily acute episodes of air pollution based on 3 category levels - "attention", "alert" and "emergency" - for each pollutant. Micro-level exposure: socio-environmental data. At the micro-level we focus on 5 socio-environmental hazards in Sao Paulo which have differential im- pacts on households in the city. These indicators of environmental health hazard were selected by local planners, academics and service agency staff in a forum, on the basis of their relationship to micro-level environmental health problems in the city (as perceived by participants). Measured at the ecological level ( ecological in the sense of using geographical data to compare environmen- tal conditions between districts within the city), the data used to compose an index of socio-envi- ronmental deprivation in the city were "income" = average per capita income; "education" = per- centage of illiterates and people who did not com- plete primary school; "sewerage facilities" = per- centage of houses linked to central sewerage; "water supply" = average per capita water con- sumption; and "housing" = average number of persons per house. Using a complex of indicators to mea,;ure socio-environmental hazard reflects both the opinions of planners and policy-makers in Sao Paulo and a concern about individual data quality. Rapp. trimest. statist. sanit. mond., 48 (1995) To analyse the effects of differentials in envi- ronmental conditions within the city, differences in conditions and health between areas in the city can be mapped, provided that geographical refer- ences for data are compatible. Sao Paulo City has until recently been divided into 56 districts and sub-districts, the traditional geographical bases for census data and vital registration. Some environ- mental service agencies use their own system to divide the city: for example SABESP (the water and sanitation agency) uses 310 "real estate sectors" . The data used for monitoring micro-level envi- ronmental problems in Sao Paulo can be obtained from the following sources: (i) Pesquisa Origem-Des- tino provided disaggregated data on "income" and "education" for each district and sub-district; (ii) data on the "water supply" and "sewerage facilities" variables were collected at SABESP; and (iii) the 1991 Population Census gave us data on housing density. Micro-level data on household or neighbourhood conditions can also be obtained through surveys organized by state or metropolitan departments or by IBGE. National household sur- veys carried out by IBGE started in 1981 with the theme "health", and an annual survey has been carried out each year since.' IBGE data are accessi- ble through publications, disks, reports and special tabulations. The national census also collects socio- environmental data every 10 years. Data quality Denominator (Population) Monitoring environmental and health conditions, particularly over time and within small areas, re- quires an understanding of population mobility, since quite often an individual will face a risk in one location but may fall ill or die in another area. The linkage of data on a particular environment with the health of those who, at any one time, live in that environment is therefore difficult. No rou- tine information about previous residence (i.e., specific location) is available routinely in Brazil. Data on population mobility within Sao Paulo are largely unavailable, but may be derived from the 1991 census. In terms of completeness of denominator data, there is an overall 5-10% under-enumeration in the Brazilian census. This under-enumeration is distributed unevenly by area and age groups. The homeless in Sao Paulo are a potential source of substantial under-enumeration in census counts. This has the effect of creating an under-estimation of denominators, with particular impact on data 1 1982 - education; 1983 - national insurance and occupation; 1984 - fertility; 1985 - children; 1986 - access to health service, nutrition, membership in community organizations and birth control; 1988 - political participation and energy facilities; 1989 - labour market Wld hlth statist. quart., 48 (1995) from central areas of the city where higher concen- trations of the homeless attempt to subsist. There are claims in SPC that the 1991 census under- enumerated considerably, although this claim is refuted. Numerator (mortality and morbidity data) Routine vital registration data on deaths in Sao Paulo are excellent overall ( 10). They are estimated to be 99% complete, with only limited area-specific problems. There is also satisfactory accuracy in the reporting of the main cause in the death certificate (11). Area-based data, particularly in the central areas of Sao Paulo, are affected by the deaths of the homeless, who are not registered in the census, but appear in mortality figures in civil registers (cart6rios) in the centre of the city, making rates appear high in these areas. Peripheral areas in the metropolitan area, such as the District of Parelhei- ros in SPC, may also be affected. Exposure data Macro environment: air pollution. Air pollution is var- ied in its nature - it can be indoor, outdoor or personal. In terms of health-threatening pollution in the short-term, tobacco is probably the main air pollutant in urban areas to cause respiratory dis- eases ( 12). It also contributes to many other diseas- es outside the respiratory system. Indoor pollution is another factor in health problems. Institutional geographical information on these two sources of pollution is scarce in SPMA. Air pollution monitor- ing activities in the SPMA collect data only on outdoor pollution. The monitoring stations in Sao Paulo gauge only very localized problems, reflect- ing each monitoring station's surroundings. Most stations are located along busy roads and are concentrated in the central areas of the city. In terms of completeness of air pollution moni- toring, there is also variation among the monitor- ing stations in continuity of data collection. For example, Lapa station monitored 360 days in 1992 whereas Santana monitored only 251 days that year. In addition, not all stations monitor the same type of pollutant. In SPMA only one automatic station (Parque Sao Pedro) and the 2 mobile labo- ratories are able to measure all common pollut- ants. The severity of the problem over a larger area cannot be assessed with a sparse set of observations. Monitoring should be complemented with spot surveys, but financial limitations have restricted such surveys. Socio-environmental data. Routine data on micro- level environmental conditions in Sao Paulo are, on the whole, very good and a large number of sources (mostly academic studies) are available to cross-check and corroborate data. Limitations of routine information on socio-environmental con- ditions are related to the level of disaggregation 99 permitted by existing data sets and to the sensitivity of existing indicators. With respect to the level of disaggregation, the division of a city (with a population of over 9 mil- lion) into 56 districts and sub-districts avoids gross and misleading aggregation of statistics, but still leaves some degree of data aggregation, some of which conceals differentials in environmental con- ditions and possibly health impacts within districts. The 1992 revision of internal city boundaries into 96 districts was intended to solve this to a degree. Adoption of the new boundaries has yet to be implemented by several agencies, most notably the vital registration system, which still uses 56 divi- sions for its data analysis. A more difficult problem is related to the sensi- tivity of existing data to reflect accurately environ- mental quality at the micro-level. To give an exam- ple of problems with data sensitivity in Sao Paulo, and using access to sanitation, despite evidence of some differentials between areas within the city, routine information on "access to sewerage facili- ties" do not give an accurate picture of distinctions in service quality between the extremely vulnerable and the privileged in the city. SABESP data indi- cate that in more than 50% of the districts and sub- districts in Sao Paulo, 90% of the households have "connections" to sewerage facilities. In contrast, and at greater degree of specificity, the results of a major household survey for the metropolitan re- gion showed that only 40% of the Javela house- holds have access to sewerage facilities and 25% of the corticosm households have no access to this ser- vice (5). Aggregate data on districts containing small pockets of Javela households and corticos do not reveal enough about conditions there. Existing health and environment monitoring Monitoring of micro-level environmental conditions Despite the fact that over 40% of the population are known to be living in poverty, environmental problems at the micro-level in Sao Paulo have not, until recently, been "monitored" in the sense of (publicly) documenting socio-environmental con- ditions and using monitoring data to act. However, to the degree that each agency is aware of the coverage and quality of its particular environmen- tal service, individual aspects of environment at the micro-level have been monitored. For example, m Corti~os or rented rooms are an "illegal" housing category in the City of Sao Paulo, resembling the tenement slums of industrialized nations. They are the oldest, most popular kind of housing, mostly in the centre of the city. Families live in one or two rooms and share common kitchens and bathrooms. Data relative to cortifos are controversial, since there is no consensus on the definition. They are also difficult to count because they are hidden in other categories in population censuses and municipal registers. Also, there has been no comprehensive survey in Sao Paulo of environmental conditions in corti~os. 100 SABESP officials have extensive data and are aware of the distribution of their services and service quality. Similarly, the Housing Secretariat moni- tors the degradation of housing stock into corticos and maps and monitors the development of Javelas. Each agency monitors conditions for its own purposes, often linked to maintenance of agency-specific environmental and professional standards, and also to financing of service delivery. Data have not been shared between agencies until recently. Despite limited communication among agen- cies about the socio-environmental conditions which they each document individually, there have been several initiatives over the past decade to compile broad-ranging data on socio-environmen- tal conditions. The municipal planning secretariat (SEMPLA) compiles and has access to diverse in- formation on services and uses it to estimate future needs. SEMPLA does not, other than in a discre- tionary sense, have control over the delivery of services to areas with poor environmental facilities. In 1993, a collaborative initiative was set in place, involving SEADE (the state statistical agen- cy), SEMPLA ( the municipal planning secretariat), SABESP (water and sanitation), the municipal Housing Secretariat, Sao Paulo State Education Secretariat, and the London School of Hygiene and Tropical Medicine ( 1). This initiative aimed to create a linked data base, which monitored socio- environmental conditions within the city for one year (1992-93), both through developing compati- ble maps of the distribution of health and environ- mental conditions by district and through the con- struction of an index of socio-environmental depri- vation based on routine, existing data. The index measured relative socio-environmental conditions between districts and established 4 zones of socio- environmental qualityn which were then linked to health conditions within the city. According to this analysis, 43.8% of Sao Paulo's population live in the quartile of districts with the worst socio-environmental conditions (Zone 1). They have a low level of education and income, have limited access to sewerage facilities, consume less water per capita and live in higher-density housing. Zone 4 has the best socio-environmental conditions in the city, with 9.2% of Sao Paulo's population. Zones 2 and 3 contained 34 and 45% respectively. Tabl.e 1 shows that large differentials in condi- tions exist among areas in Sao Paulo. For example, on average, per capita income is over 3 times high- er for residents of Zone 4 compared to income for residents of Zone 1. Differentials in education exist also, with levels of illiteracy ranging from 28.1 % in n Adapting a method developed by the United Nations Human Development Report, 1992 to derive a Human Development Index. Rapp. trimest. statist. sanit. mond., 48 (1995) Table 1 Socio-environmental differentials between zones in Sao Paulo Tableau 1 Differences socio-ecologiques entres les zones de Sao Paulo Zones Average per capita Average% of illiterates Average % of houses Average per capita Average number income (minimum & incomplete primary linked to the general water consumption (ml)- of people per house - wages) - Revenu school - Pourcentage network - Pourcentage Consommation moyenne Nombre moyen moyen par habitant moyen d'analphabetes moyen d'habitations d'eau par habitant (m3) d'occupants (salaire minimum) et de personnes n'ayant reliees au reseau par habitation pas !ermine le cycle commun de l'ecole primaire Zone 1 2.1 40.0 64.3 4.3 3.8 Zone 2 3.2 35.5 81.8 6.2 3.7 Zone 3 4.3 32.8 94.2 11.5 3.4 Zone 4 7.2 28.1 98.7 20.6 2.8 Source: Ref. (1) (from Ref. (15); SABESP, 1993; FIBGE, 1992)-Ref. (1) (d'apres Ref. (15); SABESP, 1993; FIBGE, 1992) Zone 4 (the best) to 40% in Zone 1. Differentials in water consumption are striking: in Zone 1 only 4.3 m3 of water are consumed per day per capita. In contrast, over 20 m3 are consumed per day per capita in Zone 4. The majority of districts in Zone 1 (the most deprived) are on the periphery of the city, while the centre houses the most privileged ( on aver- age). The distribution of socio-environmental con- ditions should be borne in mind as we discuss the monitoring of the macro-environmental problem of air pollution. Macro-environmental monitoring: air pollution Air pollution monitoring started in 1965 in 3 in- dustrial municipalities of the SPMA (9). Monitor- ing in Sao Paulo City started 3 years later with the measurement of sulphur dioxide (S02) and smoke levels at only one site (13). There were 14 sites in 1977 (reduced to 6 in the early 1980s) with 3 of them providing information to Global Environ- ment Monitoring System (GEMS). In 1981, the State of Sao Paulo Environmental Agency (CETESB) installed automatic monitoring devices at 25 new sites (including Cubatao, a city outside the SPMA), an improvement on the previous man- ual method. In 1994 the monitoring network consisted of 12 automatic and 6 manual stations in SPC and 10 automatic and 1 manual stations in other munic- ipalities of the SPMA. There are also 2 mobile laboratories which operate automatically. The auto- matic network measures the following features: re- spirable particles (PM10), sulphur dioxide (S02), oxidesofnitrogen (NOx),ozone (03),carbonmon- oxide (CO), hydrocarbons (HC), wind speed (WS) and direction (WD), humidity (H) and tempera- ture (T). The manual network measures S02 and smoke levels, while at 9 automatic sites, suspended particulate matter (SPM) is measured manually by the high volume samplers. Tabk 2 shows the config- uration of the automatic network in the SPMA. Wld hlth statist. quart., 48 (1995) Emissions inventory is an important activity in the air pollution control and monitoring process. Emissions are associated with power generation and industry, motor vehicle traffic and domestic solid fuel. Sao Paulo is considered to have good availability of emissions inventories as shown by the Urban air pollution in megadties of the worU ( 13). Tabk 3 presents a summary of the relative con- tribution of various emission sources and their re- spective pollutants. Topography and climatic characteristics have an important influence on air quality as they contrib- ute to the dispersion of air pollutant emissions. Varied topography, proximity to the sea, surround- ing mountains, the "street canyon" effect (busy roads with tall buildings in both sides of the road preventing dispersion of low level emissions) and thermal inversions are the main factors which influ- ence the dispersion of air pollutants in Sao Paulo. Frequently during the winter du in Sao Paulo (May to August), inversion layers form above the ground and trap pollutants, which in tum aggravates the thermal inversion. CETESB has a special scheme for monitoring air pollutant dispersion in winter. CETESB has data on first and second maximum daily pollutant levels, arithmetic and geometric means, and number of monitored days since 1988. There is also information on the number of days when pollutant levels were above national stan- dards and indications whether these levels trig- gered the categories of critical points of "atten- tion", "alarm" and "emergency". Twenty-one fixed stations monitor respirable particles. In 1992, 15 of these reported days when levels were above national standards (50 g/m3).0 The highest annual arithmetic mean was found in O There is variation between the stations in terms of number of days above the national standards. For example, Guarulhos' levels were above national standard for 15 days, Cerqueira Cesar's for 11 days, Mo6ca's for 7 days and Santana's for 2 days. Cubatao, an area outside the SPMA, had levels above national standards for 37 days. 101 Table 2 Configuration of the automatic network in the SPMA Tableau 2 Configuration du reseau automatique dans !'agglomeration urbaine de Sao Paulo NQ Station location - PM10 S02 NO N02 co HC 03 Humid· Temper- Wind Wind Emplacement ity- ature- speed- direction- de la station Hurni- Tempe- Vitesse Direction dite rature du vent du vent 01 Parque D. Pedro x x x x x x x x x x x 02 Santana x x x x 03 Mo6ca x x x x x x x x 04 Cambuci x x 05 lbirapuera x x x x 06 N.Sra. do O x x 07 S. Caetano do Sul x x x x 08 Congonhas x x x x x x 09 Lapa x x x x x 10 Cerqueira Cesar x x x x x 11 Penha x x 12 Centro x x 13 Guarulhos x x x x 14 Sto Andre - Centro x x x x 15 Diadema x x 16 Santo Amaro x x x x 17 Osasco x x x x 18 Sto Andre - Capuava x x x x 19 S. Bernardo do Campo x x x x 20 Taboao da Serra x x 21 Sao Miguel Paulista x x x x 22 Mau a x x 26 Mobile lab. II - x x x x x x x x x x x Lab. mobile II 27 Mobile lab. I - x x x x x x x x x x Lab. mobile I Source.· Ref. - Ref. (9) Table 3 Summary table of the relative contribution of different emission sources and respective pollutants Tableau 3 Tableau recapitulatif de !'importance relative de differentes sources dans !'emission de polluants Sources - Sources Pollutants - Polluants Vehicles - Vehicules Industry- Industries Other - Autres Total - Total co 94% 3% NOX 92% 7% S02 64% 36% PM 10 40% 10% Source: Ref. - Ref. (9) Cerqueira Cesar (92 g/m3) reflecting traffic prob- lems in the area. Smoke level is measured at 7 sta- tions: only one of the stations had an arithmetic mean smoke level above the national standard in 1992. Suspended particulate matter (SPM) is as- sessed at 9 sites in the SPMA, using the high vol- ume sampler collecting a 24-hour sample every 102 3% 100% 1% 100% 0% 100% 50% 100% 6 days. At 4 of these, the annual geometric mean levels of pollutant were found to be above the national standard. CETESB data indicate a decline in the levels of the 3 pollutants mentioned over the last 10 years (9). For sulphur dioxide, none of the monitoring sites reported annual arithmetic means above stan- Rapp. trimest. statist. sanit. mond., 48 (1995) <lards in 1992 and data show decreasing levels of this pollutant for all sites over the last 10 years. The case of carbon monoxide is more serious: one of the sites (Cerqueira Cesar) reported 78 days when for 8 hours, CO levels were above national stan- dards. On one of these days pollutant levels reached "attention" critical status. There is no clear, observable trend over time for carbon mon- oxide. None of the sites had representative data for nitrogen dioxide. Current health and environment data linkage activities This section documents briefly the results of envi- ronmental and health data linkage, looking first - at the micro-level - at the health effects of the socio- environmental risks faced by the population of Sao Paulo. Macro-environmental monitoring and its linkage to health data are summarized. Micro-level environmental problems: Health and environment differentials We begin with the results from the recent data linkage of environment and health data in Sao Paulo ( 1). Results reflect the inequalities in condi- tions between the central, intermediate and per- ipheral areas of the city of Sao Paulo. They most strongly indicate the impact of an environment of poverty on the population, particularly for those who live on the outskirts of the city, both with respect to material symptoms of poverty in terms of access/non access to public services and to the added environmental risks attached to living in situations of social disadvantage. The study results also suggest that Sao Paulo's economic strength has not been able to provide benefits for a considerable proportion of its inhab- itants. Fifty-nine per cent of Sao Paulo's inhabit- ants (5 664 OOO people) live in precarious housing (14) and approximately 16% of the people (1 536 OOO people) have no access to sewerage fa- cilities (SABESP, internal files, 1993). In a city which needs both a skilled, blue-collar workforce and an efficient army of white-collar workers to maintain good-quality industrial output, 33% (3 168 OOO) of its citizens are illiterate or have not completed primary school (15). The study of socio-environmental zones for the City of Sao Paulo showed that approximately 44% of the population live in areas with the worst per capita income, with least access to sewerage facili- ties, lower water consumption per capita, highest population densities and the least education. Link- ing this information to mortality data revealed that people living in the worst areas had considerably higher levels of mortality than those living in areas with better socio-environmental conditions. Looking at Sao Paulo's health differentials by age group, there is significant room for improve- ment in the 0-4 year-old group, probably through Wld hlth statist. quart., 48 (1995) preventing deaths by diarrhoea and pneumonia. Overall, respiratory and infectious disease account for the majority of deaths in this age group. Differ- entials between the zones are striking. Respiratory and infectious mortality rates are (respectively) 3.8 and 4.4 times greater for children in Zone 1 ( the least privileged) than in Zone 4. It should be noted that infectious diseases represented only 2.6% of all deaths in Sao Paulo for 1992. However, these figures indicate that despite its status as one of the most successful examples of a city in an industrial- ized developing country, Sao Paulo has yet to con- quer preventable child deaths in vulnerable popu- lation groups. Perhaps the most alarming differential is in homicides which account for major excesses in mortality between socio-environmental zones. Vio- lent deaths are a particular cause of concern for the poor: in 15-44 year olds in the worst socio- environmental conditions there were 3 428 violent deaths in 1992 (or 16.5/10 OOO) compared to 326 deaths in Zone 4, the wealthiest (7.8/10 OOO). This could be described as an "epidemic of vio- lence" affecting the poor of Sao Paulo most severe- ly. Overall in 1992 there were 3 759 deaths due to homicides (8/10 OOO population). If we consider only males aged 15-24, the mortality rate for homi- cide was 19.5/10 OOO in 1992. This rate is just below that in the United States (21.9), which has the highest rate for this age group of males among industrialized nations (16). It is probable that this "epidemic" is related to the environment of social disadvantage experienced by many in Sao Paulo. Looking at older age groups, mortality patterns in the 45-64 year-old group show evidence of pre- mature deaths from traffic accidents, cerebrovascu- lar diseases and hypertension. There are significant differentials between socio-environmental zones. Increasing levels of mortality from diseases of the circulatory system in developing countries are often thought to indicate transition from diseases of pov- erty to diseases of affluence. Data from Sao Paulo corroborate increasing evidence that circulatory diseases cannot be called diseases of affluence: in Sao Paulo, as in developed countries, their impact is felt most severely by the poor. Data on differential mortality from circulatory diseases in Sao Paulo in 1992 indicate that in socio-environmentally disad- vantaged Zones 1 and 2, residents experienced significantly higher rates of mortality compared to their wealthier neighbours in Zones 3 and 4. Interestingly, differentials in mortality appear to diminish in the 65+ age group. It seems that the elderly die at similar rates whichever zone they reside in. Ad hoe studies linking household environmental data with health data. Household-level differentials shown by this data linkage exercise (1) are supple- mented by evidence from other studies. For exam- 103 pie, Monteiro ( 17) found that downward trends in infant mortality in the City of Sao Paulo during three decades (1950-1979) correlated strongly with trends in real minimum wages, improvements in water supply and access to health care. Other stud- ies corroborate this broadly ( 18-20). A Municipal Health Authority ReportP notes that mortality from external causes (road acci- dents, homicides and suicides) was the major cause of death in the 5-39 age group. Differentials be- tween sub-districts are shocking: the mortality rate from homicide is 11.3 per 100 OOO in Jardim Paulis- ta (a rich area) and 69.2 per 100 OOO in Itaim Paulista (a poor area). Macro-environment: air pollution and health data linkage Currently, assessment of air quality consists of ex- amining ambient air quality against established guidelines. This system is not based on previous local linkage of health and environment data and establishment of local standards but works on a "risk analysis" basis, where only environmental data is monitored and public health risk is extra- polated from international data and standards. The main aim of monitoring air quality in Sao Paulo is to trigger emergency actions when pollut- ants in the atmosphere reach levels that may (ac- cording to international criteria) cause risk to pub- lic health. Control and monitoring activities in the SPMA are based on standards defined in 1976 by a State decree (9), the same standards set by the United States Environmental Protection Agency (EPA) (21). There have been some critics of this approach, who argue that establishing standards based on foreign data suggests that the adverse effects of air pollution on human health are exclusively depen- dent upon the air pollutant concentration, and are not influenced by local conditions. q In reality, local climate, socio-environmental conditions and lifes- tyles in Sao Paulo are all likely to affect the relative influence of air pollution on health. Although local air pollution and health data are not linked routinely, several ad hoe studies have been undertaken in Sao Paulo to look at links between pollutant levels and morbidity and mortal- ity (22-24).r The studies by Ribiero (22) and Sobra}r assessed pollutant levels in terms of respiratory, infectious P Issued by the Sao Paulo Municipal Health Authority and presented at the 1992 II Brazilian Conference on Epidemiology. q Rancevas, S. with the English title, Environmental risks due to the use of alternative farms of energy. r Sobral, H.R Poluiµi.o do are doen(:aS respirat6rias em crian(:aS da Grande Sii.o Pauw: um estudo de geografio. midica. Tese de Doutorado, Sao Paulo, Departamento de Geografia, Faculdade de Filosofia, Letras e Ciencias Humanas, Universidade de Sao Paulo, 1988. 104 or ventilatory symptoms in children, while the oth- er two (23,24) seem to show correlations between increased levels of particulate matter, S02 and CO and cases of respiratory and cardiovascular disease admitted as emergencies at local hospitals. Finally, Saldiva et al., through a time series ecological study, found significant association between mor- tality in the elderly and respirable particulates from pollution (25), and robust association be- tween child mortality and NOx levels (26). Conclusions: potential linkage, monitoring and policy activities One of HEADLAMP's main aims is to provide decision-makers with the necessary tools to moni- tor environmental health problems and to assess the effect of their policies. There is an assumption that action will be facilitated by data linkage and the development of indicators. Environmental health measures will "give explicit policy-related information". This final section briefly discusses the potential policy implications of the HEAD- LAMP idea. Current environmental policy and plans Air pollution. Sao Paulo has achieved some success in control of its macro-environmental problems. Control of industrial pollution, dating back two decades, (high-sulphur fuel oil was replaced by low-sulphur fuel, biomass, natural gas or electrici- ty) has reduced the emissions of S02 by a factor of over 5 since 1976 (13). This change has allowed for considerable improvement in the SPMA's air quality. Motor vehicle traffic is now responsible for the largest proportion (64%) ofS02 emissions (9). The National Alcohol Programme imple- mented in 1979 brought important changes in air quality in Brazil. In 1989 half of the motor vehicles in the SPMA were running on alcohol and gasoline and 22% of vehicles used alcohol alone. Overall it can be said that the introduction of alcohol de- creased emission of CO, NOx, HC, S02, increased amounts of aldehyde and eliminated emissions of lead. Since 1979, CETESB has done special surveys to assess these changes. It is likely that increases in motor vehicle registrations and in the proportion of vehicles which run on diesel mean that current reforms will not contribute much to further im- provement in vehicle pollution. The Programme for Controlling Air Pollution due to Motor Vehi- cles is an attempt to solve the problem and it is expected to be in full effect by 1997. Socio-environmental conditions. Disparities in socio- environmental conditions between groups within Sao Paulo have been the subject of discussion for s These causes represented only 1.5% of all causes of death. Rapp. trimest. statist. sanit. mond., 48 (1995) many years in the city (27, 28). Significant steps have been taken towards improving the coverage of basic environmental infrastructures within the city. To a large extent, this has been achieved: piped water and sanitation facilities extend into many of the favelas and the corti~os; and primary health initiatives have increased coverage by immu- nization campaigns within the city. The combina- tion of these public health actions has had a marked effect on infectious diseases throughout the city, particularly among children. However, as we have discussed, disparities in conditions be- tween groups within Sao Paulo still exist and these appear to have significant ramifications for the health profiles of both children and adults. Future planning. Sao Paulo is at an opportune mo- ment to develop policy-oriented monitoring sys- tems of its environmental and health conditions. For the first time, a comprehensive plan aimed at sustainable development of the city, has been put forward by the authorities. The Metropolitan Plan for Greater Sao Paulo (2) stresses the need to look for long-term, not short-term solutions to the city's problems. The plan attributes 4 dimensions to development: (i) economic growth; (ii) social equity; (iii) preservation and improvement of qual- ity of life; and (iv) protection of natural resources. The main aim of the plan is to guarantee the sus- tainable future development of the metropolis. Within this context there is a rising popular concern for the quality of Sao Paulo's macro- and micro-environments. The consequences of in- equalities in socio-environmental conditions are clear from documented health data; the impact of these and the potential consequences of poor air quality are widely reported in the popular press and academic circles. At the macro level, according to recent re- search, the population perceives poor air quality as a principal aspect of environmental degradation in the SPMA (28). Jacobi found that air pollution was perceived as the main environmental problem of the city, followed by urban violence and the short- age of health services. It is important to note that environmental priorities, and definitions of the problem, differed substantially according to the socio-economic status of the respondents. In the wealthiest strata, 71 % of householders ranked air pollution as their number one problem, referring to vehicular emissions as their main air quality concern. In contrast, 46% of households in the poorest strata perceived air pollution to be their major environmental problem - these households were referring to problems of air quality in terms of dust, smoke and foul smells. As the case of Sao Paulo shows, there are exten- sive data available on environmental conditions and health in the metropolis. Routine environ- mental data have not, until recently, been linked to health outcomes. Wld hlth statist. quart., 48 (1995) In the case of our macro-environmental risk, air pollution, monitoring is carried out currently on the basis of risk analysis relying on international guidelines to set local standards of suspected health effects. Ad hoe studies of air pollution and health suggest local risk, hut data linkage has not been established routinely. For micro-environmental problems, ad hoe studies have documented differentials in health and environment between groups within the city for some years, but until recently the linkage of diverse, routine data has not been accomplished. The intra-urban environment and health study in Sao Paulo demonstrated the extent of micro-envi- ronmental problems and health effects of these on the population of the city. Perhaps more impor- tantly, the study's methods can be seen as a pilot attempt to establish the feasibility of linking diverse environmental and health data and developing routine communication between diverse data sources. By working directly with selected planning and environmental service agencies and develop- ing linkages over time, the study has facilitated environmental and health data linkage on a more systematic basis for the future. Despite Sao Paulo's achievements in estab- lishing good information systems, there is still po- tential for improvement in information systems. In order to understand the priorities of different groups within the city, planners have expressed a need for the development of more compatible and interlinked data systems, for the express purpose of creating integrated urban policies. This fits closely the recommendations of Agenda 21, "to establish, as appropriate, adequate environmental monitor- ing capacities for the surveillance of environmen- tal quality and the health status of populations". In order to move towards integrated information sys- tems, Sao Paulo still needs to develop its capacities for linked monitoring of environmental and health problems. Among the priorities are the cre- ation of geographically compatible data bases for environmental and health sources and the estab- lishment of a routine system for communicating and publicizing linked health and environmental data. Stepping back to the macro level, Sao Paulo seems now to have an operational risk analysis sys- tem and to have acted concertedly on some forms of air pollution. Whether this has been achieved on the basis of actions guided by monitoring or on the basis of problems perceived at a political level, is not clear. Given HEADLAMP's thesis that health and environmental monitoring is a necessary cata- lyst for environmental change, this is an important unanswered question. Despite monitoring of the environmental prob- lem of air pollution at the macro-level, there are also still gaps in understanding of air pollution and its specific effects on the population of Sao Paulo. 105 For example, there are still significant questions related to the specific interactions of the local cli- mate with fluctuations in pollution and effects on health. Perhaps more importantly, there is almost no evidence of the relationship that the macro and micro problems may have with each other. There is international evidence, for example, that the poor may face higher health risks from exposure to air pollution (4). It is not clear whether this is true in Sao Paulo. Finally, and in relation to HEADLAMP's the- sis that monitoring is followed by policy, it should be stressed that it is not clear whether monitoring an environmental problem (and/or its health ef- fects) has a substantial impact on environmental policy, nor if it does, how this occurs. It has been argued that only one of the four key mobilisers of policy change is technical advice based on "infor- mation, analyses and options" (29). How might the complex interaction of informa- tion and policy development relate to Sao Paulo and its environment? For the macro-problem of air pollution, Sao Paulo is in a critical stage. Risk anal- ysis suggests that the city centre is the most vulnera- ble area and that vehicular pollution is the culprit. Vehicles in Sao Paulo are owned mostly by the wealthy. Will the wealthy willingly compromise their expression of economic gain - the car - for health benefits? The historical short-term reaction of the wealthy to air pollution in the centre of cities is to try to retreat to insulated suburbs, commuting (mostly by car) to a centre which may be increas- ingly characterized by office blocks and run-down housing lived in by the poor. It is not clear if this is the future for Sao Paulo, nor what role monitoring of the problem might play. Finally, monitoring of health and environment linkages has to be set within the context of public health priorities facing populations in the short and long term. Policies must be seen in the context which shapes them. To some extent, the way in which policy makers respond to the micro-level environment of poverty faced by many of Sao Pau- lo 's citizens will inform responses to macro prob- lems. In the short term Sao Paulo's health pro- file is characterised by diseases of the elderly (heart disease and neoplasms), with an adult health pro- file dominated by diseases of the social environ- ment - violence being the major threat. Currently, violence ranks second to air pollution as a major environmental problem perceived by the rich and poor in the city (28). Both are perceived urban environmental threats to health, but it is violence that has the more substantial impact on population health at present in Sao Paulo, with significantly higher risks among the poor. At present, the wealthy in Sao Paulo ensure their safety from the environmental health threat of violence with security systems. In terms of envi- ronmental strategy, it is a not dissimilar attitude to 106 that of wealthy 19th-century Londoners who built parks to insulate themselves from the ill-under- stood "miasmic" diseases of their poorer neigh- bours (30). There is a danger that both the complex micro- environmental problems and the macro-environ- mental questions may be dealt with in the same way. Air pollution may have substantial effects on people's health in Sao Paulo, but the question remains: who pollutes and who pays and will pay with their health? Given the current inequalities in the structure of Sao Paulo's society, will monitor- ing the environmental impacts of air pollution re- ally catalyse the rich into curbing their vehicle use in order to protect the health of the whole society? It is this question which is at the heart of the debate on monitoring environment and health - a ques- tion which faces many cities. Improving methods to monitor problems is just one step - an equally fundamental question is how to use that informa- tion to influence policy. Summary This article addresses the development of data linkage methods for the analysis of urban environmental health problems and the development of appropriate policies and discusses, based on existing experience of data linkage in Sao Paulo (Brazil), the potential for routine environmental health monitoring and management in a major developing country industrial centre. The article looks briefly at two major environmental health problems in Sao Paulo: first, air pollution which has potential impacts on health of the whole population; and second, environmental differentials in conditions between groups within cities, which have substantial health ef- fects on the economically deprived. The article argues that the health impact of environmental differentials in Sao Paulo is large, but unmonitored as a serious envi- ronmental health threat. In contrast, air pollution is monitored routinely, although its health effects are rela- tively small at present. The paper concludes with a discussion of policy implications of environmental health monitoring -which potentially require a substan- tial shift in attitudes of the urban wealthy. Resume Sante et environnement a Sao Paulo (Bresil) : methodes d'appariement des donnees et politique adaptee Cet article traite de la mise au point de methodes de raccordement des donnees pour !'analyse des proble- mes de salubrite de l'environnement en milieu urbain et de la conception de politiques appropriees. A partir de !'experience acquise a Sao Paulo en matiere de raccor- dement des donnees, ii etudie les possibilites de sur- veillance continue et de gestion de !'hygiene de l'envi- ronnement dans un grand centre industriel d'un pays en developpement. Deux grands problemes de salubrite Rapp. trimest. statist. sanit. mond., 48 (1995) de l'environnement a Sao Paulo sont brievement expo- ses : d'une part, la pollution atmospherique, qui a des effets potentiels sur la sante de toute la population, et d'autre part, les conditions ecologiques plus ou mains bonnes dans lesquelles vivent les differents groupes de citadins, qui ont d'importantes consequences sanitai- res chez les classes defavorisees. Malgre son ampleur, l'impact sur la sante des differences ecologiques a Sao Paulo ne fait objet d'aucun controle, alors qu'il s'agit d'un grave danger pour la sante. En revanche, la pollu- tion atmospherique est surveillee de maniere systema- tique, bien que ses effets sur la sante soient a l'heure actuelle relativement faibles. L'article se conclut par un examen de la strategie qu'implique le controle de la salubrite de l'environnement - par exemple, un changement profond d'attitude chez les citadins plus favorises. References/References 1. Stephens, C. et al. Environment and health in developing countries: An analysis of intra-urban mortality differentials using existing data in Accra (Ghana) and Sao Pau/,o (Brazil) and analysis of urban data of four Demographic and Health Surveys. London, London School of Hygiene and Tropical Medicine, 1994. 2. EMPLASA. Piano metropolitano da grande Sao Pau/,o: 1993- 2010. Sao Paulo, EMPLASA, 1994. 3. Funda(:io SEADE. 0 novo retrato de Sao Pau/,o. Sao Paulo, Fun~o SEADE, 1992. 4. Leitmann, J. Siio Pau/,o environmental projif.e. Washington, UNDP/WORLD BANK/UNCHS, 1991. 5. Funda(:io SEADE. Pesquisa das condif(ies de vida na Regiiio Metropolitana de Sao Pau/,o. Sao Paulo, Fundacao SEADE, 1992. 6. Sobral, H.R. W, Be Silva, C.C.A.S. Balanco sobre a situ~ao do meio ambience na metr6pole de Sao Paulo. Sao Pau/,o em perspectiva, 3(4): 75-81 (1989). 7. Funda(:io Instltuto Brasileiro de Geografia e Estatistlca. Censo demografico de 1991: analise preliminar. Rio de Janeiro, FlBGE, 1992. 8. Moraes, I.H.S. Diretorio de bases de dados de interesse da sali.de. Rio de Janeiro, Fundacao Nacional de Sau.de, 1993. 9. CETESB. R.elatorio de qualidade do ar no Estado de Sao Pau/,o. Sao Paulo, CETESB, 1992. 10. Junqueira, L. et al. Gestao publica no setor sali.de: um piano estrategico. Sao Paulo, Fundap, 1992. 11. Laurenti, R. &Jorge, M.H.M. 0 atestado de 6bito do ponto de vista medico. In: Laurenti, R. &Jorge, M.H.M. 0 atestado deobito. CBCD, Serie Divulgacao, pag. 7-36, Sao Paulo, 1987. Wld hlth statist. quart., 48 (1995) 12. Duchiadi, M.P. Poluicao do are doencas respirat6rias: uma revisao. Cademos de saude publica, 8(3): 311-330 (1992). 13. WHO/UNEP. Urban air pollution in megacities of the woruL Oxford, Blackwell, 1992. 14. Jacobi, P. Habitat and health in the municipality of Sao Paulo. Environment and urbanization, 2: 33-45 (1990). 15. SEMPLA. Base de dados para o planejamento. Sao Paulo, SEMPLA, 1992. 16. Cohen, L. Be Swift, S. A public health approach to the violence epidemic in the United States. Environment and urbanization, 5(2): 50-66 (1993). 17. Monteiro, C.A. Contribuicao para o estudo do significado da evolucao do coeficinete da mortalidade infantil no Municfpio de Sao Paulo (SP), Brasil nas tres filtimas decadas (1950-79). Rev. sali.depublica, 16(1): 7-18 (1982). 18. Monteiro, C.A. Be Benicio, M.H.A. Estado nutricional e renda. Sao Pau/,o em perspectiva 1(3): 67-70 (1987). 19. Ferreira, C.E.C. Mortalidade infantil: a manifestacao mais cruel das desigualdades sociais. Sao Pau/,o em perspectiva, 3(3 ): 24-29 (1989). 20. Ferreira, C.E.C. Be Saad, P. A questao da mortalidade. In: Fundacao SEADE, O jovem na Grande Sao Pauw. Sao Paulo, Fundacao SEADE, 1988. 21. Environmental Protection Agency. National air quality and emissions trends report. North Carolina, EPA, 1992. 22. Ribeiro, H.P. Estudo das correlacoes entre infeccoes das vias aereas superiores, bronquite asmatiforme e poluicao do ar em men ores de 12 anos em Santo Andre. Pediatriapratica, 42: 9 (1971). 23. Mendes, R. Be Wakamatsu, C. T. Avalia~ dos efeitos agudos da polui~ao do ar sobre a saude, atraves do estudo da morbidade diaria em Sao Caetano do SuL Sao Paulo, CETESB, 1976. 24. Rummel. D. et al. Infarto do miocardio e acidente vascular cerebral associados a alta temperatura e mon6xido de carbono em area metropolitana do sudeste do Brasil. R.evista de saude publica, 27(1): 15-22, 1993. 25. Saldiva, P.H.N. et al. Air pollution and mortality in elderly people: a time series study in Sao Paulo, Brazil, Archives of environmental health (forthcoming). 26. Saldiva, P.H.N. et al. Association between air pollution and mortality due to respiratory diseases in children in Sao Paulo, Brazil: a preliminary report, Environmental research (forthcoming). 27. Cohn, A. et al. A Sali.de como direito e Como Servi~. Sao Paulo, Cortez/CEDEC, 1991. 28. Jacobi, P. Be Camara, L.M. Environment probl.ems facing urban households in the city of Sao Pau/,o. Sao Paulo, Stockholm Environment Institute/Cedec, 1993. 29. Grindle, M.S. Be Tbomas,J .W. Publicchoii:es and policy change. The political economy of reform in developing countries. Colorado, Westview Press, 1992. 30. Lucki, B. Pollution and control: a histary of nineteenth am.tury London, London, Adam Hillgar Publications, 1989. 107 Health and Environment Analysis for Decision-Making (HEADLAMP): field study in Accra, Ghana Jacob Songsorea & Greg Goldsteinb Introduction Background of the study According to the World Health Organizations Global Strategy for Health and Environment, "the subject of 'health and environment,' studied by the WHO Commission, encompasses the health consequences of interaction between human populations and the whole range of fac- tors in their physical and social environments. The physical environment includes both natural and man-made elements, consisting of physical, chemical and biological factors. The social envi- ronment includes the social conditions (values, customs, beliefs, etc.) and structures - for exam- ple, those affecting access to employment and education - that determine the distribution of health risks and health-sustaining benefits. It has been made clear by the WHO Commission that health and environment are related subjects needing to be considered in the broad context of overall development, ... ".c In order to monitor progress in environmental health interventions, it is important to develop contextually relevant environmental health indi- cators based on health and environment moni- toring data and data linkage analysis. Data linkage methods for the development of environmental health indicators should be understood as indis- pensable tools for policy making and manage- ment (Epidemiologi,c methods for linking health and environment monitoring data for decision making, draft WHO document). This is the rationale for the current Health and Environment Analysis for Decision-Making (HEADLAMP) field study. This Accra study is a component of a HEADLAMP ex- ploratory study. This article examines the existing situation with regard to health, environmental and demo- graphic data within Accra and following from this, evaluates the state of the art of existing health and environment data linkage within the city. a Associate Professor of Urban and Regional Development, Department of Geography and Resource Development, University of Ghana, Legon-Accra, Ghana. b Coordinator, Healthy Cities Programme, Division of Operational Support in Environmental Health, WHO, Geneva. c World Health Organization, WHO global strategy far health and environment, Geneva, document WHO/EHE/93.2, 1993. 108 Defining the city of Accra Currently, there are two definitions of the city of Accra in use: (i) the Accra District or the Accra Metropolitan Area (AMA), which consist of the city of Accra, lying within the 1963 municipal boundary of Accra. Indeed, the Accra District or Accra Metropolitan Area is contrasted with other po- litico-administrative units in the Greater Accra Region including the Terna and Ga Districts. This narrow definition, currently in use for most administrative and budgetary purposes, does not correspond to the actual boundaries of the urbanized area. This has negative impli- cations for urban planning and development as large areas of the city are effectively ig- nored; and (ii) the Greater Accra Metropolitan Area (GAMA). While the boundaries of the city of Accra were largely established in 1953 (before Ghana's political independence) with some minor adjustments in 1963, much of the growth and expansion of the city has taken place since political independence, with the result that large, urbanized areas lie outside the jurisdiction of the Accra Metropolitan Area while economically, physically and func- tionally Accra, Terna and Ga Districts have become one, single, integrated metropolitan region. Urban planners confronted with de- veloping a strategic plan for the national capi- tal now talk in terms of the wider urbanized metropolitan region consisting of two nuclei, i.e., Accra Metropolitan Area and Terna Dis- trict containing the industrial satellite town- ship of Terna together with Ga District which has received much of the urban sprawl occur- ring beyond the congested core area of Accra. This new urban reality is what is now termed the Greater Accra Metropolitan Area (GAMA). GAMA defines the effective boundary of the capital city which, with a projected 1990 popula- tion of 1. 7 million, is the largest urban agglomera- tion in Ghana (1). Although about 75% of the total population is located in Accra District, the more rapid growth rates are found in the industrial Dis- trict of Terna and the peri-urban District of Ga (2). Consistent with the new vision of planners manag- ing urban growth and development, any future Rapp. trimest. statist. sanit. mond., 48 (1995) HEADLAMP activity should adopt GAMA and not be limited to AMA which reflects only a partial reality of the city. Data availability and quality at the city and sub- area levels Introduction In Accra, as in most low-income developing coun- tries, there is no coherent management informa- tion system covering environmental, demographic and health aspects of the city. However, these cate- gories of data are collected either routinely or on an ad hoe basis by different agencies within the city. The existing data bases are of differing quality but can be adapted for the purpose of health and environment data linkage. Health monitoring data There are two main types of health data bases of vital importance: mortality and morbidity data. Mortality data sources. The major agency responsi- ble for compiling routine mortality data is the De- partment of Births and Deaths in the Ministry of Local Government. The information exists in regis- tries and the Department's head office and is compiled on an annual basis. In addition to this raw data, the Department publishes Annual Re- ports. Supplementary data can be obtained from the Department of Health of the Accra Metropoli- tan Assembly. These include unpublished data on burials and mortuary dispatches together with An- nual Reports. The Criminal Investigation Depart- ment (CID) of the Ghana Police Service sometimes has data on those found dead by the police, either in the streets or in accommodation with no rela- tives. These are classified as "unknowns". Police records include age, sex, cause of death and place of residence. These other sources of data help fill gaps and achieve a more complete coverage of mortality data. The mortality data are derived from the regis- ters of births and deaths. The latter are compiled manually using cause-of-death certificates issued by medical practioners. This information is enhanced by interviews held with the relatives of the deceased for legal reasons. The available information in- cludes: age at death; primary cause of death; sex; place of residence; place of death; and date of death. Mortality data quality. The main problem with the quality of mortality data often results from ~nder- reporting which is most acute with regard to mfant deaths and deaths of children under 5 years. Ac- cording to Stephens et al., under-reporting for this age category could be as high as 50%. Reporting of deaths for those over 5 years is approximately 75% accurate (3). These comments notwithstanding, Wld hlth statist. quart., 48 (1995) mortality data are of a far better quality than mor- bidity data within the city. Morbidity data sources. The Ministry of Health col- lects data on morbidity through the following agencies: - Health Research Unit; - Metropolitan Health Office; - Regional Health Office; Korle-bu Teaching Hospital departments; and - Centre for Health Statistics. Routine morbidity data are collected from health stations and hospitals and submitted for collation on a weekly basis for certain communica- ble diseases of national importance and on a monthly basis for all diseases. These data do not capture those cases reporting to private hospitals, mission hospitals and quasi- government facilities such as police or military hos- pitals, etc., nor those at Korle-bu Teaching Hospi- tal which are reported separately. Very often, it is possible to get information on age, sex, disease, place of residence and date of reporting at the health facility from individual pa- tient's cards and daily summaries of out-patient department attendance records. Morbidity data quality. Because of the compulsory registration of deaths, mortality data, despite the shortcomings enumerated above, tend to be more accurate than morbidity data. For example, it is estimated that about 50% of the population in Accra rely on private practitioners or treatment by traditional healers, and self-medication through the direct purchase of drugs at pharmacies and from drug peddlers. Whereas data from private hospitals can easily be obtained as the system of record-keeping is similar to that in use at govern- ment hospitals, there are no records on self-medi- cation at drug stores and traditional healers. Conclusion. Morbidity and mortality data collected at the lowest level are collated in summary form to reflect the burden of ill-health either at the region- al or district level. Such data are never disaggregat- ed by residential area or by other smaller geo- graphical units to enhance the analysis of health and environment data linkages within the city. Environmental monitoring data Data sources. This is the weakest link in routine data collection at sub-area level. There is now available a rich source of environmental data from household surveys or ad hoe studies. The institutions from which these data can be obtained include: (i) Accra Planning and Development Pro- gramme; (ii) Ministry of Works and Housing; 109 (iii) Accra Metropolitan Assembly, especially the Waste Management Department; (iv) Ghana Water and Sewerage Corporation (GWSC); (v) University of Ghana; (vi) Water Resources Research Institute of the CSIR; (vii) Ministry of Environment, Science and Tech- nology (MEST); and (viii) other independently conducted research. Routine data on water quality can be found at GWSC for the metropolis as a whole. Some data (although often incomplete) on waste manage- ment is also routinely collected at the city level and sometimes disaggregated by the 6 sub-metropoli- tan districts of the AMA. Quality data, sometimes on a residential-area basis, can be found in the following studies: (i) 1990 - Housing Needs Assessment Study; (ii) 1991 - Report on 17 Deprived Areas of Ac- cra; (iii) 1992 - Strategic Plan for GAMA; (iv) 1989 - Environmental Study of Accra Metro- politan Area; and (v) 1988 - Urban Utilities and Municipal Services. All these studies were commissioned and exe- cuted for the Accra Planning and Development Programme in support of the Strategic Plan for the Greater Accra Metropolitan Area. Other useful studies that were independently carried out in- clude: (i) 1993 - Environmental Problems and the Ur- ban Household in the Greater Accra Metro- politan Area ( GAMA) - Ghana. A collabora- tive study by the University of Ghana and the Stockholm Environment Institute and its Data Base; (ii) 1991 - Environmental Profile of Accra by Amuzu and Leitmann; and (iii) 1988 - Ghana Living Standards Surveys by the Ghana Statistical Service. Data quality. Although the data quality may be very good for independent studies, the environmental monitoring data are derived from different sources for different periods and covering different popu- lations. • While environmental information is derived largely from household surveys, data on popula- tion, mortality and morbidity are based on re- corded events in the whole population thereby creating problems for area-based data linkage for health and environment data (3). • Disaggregated, area-based routine data on the environment are at a very low level of develop- ment within the various institutions concerned with environmental health. Until recently, 110 there were hardly any useful scientific indica- tors of the state of the environment for sub- units in relation to water quality, supply or con- sumption, sanitation, garbage collection, sew- age disposal, etc. Demographic monitoring data Data sources. The single most important institution responsible for demographic data collection is the Census Office of the Ghana Statistical Service of the Ministry of Finance and Economic Planning. The most useful sources of demographic data on regional and district basis can be derived from routine census publications: 1960 - Population Census; 1970 - Population Census; 1984 - Population Census and Projections; 1989 - Demographic and Health Survey; and 1987-88 - Ghana Living Standards Survey. But by far the most useful publications for the purpose of linking health and environment data can be found in the 1960 and 1970 special &port 'A' on Large Towns, i.e., towns with populations of 10 OOO or more. This special report for the 1984 Census has not yet been published. Within these special reports, data for the Metro- politan Area are disaggregated at three levels: • At the lowest level of disaggregation is the Enu- meration Area (EA). There are over 700 EAs for the AMA alone. Each urban EA is supposed to contain 1 OOO persons. • The EAs are aggregated at the next level into statistical Areas (SAs). These are created by combining geographically contiguous EAs with apparently homogeneous physical and social characteristics. Most of the names given are known features of the area and several SAs bear the traditional name of the area though the boundaries of the two may differ. • Using information at the EA level it was possible to aggregate EAs into the 55 residential areas provided by the Survey Department for the De- mographic Studies and Projections for Accra Metropolitan Area (AMA) study in support of the development of the Strategic Plan for GAMA. This framework was subsequently used by later studies. Data quality. The data are of good quality and in- clude information on age, sex, birth place, school attendance, ethnicity/nationality and economic activity. This demographic base-line data is vital for most other analyses. The main problem arises from the fact that censuses are undertaken in 10-year cycles. This, therefore, entails population projections if demo- graphic information is to be brought up to date. The underlying assumptions of these projections Rapp. trimest. statist. sanit. mond., 48 (1995) are often debatable, especially with regard to mi- gration and natural increase. Given the dynamics of inter-residential population mobility within the metropolis and rural-urban migration, the data de- rived from population projections have to be treat- ed with care, and even more care when the health and environmental data are derived from different time-periods. Health-environment data linkages Introduction Notwithstanding the problems enumerated above a number of studies have already been undertaken linking health and environment data with some success. These include the following: • Health Differentials Study; • Household Environmental Study; • Cholera mapping; and • Community-Based Environmental Manage- ment Information System (CEMIS). Case 1: Health Differentials Study One major socio-ecological (geographical) study linking environmental and health effects data for the Accra Metropolitan Area is the work by Ste- phens et al. (3). This study used routine mortality data collected in the city together with accumulat- ed socio-environmental data from household sur- veys as the basis of an ecological study of the city comparing different residential areas and broad ecological zones of the city. The first step was the classification of different residential areas using composite socio-environmental data. The next step was an analysis of mortality differentials among the socio-environmental zones identified in the first step. Composite socio-environmental zones in Accra. The study used the stratification of socio-environmen- tal zones adopted by the Accra Planning and Development Programme of the Town and Coun- try Planning Department. Seven zones have been identified using 3 main indicators in a very gener- al manner (Box 1). These indicators are as fol- lows: - average household income by residential area; - population density by residential area; and - age/ ethnicity of residential area. Using this classification of residential zones as a starting point, a more detailed discussion of the socio-environmental conditions of these 7 zones was undertaken by making detailed use of the 1990 Housing Needs Assessment Study data base, devel- oped by Housing and Urban Development Asso- ciates, 1990. This was supplemented by other sources. The findings tended to confirm the deprived nature of Zones 1 and 2 where about 63% of the Wld hlth statist. quart., 48 (1995) population resides. These zones contain the 17 most deprived areas in the AMA. Mortality differentials between socio-environmental zones. The residential areas and the socio-environ- mental zones in which they are found were then related to the distribution of mortality in Accra. The hypothesis was that areas with theworstenvironmen- tal and socio-economic conditions had the worst health burdens in terms of mortality indicators. The following mortality indicators were exam- ined by residential area: - crude mortality rates ( deaths per 1 OOO popula- tion); - infectious and parasitic disease mortality (per 10 OOO); - respiratory disease mortality; and - circulatory disease mortality. The general conclusion from the patterns is that there is a high correspondence between areas of socio-environmental deprivation and high rates of mortality especially with regard to infectious and parasitic diseases. Despite the correlation noted above it must be emphasized that these rates are based on small areas in terms of population and number of events. There was no up-to-date age and sex information at this low level of disaggregation and yet age, especially, could be a basis of vulnerability. For example, areas with a higher proportion of vulner- able age groups within the population could expe- rience higher mortality levels thereby confounding the effects of poor environmental conditions. This defect was corrected at the broader socio- environmental zone level with the use of estimated zone-based age distribution. The consistency of the above patterns was tested using cause, age-specific and age-adjusted analysis of mortality differentials by zone. For this purpose, the 7 socio-environmen- tal zones initially identified were collapsed into 4 zones as follows. Zone 1 - High Density Indige- nous Sector (HDIS); Zone 2 - High Density Low Class Sector (HDLCS) ; Zone 3 - Middle Density Indigenous Sector (MDIS); and Zone 4 - Middle- High Class Sector (including the initial MDMCS, LDMCS, LDHCS and LDNDS). The following mor- tality indicators were used: (i) all-cause mortality rates by zone, age-adjusted and by age group (per 10 OOO); (ii) cause-specific mortality rates (per 10 OOO) by zone for those under age 14 ( using infectious, respiratory and circulatory disease classifica- tions); (iii) cause-specific mortality rates by zone as in (ii) for ages 15 to 44; (iv) cause-specific mortality rates by zone for those over age 45; (v) cause-specific age-adjusted mortality rates by zone and cause (i.e., circulatory, infectious and parasitic, and respiratory); and 111 Box 1 Definition of socio-environmental zones in Accra 1992 Zones Description HDIS High density indigenous sector Zone 1 HDLCS Zone 2 These areas are the oldest sections of Accra. They house "indigenous" communities - mainly the original Ga townships with family compound houses and similar history and culture. Population very dense: growth rates now low. Characteristically low incomes mainly from fishing. Very poor levels of infrastructure. High density low-class sector Areas are characterised by very high densities.low income population: high percentage of population are migrants. Ethnically diverse. Extremely poor infrastructure conditions. High growth rates. Most areas are low-lying; easily flooded. Housing is sometimes temporary wooden shacking. MDIS Medium density indigenous sector Zone 3 Shelters people who otherwise have been living in the HDIS but have moved out because their lot has improved. Incomes are marginally higher than HDIS and HDLCS. Densities not as high as in HDIS. Many migrants also live here. Infrastructure poor to adequate. MDMCS Medium density middle-class sector Zone 4 Started as LDHCS but has been overcome by rapid urbanization. Residential quality and services are good. Housing people with primary education or better: incomes are medium but slightly lower and densities are higher than LDHCS or LDMCS. LDMCS Low density middle-class sector Zone 5 LDHCS Zone 6 LDNDS Zone 7 Started as state-owned estates for government staff. With time the quality of the estates has deteriorated. Densities are relatively low, as are growth rates. Population is middle income; infrastructure conditions are adequate. Low density high-class sector This area is populated by high socio-economic status people with high levels of education and wealth. It has low density, low growth and has adequate infrastructure and services. Low density newly developing sector Newly developing settlements usually on the city fringe. Some evidence of lack of basic infrastructure but housing facilities are usually adequate. These areas are populated by newly middle income groups seeking to develop property. Growth rates are rapid. Source: Ref. (3), p. 20. Residential areas covered Osu, Karle dudor, Nungua, Old Teshie, La, James Town, Chorkor, Adedenkpo, Old Dansoman, Karle gonno. Nima, Accra New Town, Sukura/Russia, Tudu, Saban Zongo, Mamobi, Madina, Accra Central. Abossey Okai/Mataheko, Darkuman, Abeka, Mamprobi Kokomlemle, Kpehe, New Mamprobi, Bubuashie, Adabraka, North Odorkor. Asylum Down, Avenor, Alajo, Teshie-Nungua Est, Kotobabi, Ablenkpe, Dzorwulu, Lartebiokorshie, South Labadi, Teshie Camp, Burma Camp. Teshie, Dansoman Est, Tesano, South Odorkor, Kaneshie, Ministries, New Dansoman, North Industrial, North Kaneshie, South Industrial. East/West Ridge, North Labone, Ringway Est, Cantonments, Airport. East Legon/Shiashi, Legan Village, Achimota. (vi) "excess mortality" rates for other zones using the ideal state of health found in Zone 4, the least deprived in the socio-environmental sense. tality rates from infectious and parasitic, circulatory and respiratory diseases" (3 p. 46). Overall, the relationships identified at the resi- dential area level have been corroborated at the broader level of the 4 socio-environmental zones after adjusting for age. "Proportionally, 67% of Accra's population live in areas with excess mor- 112 The import of this major geographical, ecologi- cal study is that it demonstrates the policy rele- vance of attempting health and environment data linkage on a more sustained basis as a guide to planning. Such ecological studies aim at describing prob- lems in terms of areas of risk and the proportion of the whole population affected. They can operate Rapp. trimest. statist. sanit. mond., 48 (1995) Box 2 Stratification by residential sector, Accra Stratum High-density indigenous sector (HDIS) Low-density high-class sector (LDHCS) Middle-density middle-class sector (MDMCS) Low-density middle-class sector (LDMCS) Low-density newly developing sector (LDNDS) Middle-density indigenous sector (MDIS) High-density low-class sector (HDLCS) Rural fringe (RF) Total Source: Ref. (4), p. 13. on a large scale to estimate the population at risk. For this reason, they are more usefully seen as macro-planning tools for environmental health policy, while individual data are more useful for the subsequent micro-level understanding and policy. "A serious limitation in conducting these stud- ies concerns the measurement of exposure in indi- viduals. Routine monitoring provides average ex- posures for geographic regions, and all persons within the relevant area are assumed to have expo- sures that equal the average exposure. Another limitation of routinely collected data is that expo- sure and health outcome data are recorded sepa- rately, making it impossible to link them directly".d Whereas ecological studies are useful, the above observation is in a sense their Achilles' heel. They are, nevertheless, of special significance in devel- oping countries where individual-level studies hardly exist. Case 2: Household Environmental Study Introduction. Another important study which equal- ly showed the relevance of environment and health data linkage is the study of "Environmental prob- lems and the urban household within the Greater Accra Metropolitan Area". While its point of depar- ture was the environment, it did come out with startling, although epidemiologically consistent, conclusions relating environmental risk factors to health outcomes. This was based on a household survey in 1991 (2, 4). Methodology. The field work for the study included the following: (i) a detailed and structured questionnaire sur- vey of 1 OOO representative households to- d World Health Organization. lnf<mMl consultation on Health and Environment Analysis j<Yr Decision-making (HEADLAMP) methods andfiel,dstudies-Summaryreport. Geneva, WH0, 1994 (WHO/ EHG/94.15), p. 3. Wld hlth statist. quart., 48 (1995) Sample (%) Sample size 17 2 5 11 3 12 46 5 100 170 20 50 110 30 115 455 50 1 OOO gether with physical tests of water quality and exposure to air pollution for a sub-set of 200 households; (ii) selected focus group discussions with commu- nity groups in 14 low-income neighbour- hoods; and (iii) unstructured discussions or interviews with policy makers and implementors. The first step in the sampling procedure in- volved a proportional stratification according to the residential categories outlined in Box 2, an adaptation of strata employed in the 1990 Housing Needs Assessment Study by Housing and Urban Development Associates for the AMA. This study, however, covered the larger Greater Accra Metro- politan Area consisting of the AMA ( or Accra Dis- trict), Terna District and Ga District. Consequently, a new residential category known as the "rural fringe" was included. The sample was apportioned across the sectors according to the relative number of households residing in each stratum. The background demographic information on households, environmental problem areas cov- ered, and health problems covered are shown in Box 3. The main health problems covered were diarrhoea and acute respiratory infections among children under 6, and respiratory problems such as cough, sore throat and hoarseness among the prin- cipal homemakers. Ecological/area variations in environmental risk factors and health outcomes. Results of our analysis of envi- ronmental risk factors showed area based-varia- tions in environmental conditions among the 3 districts within the wider GAMA, and more espe- cially in 8 socio-ecological zones or residential strata which were subsequently collapsed to 6 for analysis. More importantly, there were even clearer rela- tionships between household wealth and environ- mental burdens since the residential sectors were rather variegated and not as homogeneous as the 113 Box 3 Selected topics covered in household environment surveys in Accra, Jakarta and Sao Paulo Background information: Household size and age structure Indicators of income/wealth Gender of household head Education (principal male & female) Migratory status (principal homemaker) Type and quality of residence Size of residence and plot Tenure of residence Time householders spend at home Water: Type of water supply by use Ease of access to drinking water supply Water storage practices Water filtration or boiling practices Water supply disruptions Sanitation & hygiene: Type of toilet Toilet sharing Toilet use practices (e.g. use of toilets by children) Indications of unhygienic toilets Hand cleansing practices of principal homemaker Pests: Presence of flies in kitchens and toilets Mosquito biting Animals kept at home Rodent problems Cockroach problems Housing problems: Crowding Damp problems Building materials Indoor air pollution: Fuels used for cooking and heating Location of cooking place Cooking practices Pesticide use Smoking practices Food contamination: Food storage practices and facilities Food preparation practices Indications of poor food hygiene Use of food vendors Health: Children's diarrhoea problems Children's respiratory problems Respiratory problems of principal homemaker Source: McGranahan, Gordon, Household survey as tools for assessing environ- mental problems in /ow-income cities. paper presented at the WHO Consultation on Environmental Health Indicators, Ousseldorf (1992), page 4. 114 wealth groups identified in the survey. Wealth was used as a summary description of the socio-eco- nomic status of households. This indicates the pos- sibility of self-protection from larger neighbour- hood problems, at the household level. When relating health indicators to ecological area, the associations were also evident but even more so when health outcome data or indicators were linked to household wealth or other socio- environmental characteristics. Identifying high risk factors from logistical regression analysis. The advantage of household survey meth- odology is that it provides detailed information on the morbidity factors, risk factors and underlying population characteristics of interest thereby allow- ing for simultaneous analysis and test of cause- effect linkages. The application of a logistical regression analy- sis enables the analyst to identify the more sensitive environmental risk factors directly linked to health outcome rather than a broad discussion of environ- mental influence. This is of immense importance in developing environmental health indicators that are specific to the city. The focus was on notifi- able communicable ill-health problems in the me- tropolis and these health burdens were monitored for children and women, who are among the most vulnerable in Ghanaian society. The results suggest the following classification of environmental risk factors: • socio-economic variables; • access variables; • service efficiency variables (related to access variables) ; • crowding variables; • ecological/vector prevalence variables; and • behavioural variables ( 4 p. 25). Prevalence of diarrhoea and acute respiratory infection among children under 6 and respiratory problem symptoms in the principal female home- makers showed variation across these variables. Box 4 shows the approximate relative risk of envi- ronmental factors monitored for the various cate- gories of ill-health. The major limitation is that the survey is cross- sectional and not based on routine data that can be used more regularly to monitor trends over time. The range of ill-health conditions was also limited. Case 3: Cholera mapping The mapping of outbreaks of cholera within the Accra Metropolitan Area using routinely-collected, hospital-based data has been pioneered by Dr. De- rek Aryee, the Senior Medical Officer of Health in charge of the Adabraka Polyclinic in 1993. The basic geographical units used for the sim- ple mapping of the frequency of cases are the residential areas already identified in earlier stud- Rapp. trimest. statist. sanit. mond., 48 (1995) Box 4 Approximate relative risk of environmental factors in Accra Environmental factor Approximate relative risk A. Risk factors with respect to diarrhoea among children under age 6 Pot used for storing water 4.3 3.1 2.7 2.6 2.2 2.1 2.1 2.0 Water interruptions occur regularly Toilet shared with more than 5 households Prepared food purchased from vendor Water stored in open container Neighbourhood children defecate outdoors Many flies in kitchen during interview Hands not always washed before food is prepared B. Risk factors with respect to symptoms of acute respiratory problems among children under age 6 Children often present during cooking 2.6 2.4 2.3 2.2 1.8 1.8 1.7 Many flies in kitchen during interview Less than 4 m2 per person in most crowded sleeping room Water supply interruptions occur regularly Mosquito coils used Cooking never done outdoors Roof leaks during rains C. Risk factors with respect to symptoms of respiratory problems among principal women of the householdsa Pump-spray insecticide used 3.5 1.6 1.5 1.4 1.1 Water supply interruptions occur regularly Roof leaks during rains Cooking never done outdoors Cigarettes smoked per day.b • Data for only one woman per household are included. That is, the woman who was interviewed and for whom more detailed information is available, on morbidity and education, for example. b The relative risk factor for this environmental factor is determined per cigarette. Source: Reis. (2, 5). Note: All of these factors were statistically significant (greater than 95% confidence) in a logistical regression. Control variables (for example, wealth, age of principal women, number of children under age 6) were included but are not presented. The approximate relative risk, or odds ratio, is the odds of having the symptoms if the factor is present divided by the odds of having the symptoms if the factor is absent. ies which are contained within identified socio- environmental zones in the city. Most of the report- ed cases are clustered around the deprived resi- dential areas in the western parts of Accra. They include Sukura, Sabon Zongo, Bubuashie, Kanesh- ie and parts of central Accra. These are some of the areas identified by all other studies as having the worst environmental conditions and a high level of infectious and parasitic diseases. These areas in particular had suffered from acute water shortage in 1993 and it is therefore not surprising that the worst effects of the outbreak were registered there. The regular "water hunts" in times of water scarci- ty coupled with poor water handling practices by children who carry the water are some of the relat- ed factors. According to Dr Aryee; "This brings home very strongly the use of mapping for purposes of linking and monitoring of environmental and health data for the purposes of decision making" (Aryee, Cholera Mapping in Accra, unpublished, p. 2). Mapping can be a powerful tool for identify- ing cells of disease occurrence and for directing and monitoring interventions. W/d hlth statist. quart., 48 (1995) The above example of very simple disease map- ping (and indeed any health parameter mapping) shows the heuristic value of such an approach once the underlying environmental conditions of the unit areas are known. This goes to support the case for health and environment data linkage on a more sustained basis especially with regard to noti- fiable communicable diseases. Case 4: Community-based Environmental Management Information System (CEMIS) The initiation of the Ghana Country Project on Community-Based Environmental Management Information System (CEMIS) has come at a time of increased concern for the relationship between environment, habitat and health and the sustain- able management of human settlements in Ghana. In order to supplement government efforts in this direction for GAMA, the UNCHS (Habitat) and the Ministry of Environment, Science and Tech- nology plan to support and pilot test the CEMIS framework in one low-income community within the Accra Metropolitan Area. 115 CEMIS focuses on the development of a com- munity-based environmental management infor- mation system based on indicators relating envi- ronmental health to housing conditions. The sys- tem is intended to allow the low-income communi- ty itself to identify potential human-settlement-re- lated health deficiencies and to formulate and monitor appropriate interventions. It is hoped that in the end a healthy housing and environmental condition can be achieved which is reflected in a decrease in the disease burden within the commu- nity (CEMIS, Accra, unpublished).d Although the focus is on self-driven community initiatives it is obvious that the availability of a city- wide system of environmental health indicators will enhance community self-assessment of environ- mental risk factors and the setting of priorities for intervention. Application of health and environmental monitoring data to planning and policy making Some of the potential benefits of health and envi- ronmental monitoring data and environmental health indicators within the Greater Accra Metro- politan Area for both health policy formulation and urban management are: (i) It will fill the gaps in our knowledge about intra-urban health burdens. (ii) It will help make decisions on priorities for improvements in health and human settle- ment. (iii) It will help develop action plans for imple- mentation. (iv) The development of environmental health indicators within the metropolis based on envi- ronment and health data linkage will be of great value in monitoring progress in en- vironmental health management and surveil- lance of notifiable communicable diseases. (v) It will provide planners with updates on the human settlement and environmental health situation in the various ecological zones, thus functioning as an "early warning" system for urban environmental and health problems and thereby enhance diaster preparedness, etc .. (vi) It helps planners to incorporate any trends in environmental and health conditions into the strategic plan of the metropolis and to plan ongoing interventions. (vii) It aims at providing planners, community members and other actors with necessary in- formation which they can utilize in planning, implementing and managing strategic inter- ventions aimed at improving housing and en- vironmental conditions in human settlements (3,4).e e Ochola, L. et al. Framework j<YT community-based environmental management inf<Yrmation system (CEMIS), Nairobi, UNCHS (Habitat), 1994. 116 Conclusion: feasibility of routine linkage of health and environment indicators in Accra in terms of availability of routine data Notwithstanding the above problems, it is safe to conclude that it might be possible to convert all the baseline studies on environmental, housing, demo- graphic and health conditions into a GIS database. Since routine mortality data exist this can be moni- tored over time. Routine morbidity data with some adaptations could also be applied. Baseline data on environmental conditions and demographic pat- terns can be updated from time to time using rou- tine census information, ad hoe studies and up- dates on the situation in particular communities following planning interventions. Disease mapping, and indeed health parameter mapping, should be promoted in the Ministry of Health as an aid to disease monitoring and as a tool for directing interventions and monitoring the ef- fects of such interventions. There is a need for close collaboration be- tween the Health Unit and other sectors involved in environmental monitoring, to enable health and environment data linkage and monitoring. A co-ordinating committee comprising all the relevant sectors with an interest in this type of information should be set up and the roles of specific agencies in respect of environmental and health data collection, analysis and dissemination clarified. Success will be enhanced if the system for numbering houses is updated so that one can lo- cate any individual or household within a census enumeration area, locate that enumeration area within a residential area, locate residential areas within a census district and locate these within the socio-environmental zones at higher levels of spatial resolution. This will allow the analysis of geocoded health and environment information at all levels, from the household to the wider city level. Since GAMA is being adopted as the new plan- ning area, residential area mapping should be un- dertaken for Terna and Ga Districts to provide a comprehensive set of residential areas for the geoc- oding of information. This can be achieved through aggregations of EAs. These should all be classified into the 8 socio-ecological ( environmen- tal) zones for more general analysis. As much as possible future sub-divisions of EAs for census purposes should be contained within the identified residential areas. For more detailed analysis, household data based on house numbers and EAs could be used. These lower-level units will allow more refined mapping of patterns which need not always coin- cide with predetermined broad residential areas or ecological zones especially when analysing the spa- tio-temporal diffusion of epidemic diseases or other health parameters. Rapp. trimest. statist. sanit. mond., 48 (1995) The co-operation of different agencies in data gathering including the APDP and the Census Of- fice is critical for the establishment of sustainable health and environment data linkage in GAMA. Summary This field study assesses the feasibility of routine data linkage of health and environmental indicators in Accra, Ghana. In Accra, as in most low-income developing countries, there is no coherent management information system available covering environmental, demographic and health aspects of the city. These categories of data are, however, collected either routinely or on an ad hoe basis by different agencies within the city. These data bases, even though of variable quality, can be adapted for the purpose of health and environment data linkage. A number of studies have already been undertaken within Accra linking health and environment data with a good measure of success. These include: an intra- urban environment and health differentials study using mortality data; a household survey of environmental problems and the urban household in which morbidity data and environmental data were linked, also indicat- ing intra-urban differentials together with the isolation of high risk factors from a logistical regression analysis; simple cholera mapping; and an initiative to pilot-test a community-based environmental management infor- mation system in a low-income community in which health and environmental data will be linked. All these studies have created a greater awareness of the need for such data linkage and of the value of such studies for the sustainable management of the city. Resume Analyse sante et environnement pour la prise de decisions (HEADLAMP): elude de terrain a Accra (Ghana) On a etudie la possibilite de raccorder les donnees ordinaires des indicateurs sanitaires et environnement aux Accra (Ghana). A Accra, comme dans la plupart des pays en developpement a faible revenu, ii n'existe pas de systeme coherent d'information gestionnaire sur Wld hlth statist. quart., 48 (1995) l'environnement, la demographie et la sante dans la ville. Toutefois, ces differentes categories de donnees sont reunies soit systematiquement, soit de maniere ponctuelle, par plusieurs organismes implantes dans la ville. Bien que de qualite variable, ces bases de don- nees peuvent etre adaptees en vue d'un raccordement des donnees sur la sante et sur l'environnement. Les etudes raccordant les donnees sur la sante et sur l'environnement deja menees a Accra ont ete relative- ment fructeuses: etude des differences intra-urbaines en matiere d'environnement et de sante a l'aide de donnees sur la mortalite; enquete sur les problemes lies a l'environnement affectant les families citadines par le raccordement des donnees sur la morbidite et sur l'environnement, indiquant les differences intra-urbai- nes et isolant les facteurs de haut risque grace a une analyse de regression logistique; etablissement d'une carte du cholera; projet pilote de systeme d'information communautaire pour la gestion de l'environnement dans une collectivite a faible revenu, dans lequel sont raccordees les donnees sur la sante et sur l'environne- ment. Ces initiatives ont fait prendre conscience de la neces- site de raccorder les donnees en question et de !'impor- tance d'etudes de ce type pour une gestion durable de la ville. References/References 1. Accra Planning and Development Programme, UNDP, HABITAT, Strategic pi.an far the Greater Accra Metropolitan Area, Vol. I, Accra, 1992. 2. Benneh, G. et al. Environmental probl,ems and the urban lwu.sehold in the Greater Accra Metropolitan Area (GAMA) - Ghana, Stockholm, Stockholm Environment Institute (1993). 3. Stephens, C. et al. Environment and health in developing countries: an analysis of intra-urban differentials using existing data, London, London School of Hygiene and Tropical Medicine, 1994. 4. Songsore, J. &: McGranahan, G. Environment, wealth and health: towards an analysis of intra-urban differentials within the Greater Accra Metropolitan Area, Ghana. Environment and urbanization, 5(2): 10-34 (1993). 5. McGranahan, G. &: Songsore, J. Wealth, health and the urban household: weighing environmental burdens in Accra,Jakarta, and Sao Paulo.Environment, 36(6): 4-11, 40-45 (1994). 117 The effect of outdoor air pollution on mortality risk: an ecological study from Santiago, Chile Manuel Salinasa & Jeanette Vegab Introduction Santiago, the capital city of Chile, has special envi- ronmental conditions. It is located in the plain zone of the Central Valley, with a median altitude of 500 m. The Metropolitan Area is limited to the east by the Andes Mountains, and to the north by a chain of peaks (San Cristobal and Manquehue). The winds are 60% SW with a mean speed of 3 to 4 m, and 20% NNE with a mean speed of 1 to 2 m. The climate is predominantly dry in summer and cold in winter with few rainy days (annual mean rainfall of 330 mm). The region is affected by the subtropical high pressure phenomenon known as the South Pacific Anticyclone, which generates a dynamic atmospheric thermal inversion layer that, in winter, can stay as low as 200 to 300 m above ground level and is relatively stable ( 1). This combination of factors leads to poor atmo- spheric ventilation and therefore to frequent air pollution episodes of great concern to the popula- tion because of the potential health risk from a variety of pollutants that frequently reach high lev- els during the cold seasons. Special attention is given to the high prevalence of respiratory diseases during the winter among children and the elderly, which are attributed by the media and general public to the effects of air pollution. In Chile, respiratory diseases constitute the third leading cause of general population mortal- ity, and for infant mortality they follow perinatal and congenital diseases (2). Most of the paediatric deaths are due to acute respiratory infections among malnourished and socially deprived chil- dren. Respiratory diseases also account for the ma- jority of hospital use and for about 50% of chil- dren's visits to primary care centres. There exists abundant evidence on the deleteri- ous effects of air pollution on health. Acute epi- sodes of air pollution have been linked to mortality in the Meuse Valley (Belgium) in 1930, in Donora (Pennsylvania, United States of America) in 1948, and in London (United Kingdom) in 1952 (3, 4). In relation to the health effects of low concentra- tions of air pollutants, there appears to be no de- tectable threshold, as seen in the analysis of daily mortality in the Philadelphia and Steubenville a Epidemiologist, Catholic University of Chile, Santiago, Chile. b Epidemiologist, Pan American Health Organization, Santiago, Chile (E-mail: Vega@Paho.org). 118 communities in the United States (4). Up to a 4% increment in daily mortality counts has been asso- ciated with each 100 µg/m3 increase in suspended particulate matter per cubic meter, estimated for the previous day (3-5). A recent review and meta- analysis of 12 studies on mortality and air pollution estimated a relative risk of 1.06 (95% confidence interval 1.05 - 1.07) for a 100 µg increment in total suspended particles. The author concluded that the most reasonable interpretation of results is causal (6). Different Chilean studies have analysed the re- lationship between morbidity and mortality and air pollution. For morbidity, an ecological study showed differences in the proportion of daily visits to primary care centers for respiratory conditions in the capital compared to a control city located 45 miles (70 km) to the north, without the heavy emission rate of air pollutants of Santiago. Never- theless, this study did not measure actual air pollu- tion in the control city.c In another study, the same cities were compared as part of an epidemiological monitoring system for health effects of air pollu- tion. The study showed a higher proportion of bronchial obstructive diseases in Santiago during winter, and a higher incidence of pneumonia in the control city during spring. Sulphur dioxide and particulate matter (PM10) were measured in the control city in a 2-month period during which the air quality standards were not exceeded. In Santiago they were exceeded frequently for sus- pended particles, ozone and carbon monoxide.cl With regard to mortality, studies conducted in Chile have found an association between daily mortality and air quality data, controlling for con- founders such as mean temperature and humidity. These studies have used linear regression models assuming a normal distribution for the outcome variable. The purpose of the present study was to deter- mine the effect of air pollution on daily mortality in Santiago and to analyse whether the geographi- cal distribution of the risk of death within the Metropolitan Area of Santiago was due to the bad quality of urban air. c Epidemiol.ogical study on effects of air pollution, Final Report, Metropolitan Region Government, Santiago, December 1989. d Epidemiol.ogical monitoring system for the effects of air pollution in Santiago, Final Report, Special Commission for the Control of Air Pollution, Santiago.June 1993. Rapp. trimest. statist. sanit. mond., 48 (1995) Data and methods The present study analysed mortality in Greater Santiago from 1988 to 1991, extracting data from the records at the National Institute of Statistics. For each death, the age, municipality of residence, and cause of death were registered. First, the risks of death in the 32 municipalities of the province of Santiago plus 2 municipalities (San Bernardo and Puente Alto) included geo- graphically in the urban region of Santiago were compared. According to the last census, in 1992, this area has 4 756 700 inhabitants, with the popu- lation per municipality ranging from 41 100 to 328 900. Standardized mortality ratios (SMR) were calculated for each municipality using as the stan- dard the population of Chile and the age-specific mortality rates for each year under analysis. Age distribution for municipalities was obtained from the Demographic Annual Reports edited by the National Institute of Statistics.e,f,g,h Deaths due to injuries and poisoning (ICD 800 to 999) were ex- cluded. Subsequently, a correlation analysis was per- formed between SMRs, infant mortality rates and proportion of the population living under the pov- erty level for each of the municipalities. For this purpose, data were obtained from the last official survey on social conditions, carried out every two years by the Ministry of Planning.i To analyse the risk of mortality for specific res- piratory system diagnoses, we arbitrarily defined 14 population zones taking into account their pop- ulation size and geographical location. The aim of this procedure was to stabilize incidence rates across geographical zones through a merger of communities of larger population size. Specific mortality rates for pneumonia (ICD 480 to 487), chronic obstructive pulmonary disease (COPD) (ICD 491, 492, 496) and asthma (ICD 493) were calculated for each zone. In addition, in order to analyse seasonal differ- ences in the geographical distribution of risk, we calculated monthly SMRs by zone, using the coun- try's age-specific mortality rates for the same month of the year. Maps and graphs were elabor- ated to show temporal and geographical trends of risks in the general and specific mortality, using the Epimap software (7). Finally, a multiple regression analysis was done assuming a Poisson distribution model. The de- e Boktin de demografia. Instituto Nacional de Estadisticas, Santiago, 1988. f Boktin de demografia. Instituto Nacional de Estadisticas, Santiago, 1989. g Boktin de demografia. Instituto Nacional de Estadisticas, Santiago, 1990. h Boktin de demografia. Instituto Nacional de Estadisticas, Santiago, 1991. i Encuesta nacional de condiciones socioeconomicas. Ministerio de Planificaci6n, Santiago, 1992. Wld hlth statist. quart., 48 (1995) pendent variable was daily deaths counts occurring in Greater Santiago (excluding injuries and poi- sonings). The independent variables were levels of outdoor air pollutants, daily mean temperature, and humidity. These data were obtained from the Air Quality Monitoring Network ofSantiagoj Each observation corresponded to one day in the period betweenjanuary 1, 1988 and December 31, 1991, with the following variables: • number of non-violent deaths; • relative humidity(%); • temperature, in degrees Celsius; • previous day's temperature, in degrees Cel- sius; • suspended particles less than 10 µm and greater than 2.5 µm of aerodynamic diameter, ex- pressed in µg/m3; • suspended particles less than 2.5 µm of aerody- namic diameter, expressed in µg/m3; • carbon monoxide, maximum moving 8-hour average, expressed in parts per million (ppm); • sulphur dioxide, daily mean, expressed in µg/ m3; and • ozone, maximum hourly concentration, ex- pressed in µg/m3. Data on pollutants and meteorological vari- ables came from 5 monitoring stations located in the city. The data included in the analysis were daily means from the different stations for each day, considering missing values as non-existent. Data analysis was carried out using STATA statisti- cal softwarek, controlling for co-linearity and inter- action among regression variables. Results Standardized mortality ratios, infant mortality rates, and proportion of population living below the poverty level, for each of the municipalities are shown in Tab/,e 1. Map 1 shows the SMR by munici- pality in Greater Santiago. SMRs for the urban municipalities were in general under 100, which was to be expected considering that there is a strong centralization of the services as well as better living conditions in the capital in comparison with the rest of the country; nevertheless, 8 of the mu- nicipalities have SMRs above 100, 7 of which are located in the inner zone of the city. In this zone, air pollutants were more concentrated than in the outer zones of the city. This was due to the high number of mobile sources circulating in this area, suggesting an association between environmental air pollution and risk of death. j Red de monitareo ambiental de calidad del aire. Servicio de Salud de! Ambiente, Ministerio de Salud, Santiago, 1988-1991. k STATA (1993):CornputingResourceCenter, 1640FifthStreet, Santa Monica, California, United States of America. 119 Map 1 Standardized mortality ratio (SMR), municipalities of Greater Santiago, Chile, 1988-1991 Carte 1 Indices comparatifs de mortalite (ICM) dans les municipalites de l'agglomeration de Santiago, Chili, 1988-1991 SMR-ICM CJ 39.7 - 48.3 CJ 48.4-65-7 65.8- 83.1 .. 83.2-91 .9 92.0-100.6 .. 100.7-109.3 .. 109.4-118.0 .. 118.1-126.7 .. 126.8-135.6 WH095374 Map 2 Standardized mortality ratios (SMR) for pneumonia, zones of Greater Santiago, Chile, 1988-1991 Carte 2 Indices comparatifs de mortalite (ICM) due a la pneumonie, zones de l'agglomeration de Santiago, Chili, 1988-1991 SMR-ICM CJ 74.0-79.4 CJ 79.5 - 84.9 85.0- 90.4 90.5- 96.0 t• 96.1 -101 .5 1111 101 .6-107.0 - 107.1 -112.6 .. 112.7-118.1 .. 118.2 - 123.6 .. 123.7-129.3 WH095375 120 Municipalities of Greater Santiago 1. Cerrillos 18. Maipu 2. Cerro Navia 19. Nunoa 3. Conchali 20. P.A. Cerda 4. El Bosque 21 . Penalolen 5. E. Central 22. Providencia 6. Huechuraba 23. Pudahuel 7. lndependencia 24. Puente Alto 8. La Cisterna 25. a. Normal 9. La Florida 26. Quilicura 10. La Reina 27. Recoleta 11 . Las Con des 28. Renea 12. La Granja 29. San Bernardo 13. Lo Barnechea 30. San Miguel 14. Lo Espejo 31. San Ramon 15. Lo Prado 32. Santiago 16. La Pintana 33. San Joaquin 17. Macul 34. Vitacura lo Barnachaa Zones of Greater Santiago, Chile Zones de !'agglomeration de Santiago, Chill 1. North 2 - Nord 2: Huechuraba/Quilicura 2. North 1 - Nord 1: lndependencia/ Recoleta/Conchali 3. North-west- Nord-Quest: Renca/C. Navia/Lo Prado 4. Downtown 1 - Centre-ville 1: Santiago 5. Downtown 2 - Centre-ville 2: Est. Central/Q. Normal 6. West - Quest: Pudahuel/Maipu 7. South 1 - Sud 1: Cerrillos/PA Cerda/Lo Espejo 8. South 2 - Sud 2: El Bosque/San Bernardo/La Plntana 9. East 1 - Est 1: Providencia/Nunaa 10. East 2 - Est 2: Las CondesNitacura/La Reina 11. East - Est: Macul/Penalolen 12. South-east 1 - Sud-Est 1: La Cisterna/San Miguel/San Ramon/La Granja/San Joaquin 13. South-east 2 - Sud-Est 2: La Florida/Pie. Alto Rapp. trimest. statist. sanit. mond., 48 (1995) Map 3 Standardized mortality ratios (SMR) for chronic obstructive pulmonary disease, zones of Greater Santiago, 1988-1991 Carte 3 Indices comparatifs de mortalite (ICM) due aux maladies pulmonaires obstructives chroniques, zones de !'agglomeration de Santiago, Chili, 1988-1991 SMR-ICM ~ 68.0-88.5 ~ 88.6-93.3 ~ 93.4-93.7 lill 93.8-111 .7 i1l 111i~d 111 .8 - 119.4 IJlil 119.5-119.8 .. 119.9-122.0 .. 122.1 -123.0 .. 123.1-161 .5 1111161 .6 WH095376 Map 4 Standardized mortality ratios (SMR) for asthma, zones of Greater Santiago, Chile, 1988-1991 Carte 4 Indices comparalifs de mortalite (ICM) due a l'asthme, zones de !'agglomeration de Santiago, Chili, 1988-1991 SMR-ICM ~ 32.7-47.8 ~ 47.9-52.1 52.2-88.2 (111 aa.3 IMl1 88.4 - 93.o 1111 93.1 -109.7 .. 109.8-112.9 .. 113.0-133.3 .. 133.4-144.5 WH095377 Wld hlth statist. quart., 48 (1995) Lo Barnechea 121 Table 1 Standardized Mortality Ratio (SMR), Infant Mortality Rate (IMR), proportion of population under poverty level by municipalities of Greater Santiago, 1988-1991 Tableau 1 lndice comparatif de mortalite (SMR), taux de mortalite infantile (TMI) et proportion de la population vivant en dessous du seuil de pauvrete dans les differentes municipalites de !'agglomeration de Santiago, 1988-1991 Municipality - SMR IMR- Poverty- Municipalite TMI Pauvrete Cerrillos 62.9 17.1 20.4 Cerro Navia 90.5 16.4 35.9 Conchali 98.2 10.9 33.6 El Bosque 65.2 12.9 30.1 E. Central 112.2 10.7 23.2 Huechuraba 84.1 15.2 37.0 I ndependencia 121.2 15.3 15.8 La Cisterna 108.3 10.2 19.0 La Florida 74.5 10.6 24.2 La Reina 81.2 7.8 5.6 Las Condes 76.9 5.8 2.9 La Granja 89.1 8.6 33.6 Lo Barnechea 64.0 9.1 25.6 Lo Espejo 90.2 16.5 38.0 Lo Prado 81.0 9.5 34.2 La Pintana 81.6 14.0 42.8 Macul 79.7 12.5 20.2 Maipu 95.0 7.6 19.0 Nunoa 91.2 5.6 7.7 P.A. Cerda 85.6 11.6 38.2 Penalolen 84.9 10.4 28.5 Providencia 85.7 8.6 0.2 Pudahuel 92.5 18.1 23.3 Puente Alto 118.9 9.6 22.4 a. Normal 119.4 18.8 27.2 Ouilicura 88.2 11.9 31.6 Recoleta 94.9 15.1 24.2 Renea 90.3 16.7 30.8 San Bernardo 107.7 13.3 30.6 San Miguel 135.6 12.2 16.5 San Ramon 82.7 13.4 34.4 Santiago 120.5 13.9 21.9 San Joaquin 92.9 14.9 19.0 Vitacura 39.7 7.3 2.1 The correlation analysis between SMRs and proportion of population under the poverty level or without access to health services revealed no correlation (r = 0.07 and r = 0.08 respectively). The linear correlation coefficient was stronger when poverty was related to infant mortality rates across municipalities (r = 0.5). This indicates that in this particular city, infant mortality was somehow more related to living conditions than was general mor- tality, which was, in turn, probably more associated with environmental risk factors such as urban air pollution. The grouping of municipalities into 24 zones is shown in Map 2. Tabl.e 2 and Maps 2-4 show the specific SMRs for pneumonia, COPD and asthma 122 Table 2 Specific SMR for respiratory conditions, zones of Greater Santiago, 1988-1991 Tableau 2 lndice comparatif de mortalite imputable aux maladies respiratoires dans differents zones de !'agglomeration de Santiago, 1988-1991 Zone - Ouartier Downtown 2 - Centre-ville 2 North 2 - Nord 2 Northeast - Nord-Est West-Quest South 1 - Sud 1 Southeast 1 - Sud-Est 1 South 2 - Sud 2 Southeast 2 - Sud-Est 2 East- Est West 1 - Quest 1 West 2 - Quest 2 North 1 - Nord 1 Downtown 1 - Centre-ville 1 Lo Barnechea Pneumonia - Chronic Asthma - Pneumonie obstructive Asthme pulmonary 121.9 87.1 99.5 107.0 90.2 117.4 118.5 111.2 81.0 95.0 74.0 108.6 129.3 79.2 disease- Maladies pulmonaires obstructives chroniques 132.7 161.6 111.8 119.9 122.1 123.1 103.4 88.6 93.4 91.5 93.8 119.5 120.1 68.0 155.2 113.0 65.9 92.0 47.9 110.6 52.2 93.1 88.4 47.9 32.7 109.7 133.4 88.3 by zone. The analysis of specific respiratory SMRs by zone was consistent with the distribution of risk for general mortality. The central zones of the city were those with the highest risk for every specific respiratory cause of death. The analysis of seasonal trends in death risks among zones showed no great variations in most of the zones. A slight increase in deaths during winter appeared, which is in agreement with the tendency observed nationally. However, zones with the high- est SMR values tended to have a substantial in- crease during the cold season and corresponded to those located in the central area of the city (Figs. 1 & 2). Regression analysis of daily mortality data ver- sus air quality data are shown in Tabl.es 3 & 4. When the model included all days with available data during the 4-year period, the number of deaths was associated directly with humidity and carbon monoxide and indirectly with temperature. There was no significant association with other variables. There was only a marginal association with the concentration of fine suspended particles (under 2.5 µm). When the days with levels of fine suspended particles below 150 µg/m3 were anal- ysed separately, the suspended particles level vari- able was included in the model. Discussion The analysis of SMR within the territory of Greater Santiago showed a clear pattern in the geographi- Rapp. trimest. statist. sanit. mond., 48 (1995) Flg.1 Standardized mortality ratios (SMR) by month, zones with higher SMR, Santiago, Chile, 1988-1991 Taux comparatifs de mortalite, par mois, pour les districts ayant les taux de mortalite les plus eleves, Santiago, Chili, 1988-1991 -- --Southeast 1 - Sud-est 1 --Downtown 1 - Centre ville 1 - - Downtown.2 - Centre ville 2 - - North 1 - Nord 1 II Ill IV V VI VII VIII IX X XI XII Month-Mois Table 3 Poisson regression model for the total number of days Tableau 3 Modele de regression de Poisson pour le nombre total de jours Fig. 2 Standardized mortality ratios (SMR) by month, zones with lower SMR, Santiago, Chile, 1988-1991 Taux comparatifs de mortalite, par mois, pour les districts ayant les taux les moins eleves, Santiago, Chili, 1988-1991 140...----------------... West 2 - Quest 2 Southeast 2 - Sud-est 2 South 1 - Sud 1 South 2 - Sud 2 East-Est Northeast - Nord-est 11 111 IV V VI VII VIII IX X XI XII Month-Mois Variable Risk ratio - Rapport de risque Standard error - Erreur type 95% Cl lntervalle de confiance de 95% PM2s co Humidity - Humidite Temperature - Temperature Notes: Global X2- x2 total = 670 203 P< 0.0005 R2 = 11.7% Table 4 1.000246 1.011929 0.997519 0.977607 0.0001543 0.0017316 0.0006826 0.0014398 0.111 < 0.005 < 0.005 < 0.005 Poisson regression model (days with fine particulate matter levels under 150 µg/m3) Tableau 4 (0.999944 -1.000548) (1.008541 -1.015329) (0.996182 - 0.998857) (0.974789 - 0.980433) Modele de regression de Poisson (jours pendant lesquels le taux de particules etait inferieur a 150 µg/m3) Variable PM2.s co Humidity - Humidite Temperature - Temperature Notes: Global x 2 - x2 total = 591 734 P< 0.0005 R2 = 11.2% Risk ratio- Rapport de risque 1.000545 1.011248 0.997741 0.978814 Standard error - Erreur type 0.0002319 0.0019537 0.0007230 0.0014985 0.019 < 0.005 0.002 < 0.005 95% Cl lntervalle de confiance de 95% (1.000090 - 1.000999) (1.007 425 - 1.015084) (0.996325 - 0.999159) (0.975882 - 0.981756) cal distribution of risk of death, both for general mortality (excluding injuries and poisoning) and specific respiratory causes (pneumonia, COPD, and asthma). The SMRs were higher for the· central communities of the city, despite the fact that most socially-deprived populations live in the periphery of Santiago, towards the south and west. This phe- nomenon was supported by the lack of correlation with a variety of socio-economic indicators of pov- erty and unsanitary conditions. Infant mortality rates, on the contrary, tended to be more concor- dant with these social indicators; in general the rates were higher in those municipalities with the worst living conditions. Wld hlth statist. quart .. 48 (1995) The analysis of mortality risk by specific respira- tory causes showed that the highest values were 123 consistently situated in the inner city zones. It was possible to observe a gradient in the intensity of this pattern that is more marked for asthma mortality than for mortality from pneumonia. Considering that pneumonia is a frequent cause ofinfant mortal- ity, it is reasonable to expect that the distribution of a specific risk is partially linked to risk factors associ- ated with infant health (poverty, sanitary condi- tions, cultural characteristics of the community). COPD and asthma, in contrast, are principal causes of death in adults. In this case, the distribu- tion of specific mortality risks was not associated with variables showing a deterioration of living conditions or a lack of access to health services, but rather to chronic exposure to heavy air pollution, as is often observed in the inner zones of Santiago. The air quality monitoring network (MACAM) consists of only 5 stations and there are no data on air quality for communities located in the periph- ery. Therefore, it is not possible to assess directly the association between local SMRs and air quality measurements. However, dispersion models based on meteorological data and emission surveys have confirmed that the highest concentrations of air pollutants are found in downtown Santiago. In the central part of the city, carbon monox- ide, suspended particles and ozone frequently reach levels well above the air quality standards. To give a summary characterization of the air quality of Santiago, Tab/,e 5 shows the number of days in which air quality standards for carbon monoxide and PM10 were surpassed during 1989 in each of the 5 monitoring stations, all of them but M station located in the centre of the city. The high frequen- cy of air pollution episodes is due mainly to the elevated number of emission sources (about 12 OOO ill-maintained diesel buses, 30 OOO non-cata- lytic taxicabs and 500 OOO privately-owned cars), a situation that is aggravated during winter by the poor atmospheric ventilation of the area. Multiple regression models for daily mortality and environmental variables, revealed a significant association with carbon monoxide level (mean val- Table 5 ue from the 5 stations, 8-hour maximum mobile mean, expressed in ppm), and an association with the mean concentration of fine suspended parti- cles for those days with values below 150 µg/m3, suggesting a no-threshold effect. During the winter months Uune, July and Au- gust), SMRs tended to increase predominantly in those zones located in the city center. This finding strongly indicates that the higher SMRs observed in these zones are due to some environmental con- dition rather than the effect of demographic or social conditions. Concluding remarks The results of the present ecological study indi- cate that the effects of air pollution at the group level increase both the general and cause-specific mortality risks. The effect of air pollution on mor- tality, in the case of Santiago, overwhelms the influ- ence of socioeconomic factors, revealing striking differences in the geographical distribution of risk. Based on current scientific knowledge (8), we con- sider that there is sufficient evidence to support the hypothesis of a causal association, despite the lack of more specific data on exposure. SMRs for asthma or COPD could be used as environmental health indicators in this Metropolitan Area. The governmental agency in charge of air pol- lution control in Santiago has implemented several actions aimed to lower emission rates, in the long term and has developed a plan to manage critical air pollution episodes. This plan includes recom- mendations for the population, and temporary re- strictive measures against the principal sources, ac- cording to the level of air pollution and meteoro- logical conditions. These measures, however, do not take account of the unequal distribution of the mortality risk across the different zones within the Metropolitan Area. The present study calls for ur- gent action regarding high risk groups among the exposed communities and more effective control of urban air pollution. Carbon monoxide (CO) and particulate matter (PM 10), MACAM Network, Santiago, 1989 Tableau 5 Oxyde de carbone et PM10, Reseau MACAM, Santiago, 1989 Station A B c D M 124 co Number of incidents over the standard - Nombre de fois ou la norme a ete depassee 315 415 24 610 0 Number of days over the standard - Nombre de jours ou la norme a ete depassee 56 80 12 79 0 PM10 Peak concentration (µg/m3)- Concentration maximale (µg/m3) 339 344 345 500 346 Number of days over the standard - Nombre de jours ou la norme a ete depassee 51 39 65 76 22 Rapp. trimest. statist. sanit. mond., 48 (1995) Summary The aim of this ecological study was to investigate the effect of outdoor air pollution on the mortality risk of metropolitan inhabitants in Santiago de Chile. Cause- specific deaths by the day for the years 1988-1991 in Santiago de Chile were extracted from mortality data tapes of the National Center for Statistics. Deaths from accidents were excluded. Total and some specific res- piratory diseases deaths were compared calculating the risk of death by municipality and month of the year using age-adjusted standardized mortality ratios (SMRs) controlling for socioeconomic level. Daily counts of deaths were regressed using a Poisson model on the total and fine suspended particles, S02, CO and ozone on the preceding day, controlling for temperature and humidity. A clear pattern in the geographical distribution of risk of death, both for general mortality and specific respiratory causes (pneumonia, COPD and asthma) was found using SMR, with higher values in the most polluted areas regardless of socioeconomic and living conditions. A highly significant positive association was found be- tween total mortality and both fine suspended particles and CO level. The association remained significant for those days with fine suspended particles levels below 150 µg/dl suggesting a no-threshold effect for the total number of deaths. These results are in agreement with previously reported associations, and they add to the body of evidence showing that particulate pollution is associated with increases daily mortality. Resume Incidence de la pollution de /'air ambiant sur le risque de mortalite. Elude ecologique menee a Santiago (Chili) II s'agissait d'etudier !'incidence de la pollution de l'air ambiant sur le risque de mortalite auquel sont exposes les habitants de Santiago du Chili. Le taux de mortalite par cause et par jour enregistre a Santiago entre 1988 et 1991 a ete obtenu a partir des registres de mortalite du Centre national de statistique. Les deces par accident ont ete omis. On a compare le taux de mortalite total et le taux de mortalite imputable a certaines maladies respiratoires. en calculant le risque de mortalite par comte et par mois a l'aide d'indices comparatifs de mortalite par age tenant compte du niveau socio-econo- mique. On a calcule la regression du nombre de deces Wld hlth statist. quart., 48 (1995) par jour au moyen d'un modele de Poisson applique aux particules to tales et fines en suspension et a la teneur en S02, CO et ozone le jour precedent. et tenant compte de la temperature et de l'humidite. L'indice comparatif de mortalite (SMR) a permis de degager une repartition geographique precise du ris- que de mortalite, tant pour la mortalite generale que pour les maladies respiratoires (pneumonie, maladies pulmonaires obstructives chroniques et asthme), les valeurs maximales etant observees dans les zones les plus polluees, independamment des facteurs socio- economiques et des conditions d'existence. Une rela- tion positive est apparue tres nettement entre la morta- lite totale et le taux de particules fines en suspension et d'oxyde de carbone. Cette association est cependant demeuree importante les jours ou le taux de particules fines en suspension etait inferieur a 150 µg/dl, laissant a penser qu'il n'existe pas d'effet seuil pour le nombre total de deces. Ces resultats confirment les liaisons deja observees et prouvent, une fois de plus, que la pollution par les particules s'associe a une augmentation de la mortalite journaliere. References/References 1. World Health Organization. Majar poisoning episodes from environmental chemicals. Environmental & Occupational Epidemiology Series. WHO, Geneva, 1992. 2. Bol.etin anual de nacimientos y defunciones. Ministerio de Salud - Instituto Nacional de Estadisticas. Chile, 1990. 3. Thurston G.D. et al. Reexamination of London, England, mortality in relation to exposure to acidic aerosols during 1963-1972 winters. Environmental health perspectives, 79:73-82 (1989). 4. Schwartz J. &: Dockery D.W. Particulate air pollution and daily mortality in Steubenville, Ohio. American journal of epidemiowgy, 135(1):12-19 (1992). 5. Schwartz J. Air pollution and daily mortality: a review and meta analysis. Environmental research, 64(1):36-52 (1994). 6. Ware J.H. et al. Effects of ambient sulfur dioxide and suspended particles on respiratory health of preadolescent children. American review of respiratury disease, 133:834-842 (1986). 7. Dean J.A. et al. EPIMAP: a mapping program for IBM- compatible microcomputers. Centers for Disease Control and Prevention, Atlanta, Georgia, U.S.A., 1993. 8. Lippmann M. Morbidity associated with air pollution. In: Hutzinger, 0. ed. The handbook of environmental chemistry, 4 (PartC-AirPollution). Springer, Heidelberg 1991 (pp. 31-71). 125 Assessment of the impact of ambient air pollutants on health in Helsinki, Finland Antti Ponkaa Introduction S~ve_ral studies made in the USA and Europe withm the last decade have shown that ambient air pollutants cause short-term health effects at lower concentrations than believed earlier. These concentrations are below the guidelines given by the World Health Organization (1) and by many countries. Short-term health effects in- clude increased frequency of respiratory infec- tions, exacerbation of symptoms of chronic obst- ructive pulmonary disease such as asthma and chronic bronchitis, and direct or indirect exacerbation of symptoms of cardiovascular and cerebrovascular diseases. The outcome is usually measured by studying mortality or rates of hospi- talization. In addition, it can be assessed on the basis of rates of infection or exacerbation of symptoms of obstructive pulmonary disease or results of pulmonary function tests among non- h~spitalized patients, by diary studies or by follo- wing the number of cases or medication use in certain cohorts, or by observing the frequency of absences from day-care centres, schools and workplaces. The possible adverse health effects of ambient air pollutants have been studied for several years in Helsinki by using the frequency of absences from day-care centres, schools and work places, number of diagnoses made in the health care cen- tres, and number of hospital admissions as health indicators (2-4). Additional studies concerning ex- acerbation of symptoms of ischaemic cardiac and cerebrovascular diseases and mortality are in progress. The studies have been conducted to find out whether health effects can be observed in Helsinki, where the climate is cold and the con- centration of gaseous pollutants is low but that of particulates is relatively high, mainly due to ero- sion of street surfaces and to sand used to pre- vent vehicles from sliding. The information ob- tained has been used as the basis for decisions to diminish emissions. The purpose of the present article is to describe the collection of health, meteorological and air pollution data and the methods used for linking them in order to inves- tigate the adverse health effects of air pollution in Helsinki. a Chi~f of Environmental Health, Helsinki City Centre for the Environment, Helsinki, Finland 126 Data collection Reliable data sampling is the basic prerequisite for health effect studies. For pollutant and meteoro- logical parameters, the methods of measurement have become a rather established practice, but ob- taining reliable and extensive health indicator data is more difficult. Health data Helsinki, the capital of Finland with 500 OOO inhab- itants, has many advantages in obtaining health data. The population is very uniform, the socioeco- nomic differences are small, and the nationalized health care system is practically free with services from health centres to the University Central Hos- pital being provided to all residents. In Finland, ~ndividual communities are responsible for arrang- mg health care for the whole population. Helsinki has 25 communal health centres, handling more than 50% of all acute cases of illness requiring treatment, with patients from all social classes and age groups, while occupational medical services comprise the major part of the private medical s~ctor_in Helsinki. Information on hospitalized pa- t.J.ents is very thorough. Good statistics from munic- ipal and university hospitals, covering more than 95% of all admissions in Helsinki, are available, as are good statistics on mortality. Four sources of information have been used in the health effect studies in Helsinki. Communal health centres send weekly reports on all cases of respiratory tract infections and other common in- fections observed among the inhabitants of the city. Secondly, absenteeism has been followed in day-care centres, schools, and workplaces. More than 40% of 1-to 6-year-old children in Helsinki are cared for in about 300 communal day-care centres. The incidence of infectious diseases has been fol- lowed in 14 day-care centres among a sample of about 1 300 children; the diagnoses and frequen- cies of sick absence days are recorded. Absences ?.ue to infectious diseases among all of the approx- imately 55 OOO school children in more than 200 schools are followed up on by trained nurses. Ab- sentee rates have also been observed among the nearly.30 OOO working adults in 12 workplaces rep- resenung 10% of the population of working age in Helsinki. This data set has been used in estimating the health effects of pollutants in 1987 (4) and later in 1987-1991. The weekly frequencies ofrespirato- ry tract infections and absenteeism in 1987 are Rapp. trimest. statist. sanit. mond., 48 (1995) Table 1 Reported weekly frequency of respiratory tract infections and absenteeism in Helsinki, 1987 Tableau 1 Frequence hebdomadaire des infections des voies respiratoires et absenteisme enregistres a Helsinki en 1987 Follow-up period - (weeks) Periode de suivi (semaines) Respiratory tract infections diagnosed at health centres - Infections des voies respiratoires diagnostiquees dans les centres de sante • Upper respiratory tract infection - Infection des voies respiratoires superieures 52 • Tonsillitis - Amygdalite 52 Respiratory tract infections in children in daycare centres - Infections des voies respiratoires chez les enfants dans les garderies • Upper respiratory tract infection - Infection des voies respiratoires superieures 37 • Otitis media - Otite moyenne 37 • Tonsillitis -Amygdalite 37 • Lower respiratory tract infection - Infection des voies respiratoires inferieures 37 Absenteeism - Absenteisme • Children in daycare centres - Enfants dans les garderies 38 • School children - Ecoliers 37 • Adults -Adultes 38 presented in Tabl.e 1. The data up to 1994 are currently being analysed. One important source of health data is the reg- ister of all episodes of illnesses requiring hospital- ization. The register, which hospitals are obliged by law to keep, contains information on the dates of hospitalization and on the diagnoses and ages of patients. The data cover all municipal hospitals and the Helsinki University Central Hospital, which together treat practically all the patients re- quiring hospitalization. Quality control is achieved by checking diagnoses as follows: the patient records and diagnosis are always checked by a se- nior specialist, who is not directly responsible for treatment. The ICD-9 code number for each pa- tient is recorded by the physician after the episode requiring hospitalization, when the results of all the tests and examinations are available. The regis- ter of hospitalization has been used in Helsinki to analyse the effect of pollutants on the daily number of patients hospitalized due to exacerbation of symptoms of asthma, chronic bronchitis and em- physema (3,4). In 1987-1989, the number of ad- missions due to asthma was 4 209 and that due to Wld hlth statist. quart., 48 (1995) Mean (number Standard deviation - Range- or percentage) - Ecart-type Fourchette Moyenne (nombre ou pourcentage) 900 263 376-1514 207 39 126-296 97 28 52-173 13 5 5-27 2 2 0-7 4 3 0-11 9% 3 5-15 3% 1 1-4 1% 0.3 0.7-1.8 chronic bronchitis and emphysema 2 807. The in- formation in the death register will also be used for the same purpose. Air pollutants and meteorological variables In Helsinki, as in most European capitals, pollut- ants are measured continuously. The gaseous pol- lutants, usually S02, NO, N02, and 0 3, and PM10 can be measured automatically, but total sus- pended particulates are collected by a high volume sampler and weighted. There are 6 automatic sta- tions, 5 with high volume samplers. The monitor- ing stations have been situated so that their results are representative of the exposure of the popula- tion to pollutants. Usually the mean concentrations of pollutants measured in various stations are used in the analy- sis. Most commonly the mean 24-hour values are used, and also often the mean daily maximal 1- hour concentrations. Due to daily fluctuations and the guidelines for 0 3, 8-hour and I-hour values are usually used for this pollutant. Temperature and humidity should be taken into account as con- founders. 127 Statistical methods During recent years longitudinal studies have been used in health effect investigations. The popula- tion studied serve as their own controls, thus elimi- nating a potential source of bias in cross-sectional studies (5-14). The methodology of these studies today is quite standardized, and has been present- ed in detail in several studies by Joel Schwartz, for example (10-12). A ongoing multi-centre Europe- an Union collaborative survey aiming for a meta- analysis ofresults from several European cities uses a similar methodology (K. Katsouyanni et al., un- published manuscript). Similar methods have been used in the recent studies in Helsinki (4). In the beginning of these studies in the mid and late 1980s, finding proper statistical methods and the expertise to do these calculations was difficult, and knowledge increased by the trial and error meth- od. This demanded considerable time, and there- fore all attempts to gather information on the methodology of health effect studies are to be re- garded as valuable. The basic principle is to compare the daily val- ues of air pollution and the daily number of health outcomes, e.g., number of hospitalized patients, in a time series by using regression analysis. Model. The main points in the analysis are the selection of regression analysis, control of con- founders, including modifiers in the model, con- trol of autocorrelation, and calculation of missing data. The number of hospital admissions and deaths is usually not normally distributed, but skewed with a low mean and a long tail. In addi- tion, the daily number of events is usually small. Such cases are usually modelled using the Poisson regression (15). Usually, the effects of several pol- lutants are investigated. A separate Poisson model should be done for each pollutant. In the final analysis, the overall significance for each pollutant should be determined using the likelihood ratio statistics comparing the total model and the model without the pollutant in question. Temporal patterns and other confounding variables. Morbidity and mortality show various patterns over time. In the northern hemisphere, the frequency of respiratory infections and death rates is higher during the winter. Morbidity and mortality also vary in shorter cycles. The variation is especially pronounced between different weekdays. Hospital admissions due to exacerbation of chronic obstruc- tive pulmonary diseases for example, are most fre- quent on Mondays and least on Sundays. The sea- sonal fluctuation can be controlled for by using sin and cos terms in the model and by using dummy variables for seasons, days of the week and possibly for months. Also the autocorrelation can be effi- ciently eliminated by using these day-of-the-week dummies; the possible remaining autocorrelation can be dealt with by using autoregression. 128 Fig. 1 shows an example of step-wise controlling of some confounders. The uppermost graph pre- sents the smoothed plot of residuals of the regres- sion model of the daily number of patients hospi- talized due to gastrointestinal disorders. The sec- ond graph presents the plots after including the sinusoidal terms into the model, and the following graphs after including daily temperature, relative humidity, long term trend, and day of the week, respectively. In the lowest graph disappearance of cyclicity can be observed. Dummy variables can also be used to control for influenza epidemics, holidays and other unusual events causing exceptional use of health services or increased mortality. Sufficient control of autocor- relation can be checked by periodograms and that of cyclic trends by observing time plot of residuals after including the confounding variables. Missing values. The missing values of pollutants can be imputed by using the standard regression technique, in which each variable is used as a de- pendent variable and the predicted values from these regressions are used as estimates of the miss- ing values. As the temporal variables are included in the model, the estimates of missing values are roughly based on the weekly, monthly and yearly means. Lagged effects. Because the effect of the depen- dent variable and modifying variables such as hu- midity, temperature or influenza epidemics on health outcomes can be lagged, various lags from O to several days or cumulative lags should be tested. The best-fitting lags should be incorporated into the model. Transformation of variab/,es. The transformation of dependent and modifying variables is often re- quired. The need for logarithmic or quadratic transformation becomes evident when graphic presentation methods are used ( 15). Summary of results of studies in Helsinki Results, published in detail elsewhere (2-4), show that low levels of pollutants were observed to in- crease acute respiratory infections and exacerbate the symptoms of asthma and chronic bronchitis. In 1987, high levels of S02 were observed to be relat- ed to the high number of upper respiratory tract infections diagnosed in health centres, in a study using a standard regression analysis ( 1). No associa- tion between absences of children from day-care centres or schools or of adults from work and the levels of pollutants was found. In 1987-1989, the number of hospital admissions due to asthma was seen to be associated with various pollutants, even though their mean concentrations were low, with the exception of particulates (Tab/,e 2) (2). The number of admissions due to chronic bronchitis and emphysema has also been found to increase with rising concentrations ofS02 and N02 (3). Rapp. trimest. statist. sanit. mond., 48 (1995) Flg.1 Smoothed plot of residuals of the regression model analysis analysing the daily number of hospital admissions due to gastrointestinal disorders, Helsinki, 1987-1989 Residus du modele de regression analysant le nombre d'admissions a l'hopital par jour pour cause de troubles gastro-intestinaux, Helsinki, 1987-1989 -2 1987 1988 1989 0 "' :::, ~ -1 .... a: I -2 "' ".;§ 1987 1988 1989 :::, "C 2 ·;;; ... a: 0 -1 -2 1987 1988 1989 2 0 -1 -2 1987 1988 1989 Year-Annee WH095355 Note: The uppermost graph presents raw data; subsequent graphs show residuals after including the sinusoidal terms, temperature, relative humidity, long-term trend and day of the week into the model. - Le premier graphique reflete les donnees brutes; les cinq autres indiquent les residus apres inclusion dans le modele des termes du sinus, de la temperature, de l'humidite relative, de la tendance a long terme et du jour de la semaine. Wld hlth statist. quart., 48 (1995) 129 Table 2 Mean daily concentrations (µg/m3) of ambient air pollutants and mean values of meteorological variables, Helsinki, 1987-1989 Tableau 2 Concentrations journalieres moyennes (µg/m3) de polluants dans l'air ambiant et valeurs moyennes des variables meteorologiques, Helsinki, 1987-1993 Mean - Moyenne Range - Fou rchette Standard deviation - S02 N02 03 Total suspended particulates (TSP) - Particules totales en suspension Mean temperature - Temperature moyenne ("C) Relative humidity- Humidite relative (%) Discussion The importance of the harmful effects of ambient air pollutants even in low concentrations has be- come obvious during recent years. Statistical meth- ods and standard methodology, used routinely, have made it possible to use various existing health data time series in the local analysis of possible health hazards. These results can be used as the basis for decision-making and actions to reduce emissions of pollutants and for various cost-benefit analyses. As a result of municipal decisions in Helsinki, the emission of pollutants, especially S02 and par- ticulates, has decreased during recent years. N02 emissions are also declining slightly in spite of the increase in traffic. The results of health-effect stud- ies have been used as one basis for decisions in city planning, and emission restrictions covering indus- try and energy plants, although it is difficult to determine exactly how much weight is placed on these findings. The recent popular interest in envi- ronmental health and protection issues has prompted politicians to consider the environment more carefully. The dissemination of objective data through the mass media has also been very important. Often the results obtained in one location can be used to roughly estimate the possible health effects of pollutants in other locations. For exam- ple, the concentration of ambient air particulates, rather than their chemical composition, seems to be critical for the adverse effects: increased num- ber of respiratory infections, exacerbation of symp- toms of chronic obstructive pulmonary diseases and increased risk of death from pulmonary and ischaemic cardiac diseases. These effects have been found regardless of the source of particulates; they can be derived from the combustion of fossil fuels by vehicles, by industry or for energy production (10, 11, 16). Generalization of results from other 130 Ecart-type 19 0.2-95 12.6 39 4-170 16.2 22 0-90 13.1 76 6-414 51.6 5.4 37.0 ± 26.4 9.3 83 37-100 12.0 locations to estimate adverse effects, however, re- quires that there are no major differences in the confounding variables at the various locations_ For example, cold air may alter the effects of pollutants in certain ways. Therefore special circumstances necessitate local epidemiological studies. In Hel- sinki these circumstances include a cold climate and high concentrations of particulates originat- ing mostly from the mechanical action of cars on street surfaces and sand used to prevent accidents on icy roads. Recent studies have shown that many guideline values for ambient air pollutants are too high at present and should be re-evaluated. The most im- portant interacting factors such as climate should be taken into account. Local studies on the adverse health effects of pollutants are needed. Standardization of meth- ods, guidelines and recommendations concerning methodology, all save work and money and ensure the reliability of results. Summary Several studies from various countries within the last decade have shown that ambient air pollutants cause short-term health effects in lower concentrations than believed earlier. Obtaining reliable and comprehensive health and environmental data is difficult but is the basic prerequisite for these studies. Linkage of data on air pollution and on several health effects has been con- ducted in Helsinki since the late 1980s, using time series analysis. The uniform population, small socioeconomic differences, a practically free national health care sys- tem with high coverage and extensive health, pollution and meteorological data render such studies possible. These kinds of local studies are necessary because many confounders or modifiers cause problems for the generalization of results of studies from other locations. International standardization, recommendations and Rapp. trimest. statist. sanit. mond., 48 (1995) guidelines concerning the methodology of linkage stud- ies are needed to save work and money and to ensure the reliability of results. Resume Evaluation de /'impact des polluants de /'air ambiant sur la sante a Helsinki (Finlande) Plusieurs etudes menees dans divers pays au cours des 10 dernieres annees ant montre que des concentrations plus faibles qu'on ne le pensait de polluants dans l'air ambiant avaient des effets a court terme sur la sante. La collecte de donnees fiables et completes sur la sante et l'environnement est certes difficile mais indispensable pour realiser de telles etudes. Oepuis la fin des annees 80, on a entrepris a Helsinki de raccorder les donnees sur la pollution atmospherique et les donnees relatives a divers effets sur la sante en analysant des series chronologiques. Les etudes sont rendues possibles par l'uniformite de la population, les faibles differences socio-economiques, un systeme de sante national quasi gratuit offrant une large couverture et !'existence de nombreuses statistiques sur la sante, la pollution et les conditions meteorologiques. Les eludes locales de ce genre sont necessaires car nombre de facteurs de confusion empechent de generaliser les resultats des etudes menees dans d'autres endroits. Une standardi- sation internationale, des recommandations et des gui- des sur la methode a utiliser pour les eludes de raccor- dement permettraient d'economiser des forces et de l'argent et garantiraient la fiabilite des resultats. References/References 1. World Health Organization. Air quality guidelines for Eur&f>e. Copenhagen, WHO Regional Office for Europe, 1987. (WHO Regional Publications, European Series no. 23). Wld hlth statist. quart., 48 (1995) 2 Ponkii, A. Absenteeism and respiratory disease among children and adults in Helsinki in relation to low-level air pollution and temperature. Environmental research, 52: 34-46 (1990). 3. Ponkii, A. Asthma and low-level air pollution in Helsinki. Archives of environmental health, 46: 262-269 ( 1991). 4. Ponkii, A. &: Vtrtanen, M. Chronic bronchitis, emphysema, and low-level air pollution in Helsinki, 1987-1989. Environmental research, 65: 20 7-217 (1994). 5. Goldstein,J.F. &: Weinstein, A.L Air pollution and asthma: effects of exposure to short-term sulfur dioxide peaks. Environmental research, 40: 332-345 (1986). 6. Bates, D.V. &: Sizto, R. Hospital admissions and air pollutants in Southern Ontario: the summer acid haze effect. Environmental research, 43: 317-331 ( 1987). 7. Pope, C.A. m. Respiratory disease associated with community air pollution and a steel mill, Utah Valley. American journal of public health, 79: 623-628 ( 1989). 8. Bates, D.V. et al. Asthma attack periodicity: a study of hospital emergency visits in Vancouver. Environmental research, 51: 51-70 (1990). 9. Dockery, D.W. et al. Air pollution and daily mortality: association with particulates and acid aerosols. Environ- mental research, 59: 362-373 (1992). 10. Schwartz, J. et al. Particulate air pollution and hospital emergency visits for asthma in Seattle. American review of respi.ratory diseases, 147: 826-831 (1993). ll. Schwartz,J. What are people dying ofon high air pollution days? Environmental research, 64: 2 6-35 (1994). 12. Schwartz, J. et al. Daily diaries of respiratory symptoms and air pollution: methodological issues and results. Environ- mental health perspectives, 90: 181-187 ( 1991). 13. Sunyer,J. et al. Effects of urban air pollution on emergency room admissions for chronic obstructive pulmonary disease. American journal of epidemiowgy, 34: 277-286 (1991). 14. Katsouyanni, K. et al. Air pollution and cause specific mortality in Athens. Journal of epidemiowgy and community health, 44: 321-324 (1990). 15. McCullagh,P.&:Nelder,J.A. Generalizedlinearmodels,2nded. Chapman and Hall, London, 1989. 16. Pope, C.A. m. Respiratory hospital admissions associated with PM10 pollution in Utah, Salt Lake, and Cache valleys. Archives of environmental health, 46: 90-97 (1991). 131 Environmental health indicators and sanitation-related disease in developing countries: limitations to the use of routine data sources Peter J. Kolsky8 & Ursula J. 8/umentha/b This article explores conceptual issues in the devel- opment and use of environmental health indica- tors for disease related to water and sanitation in developing countries. First, some inherent limita- tions are examined in trying to link routinely col- lected environmental and health data. While indi- cators obtained from such "data linkage" are not promising for sanitation-related disease, environ- mental health indicators of a different sort are badly needed to reflect health aspects of sanitation. Problems which environmental health indicators might solve are presented as a spur to further thought and research, and suggestions for alterna- tives to routinely collected data are made, in full recognition of the additional research needed to validate them. limitations to routinely collected data for sanitation-related disease As described elsewherec the Health and Environ- ment Analysis for Decision-making (HEADLAMP) project represents a serious effort to link environ- mental and health data to obtain new insights into the environmental health of a population. Wher- ever possible, HEADLAMP is based on the use of data which are already routinely collected, for very practical reasons of time and cost. Since HEAD- LAMP' s inception, this approach has been seen as more attractive and plausible for some aspects of environmental health than for others. As noted in early HEADLAMP project documentation, link- ages and indicators are likely to be most useful for the monitoring of already-known cause-effect rela- tionships, such as between severe air pollution by sulphur dioxide and particulates, and morbidity and mortality in heart and lung diseases. In this section, we explore reasons why this linkage of routinely collected environmental and health data may be plausible for air pollution, but is unlikely to work well for water, sanitation and faecal contami- a Lecturer, Tropical Health Epidemiology Unit, London School of Hygiene & Tropical Medicine, London, United Kingdom. b Senior Lecturer, Tropical Health Epidemiology Unit, London School of Hygiene & Tropical Medicine, London, United Kingdom. c World Health Organization. Infurmal consultation on Health and Environment Analysis Jar Decision-making (HEADI.AMP) methods andfiel.d studies-Summary report. Geneva, WHO, 1994 (WHO/ EHG/94.15). 132 nation of the environment. In such cases, alterna- tive survey methods may be more appropriate. Air pollution, environmental quality data and health Governments, through GEMS (Global Environ- mental Monitoring System) and other initiatives, regularly collect data on air quality. Governments also regularly collect data on health statistics. Given clear epidemiological relationships between air quality and health, it appears plausible that one could impute the current burden of disease attrib- utable to poor air quality by combining these sets of routinely collected data. If this is true, it also seems plausible that one could: (a) monitor trends of increasing air pollution and declining health; (b) verify, refute or modify currently understood epidemiological relationships between the two based on systematically accumulated experience; and (c) make projections about the health conse- quences of further increases in pollution. We suspect that there are major conceptual difficulties in achieving the above, some of which are similar to the problems involved in sanitation- related disease. Pollution may not be diffused equally over whole neighbourhoods, and levels in the micro-environment may be very different from the neighbourhood average. How detailed a level of monitoring is required to pick up variations within different parts of a city, and is this monitor- ing feasible? Exposure to particulate air pollution is also strongly influenced by the domestic environ- ment. Domestic indoor air pollution is a major environmental health concern in developing countries where inadequate stoves and poor venti- lation yield high local particulate concentrations that routine monitoring of the public environment cannot detect. The problem is further complicated if "com- posite" indicators are developed from a variety of measures. If, for example, health effects of individ- ual pollutants were well understood, it is unclear how a composite index made up of, say, the sum of 3 pollutant concentrations could be used for the above purposes without fresh epidemiological studies to detect the effects of their interaction. Even then, any specific value of the composite indi- cator might be reached in several different ways, each of which may have different health impacts. (Does A+ B + C = 2A + B = 3A?) This is the price of "lost information" inherent in any simplified sin- gle index. Rapp. trimest. statist. sanit. mond., 48 (1995) Flg.1 Transmission of disease from faeces Transmission des maladies a partir des matieres f6cales PB: primary barrier - BP: barrl~re primaire SB : secondary barrier - BS: barri~re secondaire Source: Ref. (4) - R6f. (4). Nevertheless, the case of air pollution is rela- tively straightforward, in comparison with faecal contamination, because of the following two prop- erties of air quality monitoring: (i) some of the measured properties of "air quality" have a direct relationship to health; and, (ii) to a reasonable extent, the air being sampled is the air people breathe (leaving aside for the moment the individual and household varia- tions in exposure noted above.) Faecal pollution, environmental quality data and health Faecal contamination is the most important envi- ronmental health problem in the world; every year diarrhoea kills approximately 3 million children, and hundreds of millions of others suffer the debil- itating effects of schistosomiasis, infection by intes- tinal helminths and diarrhoea. There has been much recent work on the health effects of water and sanitation (1-3). Given the importance of the problem, could we use routinely collected environ- mental and health data to assess the environmental health impact of water quality and faecal contami- nation? We think not, for 4 reasons. Most faecal contamination of the environment is not monitored. A defining characteristic of faecal con- tamination is its spread through a variety of means; this is well reflected in the F-diagram (Fig. 1). Rou- tine environmental monitoring systems do not ex- ist to measure the extent of faecal contamination Wld hlth statist. quart., 48 (1995) of the environment through most transmission routes. While it is well known that contaminated drinking-water can cause epidemics, most endemic sanitation-related disease is not water-borne. In- deed, recent epidemiological literature points to the importance of (a) the roles of both safe excreta disposal (which is not routinely or accurately mon- itored) and water quantity (as opposed to quality) in reducing diarrhoeal disease ( 1, 2); (b) the impor- tance of food contamination which is not moni- toredd (5); (c) the over-riding importance of per- sonal hygiene, which varies from individual to indi- viduate ( 1); and ( d) the practical difficulties of dis- ease attribution given non-linear dose-response curves and multiple transmission routes (6). The significance of water as a disease transmission route (which is the only environmental component rou- tinely monitored for faecal contamination) de- pends upon a host of intermediate and external variables which are not routinely monitored. (In passing, we note that interactive effects with other un-monitored variables could also be a significant problem for environmental and health data link- age in air pollution.) If we restrict our attention to faecal contamina- tion of water, we can nevertheless make the argu- ment sound similar to that for air pollution. "Gov- ernments, through GEMS and other initiatives, regularly collect data on water quality. Govern- ments also regularly collect data on health statis- tics. Can we put the two together to quantify the environmental health burden of poor water and sanitation in a specific context?" The answer is still "no", as shown below. Almost all of the routinely measured environmental properties of water quality do NOT bear directly on faecal-oral disease. Biochemical Oxygen Demand, suspended solids, nitrogen, phosphorus and other routinely collected water-quality data are relevant to the downstream aquatic ecosystem, but not di- rectly to the health of the human population. Only the classic environmental health indicators of "to- tal coliform organisms", E. coli or "faecal coliform organisms" (which indicate the degree of faecal contamination of water) are plausibly linked to the epidemiology of faecal-oral disease. Even these in- dicators are epidemiologically problematic, as they are rarely pathogenic, and no clear dose-response relationship is postulated between them and the burden of faecal-oral disease. In addition, "total coliforms" are not exclusively faecal in origin, but d Esrey, S.A & Feachem, R.G. Interventions far the control of diarrhoeal diseases among young chil,dren: promotion of food hygiene. Geneva, WHO, 1989 (WHO/CDD/89.30). e Martines, J. & Hebert-Simpson, M. Improving water and sanitation hygiene behaviours far the reduction of diarrhoeal disease: the repart of an infarmal consultation, Geneva 18-20 May 1992, Geneva, WHO, 1993 (WHO/CWS/93.10, WHO/CDD/93.5). 133 also occur naturally in unpolluted soil and water, while E. colior "faecal coliforms" are not exclusive- ly human in origin, but are present in the faeces of other warm-blooded animals as well. Such indica- tors are still "classic" (and useful!) because patho- gen-specific monitoring of water is expensive, im- practical and fraught with its own conceptual diffi- culties. Routinely collected water quality data do not reflect the quality of water which is consumed. The water which is sampled for routine environmental monitoring is usually untreated river water. As noted above, it is usually sampled to gauge the "environmental health" of the downstream natural aquatic ecosys- tem, not of the human population. The water sam- pled often has a different quality from that con- sumed by the public. River water is often treated by utilities or consumers before consumption, even if only by sedimentation and storage, and this will not be reflected in the data. Even where drinking-water quality data are col- lected by the utility, this is usually done at the treatment works. In fact, drinking-water is fre- quently contaminated in its distribution, (particu- larly where supply is intermittent,) and in its stor- age. In many cases, those in urban areas most at risk from faecal-oral disease are too poor to drink from treated supplies, and are often supplied by unmonitored vendors. Such vendors draw water from a variety of sources and may add varying de- grees of contamination or treatment. (In rural ar- eas, drinking-water quality monitoring of any kind is usually impractical.) While most people in a small area breathe the same air, not everyone drinks the same water. The water supply and sanitation statistics col- lected by WHO in recent years, cooperating with UNICEF, have generally concentrated mainly on service coverage and not usage. As L. Laugeri pointed out (in a paper presented at the WHO Consultation on the Development and Use of Envi- ronmental Health Indicators in the Management of Environmental Risks to Health, 1993) these data have limited value as public health information because (a) the facilities may be present but not operational or not used; (b) the informal sector may play a large role, yet data on services provided by them are not collected; and (c) they do not describe the quality, reliability or sustainability of the services or their effective utilization. The critical role of hygiene behaviour. The simple act of washing one's hands has been shown in some cases to reduce Shigella infection by 25% (7). An influential WHO reference documente has tar- geted 3 behaviours as particularly important in controlling diarrhoeal disease: hand washing, the safe disposal of children's faeces, and the protec- tion of drinking-water quality in the home. None of these are monitored by "routinely collected envi- 134 ronmental or health data", and they can vary widely between households in the same geographic vicini- ty. Latrines do not improve health in and of them- selves, but only when both male and female adults and children use them and faecal-oral contact is accordingly reduced. Nevertheless, "latrine cover- age" data are collected rather than "latrine usage" data. Exposure data, intermediate variables and outcome data Broadly speaking, the problems oflinking routine- ly collected environmental and health data about faecal contamination of the environment in devel- oping countries fall into 3 categories; difficulties in measuring or quantifying exposure, health out- come, and intermediate variables which influence the relationship between exposure and health. Exposure data Very simply, we do not measure people's exposure to faecal contamination. We need to consider many forms of exposure: e.g., through water, through soil (via inadequate sanitation), through food, and through vectors such as flies. In addition, "exposure" has two components, both of which need to be measured to define the term accurately: the level of contamination (of the water, soil, etc.) and whether or not (and to what extent) a person is exposed to that contamination. Even where we have data on the level of water contamination, we generally do not know who is exposed to that water with any degree of certainty. It becomes clear that there is a weaker case that water quality data reflect the extent of faecal contamination to which people are exposed than that air quality data reflect any threat from the air that people breathe. Useful work on exposure to health hazards from water and faeces has been and can be done. Differences in exposure to schistosomiasis have been measured, through quantifying human con- tact with infested water bodies, and related to the level of infection (8, 9). In the sanitation context, differences in exposure to Ascaris-contaminated street sand have been shown to influence levels of Ascaris infection (C. Smith, 1994, University of London PhD thesis). However, such studies have involved measuring exposure (and infection) at the individual level, and have been done in a re- search context, not as routine monitoring. Intermediate variables A number of intermediate variables influence the relationship between exposure and outcome. As noted earlier, personal hygiene is extremely im- portant, but is not routinely monitored. Structured direct observations are increasingly being used to measure hygiene behaviours, such as hand-wash- ing and defecation behaviour (10-12). However, Rapp. trimest. statist. sanit. mond., 48 (1995) there has been no consensus over how hygiene behaviour can be accurately measured; the publi- cation arising from a recent conference on the subject is now providing useful guidance on the range of methods that can be used ( 13). In an important application of research to field practice, Almedom is now exploring ways in which project level staff and communities can measure hygiene behaviour (14). In exploring diarrhoea morbidity among children, other variables such as nutritional status and breast-feeding practice are important intermediate variables that influence the relation- ship between any environmental indicator of faecal pollution and health outcome. Health outcome Health outcome for these important diseases is not routinely monitored. Only a fraction of even the most serious cases of diarrhoeal disease end up in clinics or hospitals; estimation of diarrhoeal dis- ease morbidity and mortality from routinely col- lected data is generally not recommended, and special surveys are usually required to assess the extent of the problem. Similarly, other faecal-oral diseases (e.g., intestinal helminth infections) are not well-reflected in public health statistics. Are indicators a dead end? In summary, effective linkage between routinely collected health and environmental data for the fundamental problem of sanitation-related disease is unlikely to bear much fruit for 3 reasons: (a) the right sort of environmental data is not being collected, and in some cases may not even be capable of collection; (b) linkage relationships are complex and un- clear, involving intermediate variables, partic- ularly behavioural variables; and (c) relevant health outcome data are also prob- lematic to collect. This does not mean that there is no validity to the concept of environmental health indicators for sanitation-related disease; only that linkage of rou- tinely collected environmental and health data does not appear to be the way forward on this front. In the next section, we explore the need for environmental health indicators for water and sanitation through examples, and conclude with a preliminary exploration of the kinds of data we think would be worth collecting towards this end. The need for environmental sanitation indicators We need environmental health indicators in devel- oping countries to help set priorities for action; this is more important than any role they play in developing relationships between exposure and disease. Wld hlth statist. quart., 48 (1995) Consider a slum on the outskirts of a large city. Water supply is intermittent, and many people do not have house connections, but fetch water from public taps. Because of the intermittent supply, the water is of dubious quality; even when the water supply is on, the available quantities are lim- ited. The area's residents are largely immigrants from the countryside, where lack of space for ex- creta disposal had never been a problem. Some residents have no sanitation facilities at all, others have bucket latrines, while some have toilets con- nected to open drains. Surface water drainage is poor, and during the monsoon, flooding of the open drains is frequent. Many open plots are flooded, and residents spend a substantial amount of money on household insecticides to reduce mosquito biting. Suppose you are the municipal health officer for the above city: how do you rationally decide which environmental improvement is the most im- portant for the residents of this area? It is easy to start listing improvements that "ought to be made" and insist that "all are essential"; if the budget is limited, how can one rationally and prac- tically set priorities among them? One way might be to call in an "environmental health expert", but even this is suspect. If the expert is familiar with water supply but not sanitation, water supply will probably emerge as the priority; if familiar with mosquito control but not low-cost sanitation, then mosquito control becomes urgent; if a drainage expert, then drainage becomes the highest prior- ity. Nor can any of these opinions be easily shown to be wrong; there is no objective methodology with which to set priorities among these hazards, despite recent efforts to further this end.f In the above example, all of the environmental problems listed were those associated with "tradi- tional" infectious disease. Let us now suppose that this slum is in a heavily industrial area of a large city. Many people buy water from vendors, who draw it directly from the river into which asbestos tailings and other industrial wastes are dumped. Air pollution is high. A solid waste dump is located nearby, and many residents scavenge this site for recyclable material. It is rumoured that toxic wastes are also dumped illegally on this site. As we include environmental health problems of industrialization and the disciplines needed to assess them, we also increase the scope for profes- sional bias; industrial waste experts will naturally find that industrial waste is the principal environ- mental health problem. We suspect, in fact, that there is an industrial "bias" in environmental f WASH (Water and Sanitation for Health Project) & Pritech, Environmental health assessment: an integrated 1112thodol.ogy far rating environmental health probl.ems. WASH Field Report 436 (1993). 135 health, as industrial problems are more attractive to technical professionals than the more basic issue of faecal contamination. They also tend to be of more concern to the middle classes, whose basic water and sanitation needs have generally been met. Industrial waste problems are associated with an advanced level of development, and are less embarrassing to discuss; indeed, their existence may be viewed as a sign of industrial progress, and the decision to tackle them as social and govern- mental enlightenment. Industrial waste problems are current topics in international research jour- nals (dominated by the industrialized world) and their study and solution involve sophisticated tech- nical analysis which confers more professional sta- tus than worrying about latrines and community organization. Alternatively, we could be accused by others of a "faecal" bias, which is insensitive to the health hazards of industrialization. This may be justified, and further underlines the need for good environ- mental health indicators. For all of the advances in environmental health over the past 150 years, and the more recent resurgence of interest in environ- mental health issues since the 1960s, we still have no systematic way of ranking the potential health problems. Ideally, a rational system of environmen- tal health indicators would help us to set priorities among different improvements in an objective and systematic fashion. The examples also illustrate the dangers of not having a comprehensive set of environmental in- dicators. First, assume that a "good" environmen- tal health indicator or measure can be developed for, say, levels of small particulates in the air, and that this indicator identifies a potential problem. If an equivalent indicator of faecal contamination of the environment does not exist, policy-makers are not justified in concluding that air pollution is a higher environmental priority for this slum than sanitation, simply because they have a better indi- cator for it. Even worse, suppose water quality leaving the treatment works is taken as a proxy for "faecal contamination of drinking-water"; this could show "no risk" when the realities of inter- mittent supply, contamination in the household, and alternative untreated sources of supply could constitute real, but unmonitored, environmental health hazards. The need for indicators for "traditional" or "pre-transition" environmental health problems is clear. Their conceptual and practical develop- ment is fraught with difficulty; as we have shown, linkage between routinely collected environmen- tal and health data is unlikely to help. There is therefore a need for new approaches to these in- dicators, based on a mix of their heuristic plausi- bility and whatever light can be shed on their rela- tionship to health outcome by rigorous epidemio- logical research. 136 Some plausible candidates for environmental sanitation indicators What kinds of data could be useful to identify envi- ronmental sanitation priorities in a community? The important work of McGranahan and other colleagues in several cities, including Accra ( 15) (G. McGranahan, paper presented at WHO Con- sultation on the Development and Use of Environ- mental Health Indicators in the Management of Environmental Risks to Health, 1993) show how a number of household environmental variables and their relationships to health can be measured through household surveys. They illustrate one path to the development of environmental health indicators. Case studies have so far been carried out in 5 southern cities ( 16). On a community basis, the following types of data would seem relatively straightforward to col- lect as baseline data both for decision-making and for measuring progress during monitoring: • Access to water, e.g., proportions of the popula- tion served by public tap, yard tap, house con- nection and vendor, or distance to nearest tap • Hours/day of avail.abk piped water supply • Excreta disposal type and use, e.g., proportion of households with each main type, including whether they are shared or public facilities, and use by different household members • The price, demand for, and use of soap • Proportion of community streets with paving (an in- hospitable environment for soil helminth eggs) • Persons/room of housing • Ekvation, where flooding is a suspected prob- lem • Proportion of houses entered by floodwater and ap- proximate frequency of flooding • E. coli/100 ml of wat,eras consumed by residents, for each source • Disposal practices for chi/,dren 'sf aeces. Other data will spring to mind in any specific community, and these examples are intended more to show the kinds of data to use, rather than a specific set of indicators. The strengths of the above sorts of indicators are: (a) their plausible linkage to health; and (b) their relatively straight- forward means of collection. Data for some indica- tors may be collected periodically by the relevant authorities ( e.g., hours/ day of available piped wa- ter supply) and may involve technicians in field- work ( e.g., for price and demand for soap), where- as other data will need to be collected through periodic household surveys, group discussions and the like (e.g., excreta disposal type and use). Their drawback is that while the health linkage is plausible, quantification of the relationship will be extremely difficult. While this can in principle be done through rigorous epidemiological studies, such studies are often difficult, and the results may not be applicable outside the communities where they are attempted. Rapp. trimest. statist. sanit. mond., 48 (1995) Other types of indicators could also be consid- ered. Water and sanitation matter to most people for many other reasons besides health, such as time saving, convenience, privacy and status. People spend money on improving their living conditions, and data on the amount they spend (for example, on buying water from vendors, or as payments to nightsoil collectors) could also be collected. Such data could be used to indicate the need for ser- vices, the priority people put on them and also their willingness to pay for them. Assessment of the community's own environmental health priori- ties is a high-priority "indicator"! Despite the difficulties, systematic collection and analysis of this sort of data could indicate the directions in which environmental sanitation ser- vices should develop. If water and sanitation facili- ties appear adequate and well-used by adults, but the disposal of children's faeces is indiscrimi- nate, health promotion would be a high-priority intervention. Where acceptable sanitation facilities are only available to a small proportion of the population, water supply is only available for half an hour per day, and the levels of chemical pollu- tion appear to be approximately 10% above inter- nationally acceptable standards, the basic needs for water supply and sanitation should take prece- dence over the reduction of chemical pollution. Where water and sanitation facilities are adequate and well-used, but chemical standards are violated by a factor of 4, then tighter control of chemical pollution would seem appropriate. Many cases will constitute difficult "grey areas" where risks from different problems appear roughly equivalent. The critical point is that we cannot afford to look at specific environmental health problems in isola- tion, without understanding the risks from other environmental health problems. It is for this rea- son that simple environmental health indicators for the developing world must be developed to provide a rational basis for planning environmen- tal health interventions. Acknowledgements At the time of writing, both authors were members of the Environmental Health Programme at the London School of Hygiene & Tropical Medicine, which was funded by the Health and Population Division of the Overseas Development Administra- tion of the British government. The authors also wish to acknowledge the useful comments made by Dr Sandy Cairncross in reviewing an earlier draft. Summary This article explores conceptual issues in the develop- ment and use of environmental health indicators for basic problems related to water and sanitation in devel- oping countries. In this context, faecal contamination is Wld hlth statist. quart., 48 (1995) the most important environmental health problem. re- sponsible for the death of approximately 3 million chil- dren a year. and the infection of hundreds of millions. Good indicators would be invaluable in assessing the magnitude and source of such problems in different settings. Linkage of routinely collected data Unfortunately, there are inherent limitations in trying to link routinely-collected environmental and health data about "the faecal peril", including the following: • Most faecal contamination is not monitored be- cause it is spread through a variety of routes. • Most routinely collected water quality data do not bear directly on faecal-oral disease. Most are con- cerned with the health of the aquatic ecosystem rather than human health. • Most water quality data do not sample the water which is actually consumed. • Hygiene behaviour, which is a critical mediating factor, is rarely if ever monitored. Exposure data, intermediate variables and outcome data • The right sort of environmental data is not being collected, and in some cases may not even be capable of collection. • Linkage relationships are complex and unclear, in- volving intermediate variables, particularly behav- ioural variables. • Relevant health outcome data are also problematic to collect. Need for environmental health indicators. Despite these difficulties, we need environmental health indicators in developing countries to help set priorities for action; this is more important than any role they play in describing relationships between exposure and dis- ease. Two possible situations where such indicators would be useful are presented. Often, many environ- mental health problems cry out for attention, but re- . source constraints dictate a restricted programme fo- cused on priorities. Good indicators could help set priorities among basic sanitation measures (e.g., water supply, excreta disposal, drainage, etc.) or between basic sanitation and higher levels required by problems of industrialization (e.g., air pollution, industrial wastes, etc.). At present, such priorities are set by judgement and politics alone. The examples also illustrate the need for a comprehensive set of environmental indicators; excreta disposal is still important even if air pollution has better indicators! It would be most unwise to invest only in those activities for which good indicators exist. Pragmatic alternatives to routinely collected data If routine data are currently inappropriate, other data, plausibly linked to health, may be easier to collect than detailed health outcomes. These could include: access to water; hours/day of available piped water supply; 137 excreta disposal type and use; price, demand for and use of soap; the proportion of community streets with paving; and disposal practices for children's faeces. We stress that these are only suggestive of the sort of data to think about collecting, and are not presented as a proposed set of indicators. We need more epidemiology to link the above types of indicators to health, but we are confident they are more relevant than most routinely collected environmental data. The development, test- ing, use and evaluation of such indicators is thus an important challenge for both researchers and practitio- ners in environmental health. Resume lndicateurs de la salubrite de l'environnement et maladies /iees a /'hygiene du milieu dans Jes pays en deve/oppement: obstacles a /'utilisation des donnees recueillies systematiquement Cet article traite sous l'angle theorique de la mise au point et de !'utilisation d'indicateurs de la salubrite de l'environnement pour faire face aux principaux proble- mes lies a l'eau et a !'hygiene du milieu dans les pays en developpement. Dans ces pays, le plus important des problemes de sante lies a l'environnement est la conta- mination par les matieres fecales: environ 3 millions d'enfants meurent chaque annee et des centaines de millions de sujets sont infectes. De bons indicateurs seraient d'une valeur inestimable pour mesurer l'am- pleur et connaitre l'origine de ces problemes dans differents cadres de vie. Appariement des donnees recueillies de maniere systematique Malheureusement, ii existe certains obstacles a l'appa- riement des donnees sanitaires et ecologiques re- cueillies de maniere systematique sur le danger lie aux excreta, notamment: • La contamination par les matieres fecales fait rarement l'objet d'une surveillance continue. • La plupart des donnees sur la qualite de l'eau recueillies de maniere systematique ne portent pas directement sur les maladies transmises par voie orate a part,r des matieres fecales. Elles concer- nent la salubrite de l'ecosysteme aquatique plutot que la sante des hommes. • La plupart des donnees sur la qualite de /'eau ne concernent pas l'eau effectivement consommees. • L 'hygiene, qui joue un role determinant dans la transmission, ne fait qu'exceptionnellement /'objet d'une survetJ/ance. Donnees sur /'exposition, variables intermediaires et donnees sur /'impact Les problemes peuvent etre redefinis selon les con- cepts epidemiologiques traditionnels d'exposition, d'impact et de variables intermediaires: 138 • Les donnees appropriees sur l'environnement ne sont pas recueillies, voire dans certains cas, impos- sibles a recueillir. • Les rapports de correlation sont complexes et flous, car ii y entre des variables intermediaires, notam- ment des variables de comportement. • II est egalement difficile de reunir des donnees pertinentes concernant l'impact sur la sante. Necessite d'indicateurs sur la salubrite de /'environnement Malgre ces difficultes, les pays en developpement ont besoin d'indicateurs de la salubrite de l'environnement pour fixer leurs priorites en matiere d'action sanitaire; c'est la une application bien plus importante que la mise en evidence d'une liaison entre un risque et une mala- die. De tels indicateurs s'avereraient utiles dans nombre de cas, dont deux sont etudies dans !'article. Bien souvent, en raison d'un manque de ressources, les problemes de salubrite de l'environnement exigeant une attention soutenue ne font l'objet que de program- mes restreints axes sur des priorites. De bons indica- teurs pourraient permettre de fixer des priorites entre les diverses composantes de l'assainissement de base (par exemple l'approvisionnement en eau, !'evacuation des excrements et des eaux usees, etc.), ou entre l'assainissement de base et !'attenuation de problemes plus complexes lies a !'industrialisation (pollution at- mospherique, dechets industriels, etc.). A l'heure ac- tuelle, ces priorites resultent de choix arbitraires et politiques. Les exemples fournis mettent egalement en evidence la necessite d'une serie complete d'indica- teurs touchant a l'environnement; !'evacuation des ex- crements conserve toute son importance meme si les indicateurs utilises pour mesurer la pollution atmosphe- rique sont plus precis! II serait tout a fait deraisonnable de n'investir que dans les activites pour lesquelles on dispose de bons indicateurs. Solutions pratiques pour remplacer Jes donnees recueillies de maniere systematique Si la collecte systematique de donnees detaillees rela- tives a l'impact sur la sante est insuffisante, on peut reunir d'autres donnees vraisemblablement liees a la sante: par exemple, acces a l'eau; nombre d'heures/ jours d'approvisionnement en eau potable; type de systeme d'evacuation des excrements et utilisation; prix, demande et consommation de savon; proportion de voies publiques couvertes d'un revetement; prati- ques d'evacuation des excreta des enfants. Ce ne sont la que des exemples des donnees qu'il est possible de reunir et non une liste fixe d'indicateurs. D'autres tra- vaux epidemiologiques sont necessaires pour etablir un lien entre les indicateurs et la sante, mais tout porte a croire qu'ils sont plus pertinents que la plupart des donnees sur l'environnement recueillies de maniere systematique. La mise au point, l'essai, !'utilisation et !'evaluation de tels indicateurs doivent done tenir une place importante dans les travaux des chercheurs et des praticiens s'interessant a la salubrite de l'environne- ment. Rapp. trimest. statist. sanit. mond., 48 (1995) References/References 1. Esrey, S.A. et al. Effects of improved water supply and sanitation on ascariasis, diarrhoea, dracunculiasis, hook· worm infection, schistosomiasis, and trachoma.Bulletin of the Worl.d Hea/J,h Organization, 69(5): 609-621 (1991). 2. Esrey, S.A. et al. Interventions for the control of diarrhoeal diseases among young children; improving water supplies and excreta disposal. Bulletin of the Worl.d Heal.th Organization, 63(4): 757-772 (1985). 3. Cairncross, A.M. Water and Sanitation: An agenda for research. Journal of tropico.l medicine and hygiene, 92(5): 301- 314 (1989). 4. Kawata, K. Water and other environmental interactions- the minimum investment concept. American journal of clinico.l nutrition, 31(11): 2114-2123 (1978). 5. Black, R.E. 8c Lanata, C.F. Epidemiology of diarrhoeal diseases in developing countries. in: Blaser, MJ. et al. ( eds.) Infections of the gastro-intestinal tract. New York: Raven Press ( in press). 6. Briscoe, J. Intervention studies and the definition of dominant transmission routes. Americo.n journal of epidemiowg;y. 120 < 3) : 449-455 < 1984) . 7. Khan, M.U. Interruption of shigellosis by handwashing. Transactions of the Ruyal Society of Tropical Medicine and Hygiene, 76(2): 164-168 (1982). 8. Bundy, D.A.P. 8c Blumenthal, U.J. Human behaviour and the epidemiology of helminth infection: the role of behaviour in exposure in: Barnard, CJ. & Behnke, J.M. Wld hlth statist. quart., 48 (1995) (eds). Parasitism and host behaviour. London, Taylor and Francis, 1990. 9. Wtlkins, H.A. et al. Resistance to reinfection after treatment of urinary schistosomiasis. Transactions of the Ruyal Society of Tropico.l Medicine and Hygiene, 81 ( 1): 29-35 ( 1987). 10. Stanton, B.F. et al. Twenty-four hour recall, knowledge- attitude-practice questionnaires, and direct observations of sanitary practices: a comparative study. Bulletin of the Worl.d Health Organization, 65(2): 217-222 (1987). 11. Curtis, V. et al. Structured observations of hygiene behaviours in Burkina Faso: validity, variability, and utility. Bulletin of the Worl.d Heal.th Organization, 71 ( 1): 23-32 ( 1993). 12. Huttly, S.R.A. et al. Structured observations of hand washing and defecation practices in a shanty town of Lima, Peru. Journal of diarrhoeal disease research 12(1): 14-18 (1994). 13. Boot, M.T. 8c Cairncross, S. (eds.) Actions speak: the study of hygiene behaviour in water and sanitation projects. The Hague, IRC International Water and Sanitation Centre and London School of Hygiene & Tropical Medicine, 1993. 14. Almedom, A. 8c Odhiambo, C. The rationality factor: choosing water sources according to water uses - an example from Kenya. Waterlines, 13 (2): 28-31, (1994). 15. Benneh, G., et al. Environmental problems and the urban househol.d in the Greater Accra Metropolitan Area (GAMA), Ghana. Stockholm, Stockholm Environment Institute, 1993. 16. Stockholm Environment Institute. Urban environments and human welfare in southern cities: lessons from five case studies. Stockholm, Stockholm Environment Institute, 1993. 139 An epidemiological perspective on environmental health indicators Harris Pastidesa The descriptive and analytical tools of epidemiology have become widely recognized as fundamental methods for determining whether environmental exposures have resulted in adverse effects on hu- man populations. Epidemiological methods are also important for helping to determine which spe- cific target populations, defined geographically or by personal characteristics, may be at risk. In numer- ous instances such information has been influential in contributing to the rational planning of public health policies and specific intervention programs aimed at disease prevention. A relevant example of this influence would be the determination of lead exposure limits for children. These limits have been lowered recently as new epidemiological research studies have identified developmental deficits at biological lead levels once thought to be safe. Once interventions or policies have been implemented, the quantitative tools and study designs of epidemi- ology have also been used to evaluate their success in relation to their specific goals. Unfortunately, the methods of epidemiology, especially the analytic research studies, are some- times impractical in terms of directly responding to a perceived environmental health problem. This is because they usually take a long time to execute, can be very costly, and often require specialized skills which may not be widely available in certain parts of the developing world. These and other realities have fuelled much recent discussion about the need to develop tools and measures for moni- toring and assessing a population's health sus- ceptibility and response to environmental factors, and for helping to guide appropriate interven- tions. These measures have been labelled environ- mental health indicators. Indicators have been de- fined in numerous ways, most of which are closely related. For the purposes of this paper, the work- ing definition of an indicator is a summarized mea- sure of a hazardous exposure or health outcome which has been derived through the interpretation of routinely collected or otherwise available infor- mation for the purpose of planning or evaluating population-based interventions, or developing health policy. As discussed by Iqellstrom and Corvalan else- where in this issue (1), environmental health indi- a Professor of Epidemiology, Department of Biostatistics and Epidemiology, School of Public Health, University of Massachusetts at Amherst. 140 cators should be relatively easy to obtain, and should be understandable to policy makers and by the public at large. Most importantly, environmen- tal health indicators must be valid surrogates for established ( or very likely) causal relationships be- tween environmental exposure and disease out- comes. Since the indicator will rarely be a measure of the precise causal risk factor or of the disease itself, it must be a valid substitute for these. If not, it will incorrectly inform public health officials plan- ning possible interventions to reduce morbidity or mortality. The purpose of this article is to provide a framework for relating environmental health indi- cators to the methods of epidemiology and to pro- vide some guidance for selecting and evaluating the appropriateness of proposed environmental health indicators. Before a decision is made to adopt a specific environmental health indicator, the following questions should be addressed: (i) Does the pro- posed indicator truly represent an underlying caus- al relationship between an environmental expo- sure and a health consequence? and (ii) Is the proposed indicator a reasonably valid estimate of the underlying causal factor? Does the indicator reflect an underlying causal mechanism? Just as any therapeutic measure for an individual suffering from an illness should target the un- derlying cause of the illness if it is to result in recovery, so must an indicator relate to an un- derlying cause of disease in the population. The public health community has not infrequently been misled into believing that a non-causal, sta- tistical association identified in some research study was the basis or justification for an action. For example, in the early 1980s scientists and government officials in the United States and Europe called for the banning of amyl nitrate (used as a recreational drug) after a study dem- onstrated a strong association between use of this drug and acquired immune deficiency syn- drome (AIDS) (2). At the time, the quest for some method to forestall the growing epidemic was so strong that officials felt a need to act. This was later judged to be premature in that amyl nitrate use was eventually found to be only a correlate of risky sexual behaviours - the un- derlying source of contact with the human im- muno-deficiency virus (HIV). Rapp. trimest. statist. sanit. mond., 48 (1995) Similarly, an indicator which is not a manifes- tation of a direct causal mechanism should not be used to guide the development of a public health intervention; trying to remedy a perceived prob- lem which is only a correlate of a true risk factor may well be a waste of time and resources. For example, it may be tempting to consider a sum- mary of the average daily measurements of sul- phur dioxide reported by a town's monitoring station as an environmental health indicator since the data are easily accessible and since respiratory illness has been correlated with sulphur dioxide levels. However, the true underlying relationship between respiratory illness and air pollution may involve particulate levels, which sometimes hap- pen to be correlated with sulphur dioxide levels, since they are often emitted from the same pol- lutant source. If so, an intervention strategy or policy decision to reduce sulphur dioxide levels, by levying fines on local industry for example, would be ineffective. Similarly, a decision to mon- itor future sulphur dioxide levels as an indicator of some health-based intervention would miss the mark. A different side of this issue can also be illus- trated. Examining trends in cigarette sales over time may be proposed as an indicator of whether a mass media campaign to reduce the prevalence of smoking will result in a reduction in disease mor- bidity and mortality once the adequate latency pe- riod has elapsed. Cigarette sales data satisfy the criteria of being easily available and simple to un- derstand. This indicator also satisfies the require- ment that it reflects an underlying causal mecha- nism since smoking is known to cause a variety of respiratory diseases. If cigarette sales are reduced and the smoking prevalence in turn is reduced, there is little doubt that disease rates will decline eventually. There is one caveat, however, with re- spect to the inference that the mass media cam- paign was responsible for the decline in smoking. If there were other simultaneous influences on smoking behaviour (e.g. a tax on cigarettes intro- duced at the same time as the campaign), it might be impossible to determine the true cause of the disease reduction without performing a study at the individual level. The biases represented in the examples above may be described as confounding. As is well known, confounding can occur in situations where data have been collected at the individual level as well as at the ecological level. While confounding seems to be adequately accounted for when time- series analyses of ecological data are performed (3, 4), it remains an important potential problem when the data are all from one point in time ( cross- sectional). In terms of a practical strategy, it is necessary for health officials to insist that epidemi- ologists or other public health scientists provide evidence that the indicator being proposed will not Wld hlth statist. quart., 48 (1995) be susceptible to the effects of confounding. The evidence can be from the published epidemiologi- cal and related scientific literature. Alternatively, a small pilot study can be designed for the local context in order to help disentangle the potentially complex web of environmental factors which may be acting simultaneously. Of course, it should be kept in mind that new information concerning potential confounding by previously unsuspected factors can become available at any time. Is the indicator a valid estimate of the causal relationship? Since measurement error is intrinsic to all estima- tion (5), an attempt should be made to evaluate the amount of error associated with a proposed indica- tor, and its impact on any policy decisions to be made. The public health community needs to have more than just a general sense that an environmen- tal health indicator is valid (i.e., that it is closely measuring what it purports to measure). One must keep in mind that the indicator is usually only a surrogate measurement. Surrogates are used in many epidemiological studies, including analytical ones. Take, for example, a case-control study of whether dietary constituents are causally related to colon cancer mortality. Since the actual subjects are deceased, dietary histories might be collected from surviving family members. The accuracy and precision of the final measure of association (i.e., the odds ratio) between diet and colon cancer will depend on two factors: first, the accuracy with which surrogates can report the food intake pat- terns of their deceased relatives; and second, whether the relatives of cases and of controls re- port dietary histories with a similar degree of accu- racy. Given that it would be unrealistic to expect perfect reporting, epidemiologists have come to tolerate minor, and perhaps modest, levels of mis- classification ( of exposure and disease) in their studies. This analogy applies to indicators as well. As a general but arbitrary standard, studies with misclassification rates of up to 10% should not result in a significantly biased estimate of risk or benefit. One can take the example of one of the most widely used health indicators of this century, the infant mortality rate (IMR). The IMR is widely used as an indicator of the general health status of a population and, in particular, of the level of health care services, nutrition and household safety. In this sense, it is like all indicators, in that it is serving as a surrogate measure for broader parameters. The attractive features of the IMR are: (a) it is routinely collected and reported; (b) it is available in all parts of the world thus facilitating interna- tional comparisons; and (c) it is quite accurate in that the required numerator and denominator 141 data are, generally, of acceptable quality in nearly all parts of the world. This is because infant deaths (the numerator) are serious enough that they are reported, and the assessment and reporting of birth rates (the denominator), is also required for many reasons besides health. Of these three advan- tages, accuracy is of course the most important in terms of its use as a health indicator. The main disadvantage of the IMR as an indi- cator is that it is several levels of interpretation away from determining which specific policies or interventions would most directly and quickly lead to a reduction in infant deaths. For example, would maternal education about rehydration therapy be more effective than disease vaccina- tion? Obviously, there can be no answer to this or related questions without detailed analysis of the component causes of infant mortality. Only after breaking down the indicator into smaller, discrete components, or examining it in conjunction with other relevant measures, can the necessary debate about alternative intervention strategies take place in a rational way. Unlike the IMR, a proposal to use paediatric hospitalization for asthma as an indicator for as- sessing a city's efforts in controlling dangerously high levels of air pollution associated with meteo- rological inversions, is more vulnerable to misin- terpretation, for several reasons. First, the diagnos- tic criteria for asthma are frequently not standard- ized; for example, one doctor's diagnosis of asthma may be another's diagnosis of bronchiolitis. Sec- ond, the hospitalization rate may be a general marker of disease severity but in communities with poor access to hospitals (because of location, pov- erty, poor education, etc.), significant underesti- mation of severe asthma may occur. Finally, given the large number of other precipitating factors for asthma, including allergic responses, smoke, and other indoor pollutants, the assumption that asth- ma is predicted by ambient levels of pollution may be wrong. In summary, environmental health indicators can be of great practical value since they are based on routinely available data, or data which can be easily collected and summarized. Their value should be especially significant in terms of provid- ing an empirical basis for making health policy decisions. However, the advantages they provide can be severely outweighed when they are not based on causal associations between environmen- tal exposures and disease outcomes. Since public health officials, including those responsible for health policy, cannot be expected to discriminate between the complicated causal connections which exist for environmental diseases, epidemiol- ogists and other public health scientists must be willing to contribute to the development and eval- uation of potential environmental health indica- tors in a substantive way. 142 Summary There is a great amount of ongoing discussion about the need to develop new ways to assess and monitor a population's disease susceptibility to environmental factors. The ultimate goal in developing these tools, called environmental health indicators. is to increase the public health community's capacity for implementing interventions to prevent disease. Much of the discussion focuses on the requirement that the indicators be rela- tively easy and quick to apply. However. in the rush to find useful existing indicators. or to develop new ones. there is the danger that certain other important attributes of the indicator may be overlooked. These include: (a) whether the indicator truly represents an underlying causal relationship between an environmental exposure and a health consequence; and (b) whether the pro- posed indicator is a reasonably valid estimate of the underlying causal factor. This article provides a framework for relating environ- mental health indicators to the methods of epidemiology including some guidance for selecting and evaluating the appropriateness of proposed environmental health indicators. Examples are given which demonstrate how environmental health indicators can lead to a biased interpretation of underlying associations between envi- ronmental factors and the potential for disease when they are improperly conceived. These problems can be avoided by employing routine epidemiological con- cepts and methods as indicators are developed and evaluated. Resume lndicateurs de la salubrite de l'environnement: perspectives epidemiologiques On debat beaucoup a l'heure actuelle de la necessite de mettre en place des mechanismes nouveaux pour eva- luer et surveiller !'influence de facteurs lies a l'environne- ment sur la sensibilite d'une population a une maladie. En mettant au point ces instruments, appeles indica- teurs de la salubrite de l'environnement, on vise a developper les moyens de sante publique en matiere de prevention de la maladie. On insiste beaucoup sur le fait que les indicateurs doivent etre d'une utilisation relative- ment facile et rapide. Toutefois, dans la hate de trouver des indicateurs utiles parmi ceux deja existants ou d'en mettre au point de nouveaux, ii faut eviter de sous- estimer les autres caracteristiques importantes qu'ils doivent presenter, a savoir: a) qu'ils representent verita- blement un lien sous-jacent de cause a effet entre un facteur lie a l'environnement et un probleme de sante, et b) qu'ils fournissent une estimation satisfaisante du facteur causal sous-jacent. Cet article fournit un cadre pour associer les indicateurs de la salubrite de l'environnement aux methodes epide- miologiques et, notamment, une orientation sur le choix de ces indicateurs et !'appreciation de leur opportunite. Plusieurs exemples montrent comment des indicateurs de la salubrite de l'environnement mal corn;:us peuvent conduire a une interpretation erronee des associations Rapp. trimest. statist. sanit. mond., 48 (1995) sous-jacentes entre des facteurs lies a l'environnement et le risque de maladie. On pourra eviter les pieges en ayant recours aux concepts et methodes epidemiologi- ques traditionnels pour mettre au point et mesurer les indicateurs en question. References/References 1. IgeUstrom, T. &: Corvalan, C. Framework for the development of environmental health indicators, World health statistics quarterly, 48(2): 144-154 (1995). IgeUstrom, T. &: Corvalan, C. Schema pour la mise au point d'indicateurs de la salubrite de l'environnement [resume], Rapport trimestriel de statistiques sanitaires mondiaks, 48(2): 153 (1995). Wld hlth statist. quart., 48 (1995) 2. Greenland, S. &: Brenner, H. Correcting for non-differential misclassification in ecologic analyses. Applied statistics, 42: 117-126 (1993). 3. Greenland, S. &: Robins, J. Ecologic studies: biases, misconceptions, and counterexamples. American journal of epidemiol.ogy 139: 747-759 (1994). 4. Kelsey,J.L, et al. Methods in obseruational epidemiol.ogy. Oxford University Press, New York, 1986. 5. Vandenbroucke, J.P. &: Pardoe(, V.P.A.M. An autopsy of epidemiologic methods: the case of "poppers" in the early epidemic of the acquired immunodeficiency syndrome (AIDS). American journal of epidemiol.ogy, 129: 455-57 (1989). 143 Framework for the development of environmental health indicators Tard Kje/lstroma & Carlos Corvalanb Introduction The term "indicator" has become widely used in documents from international agencies and scien- tific groups in the last few years. One explanation is the active debate on indicators that was spurred by the recommendations in Chapter 40, "Informa- tion for decision-making", of Agenda 21 (1). In this chapter it was stated that, "indicators of sustain- able development need to be developed to provide solid bases for decision-making at all levels and to contribute to a self-regulating sustainability of inte- grated environment and development systems." Countries, international governmental and non- governmental organizations were called upon to develop the concept of indicators of sustainable development. The Statistical Division of the United Nations was given a special role to support this work and to promote the increasing use of such indicators. The World Health Organization (WHO) is con- tributing to the development and promotion of indicators related to the health status of popula- tions, a very important aspect of sustainable devel- opment highlighted in Chapter 6 of Agenda 21, "Protecting and promoting human health". In re- lation to the specific environmental aspects of de- velopment, WHO has a particular interest in what may be called "environmental health indicators" or EHis. This article will analyse the concepts be- hind this term and the potential application of such indicators in environmental health manage- ment field work. The specific characteristic of an environmental health indicator (EHi) is that it somehow provides information about a scientifically based linkage be- tween environment and health. Thus, an indicator which purely describes the state of the environ- ment with no obvious link to the health impacts of the environment could not be considered an EHi. In the same vein, a pure health status indicator with no obvious linkage to environmental causa- tion of health deterioration ( or health improve- ment), could not be considered an EHi. By envi- ronment, as used in this article, we include not only the general environment to which everyone is a Director, Office of Global and Integrated Environmental Health, World Health Organization, Geneva. b Scientist, Office of Global and Integrated Environmental Health, World Health Organization, Geneva. 144 exposed, but also specific environments, such as the workplace or the domestic environment, where people spend a significant proportion of their time. Further, we include among environmental hazards not only the immediate biological, chemi- cal or physical factors that affect health, but also to the underlying social, economic and technical con- ditions that create environmental health problems. Ever since the WHO Programme for the Pro- motion of Environmental Health was established almost 50 years ago, the development of methods and practical applications of the measurement of environmental health status has been an important concern. The initial priority was to provide infor- mation on basic issues of drinking-water, sanitation and shelter. Still, two of the most widely used indi- cators of environmental health status in a commu- nity are the percentages of a population that have access to drinking-water and sanitation. In the 1980s the term "environmental epidemiology" was introduced to identify measurement activities of a more sophisticated nature, incorporating quantification of the linkages between environ- mental exposures and health impacts (2). The first WHO meeting dealing specifically with EHis was held in Dusseldorf in l 992c and since then our work in this area has intensified. This article brings together ideas and analysis inspired by the material and discussion on indicators in conjunction with the HEADLAMP project, described in other arti- cles in this issue. Environmental health indicators in the context of sustainable development Sustainable development has been defined as "development that meets the needs of the present without compromising the ability of future genera- tions to meet their own needs" (3). An indicator that measures sustainability should therefore focus on this definition. It should somehow measure how a component of development meets current "needs", while at the same time indicating to what extent the "needs" of future generations are met. A large number of statistical measures and vari- ables have been listed as potential sustainable- c World Health Organization. WHO consultation on the devel- opment and use of environmental heal.th indicatOTS in the management of environmental risks to human heal.th. Dussel.dorf, 15-18 Deumber, 1992. WHO, 1993 (WHO/EHE/93.3). Rapp. trimest. statist. sanit. mond., 48 (1995) development indicators (SDis) in recent docu- ments from international agencies or scientific groups (4,5),M but most of the proposed indica- tors do not reflect the sustainability aspect. Eco- nomic performance indicators, such as GNP or annual GNP increase, tell us nothing about the ability of future generations to sustain such a GNP or to surpass it. In fact, one could speculate that a high GNP today may be the direct cause of a low- ered GNP tomorrow if natural resources are de- pleted and the high current GNP has been created at the expense of the community's future pro- ductivity. In addition, economic performance in itself can not be the ultimate aim of sustainable development. Human health and welfare, biodi- versity protection and global ecosystem health are the key objectives of sustainable development as concluded in Agenda 21 (1). Most environmental indicators (e.g., air quality) or health indicators (e.g., life expectancy) provide no information about sustainability as such, but they are at least essential elements of the community well-being that is implied in the "meeting of community needs". Some environment and health indicators can be interpreted more directly in relation to sustainability. For instance, an indicator of soil quality and soil stability could be interpreted as directly linked to future agricultural productivity and the ability of future generations to meet their needs. Ideally, this indicator should also incorpo- rate an element of change in soil quality with time. Similarly, an indicator of the occurrence of infec- tious disease in a community could be interpreted in relation to likely health problems in the future, as a high rate of infectious disease is now a founda- tion for future infectious disease occurrence. Static versus dynamic indicators An indicator can be measured as a point estimate at particular time or it can be measured as the change during a time period. One can apply the terms "static" and "dynamic" indicators to the two types. For example, the information contained in the indicators differs if a desertification indica- tor describes the proportion of the area of a coun- try that is classified as desert, or if the indicator describes the annual change in that proportion. The best indicator of sustainability would need to d UNEP /RIVM. An overoiew of environmental indicatim: state of the art and perspectives. 1994 (UNEP/EATR.94-01; RIVM/ 402001001). e Scientific Committee on Problems of the Environment. Environmental indicatim: a systematic approach to measuring and reporting on the environment in the context of sustainabkdevelnpment. Paper presented at the workshop on Indicators of Sustainable Development for Decision-Making, Ghent.January 9-11, 1995. f The World Bank. Monitoring environmental progress (draft). Environment Department, The World Bank, Washington, 1994. Wld hlth statist. quart., 48 (1995) include the dynamic aspect of desertification, while at the same time describing how the current situation fits with the requirement to "meet the community needs". In the field of health indicators, available epide- miological tools provide a number of options for defining static and dynamic indicators. The occur- rence of ill health ( or good health) in a population is basically measured either as the number of exist- ing cases of a disease - "prevalence" - or as the number of new cases of the disease occurring in a set time period, "incidence" (6). If 100 people in a specific population are suffering from a specific disease (prevalence) and 200 new cases (inci- dence) occur in a specific week, the new cases could be linked to changes in the environmental conditions during the week. As the number of cases depend on the size of the population studied, prevalence and incidence are mostly measured as "rates" (e.g., number of cases per 1 OOO popula- tion). Measurements of prevalence or incidence need to be interpreted as static, unless a compar- ison rate ( expected "background" rate or the rate at some earlier time) is available. Explicit dynamic indicators would be the change over time of prevalence rate (e.g., the an- nual percentage increase of the deafness preva- lence rate) or the incidence rate (e.g., change in annual lung cancer mortality rate over a 10-year period). The concept of static and dynamic indicators can be applied to any of the developmental, eco- nomic, environmental or health variables needed to describe and monitor sustainable development. Until now, the majority of indicators proposed for measuring sustainable development have been of the static type. Descriptive versus analytical indicators Another fundamental feature of an indicator is the extent to which it reflects cause-effect relation- ships. In the definition of sustainable develop- ment, a cause-effect linkage between the activities of the current generation and the fate of future generations is clearly expressed. An indicator that merely presents the current state of an environ- mental feature, such as the concentration of ni- trates in drinking-water, provides less information for the assessment of sustainability than an indica- tor measuring the proportion of the nitrate level in drinking-water due to agricultural use of fertilisers. Borrowing from the terminology of epidemiol- ogy one can define indicators as "descriptive" or "analytical", the latter reflecting an exposure- effect relationship. It makes a big difference for decisions concerning environmental health man- agement actions, if the high level of nitrate in drinking-water is related to agricultural run-off rather than being due to natural seepage from soil. In practice, the identification of exposure-effect 145 relationships is mostly based on repeated measure- ments of the environmental and health variables and the analysis of relationships, using epidemio- logical techniques. It is often difficult to encapsu- late interpretable information of this type into one single indicator. Measures of population impact in epidemiolo- gy, such as population attributable risk (PAR), can be understood as one type of "analytical" indica- tors. The PAR is a measure of the excess rate of disease in a population which is attributable to an exposure. Even if the observed association between an exposure and a health outcome is small, if a large proportion of the population is exposed, the total number of affected persons can be large, and of concern for decision-makers. Causal chain for environmental health linkages In the field of environmental health the causal relationships of greatest interest are those that link human exposure to environmental hazards to spe- cific health effects in the exposed population. Lead poisoning is diagnosed by the verification of high lead exposure in the individual and the verifi- cation of specific signs of ill health, such as behav- ioural disorders or anaemia in the same individual. For environmental health management purposes it is important to know what caused the high expo- sure. Did the high lead exposure emanate from food, air, drinking-water or dust inside the house? Did the lead in the food originate from lead in petrol (gasoline) or lead pollution from a factory? The answers to these questions lie in an accurate description of the "causal chain" as described in Box 1. The concept of the causal chain has been wide- ly applied already in the debate about sustainable development indicators, by the use of the frame- work for environmental indicators developed by the Organisation for Economic Co-operation and Development (OECD) (7). The Driving Forces- Pressure-State-Effect framework in Box 1 is a modi- fication of the OECD framework that highlights the health impact component of the causal chain. Within this framework more detailed steps in the causal chain have been shown in the examples in Box 1. This provides a better description of the different items that could be used as descriptive indicators, and the linkages that could be the basis for analytical indicators. The development activity (e.g., industry or agriculture) or the daily activity of a population (e.g., defecation) leads to a certain amount of chemical or biological waste that could pollute the environment. This first step in the caus- al chain (Box 1) could be labelled the "driving forces" behind environmental pressures and change. Population size or density should be con- sidered as one of the major driving forces. The amount of waste produced by the popula- tion and the development activities can, if it is not 146 contained, cause em1ss1ons to the environment (air, water or soil), which can be quantified (col- umns 2 and 3 in Box 1). These emissions are the basis for measured environmental levels of the pol- lutant, but the actual levels depend on dilution in the environmental medium, direction of flow of this medium (e.g., wind direction from a chim- ney), and the persistence of the pollutant in the environmental medium. Only if significant pollut- ant levels occur in the environment near people will human exposure occur, and exposure levels will depend on the time spent in different parts of the environment ( e.g., indoors and outdoors). The actual dose inside the body depends on the pollut- ant actually reaching the target organ, which de- pends on absorption and distribution inside the body ( columns 4 to 6 in Box 1). Health effects can range from early warning signs to clinical disease and finally death ( columns 7 to 9 in Box 1). Intervention to protect human health can be made at each of these steps in the causal chain. Analytical environmental health indi- cators that quantify the impact at each step in the causal chain would be particularly useful. They would highlight where an intervention aimed at protecting human health would be the most effective. Specific versus composite indicators In the transition from raw data to information on which to base environmental health management and decision-making, one can distinguish between indicators which provide information on one spe- cific item ( e.g., one pollutant) or on a range of items of similar characteristics ( e.g., of several pol- lutants combined). Consider air pollution as an example. Short interval measures (say, hourly) ofa given pollutant, such as sulphur dioxide, may be combined to provide a daily ( or other time period) average or other summary statistic. This observed average may then be compared to maximum per- missible levels or be used to estimate the health impact. Such information on a given pollutant can be considered as a specific indicator, which in turn is associated with specific health effects. Following the same air pollution example, one could take the averages of several air pollutants and combine them in a particular way so as to derive a composite indicator. This new indicator would describe, in very general terms, the combined air pollution situation of a town, city or region. It would also be possible to combine all pollutants (say in water, food, soil and air) into an overall pollution index to describe the environmental situation of a partic- ular geographical area. The usefulness of specific indicators or compos- ite indicators depends on their interpretation. A decision-maker may be interested in the overall air pollution level of a city, and this could be provided in a single composite air pollution index. If this Rapp. trimest. statist. sanit. mond., 48 (1995) ~ :::,- § "' iil" ~ ,.... - <> § ";:i. t - . . . to ~ ""' ...... Bo x 1 Ex am ple s of en vir on m en tal h ea lth in dic ato rs w ith in the D PS EA fr am ew or k Dr ivi ng fo rce Pr es su re Ca us al ch ain 1. 2. 3. Ty pe o f Am ou nt or s ize Em iss ion s de ve lop me nt or of pro du cti on Po llu tan t t yp e hu ma n ac tiv itie s Ch em ica l Us e o f le ad a s a Am ou nt of lea d To ns o f le ad (e. g.,l ea d) pe tro l a dd itiv e us ed fo r t his em itte d fro m pu rpo se ca rs Ph ys ica l En erg y Am ou nt of Ca lcu lat ed (e.g ., i on izi ng ge ne rat ion ra dio ac tiv e em iss ion s at ra dia tio n) (nu cle ar re ac to r) m ate ria l u se d nu cle ar fa cil itie s Mi cr o- Se wa ge Am ou nt of w as te Am ou nt of bio log ica l ge ne rat ion pro du ce d un tre at ed (e.g ., w at er eff lue nt co nt am ina - tio n) 4. En vir on me nt al lev els lea d co nc en tra tio n in air Ra dia tio n lev els in air , w ate r, foo d Co lifo rm s in w ate r, foo d Sta te Ef fec ts 5. 6. 7. 8. 9. Hu ma n Do se Ea rly e ffe cts La te eff ec ts De ath d ue to : ex po su re ca lcu lat ed Le ad in b loo d Be ha vio ur al An ae mi a; En ce ph alo pa thy ; pe rso na l dis or de rs; Inc rea se in b loo d Ac ute le ad ex po su re to le ad Re du ce d IQ pre ss ure po iso nin g fro m a ll s ou rc es Ca lcu lat ed Pe rso na l Ch rom os om al Ge ne tic d efe cts ; Ac ute ra dia tio n ex po su res : do sim ete rs; ab no rm ali tie s Le uk ae mi a; sic kn es s; w or ke rs; Ur ine ; Ca nc er Ca nc er ne ar by re sid en ts Fa ec es Es tim ate d Se rum a na lys is Di arr ho ea , f ev er, Ch ole ra, De ath fr om ex po su re to fo r H ep ati tis A na us ea He pa titi s A, de hy dr ati on co nt am ina ted an d typ ho id; typ ho id, foo d/w ate r Fa ec es fo r dy se nte ry, ch ole ra , s hig ell a ga str oe nte riti s index is within pre-determined guideline values, this may be sufficient to decide that no pollution control action is required. If the index is higher than the guideline value, then specific indicators for each pollutant may be used to identify the source and the specific health concerns of the air pollution problem. In the risk-communication and decision-making process it may be difficult to use all the specific indicators. The composite indicator provides a summary view of the environmental health situation, which makes the interpretation easier. Composite indicators can be constructed as weighted averages based on each pollutant's po- tential harmfulness, and they can take account of the characteristics of the population exposed (geo- graphical distribution, age distribution, etc.), in order that the indicator value becomes closely cor- related with the expected health impact. The basis for such a composite indicator is a health risk mod- el. Modelling and risk analysis are developed and applied particularly for evaluation of the cost-effec- tiveness of different pollution control options. Towards a definition of environmental health indicators Indicators are used in different fields to describe the situation in question ( especially geographical distribution), to demonstrate trends, to provide comparisons and to show the extent to which ob- jectives are being achieved. The assumptions in this case are that the indicator in question is valid (i.e., it measures what was meant to measure); it is reliable, meaning that its measurement is not greatly affected by random error; it has wide and representative coverage of the population or area of interest; and if comparisons over time are re- quired, they can be obtained in an on-going and periodic fashion. Many authors have described these and similar specific criteria for selecting or developing indicators (8).d In addition to some of these characteristics, among the most important criteria relevant to environmental health indica- tors (EHis) are that they should be based on estab- lished ( or plausible) associations between environ- ment and health; that they should use existing or easily collected data; that they are easily under- stood by decision-makers and non-specialists; and that specific preventive actions can be guided by them. Environmental health indicators have been de- scribed in the context of HEADLAMP as specific variables which give explicit policy-related infor- mation on the state of, and trends in, environmen- tal health.g It is necessary to develop this descrip- g World Health Organization. Informal consultation on Health and Environment Analysis fur Decision-making (HEADLAMP) methods andfiel,dstudies-Summaryreport. Geneva, WH0, 1994 (WHO/ EHG/94.15). 148 tion further, bearing in mind that, unlike health or environment indicators on their own, EHis have additional complexities. These stem from the fact that an "environmental exposure - health effects" link is assumed, but this link may not hold for all persons, nor groups of persons, at all times. There is a variability in susceptibility among individuals, determined by genetic factors, combined exposure to other pollutants, and occurrence in individuals of diseases that make them susceptible to the envi- ronmental factors included in the indicator. In some situations the main impact of the environ- mental factor may be the worsening of pre-existing ill health, e.g., air pollution triggering asthma at- tacks. Several other working definitions of EHis have been proposed. For example, EHis have been de- scribed as: • information on environment and health which may be used in making decisions and in man- agement for the protection and promotion of human health.c • designed to clarify environmental influences on human health and well-being. The informa- tion is to serve as an aid for decision-making in environmental and health management. This presupposes that a plausible link between envi- ronment and health is established;c • a parameter or value derived from parameters, which points to/provides information about/ describes the state of environment in its rela- tion to human health with a significance ex- tending beyond that obtained directly from the observed properties.h While these definitions have been developed within very specific contexts, they serve the pur- pose of providing an approximation to the concept we aim to grasp. The two key elements in all these definitions are the environment-health link and decision-makers' actions to improve environ- mental health problems. A conceptual framework for environmental health indicators One of the most widely used frameworks for envi- ronmental indicators is the Pressure-State-Response model proposed by the OECD (7). In this frame- work, pressures refer to socio-economic activities and associated processes or products ( e.g., emis- sions), which impact upon the environment. Re- cently, the term driving forces has been used as an alternative to the term pressures. It would seem logical to use the term driving forces for the socio- h Kuchuk, A.A. & Merineau, R. Environmental health indicaton used in the Health and Environment Geographicallnfurmation System (HEGIS). European Centre for Environment and Health, 1994 (WHO/EUR/ICP/CEH 257/6). Rapp. trimest. statist. sanit. mond., 48 (1995) Fig. 1 A conceptual framework for the development and identification of environmental health indicators Cadre theorique pour la mise au point et la selection d'indicateurs de la salubrite de l'environnement 1 Driving force Forces matrices 2 Pressure Pression 5 Actions 3 State Etat economic activities and to use the term pressures for the environmentally harmful products that may emerge from the socio-economic activities (the ex- tent of pressures from a specific driving force can be modified by social and technical interventions). Pressures cause changes in the state of the environ- ment ( e.g., air quality, water quality or the status of natural habitats), leading in tum to the adoption of responses, such as environmental policies or tech- nological innovations, by society. This model does not translate readily into the environmental health context for it omits the im- portant element of the consequences of changes in the state of the environment for health ( e.g., changes in health status or quality of life). As the Environmental Indicators Team from the Environ- mental Protection Agency of the United States of America (USEPA) has suggested, it is appropriate to add an effects component to the framework.i Recognition of effects as a separate component of the framework has consequences for the type of response, that may occur and its aims. The term response may be construed as a passive reaction to the pressures, states and effects related to the envi- ronment. The purpose of EHis is to encourage informed decision-making for targeted actions to protect health. Therefore, we prefer the term ac- tions instead of response. An additional reason is that epidemiologists use the term response to mean the probability of an effect, while toxicologists use the same word to mean the effect itself. The framework for environmental health indi- cators thus becomes driving Jorces-pressure-state- effects-action or DPSEA (Fig. 1). Box 1 lists a few examples of chemical, physical and biological haz- ards in relation to this framework. Within this context, EHis can be exposure- based, health outcome-based or linkage-based. If exposure-based, indicators can be built on, for ex- ample, the percentage of persons exposed to mon- i United States Environmental Protection Agency. A conceptual framework to suppMt the deve!,opmmt and use of environmental infarmation (in preparation). USEPA, 1994. Wld hlth statist. quart., 48 (1995) 4 Effect Ettel .4 itored levels of a pollutant exceeding standard guideline values (e.g., WHO guidelines for maxi- mum recommendable exposure to a contami- nant). If health outcome-based, indicators can fo- cus on specific health outcomes which are sensitive to short term changes ( daily, weekly or monthly) in exposure levels. Linkage-based indicators can be obtained from routine, on-going linkages of health and environmental monitoring data (e.g., time se- ries analysis of air pollution and respiratory condi- tions). Exposure-based indicators have the problem that the closer one gets to the "true" levels of human exposure, the more difficult it becomes to obtain these estimates. Thus, in some cases the question to answer may be how one estimates the adverse health effects given known levels of pollut- ants (e.g., application of dose-response functions to estimate health outcome based only on expo- sure). In other situations, however, the particular demographics and combination of confounding variables may preclude the application of dose- response functions obtained from external popula- tions. In these cases, the critical question may be, given observed health trends, how can one deter- mine if certain environmental pollutants are linked to this trend, and to what extent? Environmental indicators with health linkage Environmental indicators have been described as "a measurement, statistic or value that provides a proximate gauge or evidence of the effects of envi- ronmental management programs or the state or condition of the environment" (9). Issues relating to health are just a few of the many reasons for collecting environmental indicators. Other rea- sons include the impact of environmental pollu- tion on agriculture, forests, rivers and lakes. Thus, the collection of data on air pollution emissions and concentrations, organic and inorganic water pollution, stratospheric ozone, natural resources, waste production, climate change, etc., is not per- formed specifically for health-related purposes and this makes the available information of limited val- ue in the development of EHis. In the context of 149 8012 Environmental contaminants with potential human health impact Substance Class• Indicators and medium Air quality indicators Sulphur dioxide Nitrogen dioxide Particulates Ozone Carbon monoxide 2 Water quality indicators Drinking-water quality Multi-media and other indicators Volatile organic 2 compounds (VOCs) Polyaromatic 2 Hydrocarbons (PAHs) Metals and trace elements 2 Persistent organic 3 chemicals Pesticides 1 2 3 Nitrates, etc. 2 Pathogens and allergens 1 3 Radiation 1 1 2 150 Concentration in air Concentration in air Total suspended particulates (TSP); particulate matter in the respirable size range (less than 1 Oµm, PM10); Concentration in air Concentration in air Concentration in air Hardness; Water colour; Taste; Acid level (pH); Conductivity/TSS; Biochemical oxygen demand(BOD); Volatile organic compounds(VOC); Total organic compounds(TOC); Nitrates, nitrites; Phosphates Concentration of specific VOCs in air and water Concentration of benzo(a)pyrene in air and food Concentration of cadmium (Cd), lead (Pb), arsenic (As) and mercury (Hg) in human tissue; Concentration of aluminium (Al) in drinking-water Concentration of polychlorinated biphenyls (PCBs). dioxins, etc. in human tissue Concentration in food; Concentration in soil, water; Concentration in human tissue Concentration of nitrate, nitrite, phosphate, etc .• in surface water; Concentration in groundwater, food Foodborne pathogens; Waterborne pathogens; Airborne allergens (e.g., pollen); Indoor allergens Activity of radon in household air; Solar radiation; Radiation equivalent of food Proxy/surrogateh Exceeding WHO or national guidelines; Emissions; Use of coal for domestic heating/cooking Exceeding WHO or national guidelines; Emissions; Use of gas for domestic heating/cooking; Traffic density Exeeding WHO or national guidelines; Black smoke; Emissions of TSP; Use of coal Emissions; Traffic density, city gas usage Water treatment Emissions; Petrol usage Small-scale wood and coal burning; Traffic density Concentration in air, water, soil, food; emissions Concentration in air, water, food; Emissions; Production/consumption Pesticides use; Sales; Land use Fertilizer usage; Additive use Concentration; Food hygiene; Land use/vegetation; Water treatment; Wastewater treatment; Humidity; Housing quality; Geology; Sunshine/cloudiness Rapp. trimest. statist. sanit. mond., 48 (1995) Box 2 (continued) Substance Exposure to tobacco smoke Class• Indicators and medium 3 Cotinine in urine Proxy/surrogateb Particle concentration in indoor air; Mutagenicity of air; Tobacco consumption; Smoking controls in public buildings, etc. Nuisances 3 Nuisance caused by odours; Complaints; Waste treatment; Complaints; 3 Noise levels in home; 3 Traffic noise Noise emissions; Traffic density a Priority classes: Class 1: descriptors that are highly relevant to health, for which harmonized definitions and measurement methods have been developed, and for which data collection already exists in most of developed and many developing countries; Class 2. descriptors that are relevant to health but for which harmonized definitions and methods are not yet available, and/or for which only limited data exist; Class 3: descriptors that are potentially relevant to health and/or of public concern, but for which no harmonized techniques and definitions yet exist. and for which measurement is either technically difficult or expensive. b Many of the preferred indicators are not amenable to direct measurement (at sufficient coverage), hence proxies and surrogates were defined where feasible. In each case the specific procedure for converting descriptors to operational indicators must still be outlined in detail. Source: Ref. (17). EHis we are concerned with the degree of expo- sure to human beings, and the human health im- pact of such exposure. Environmental pollution without direct or potential human exposure, is by definition not covered by EHls. Indicators of exposure are those which measure the potential of a substance or microbiological or- ganism to enter the human body through contam- inated air, water, food and soil. Examples of con- taminants with plausible or known links with hu- man health impact are listed in Box 2, based on a report of a WHO consultationj The difficulty with environmental indicators is that the presence of pollutants in the environment does not translate automatically into health out- comes. Similarly, the incidence of many environ- mentally related diseases cannot be easily traced back to specific environmental exposures. Only individual-level epidemiological studies are able to establish reliable links between exposures and health outcomes. Such studies, however, defeat the purpose of using easily collected or available statis- tics from which to derive the relevant indicators. Certain ecological methods in epidemiology, for example time series analyses, are able to make group-level linkages using existing aggregate level data. We call this method of obtaining an EHi "linkage-based indicators". Other ecological methods, such as geographical linkages, or joint health and environmental data analysis using Geo- graphical Information System (GIS) techniques, may also help to point out trends or associations which may require further scrutiny. i World Health Organization. Environment and health indicators for use with a health and environment geographical information system (HEGIS)forEurope. &port on a WHO consultation. Bilthoven 11-13 March, 1993. WHO, 1993 (WHO/EUR/ICP/CEH 246). Wld hlth statist. quart., 48 (1995) Health indicators with environmental linkage Health indicators have been used extensively to monitor the health of populations. The "Health for All" policy involves monitoring of progress to- wards attaining a minimum health level for all per- sons by the year 2000 and provides numerous ex- amples of health indicators monitored on a global scale. According to the third progress report on Health for All, "monitoring is the continuous fol- low-up of activities to ensure that they are proceed- ing according to plan, so that if anything goes wrong, immediate corrective measures can be tak- en. The information gained from monitoring is used for evaluation. Evaluation is the systematic assessment of the relevance, adequacy, effective- ness and impact ofa health programme".k The health and environment link is also a prominent part of the Health for All policy. Impor- tant environmental health issues such as access to water and sanitation, acute and chronic exposures to chemicals, population exposed to unacceptable levels of contaminated air, housing issues (and also environmental issues with a less direct link to health, such as loss of biodiversity, deforestation, soil degradation and global warming), are dis- cussed in the lmpl,ementation of the gwbal strategy for Health for All by the Year 2000 (JO). The Swedish environmental protection agency has compiled a tentative list of environment-relat- ed diseases (11). This list includes certain cancers ( especially lung and skin, particularly in children); respiratory disease ( chronic bronchitis, pulmonary emphysema, bronchial asthma, hyper-reactivity); allergic diseases (atopic allergies and symptoms occurring in connection with atopic diseases, k World Health Organization.Implementation of strategies for health for all uy the year 2000. Third monitoring of progress. Common framework. Geneva, WHO, 1993 (WHO/HST/GSP/93.3). 151 8013 Potential signs of population exposure to environmental contaminants A. Diseases/health problems identifiable through existing health reporting systems (and potential sources of information) Low birth weight Birth defects Spontaneous abortions Chronic respiratory disease in children Active leukaemia in children Acute granulocytic leukaemia in adults Aplastic anaemia Asthma in children Dermatitis and dermatoses Skin cancer Malignant melanoma Lung cancer in nonsmokers Bladder cancer in nonsmokers Primary liver cancer in nondrinkers Vital statistics Vital statistics, hospital discharges, birth defect registries Hospital discharges Hospital discharges Cancer registries, vital statistics, hospital discharges Cancer registries, vital statistics, hospital discharges Hospital discharges, vital statistics Hospital discharges Hospital discharges Cancer registries, hospital discharges Cancer registries, hospital discharges, vital statistics Cancer registries Cancer registries Cancer registries B. Disease/defects not usually identifiable through existing health reporting systems Acute sensory irritation (eye, respiratory, olfactory) Developmental defects Hearing loss in children Chromosome defects C. Deviation from normal biological functions requiring special surveys to detect Neurological function Immunological function Renal function Cardiac function Haematologic function Respiratory function Reproductive function Liver function Auditory function D. Indicators of body burdens potentially due to environmental exposures Blood lead (ZPT)a Heavy metals in blood, urine, hair, nails Carboxyhemoglobin Organophosphates ( cholinesterase )b PCBs and PBBs (polychlorinated and polybrominated biphenyls) Other pesticides Adductsc a ZPT, zinc protoporphyrin, an easily measured metabolite involved with the structural materials of haemoglobin. Lead impairs the use of ZPT and hence elevated ZPT in screening tests indicates a likelihood of a lead body burden. b Cholinesterase is an enzyme that is specifically blocked by organophosphate pesticides, and a decrease in cholinesterase may reflect a relatively recent exposure to such pesticides. c Adducts are combinations of pollutants with one or more molecules in the body, such as DNA or haemoglobin, which tend to persist and can be detected at very low concentrations. Source: Ref. (22). namely asthma, hay fever, conjunctival catarrh and eczema); cardiovascular disease; effects on repro- duction (miscarriage, late intrauterine death, neo- natal and perinatal death, low birth weight, various malformations and chromosome abnormalities); and diseases of the nervous system ( organic psy- chosyndromes and dementia - Alzheimer's dis- ease, Parkinson 's disease, amyotrophic lateral 152 sclerosis, multiple sclerosis, peripheral nervous dis- ease in combination with polyneuropathy). Not all cases of these diseases are due to environmental exposures. The term "sentinel health event" has been applied to cases of disease that in a particular situation appears out of the ordinary, and can be potentially linked to an external factor. A sentinel health event serves as a warning signal that the Rapp. trimest. statist. sanit. mond., 48 (1995) quality of preventive or medical care may need to be improved. Examples of sentinel health events include infant or maternal deaths as indicators of the adequacy or quality of prenatal or maternal health care. The concept of sentinel health events has been adapted for use in occupational health. Currently more than 50 conditions are considered as "sentinel health events (occupational)". These include, for example, asbestosis and mesothelio- mas (as indicators of asbestos exposure), silicosis, heavy metal poisoning, leukaemia (as an indicator of exposure to ionizing radiation or benzene), methaemoglobinaemia, extrinsic asthma and pesti- cide poisoning ( 12). A preliminary list of environ- mentally-related sentinel health events has also been devised ( 13). These include conditions that are clearly identifiable; those that potentially indicate exposure to environmental contamina- tion; and those which are indicators of body bur- dens, potentially due to environmental exposures (Box 3). Conclusions Environmental Health Indicators presuppose a link between specific environmental exposures and health outcomes. This makes the identifica- tion of EHis quite complex. While only epidemio- logical studies with data collected at the individual level can establish sound environmental health as- sociations in a given place and time period, sim- plicity and cost-effectiveness call for the use of population-level environmental or health data to derive EHis from known environment-health rela- tionships. If reliable environmental data are avail- able, these are simpler to use than health outcome information, and they can be converted into reli- able EHis if appropriate estimates are obtained of the potential population exposed. Once identi- fied, routine or periodic EHI data collection can provide useful information for decision-makers as targeting and evaluation tools in their efforts to protect human health. The framework for environmental health indi- cators proposed in this article is intended to aid decision-makers in exploring specific environmen- tal health problems through the causal chain from source to health effect, and thus identifying areas where action is most efficient. Implicit in the con- cept is that actions should be taken at every level, that is, on developing policies to avoid the driving forces behind environmental health hazards, on re- ducing environmental health pressures in the form of dangerous products and emissions, on repairing the state of the environmental health situation, and on rectifying health effects if they occur. There is a clear need for further research in the area of EHI development and also in identifying more sensitive and selective markers of the health effects of environmental exposures. However, for Wld hlth statist. quart., 48 (1995) many existing problems, the proposed framework should be a useful tool to describe environmental health problems at a specific place and time. With this information, decision-makers can formulate actions to improve environmental health problems at different points in the causal chain. Summary Environmental health indicators provide information about scientifically-based linkages between environ- ment and health. This information can be used for environmental health management and decision- making. Environmental health indicators are rendered more complex than either environmental indicators or health indicators because they must take account of factors such as the variability in susceptibility in individ- uals and variability in co-exposures. Such variability implies that any links that are defined may not apply to all individuals or groups at all times. Individual-level epidemiological studies can contribute to establishing environmental health relationships for particular places and time periods. However, cost-efficiency demands that aggregated data and known environment and health relationships be used to derive these indicators. Environmental health indicators can therefore be con- structed by linking aggregated data, or by identifying environmental indicators with a health linkage, or health indicators with an environmental linkage. The framework for environmental health indicators pro- posed here is an adaptation of the Pressure-State- Response framework. Its first level consists of driving forces, which create pressures on the environment. These in turn alter the state of the environment by increasing existing exposures or introducing new ones, which produces a measurable health effect. In order to rectify the problem, actions (i.e., environmental health management) must be undertaken at each level. Thus the framework becomes the Driving-force - Pressure - State - Effects - Action (DPSEA). Resume Schema pour la mise au point d'indicateurs de la salubrite de l'environnement Les indicateurs de la salubrite de l'environnement four- nissent des renseignements sur les liaisons scientifi- quement etablies entre l'environnement et la sante. Ces renseignements sont utiles pour controler !'hygiene de l'environnement et prendre des decisions. Les indica- teurs de la salubrite de l'environnement sont plus com- plexes que ceux portant uniquement sur l'environne- ment ou la sante, car ils tiennent compte de facteurs tels que les ecarts de sensibilite entre les individus ou groupes a tout moment. Des etudes epidemiologiques realisees au niveau individuel peuvent permettre d'eta- blir des liaisons entre l'environnement et la sante en un lieu precis et pendant une periode donnee. Cependant, pour des considerations de rentabilite, ces indicateurs sont calcules a partir de donnees groupees et refletant des liaisons connues entre l'environnement et la sante. 153 On peut done mettre au point des indicateurs de la salubrite de l'environnement soit en raccordant des donnees groupees, soit en identifiant des indicateurs environnementaux en rapport avec la sante ou inverse- ment. L'orientation fournie ici sur les indicateurs de la salubrite de l'environnement s'inspire du schema pression-etat- reaction propose par l'OCDE. Au depart, des forces matrices exercent des pressions sur l'environnement. Celles-ci modifient l'etat de l'environnement en intensi- fiant les expositions existantes ou en engendrant de nouveaux risques, qui produisent un effet mesurable sur la sante. Pour remedier ace probleme, ii taut prendre des mesures d'hygiene de l'environnement a chaque niveau. Le schema est alors le suivant: forces motrices- pressions-etat-effets-action. References/References 1. Agenda 21: Programme of Action for Sustainahl.e Devewpment. New York, United Nations, 1993. 2. Environmental Health Criteria 27. Guidelines on studies in environmental epidemiology. Geneva, World Health Orga- nization, 1983. 3. The World Commission on Environment and Development. Our common.future. Oxford University Press, Oxford, 1987. 154 4. Bartelrnus P. Towards a framework for indicators of sustainabl.e devewpment. Working paper series 7, Department for Economic and Social Information and Policy Analysis. Doc. No. ST/ESA/1994/WP.7. United Nations, New York, 1994. 5. World Wide Fund For Nature (WWF) and The New Economics Foundation. Indicators for sustainabl.e devewpment. London, 1994. 6. Beaglehole, R. et al. Basic epidemiology. World Health Organization, Geneva, 1993. 7. Organization for Economic Cooperation and Development. OECD core set of indicators for environmental performance reviews. Environmental monograph No. 83, Paris, 1993. 8. Kreisel, W.E. Representation of the environmental quality profile of a metropolitan area. Environmental monitoring and assessment4:15-33 (1984). 9. United States Environmental Protection Agency. Terms of Environment. Doc. No. EPA l 75-B-93-001. 1993. 10. World Health Organization. Impl.ementation of the global strategy for health for all by the year 2000, Second evaluation-Eighth reprnt on the wor/,d health situation. Geneva, World Health Organization, 1993. ll. Swedish Environmental Protection Agency (SEPA). Environment and public health. An epidemiological research programme. Solna, Swedish Environmental Protection Agency, 1993. 12. Mullan, R.J. &: Murthy, L.I. Occupational sentinel health events: An up-dated list for physician recognition and public healthsurveillance.Americanjoumalofindustrialmedicine, 19: 775-799 (1991). 13. Rothwell, C.J. etal. Identification of sentinel health events as indicators of environmental contamination. Environmental health perspectives, 94: 261-263 ( 1991). Rapp. trimest. statist. sanit. mond., 48 (1995) Developing indicators for environment and health John T. Willsa & David J. Briggsb The need for information Reliable, valid, up-to-date information and statis- tics are essential for supporting policy, monitoring the effectiveness of management, directing re- search and informing the general public (1). This was confirmed by the Rio de Janiero Earth Summit in 1992: "The need for information arises at all levels, from that of senior decision-makers at the national and international levels to the grass-roots and individual levels" (2). This need for informa- tion is especially acute in assessing the relationship between the environment and health; the environ- ment affects health in numerous ways, with both beneficial and detrimental effects. The environ- mentally-related aspects of health can therefore be protected through the inclusion of health consid- erations in environmental policies. To date, how- ever, health considerations have been very poorly represented in environmental policy. For example, the fifth European Community environmental ac- tion programme, launched in 1992, concentrated on target sectors (e.g., industry, energy, and trans- port) and themes ( e.g., waste management and the urban environment). While health was given some consideration, priority was given to general envi- ronmental issues. The same is also true of many national environmental policies; for example in the United Kingdom (3) and Italy (4). Conversely, several recent international policy developments encourage the integration of environmental and health concerns. Principle 1 of the Rio Declaration on Environment and Development, for example, states that "Human beings are at the centre of concerns for sustainable development. They are entitled to a healthy and productive life in harmo- ny with nature" (2). In addition, the World Health Organization (WHO) Global Strategy for Environ- ment and Health identifies three key objectives for environment and health; • achieving a sustainable basis for Health For All; • providing an environment that promotes health; and • making all individuals and organisations aware of their responsibility for health and its environ- mental basis. a Researcher, Institute of Environmental and Policy Analysis, University of Huddersfield, Huddersfield, United Kingdom. b Professor and Director, Institute of Environmental and Policy Analysis, University of Huddersfield, Huddersfield, United Kingdom. Wld hlth statist. quart., 48 (1995) Decision-makers at the international, national and local levels therefore need information at the appropriate scale to monitor and survey the cur- rent environment and health situation and to eval- uate progress towards achieving policy targets. Indicators Indicators are one means of providing policy- makers with information; representing the end product of a lengthy information chain. Measure- ments produce raw data; the combination and publication of data produces statistics; statistics are translated and applied, creating indicators. Indica- tors have an added significance as compared to the underlying statistics and are tied to a specific pur- pose (5). The word indicator is derived from the Latin verb indicare, meaning to point out, indicate or announce. The use of indicators is well-established in many fields; for example, in the areas of poverty, depriva- tion and social policy (6, 7), ecology and nature conservation (8, 9), and in economics. Gross Do- mestic Product (GDP), has long been used as the primary measure for determining wealth and eco- nomic development nationally and internationally ( 10). Other economic indicators include the Retail Price Index (RPI), the Gross National Product (GNP) and the unemployment rate. Social indica- tors include direct measures, such as educational expenditure per student or the percentage of households living below the poverty line (11, 12) and indirect measures, such as composite depriva- tion indices (7). In all of these areas, indicators provide an important contribution to policy for- mulation and the decision-making process. They can be used to fulfil many functions, including; promoting specific policy issues, providing objec- tive baseline information, demonstrating spatial and temporal variations and monitoring the effec- tiveness of policy actions (1). The development of environmental indicators is currently ongoing at local, national and interna- tional levels, with the involvement of many govern- mental, non-governmental and international agen- cies - from individual community groups con- cerned about their local environment to interna- tional agencies responsible for evaluating national environmental performance. The Organisation for Economic Cooperation and Development (OECD), for example, was given a mandate by the G-7 Economic Summit in Paris, 1989 to "examine 155 how selected environmental indicators could be developed" (13). The OECD has chosen to adopt the "Pressure-State-Response" framework, based on the premise that "human activities exert pres- sures on the environment and change its quality and the quantity of natural resources (state). Soci- ety responds to these changes through environ- mental, general economic and sectoral policies ( re- sponse)" ( 14). Individual indicators have been devel- oped to reflect environmental pressures, the state of the environment and societal responses. A number of indicator selection criteria have also been identi- fied which embody the main requirements for en- suring indicator effectiveness. In each of 3 main categories, indicators should: Policy rel.evanu and utility for users - provide a representative picture of environ- mental conditions, pressures on the environ- ment and society's responses; be simple, easy to interpret and show trends over time; - be responsive to changes in the environment and related human activities; - provide a basis for international comparisons; - be national in scope or applicable to regional issues of national significance; - have a target or threshold against which it can be measured; Analytical soundness - be theoretically well-founded in technical and scientific terms; be based on international standards and con- sensus about its validity; - be capable of linkage with information systems, economic models and forecasting; Measurability be based on data which are available, or readily available at an acceptable cost/benefit ratio; be based on adequately documented data of a known quality; and be based on data which are reliably updated at regular intervals (14). It should, however, be recognized that these criteria represent the ideal for indicator selection; in reality, it is unlikely that indicators can be found which satisfy all of the criteria, particularly in the short-term. In addition, these criteria reflect to a large degree the purpose for which the OECD indicators are intended; namely, to highlight na- tional environmental issues and evaluate national environmental performance. Indicators developed for alternative purposes will therefore require dif.. ferent selection criteria. At the national level, indicator projects are un- derway in Canada and the Netherlands. In the Netherlands, Environmental Policy Performance Indicators are being developed to reflect the issues identified by the National Environment Policy Plan 156 (NEPP) in 1989 and the NEPP-Plus in 1990 (15). Theme indicators (e.g., ozone depletion and eu- trophication) reflect key issues and target group indicators relate to sectors of socio-economic activ- ity (e.g., agriculture and industry). Targets out- lined in the policy plans accompany the relevant indicators to illustrate trends and demonstrate the effectiveness of environmental policy. To provide decision-makers and the general public with dear, concise and unambiguous information, the indica- tors are presented in a highly aggregated format. For example, the indicator for "eutrophication of the environment" comprises combined annual emissions of phosphorus (P) and nitrogen (N) with a weighting factor of 1 for phosphorus and 0.1 for nitrogen (16). Canada is in the process of developing indica- tors which provide a comprehensive picture of the state of the environment and indicate trends to- wards the environmental goals of sustainable devel- opment. A preliminary set of indicators divided predominantly by environmental media, was pub- lished in 1991, with individual indicators reflecting environmental stresses and the state of the environ- ment for particular policy issues ( e.g., climate change, ozone depletion and freshwater quality) ( 17). Since 1992, a series of indicator bulletins have been published which present key indicators for particular environmental issues and illustrate the complexities and cyclical nature of environmental systems. For example, the indicator bulletin for stratospheric ozone depletion describes the envi- ronmental route of ozone-depleting substances and presents data for three indicators: "domestic supply of ozone-depleting substances"; "global at- mospheric concentrations ofCFC-11 and CFC-12"; and "stratospheric ozone levels over Canada" ( 18). The concept of environmental indicators has similarly been widely adopted and applied at the local level. Examples include the Jacksonville "Quality-of-Life Project", Jacksonville, Florida, United States of America (11); the Local Govern- ment Management Board (LGMB) "Sustainabili- ty Indicators Project" in the United Kingdom (19); and the "Sustainable Seattle Indicators of a Sus- tainable Community" project, Seattle, Washington, United States (20). These projects are based on local sustainable development issues and attempts to assess quality of life, individual indicators being divided by issue rather than by environmental me- dia. While not being based solely on environmen- tal issues, they do include several environmental indicators. For example, indicators used in the Jacksonville project include "river compliance with dissolved oxygen standards" and "tons of sol- id waste per capita" (11), while indicators used in the Sustainable Seattle project include the "num- ber of days per year air quality fails to meet air quality standards" and ''wild salmon runs through local streams" (20). Rapp. trimest. statist. sanit. mond., 48 (1995) Chart 1 The exposure chain and the relationship between exposure (stress) and health outcome -p 0 I i i---,., c y p r i O i--,. r i t y a r~ e a s - I Industry I E N v I R 0 N M E N T A L D 0 M A I N H D E O AM L A T I H N I Transport I STRESS Sectors I I I Energy Agriculture Waste I I I Source activity \ii • Discharge/Emission • . I Environmental media I I + ~ Ambient concentration • • • I Routes of exposure I + + + Human exposure \ii -Individual dose + Health outcome OUTCOME Indicators for environment and health I Households I While considerable progress has been made in the development of environmental indicators, the use of indicators to demonstrate relationships between environment and health is much less well devel- oped. The question of what constitutes an effective environment-health indicator can therefore be posed. In attempting to answer this question, a useful starting point is the principle that environ- mental impacts on human health operate primari- ly through physical exposure to various pollutants ( though also through exposure to other environ- mental hazards and less tangible psychological ex- posures from li\ling and working environments). The link between environment and health is thus provided by the process of exposure. Exposure itself, however, is the product of a relatively long causal chain. As Adriaanse ( 16) has stated: " ... sources or actors cause emissions to the envi- ronment, leading to certain concentrations in the living environment via dilution, dispersion and conversion reactions. Receptors (man, flora, fau- na) are exposed to these concentrations, with harmful effects". The conceptual model in Chart 1 illustrates the exposure chain in more detail and Wld hlth statist. quart., 48 (1995) Relation to Ease of Level of health outcome collection/ control compilation Indirect Simple Controllable I Natural environment I , , ,, , , More direct Complex Uncontrollable WH095356 describes the relationship between exposure (stress) and health outcome (outcome). While the majority of these exposures result from human activity, it must be recognized that some are attrib- utable to "natural", or background sources; such as the emission of the carcinogenic gas radon from particular geological formations. Based on the above model, two categories of indicators can be defined; • Health-re/,ated environmental indicators (HREI's): definable environmental conditions or trends which suggest potential health effects; and • Environment-re/,ated health indicators (ERHI's) health outcomes which suggest an environmen- tal cause, or a contribution from environmental factors. The remainder of this article is devoted to health-related environmental indicators. Developing health-related environmental indicators (HREl's) Based on the above model, the ideal HREI may seem to be exposure. Exposure, being the last link in the environmental chain, is most closely related to individual dose and health outcome. While the 157 connection between some exposures and health effects is epidemiologically well-established, the links are rarely unitary or straightforward (21); dif- ferent individuals experiencing similar exposures may exhibit a wide variety of health outcomes due to numerous factors, including variations in indi- vidual susceptibility, medium of exposure and age. In addition, different social and environmental conditions may act in concert to affect health, while single factors may contribute to a wide variety of possibly contradictory health outcomes. Health- related environmental indicators are therefore, by necessity limited to calculating exposure and can- not be used accurately to predict health outcome. To ensure their effectiveness, HREI's should also meet certain selection criteria, although it should be noted that these represent the ideal situation and their relative importance has yet to be evaluated. Inter alia, indicators should: R.ileuance • be based on environmental conditions which are amenable to change; • be based on epidemiological relationships be- tween environment and health; • be based on definable health-related environ- mental issues; Objectivity • be reliable, consistent and objective; • be sensitive or responsive to changes in environ- mental conditions; • be scientifically valid, i.e., indicate what they purport to indicate; • provide a representative picture of health- related environmental exposure; Data • show trends over time through the use of retro- spective data; • be based on data which is available at an accept- able cost/benefit ratio; and • be based on adequately documented data of a known quality. The principal problem with using exposure as an HREI is the lack of available data. Measures of exposure are extremely difficult to obtain and are only available for a small number of substances in a limited number of geographic areas; for example, where individual exposure to dioxins has been monitored as part of an epidemiological study to evaluate the relationship between dioxins and con- genital malformations (22-24). It therefore be- comes necessary to infer exposure by indirect methods -for example, from data on industrial ac- tivity in certain sectors, from emissions and from levels of pollution in the ambient environment. For example, for the issue of nitrogen dioxide pol- lution, the ideal exposure indicator is the amount of N02 that an individual is exposed to during the 158 day- at home, en route to work and at work. While this information can be obtained by personal mon- itoring, it is both difficult and costly to obtain. The following "proxy" indicators for exposure could therefore be used: source activity: number of vehicles emissions: number, type and speed of vehicles (predicted amount of N02) concentration: amount of N02 present at monitoring sites (actual amount ofN02). Proxy indicators, however, must also comply with the selection criteria identified above. Three factors are of specific importance. 1. Firstly, selection of suitable proxies depends upon the existence of strong relationships between the proxy and exposure. In reality, however, these relationships are often weak and are diminished by the complexity of the processes involved: by the effects of confounding, and by poor data quality. The further removed from exposure the proxy is, the weaker it is likely to be as a reliable indicator of exposure and hence of health effect. Total expo- sure, for example, is a product of many different exposure events at different times and different locations. Simple measures of pollution levels (e.g., average annual concentration) at a broad geographic scale, or at monitoring stations far re- moved from the area of concern, are unlikely to provide a reliable indication of exposure. Indica- tors from yet further up the exposure chain, such as emission levels or industrial activity, are likely to be weaker still. To assess the reliability of a proxy indicator, it is therefore necessary to test the strength of its correlation with the relevant envi- ronmental exposure. 2. On the other hand, indicators from earlier in the exposure sequence reflect processes which can be more readily controlled by policy. While little can be done to reduce pollution levels once they have been emitted and dispersed in the envi- ronment, relatively rigorous control is feasible at source; for example by regulating emission levels or particular industrial processes. As a result, indi- cators earlier in the causal chain tend to be more immediately relevant to policy. 3. At the same time, data are more readily available for indicators higher in the exposure chain. As has been noted, direct measurements of exposure are rare. Data on pollution levels in dif- ferent environmental media are more abundant, but still limited in geographic scope, range of pol- lutants and quality (1). Data on emissions are also widely available. Many countries now maintain na- tional emissions inventories, while a number of international inventories also exist or are under development (25). Relatively few emissions inven- tories, however, provide a complete picture of emissions via all pathways and from all sources: the Rapp. trimest. statist. sanit. mond., 48 (1995) Box 1 Indicator survey Indicator source Title Number proposed Number in use Data available Canada: Environment Canada National indicators project: 43 Yes preliminary indicators 1991 Canada: Environment Canada National indicators project: 16 Yes indicator bulletins 1992-94 Costa Rica: Edgar E. Gutierrez-Espeleta: 55 Approximated Sustainability Index problem, issue and sustainability indicators 1994 Dirgha Tiwari: Country and project indicators 38 Asian Institute of Technology EU ROST AT Eurostat pressure index project: 50 illustrative indicators 1994 Netherlands: Environmental Policy Theme and target group indicators 14 Yes Performance Indicators 1993 Nigeria: Odemerho & Chokor Aggregate index of environmental 68 quality 1991 Norway: Environment Ministry Norway environmental quality 28 indicators project 1992 OECD Preliminary environment indicators 25 Yes 1991 OECD Environmental performance review 98 indicators 1993 United Kingdom: Green Gauge: indicators for the 46 10 Yes Environment Challenge United Kingdom environment 1994 United Kingdom: A pilot environmental index 26 9 Hope/Parker Pilot Environment Index for the United Kingdom 1991 United Kingdom: Proposed positive indicators 1983 37 J. Catford, Wessex Regional Health Authority United Kingdom: Health-related indicators for a 198 Lancaster County Council sustainable community project 1994 United Kingdom: Leeds quantifiable city project 52 Leeds Quantifiable City proposed input, output & process indicators 1994 United Kingdom: Local sustainability indicators 102 Local Government Management project 1994 Board United Kingdom: Index of social deprivation 83 Townsend deprivation index United Kingdom: Measuring sustainability: 30 United Nations Association potential core indicators United Kingdom: Suggested priority environmental 16 V. Anderson: Alternative indicators 1991 Economic Indicators United Kingdom: Action for indicators: national 75 WWF/NEF Indicators Initiative performance and sectoral indicators 1994 United States: Quality of life indicators 1983-92 78 Yes Jacksonville Quality of Life Project United States: Draft indicators of a sustainable 40 20 Yes Sustainable Seattle 1993 community 1993 United States: Proposed environmental indicators 26 World Resources Institute for national reporting 1994 Wld hlth statist. quart., 48 (1995) 159 Box 1 (continued) Indicator source Title Number proposed Number in use Data available WHO: Health For All WHO Health For All statistical 27 Yes indicators ( environmentally-related indicators only) WHO: Multi-City Action Plan WHO healthy city indicators 39 WHO: HEGIS WHO consultation on developing 62 health and environmental indicators 1993 Total number of indicators majority are concerned only with emissions to the air, and many relate only to industrial point sourc- es. The reliability of many emissions estimates is therefore open to doubt, due in part to deficien- cies in the input data and the emission models used (25). Data on source activities may also prove unreliable. Identical processes in different loca- tions may produce different emissions over time due, for example, to variations in climatic condi- tions and raw material quality. Different processes for identical source activities are also likely to pro- duce different emissions. The utility of existing indicators With the large number of environmental indica- tors currently under development, there is a need to examine their utility for assessing environmental and health relationships; namely, whether they ful- fil the criteria for health-related environmental in- dicators and if so, what data is available. A survey of 26 international, national, regional and local indicator projects from around the world yielded a total of over 1 400 indicators. As Box 1 illustrates, the majority of these are at the proposal stage, while relatively few are actually in use. Out of a total of 1 411 indicators, 960 have been pro- posed, while 451 are in use. It should be noted, however, that there is considerable duplication of individual indicators between different projects. For example, 21 of the 26 indicator sets included S02 and N02 as indicators, although the definition varied between different sets - from emissions, to ambient concentration, to the number of times air quality guidelines are exceeded. Thus the actual number of issues and substances covered by the various projects is smaller than initially apparent from the number of indicators. However, while many indicators may cover the same issues and substances, the use of different definitions and de- nominators leads to non-comparability between in- dicators. For the effective use of non-comparable indicators, it is therefore necessary to refer back up the information chain to the original data on which the indicators are based. Of the 451 indicators in use, data were available for only 233. Some indicator projects are in the 160 960 451 233 early stages of implementation and data have yet to be collected; for example, the United King- dom's Local Government Management Board indicators. Other projects, such as Sustainable Se- attle in the United States and Environment Chal- lenge in the United Kingdom, have chosen to adopt a limited number of indicators from a larger range of proposed indicators. Finally, details of two indicators sets were obtained from academic pa- pers (Odemerho & Chokor's index in Nigeria (26) and the Hope/Parker index in the United Kingdom (27)) where insufficient supporting data were given. The 233 indicators (relating to 8 projects), for which data were available, are analysed in Box 2. The indicators were initially assessed for their health relevance; namely, whether they relate spe- cifically to environmental and health problems. Predictably, it was found that indicators developed for primarily environmental purposes (e.g., the OECD indicators), or as quality of life measures (e.g., the Jacksonville indicators), corresponded only partially with environmental and health is- sues. Next, the indicator coverage and level of data aggregation were examined. Local level data were available for two of the projects (Seattle and Jack- sonville), while national level data were available for the remaining 6; Canada, the Netherlands, the United Kingdom, selected OECD countries and member countries of the World Health Organiza- tion's European Region. Data for the indicators were presented for a range of years: 4 of the projects covered the 1970s to 1990s and the re- maining 4 covering the 1980s to 1990s. Of the indicators themselves, only 55% were found to correspond to the stress-outcome frame- work, 45% being in the "other indicators" catego- ry. In part, this reflects the intended purpose of the indicator projects. For example, the Jacksonville indicators are intended primarily to provide gener- al information to local decision-makers and mem- bers of the public; 91 % of the indicators relate to non-environmental issues such as education, the economy, public safety and the social environ- ment, while only 9% relate to the environment Rapp. trimest. statist. sanit. mond., 48 (1995) ~ ~ s "' ii? ~- ,... . Q Iii ?- it ....... fg ~ ~ ~ Bo x 2 An aly sis o f 2 33 in dic ato rs alr ea dy in u se Ind ica tor se ts He alt h-r ela ted ? Ca na da : I nd ica tor b ull eti ns Ye s Ca na da : P re lim ina ry ind ica tor s Ye s Ne the rla nd s: En vir on me nta l P oli cy Pe rfo rm an ce In dic ato rs Ye s OE CD : P re lim ina ry ind ica tor s Pa rtly Un ite d Ki ng do m: E nv iro nm en t C ha lle ng e - Gr ee n Ga ug e i nd ica tor s Ye s Un ite d Sta tes : J ac ks on vil le qu ali ty of life in dic ato rs Pa rtly Un ite d Sta tes : S us tai na ble S ea ttle ind ica tor s Pa rtly W HO : H ea lth -fo r-A ll i nd ica tor s (en viro nm en tal on ly) Ye s To tal s No te: N um be rs in bra ck ets de no te pe rce nta ge s o f to tal fo r e ac h ro w Ind ica tor co ve ra ge Le ve l o f a gg reg ati on Ca na da Na tio na l Ca na da Na tio na l Ne the rla nd s Na tio na l OE CD c ou nt rie s Int er na tio na l Un ite d Ki ng do m Na tio na l Ja ck so nv ille / Lo ca l Du va l C ou nty Se att le/ Lo ca l Kin g Co un ty W HO /EU RO Int er na tio na l co un trie s Da ta pe rio d No . o f s ou rc e No . o f e m iss ion No . o f c on ce nt ra tio n No . o f e xp os ure No . o f o the r ac tiv ity in dic ato rs ind ica tor s ind ica tor s ind ica tor s ind ica tor s M os tly 1 97 0s 5 (31 ) 1 (6) 6 (38 ) 0 4 (25 ) to e ar ly 19 90 s M os tly 1 97 0s 15 (3 5) 10 (2 3) 11 (2 6) 0 7 (16 ) to 1 99 0 19 80 to 1 99 1 9 (64 ) 3 (21 ) 1 (7) 1 (7) 0 M os tly 1 97 0 15 (6 0) 4 (16 ) 1 (4) 0 5 (20 ) to la te 19 80 s M os tly 1 98 0s 1 (10 ) 1 (10 ) 3 (30 ) 0 5 (50 ) to 1 99 0s 19 83 to 1 99 2 2 (3) 0 5 (6) 0 71 (9 1) M os tly e ar ly 5 (25 ) 0 0 1 (5) 14 (7 0) 19 80 s t o 19 90 s M os tly 1 97 0 25 (9 3) 0 2 (7) 0 0 to 1 99 0 77 (3 3) 19 (8 ) 29 (1 2) 2 (1) 10 6 (45 ) ( 13). However, it also suggests that there is a short- age of indicators developed specifically to address environmental and health problems. Among the indicators which do fit the frame- work, there is a bias towards the earlier categories. Source activity indicators account for 33%, emis- sion indicators for 8%, concentration indicators for 12% and exposure indicators for 1 %. While this may stem from the intended use of the vari- ous indicator sets it also reflects the shortage of data for indicators in the later categories. As previously mentioned, while data are more readily available for the earlier indicators, they are un- likely to provide a reliable indication of exposure. Considerable doubt therefore remains as to their suitability as health-related environmental indi- cators. The indicators also fail to meet a number of the identified HREI selection criteria: for example, most of the indicators are based on general envi- ronmental issues, rather than specific health-relat- ed environmental issues. As Bakkes (7) has noted, "indicators suitable for one function may be total- ly inappropriate for others". Neither are the exist- ing indicators based on epidemiological relation- ships: they are therefore unlikely to provide an accurate representation of health-related environ- mental issues. In addition, the reliability, consisten- cy, sensitivity and objectivity of existing indicators has yet to be assessed. For example, can two re- searchers using the same traffic-related emissions indicator for N02 in two different countries obtain comparable answers, or do the factors involved (e.g., vehicle type and speed) influence the process to such a degree that the reliability and consistency of the results is open to doubt? Conclusions While considerable progress has been made in de- veloping environmental indicators, the use of indi- cators to demonstrate relationships between envi- ronment and health is clearly less well developed. From the analysis above, it is clear that the majority of existing environmental indicators are unsuitable as health-related environmental indicators. In par- ticular, they are only partially health-related and provide a very poor guide to health-related envi- ronmental exposure; only 2 out of 233 indicators relate specifically to exposure. There is therefore a need to develop indicators specifically for health-related aspects of the envi- ronment. In particular, these indicators should be based on current environment-and-health policy priorities and known epidemiological relation- ships. They should also be assessed to ensure that they provide a representative picture of health- related environmental exposure and their compli- ance with the HREI selection criteria should be examined. 162 Summary This article demonstrates the need for indicators in the policy-making process and presents some of the cur- rent work on developing environmental indicators at the international, national and local levels. The paucity of indicators to demonstrate relationships between envi- ronment and health is outlined and a causal chain for physical pollutants - from source activity to emission, concentration and exposure - is presented. While indi- cators can be developed at each of these stages, indicators from earlier in the chain reflect processes which can more readily be controlled by policy; data is also more readily available for these indicators, yet only the later indicators are likely to provide a reliable guide to human exposure. A survey of 26 indicator projects from around the world yielded 8 sets which are in use and for which data exists. These were found to be only partially health-related and heavily weighted towards the source activity indicator category. Their utility as reliable indicators of human exposure is therefore seri- ously open to doubt and there is consequently a need to develop indicators specifically to assess environmental and health relationships. Resume Mise au point d'indicateurs environnement et sante Cet article demontre qu'on ne peut elaborer une strate- gie sans indicateurs et presente certains des travaux entrepris pour mettre au point des indicateurs environ- nementaux aux niveaux international, national et local. Ayant souligne la rarete des indicateurs demontrant un lien entre l'environnement et la sante, les auteurs pre- sentent une cha7ne de causalite pour les polluants physiques: la source d'activite - emission - concentra- tion - exposition. On peut certes mettre au point des indicateurs pour chacune de ces etapes, mais ceux qui mesurent les premiers elements de la cha7ne revele des processus plus faciles a ma7triser par !'application d'une strategie; les donnees de ces indicateurs sont egale- ment plus faciles a obtenir, mais seuls les indicateurs portant sur les elements suivants de la chaine sont susceptibles de fournir des indications fiables sur !'ex- position des individus. Une enquete portant sur 26 indicateurs mis au point a titre experimental dans diver- ses parties du monde a permis d'etablir huit categories actuellement en usage et pour lesquelles on dispose de donnees. Ces indicateurs ne se sont averes que partiel- lement lies a la sante et portent principalement sur la source d'activite. Leur utilite en tant qu'indicateurs fia- bles de !'exposition est done fort discutable, d'ou la necessite de mettre au point des indicateurs mesurant tout specialement les liens entre l'environnement et la sante. References/References 1. Briggs, D.J. Be Dargie, T. The comparability of national environmental statistics in Europe. IEPA Research Report 92/2. Huddersfield, Institute of Environmental and Policy Analysis, University of Huddersfield, 1992. Rapp. trimest. statist. sanit. mond., 48 (1995) 2. United Nations. Earth Summit 92: The United Nations conference on environment and deve!,opment. London, Regency Press, 1992. 3. Department of the Environment. TM UK environment. London, Her Majesty's Stationery Office, 1992. 4. Ministry of the Environment. Repurt on the state of the environment in Italy. Rome, Instituto Poligra.fico e Zecca dello Stato, 1993. 5. Bakkes, J. et al. An overview of environmental indicators: state of the art and perspectives. Nairobi, United Nations Environment Programme, 1994. 6. Jarman, B. Identification of underprivileged areas. British medicaljoumal286:1705-1709 (1990). 7. Townsend, P. Deprivation.Journal of social policy 16(2): 125- 146 (1987). 8. Kent, D. et al. Coupling wetlands structure and function: developing a condition index for wetlands monitoring. In: McKenzie, D. et al (eds) Ecological indicators: Volume 2. London, Elsevier Applied Science, 1992, pp 559-570. 9. Ludwig, J. & Tongway, D. Monitoring the condition of Australian arid lands: linked plant-soil indicators. In: McKenzie, D. et al (eds) Ecological indicators: Volume 1. London, Elsevier Applied Science, 1992, pp 765-772. 10. Anderson, V, Alternative economic indicators. London, Routledge, 1991. 11. Life in Jacksonville; QJJ,ality indicators for progress. Jacksonville, Florida,Jacksonville Community Council, 1993. 12. Alexander, L HeaUh-related indicators for a sustainab/,e community. Lancaster: Environmental Epidemiology Research Unit, Lancaster University, 1994. 13. Environmental indicators. A preliminary set. Paris, Organisation for Economic Cooperation and Development, 1991. 14. Core set of indicators for Environmental Performance Reuiews: a synthesis repurt by the group on the state of the environment, Environmental monographs No. 83. Paris, Environment Directorate, Organisation for Economic Co-operation and Development 1993. 15. Ministry of Housing, Spatial Planning and the Environment, The Netherlands. National Environmental Policy Plan 2: Summary. The Hague, Department for Information and International Relations, 1994. Wld hlth statist. quart., 48 (1995) 16. Adriaanse, A. Environmental policy performance indicators. A study on the development of indicators for environmental policy in the Netherlands. The Hague, Directorate-General for Environmental Protection, 1993. 17. A repurt on Canada's progress towards a national set of environmental indicators. SOE Report No. 91-1 Ottawa, Environment Canada, 1991. 18. Environmental indicator bulletin: Stratospheric ozone depktion. Ottawa, Environment Canada, 1993. 19. Local Government Management Board. Sustainability indicators - guidance to pilot authorities. London, Touche Ross &Co., 1994. 20. TM Sustainab/,e Seatt/,e indicators of sustainab/,e community. Seattle, Sustainable Seattle, 1993. 21. World Health Organization Regional Office for Europe. Environment and HeaUh: The European Charter and Commentary. First European Conference on Environment and Health, Frankfurt 7- 8 December 1989. European Series No. 35, Copenhagen, WHO Regional Publications, 1989, p 21. 22. Jansson, B. & Voog, L Dioxin from Swedish municipal incinerator and the occurrence of cleft lip and palate malformations. In: Rose, J. ( ed) Environmental Health. TM impact of pollutants. London, Gordon & Breach Science Publications, 1987. 23. Hassoun, E. et al. Teratogenicity of 2,3,7,8-tetra- chlorodibenzo-furan in the mouse. Journal of toxicology and environmentalhea/J,h47: 1-130 (1984). 24. Brunner, C.R. Hazardous air emissions from incineration. London, Chapman & Hall, 1985, pp 5+65. 25. Briggs, D.J. An integrated emissions inventory for Europe. IEPA Research Report No. 93/1. Huddersfield, Institute of Environmental and Policy Analysis, University of Huddersfield, 1993. 26. Odemerho, F.O. & Chokor, B.A. An aggregate index of environmental quality: the example of a traditional city in Nigeria. Applied geography. 11(1): 35-58 (1991). 27. Hope, C et al. A pilot environmental index for the United Kingdom: Results for the last decade. Statistical journal of the United Nations. ECE 8 85-107 (1991). 163 Decision-making in environmental health Eugene Schwartza & Carlos Corva/anb Introduction The ultimate goal of environmental health and epidemiology is to prevent needless morbidity and mortality by protecting people from unnecessary exposure to environmental hazards. Unfortunate- ly, despite the knowledge about harmful exposures obtained from environmental epidemiology stud- ies, preventive action is often slow in coming. This situation is worse in many developing countries where environmental and occupational exposures exceed national and international guideline levels by a considerable amount, but very little is done to rectify these trends. To reverse this trend, results from environmental epidemiological studies must be translated from theory into public health prac- tice more efficiently. This process requires epide- miology to provide the right type of data for deci- sion-making, communicating the results of envi- ronmental epidemiology studies in an easily under- standable format to those empowered to take ac- tion. Additionally, tools for decision-making based on epidemiological data need to be developed fur- ther, and the epidemiologist has a role to play in the process of finding solutions to problems. Decision-making involves choosing among alter- native ways of meeting objectives ( 1). Implicit in this definition is the notion that there are a number of alternatives, and that their effects can be measured or estimated, and that therefore there is adequate information to make an informed choice. To com- plicate matters, there may be a number of objectives in competition or conflict with each other. For example, a decision-maker's overall objectives may include protecting or improving health status, max- imizing productivity, and maximizing environmen- tal protection, while at the same time minimizing the costs of environmental controls. Further, there may be limited or uncertain information on how to measure the impact or costs of various policy alter- natives. Not infrequently, this combination of un- certainty and conflict produces diverse conclusions about the "best alternative" when viewed by differ- ent observers. The Health and Environment Analysis for Deci- sion-making project (HEADLAMP) aims at provid- ing the necessary tools for the management of a Formerly Scientist, Office of Global and Integrated Environmental Health, World Health Organization, Geneva. b Scientist, Office of Global and Integrated Environmental Health, World Health Organization, Geneva. 164 environmental health problems. The main contri- bution of HEADLAMP to the decision-making pro- cess is the creation of the information base re- quired by those responsible for defining the policy actions needed to avoid or reduce adverse environ- mental health effects. Within this context, the de- velopment and use of environmental health indi- cators are fundamental to the decision-making process, since they provide information on the state of, and trends in, environmental health, which is essential for policy-related action and for monitoring the effects of policy implementation.c Public health planning and prioritizing There is a distinction between decisions (prod- ucts) and the techniques (process) used to make them. Decisions are the final judgements regard- ing what actions should or should not be taken. Techniques include the framework or process em- ployed, such as a public debate, a consensus con- ference of experts (e.g., Delphi technique (2)), an algorithm or decision-tree that may be part of a rule-making procedure, and mathematical model- ling to forecast or simulate the future impact of various policy alternatives. Implicit in decision- making is a value system. For example, two com- mon values systems focus on maximizing either equity or utility. The egalitarian approach seeks to ensure that equal risks are borne by all sectors of the population. On a global scale, this would re- quire parity, for example, in the infant mortality rate among all nations. Within a country, it would mean similar rates among different geographical or demographic groups. The utilitarian approach seeks to minimize risks or costs and, at the same time, maximize the benefits. Because this ap- proach only places value on maximizing output per unit input, it is possible that inequalities in health between population groups may be pro- duced or even accentuated. The most common use of public health deci- sion-making is in the process of planning and pri- oritizing public health programmes. Planning is the process of determining the programme's ul- timate objectives (where do we want to be?); the c World Health Organization. lnf<mnal consultation on Health and Environment Analysis fur Decision-making (HEADLAMP) methods andfal.d studies-Summary report. Geneva, WHO, 1994 (WHO/ EHG/94.15). Rapp. trimest. statist. sanit. mond., 48 (1995) programme's current status (where are we?); and how we can get from where we are to where we want to be. Prioritizing involves valuing or ranking a set of possible programme alternatives; it involves choosing between areas of public health where resources and work efforts should be focused. Within chosen areas, this may also involve selecting from among alternative approaches. Epidemiological data can provide the informa- tion needed for national public health planning and priority setting. Epidemiological measures are commonly used for prioritizing or targeting pro- grammes based on rankings of measurements of exposure or expected health outcome (e.g., infant mortality or prenatal care). However, considerable deliberation must be given to the epidemiological measure chosen for the process of prioritization, for each epidemiological measure is itself value- laden (3). For example, relative outcome measures (e.g., relative risk) are an indicator of equity, iden- tifying populations that experience higher risk than the comparison groups. Prioritizing interven- tions based on relative measures seeks to produce parity in risk (an egalitarian goal) among all groups. Clearly, this approach may not necessarily produce the most effective use of resources if, for example, a group at high risk requires a dispropor- tionate amount of input for a marginal benefit. Alternatively, prioritizing based on the number of avoidable cases of disease or death seeks to max- imize the impact of the intervention, providing the greatest good for the greatest number of individu- als (a utilitarian goal). Further, measures such as potential years of life lost (PYLL), used either as a relative or absolute measure, add an additional value dimension by placing greater weight on younger populations. For example within a given country, population "A" might be targeted based on a relative risk measure. Alternatively, population "B" might be chosen based on an index that measures the num- ber of preventable cases, because of their expected high probability of prevention. Further, popula- tion "C" might be chosen based on PYLL, ac- counting for its skewed age structure. Obviously, other epidemiological measures will lead to other conclusions. Thus, when epidemiological mea- sures are used as the rational basis for public health planning and prioritizing, the public health agen- da will be a function of the epidemiological param- eters used in the process (3). The amount and type of information available will always be one driving force in the agenda-setting process. Community-based approach versus targeting The goal of community-based intervention is to favourably shift the entire distribution of a risk factor (e.g., blood lead level or hypertension) to the desired target level (Fig. 1). A conclusion about an incremental benefit ( e.g., decrease of 1 µg/ di of Wld hlth statist. quart., 48 (1995) Fig. 1 Goal of a community-based intervention But d'une intervention communautaire blood lead or 1 mm of mercury in blood pressure) hinges on: (a) the shape of the dose-response curve; (b) the likelihood that small decrements in exposure will produce concomitant changes in the outcome of individuals; and (c) the marginal de- gree ofpreventability at various levels of exposure. For example, the existence of a non-linear dose-response curve or "threshold effect" might yield very different beneficial results for a given intervention targeted at different levels of. expo- sure. As a further example, targeting the highest risk groups ( those in the upper tail of the distribu- tion curve) might be more difficult or more expen- sive than focusing on other segments of the distri- bution. For example, targeting a smoking-cessa- tion intervention at long-term, "hard-core" smok- ers, would likely be more difficult and expensive than focusing an intervention on those smokers who have recently taken up the habit. An effective intervention targeted on the sub- population that is represented by the upper tail of a distribution will produce a distribution with a narrower range. However, the intervention may or may not yield the most utilitarian results, depend- ing on the 3 factors mentioned above. Tools for decision-making There are a variety of tools and approaches used to make decisions. Each relies on various ways of weighing evidence or balancing actions. The fol- lowing are some examples that apply to environ- mental health. Belief in ecology Environmental actions can be based solely on ex- posure data, without evidence oflinkage to adverse human health outcomes. A belief in the protection of scarce natural resources, securing the environ- ment for future generations, knowledge about the impact of pollutants on aquatic systems or animals may be a sufficient basis for remedial intervention with this approach. Risk trading or risk substitution The paradigm often cited to describe risk trading involves the increase in "biological" risk from de- creasing the use of chemicals to preserve or protect 165 food and water. For instance, prohibiting the use of nitrates as a food preservative because of their potential for increasing cancer risk, should be weighed against the increased risk of Clostridium botulinum, a bacteria that causes a potentially fatal form of food poisoning. Similar examples are often cited with respect to chlorination or fluoridation of water, use of pesticides, use of nuclear energy for power generation, and others ( 4). Some of these trade-offs are different in different countries given existing economic and social conditions. Other trade-offs that commonly engender conflict occur when interventions designed to maximize public health benefits require restricting an individu- al's rights to privacy (e.g., compulsory vaccina- tions, use of seat belts, motorcycle helmets, STD contact tracing, etc.). In clinical medicine, the patient's preference for various outcomes can be determined by pre- senting various scenarios. For example, the patient is asked to consider scenarios that differ in their outcome probabilities or costs. An example of an environmental health application might be an ef- fort to determine a community's preference for lower environmental arsenic levels (by closing a smelter) or maintaining current levels in the face of the risks of future lung cancer. The trade-off may be potential economic insecurity and in- creased unemployment from the plant closing. Cost-benefit analysis (CBA) CBA weighs the benefits of an intervention against its costs by assigning monetary values to both. For example, a cost-benefit analysis that examines air pollution reduction sums up the costs of achieving the desired reduction. This is then compared with the sum of the benefits, which include reduced costs for medical care and improved productivity, intangibles such as avoided pain, suffering, loss of a loved one, as well as the positive impact of pollu- tion reduction on quality of life and improved visibility. The monetary value of intangibles has been measured through surveys of people's "willingness to pay", for example, to reduce their risk of disease or protect the environment for fu- ture generations. The willingness to pay to reduce risk is also the basis for purchasing insurance for a variety of relatively rare events such as floods, fires and hurricanes. As an example of a cost-benefit analysis, a re- cent report from the United States found that a 1 µg/ dl decrease in the mean blood lead level con- centrations in children would produce at least $5 billion per year in benefits (5). Savings included medical costs and the costs of remedial education. A major estimated benefit was due to increased lifetime earnings reflected by increased IQs. Among adults, benefits included the reduction of lost wages due to hypertension, heart attacks, strokes and premature mortality. 166 Among the many problems with cost-benefit analysis are difficulties in determining the scope of all costs and benefits; in placing a price on intangi- bles such as life, or pain and suffering; and in determining the "discount rate" for consequences (e.g., deaths) that are expected to occur in the future. In addition, similar CBAs performed in countries at different stages of economic develop- ment may lead to quite different outcomes. Cost effectiveness analysis (CEA) CEA weighs the marginal impact of a group of options. The "preferred" action would therefore be the one which provides the greatest effective- ness (e.g., in health status of a population) given the same investment of resources and effort. This may involve deciding among alternative forms of pollution control and comparing the monetary costs of each alternative. Distributional analysis This approach considers whether particular groups are expected to pay or bear an unreason- able share of the costs or if the benefits are widely distributed or limited to a small sub-population ( 4). The benefits of water treatment, for example, will not help those without access to piped water, and in some cases may even increase the risk of water contamination (as well as the cost of water) if water has to be purchased from vendors. Other approaches Additional potentially useful decision-making tools for use in environmental health may be found by examining the decision-making process employed in clinical medicine, as identified by Pauker & Kas- sirer (6). These include Bayes' rule, decision trees, sensitivity analysis and Markov chains. Bayes' rule, based on prior and conditional probabilities, has been used in clinical medicine to interpret the usefulness of diagnostic or screening tests in various populations. Similarly, Bayes' rule might provide guidance when determining the utility of conducting environmental exposure assessments or implementing environmental epi- demiological studies. Decision trees can help iden- tify all available choices and their potential out- comes by structuring a branching model of the alternatives. Using known or estimated probabili- ties at each branch (node) of the tree, the relative utility of various strategies can be calculated. The strategy with the highest utility would be expected to provide the best choice. This method can be applied to environmental decision-making to help examine the alternatives, and makes explicit the known or assumed outcomes in a given situation. In situations where the probabilities of an outcome are uncertain, sensitivity analysis may be useful. This type of analysis allows one to examine varying Rapp. trimest. statist. sanit. mond., 48 (1995) values of one or more variables in an effort to determine how the utility of a decision might change. The outcome of a policy can be modelled as a chain of events occurring over time. The sequence of events can be represented by a decision tree, each event having a given probability. The se- quence can be repeated to determine how the impact changes over a long period of time. This type of model can be mathematically represented and analyzed using Markov chains (7). For exam- ple, computer software using Markov chains has been used to assist with public health decision- making in the area of cancer prevention and con- trol. Given various demographic and baseline can- cer incidence and mortality data for a community, the software program CANTROL can estimate the potential impact of various levels of cancer screen- ing or smoking prevalence (8). Problems associated with the decision-making process There are numerous technical problems in the process. One of the basic conflicts arises from the inexact nature of the process while the public de- mands quick, clear answers. As Steenberg has clear- ly stated, there is no definable boundary between what is safe or hazardous, but rather a zone of uncertainty, and one is limited to describing one's position by referring to a changing "prob- ability" of an effect being produced (9). There are several other issues that need to be considered in the decision-making process, for ex- ample: • quantifying the extent to which prevention can be obtained; • extrapolating from evidence derived at high doses to determine risk at lower doses; • extrapolating from data derived from animal evidence to determine human risk; • individual susceptibility; • inability to adequately control for all possible confounders; • combinations of exposures and multiple routes of exposure; • time constraints; • models which do not always reflect the real world; • difficulty determining true probabilities; and • difficulty valuing intangibles such as quality of life. Setting clear guidelines to facilitate the deci- sion-making process is not a simple endeavour. All the items above are subject to interpretation, and even the experts are likely to disagree regarding both the weight to allocate to each of the points above and to the conclusions to be derived from each specific point. Those with decision-making responsibilities are not expected to be directly in- Wld hlth statist. quart., 48 (1995) Fig. 2 Basic process of decision analysis Analyse de decision: modele de base ~-------- Statement of problem - ...----, Enonc6 du probleme Research- Recherche Evaluation of existing information - Evaluation de l'information disponible Identification of new information required D6termination de l'information A obtenir Evaluation of options - Evaluation des options Evaluation of impact of decision - Evaluation de l'impact de la d6cision Based on - D'apres: Ref. 9. - Ref. 9. WH095349 valved in the technical processes outlined above. It is necessary, however, that decision-makers learn new ways of thinking and of evaluating information on health. They have a responsibility to understand the implications of the uncertainties in the infor- mation and the value this information has for deci- sion-making. The process of decision analysis The basic process involves stating the problem, evaluating existing information, identifying the need for new information, identifying options and alternatives, evaluating the impact of alternatives, and finally, making the decision (Fig. 2). Five char- acteristics of an effective standard-setting process have been noted by de Koning (4): 1. Involve the major parties in the community, including politicians, citizen groups, industri- al leaders and health officials. This should stimulate debate encompassing differing per- spectives and values, leading to some compro- mises being made in both goals and methods, thus ensuring broad support in the society at large. 2. Provide a mechanism through which technical and policy analysis can be generated, distrib- uted and criticized. 3. Provide a mechanism whereby the results of analyses can be presented to policy-makers and the other centres of interest in the soci- ety, to inform these groups of the costs, bene- fits, and impact of the proposals under consid- eration. 4. Provide a mechanism for conflicting interests to be heard and discussed in a controlled manner, so that divergent opinions in the society can be aired and, as far a possible, accommodated in the implementation of the proposal. 167 5. Provide a mechanism whereby the society can reach a decision and take useful action, even though such action may be less than what is "objectively" ideal. The decision-making process should be "reasoned" and "principled", explicitly stating all assumptions and evaluating the evidence about what is known and also what is unknown. Perhaps the two most pervasive problems with decision- making are the failure to make decisions in a time- ly manner and the failure to include the public in the process. Commonly, decisions or actions are delayed or deferred, in favour of review of more compelling evidence or better or cheaper control technolo- gies. But waiting too, has its costs and may have irreversible consequences (10). Decision-making is a complex scientific, social and political process. The public must make, pay for and eventually live with the results of the deci- sions that are made. Science can provide guidance but not provide all the answers. In an effort to gain the trust of the public, it is suggested that the community be involved in the design, implementa- tion and interpretation of the study. An open and participatory approach is more likely to make the results more credible and to provide time for the community to consider in advance the technical concepts, limitations and range of outcomes (11). Moreover, some questions are unanswerable be- cause of gaps in our knowledge, hence the need for dialogue with the community in order to reach a mutually agreeable solution. The different agen- das of the affected community, the scientists and decision-makers are not less important barriers than current knowledge gaps. Each of these and other stakeholders' beliefs regarding what are acceptable solutions reflect economic, profession- al, political and bureaucratic pressures in making Box 1 Factors that influence actions social decisions, for example, of how much risk is acceptable, a very difficult task ( 12). Many factors influence actions towards deci- sion-making. Some of these are outlined in Box 1. In addition, epidemiologists need a better-defined role in the decision-making process. Epidemiolo- gists should not be just researchers but also a prac- titioners of public health. As such they have an obligation to do more than publish their results. Epidemiologists have specific obligations to the study subjects (e.g., informed consent, maintain- ing confidentiality); to society (e.g., to share their information and study results); to colleagues; to employers; and to research sponsors ( 13). How- ever, the epidemiologist's role should include more emphasis on the following points:. • helping to state the problem so it is clear; • implementing studies to answer relevant ques- tions; • communicating and interpreting epidemiologi- cal data for non-epidemiologists. The public has a right to know and the epidemiologist's responsibility includes more than just publish- ing results in a journal; and • being a good citizen and member of the com- munity where they live and work. As expressed by Gordis, "[we] should not hide under the mantle of scientific objectivity and detach our- selves from critical decision-making in public health" (14). HEADLAMP proposes that decision-makers work in close collaboration with environmental ep- idemiologists and environmental health scientists in the management of environmental health prob- lems. This requires that decision-makers become involved at an early stage in the implementation of. studies or monitoring processes which have a well- defined policy action goal. Similarly, environmen- There are many factors that influence actions towards decision-making. Some of these are: • Value placed on health, human life extension, environmental protection, concern for future generations • Strength of data, extent of documentation • Public understanding of data and perceptions (acceptability) of risk • Costs of intervention; are they affordable? • Leadership: ability to persuade/motivate, negotiate, resolve conflicting goals or competing interests • Process that provides a forum for debate and permits input into public policy setting • Emphasis on planning for the future, government responsibility for protecting public from future harms • Degree of collaboration: government/business/non-governmental organizations • Regulatory process • Judicial process • Seriousness of the outcome • Involvement of mass media • Targeted message for decision-makers 168 Rapp. trimest. statist. sanit. mond., 48 (1995) tal health scientists and epidemiologists need to be committed to this process beyond their current involvement which often ends with the publication of study findings. Conclusions Decision-making is not a simple process. A deci- sion-maker must choose between competing alter- natives, and may face uncertainties at every step. These difficulties, however, are no excuse for lack of action. There is a clear gap between the sophisti- cation in public health research and that of deci- sion-making in public health and environmental issues (5). One option to rectify this gap is to in- volve epidemiologists in the process of addressing the solutions to the problems they study. This would entail a change in attitude and would have to be reflected in their training, because this role is not normally part of an epidemiologist's educa- tional and professional experience. Certainly, increasing evidence about a potential health problem would aid the decision-making process, but waiting for more evidence implies that someone has to endure suffering in the meantime (15). As noted by Bradford Hill in 1965, "All scientific work is incompkte - whether it be obseroational or experimental. All scientific work is liabk to be upset or modified b-y advancing knowkdge. That does not confer upon us a freedom to ignore the knowkdge that we already have, or to postpone the action that it appears to demand at a gi.ven time. "( 16) Summary Despite our current knowledge about harmful expo- sures obtained from environmental epidemiology stud- ies, preventive action is lacking in many fronts. To reverse this trend, results from environmental epidemi- ological studies must be translated from theory into public health practice more efficiently. This process requires epidemiology to provide the right type of data for decision-making and to communicate the results of environmental epidemiology studies in a form under- standable to the community at large and to those em- powered to take action. Tools for decision-making based on epidemiological data need to be developed further, and the epidemiologist has a role to play in the process of addressing the solutions to the problems they study. Similarly, while those with decision-making responsibilities are not expected to be directly involved with technical aspects of conducting epidemiological studies, it is necessary that they learn new ways of thinking and of evaluating information on health. They have a responsibility to understand the implications of the uncertainties in the information and the value this information has for decision-making. Decision-making involves choosing among alternative ways of meeting objectives. Often, however, there may be a number of objectives that may be in competition or Wld hlth statist. quart., 48 (1995) conflict. Not infrequently, this combination of uncertain- ty and conflict produces diverse conclusions about the "best alternative" when viewed by different observers. A decision-maker must choose between competing alter- natives, and may face uncertainties and difficulties at every step. These difficulties, however, should not serve as excuses for lack of action. While it is true that increasing evidence about a potential environmental health problem would aid the decision-making process, lack of action while waiting for more evidence may also carry significant adverse consequences. Resume Prise de decision en matiere d'hygiene de l'environnement En depit des informations fournies par les etudes eco- epidemiologiques sur les expositions nocives, les me- sures preventives font defaut dans bien des domaines. Si l'on veut remedier a cette situation, ii taut traduire les resultats des etudes eco-epidemiologiques theoriques en des mesures pratiques et efficaces de sante publi- que. Pour cela, les etudes epidemiologiques doivent fournir les donnees indispensables a la prise de deci- sions, et les resultats des etudes eco-epidemiologiques doivent etre exprimes sous une forme intelligible pour la communaute dans son ensemble et pour les decideurs. II taut developper les instruments de prise de decisions fondes sur les donnees epidemiologiques. D'un cote, les epidemiologistes doivent contribuer a trouver des solutions aux problemes qu'ils etudient; de l'autre, si l'on ne peut exiger des decideurs qu'ils participent directe- ment aux differents stades techniques des etudes epi- demiologiques, ils doivent adopter de nouveaux raison- nements de nouvelles methodes d'appreciation de !'in- formation sanitaire. lls doivent etre conscients des con- sequences pouvant decouler de !'approximation de !'information, et de l'utilite de cette information a l'egard de la prise de decisions. Prendre des decisions consiste a faire un choix entre differents moyens d'atteindre des objectifs. II arrive souvent, cependant, que certains objectifs soient en concurrence ou en opposition. II n'est pas rare que pour cette raison, et a causes de !'approximation de !'infor- mation, les observateurs parviennent a des conclusions differentes quanta la «meilleure solution». Le decideur doit choisir entre plusieurs solutions et peut, a chaque etape, etre gene par des incertitudes et des difficultes. Toutefois, ces difficultes ne doivent pas servir de pretex- te a !'inaction. S'il est vrai qu'un surcroit de preuves sur un probleme d'hygiene de l'environnement faciliterait le processus de prise de decisions, !'absence de mesures dans l'attente de preuves supplementaires peut egale- ment avoir des consequences facheuses. References/References 1. Warner, D.M. et al. Decision making and rontrol fur health administration. Ann Arbor, Health Administration Press, 1984. 2. Richey, J.S. et al. The Delphi technique in environmental assessment. Parts 1 and 2Joumalofenvironmentalmanagement 21: 147-59 (1985). 169 3. Schwartz, E. Premature mortality in New Hampshire. Morbi.dity mmtality weekly repart, 36: 765-6 ( 1987). 4. de Koning, H.W. Setting environmental standards. Guidelines JM decision-making. Geneva, World Health Organization, 1987. 5. Schwartz, J. Societal benefits of reducing lead exposure. Environmental research, 66: 105-124 (1994). 6. Pauker, S.G. &:Kassirer,J.P. Decision analysis. NewEngland journal of medicine, 316: 250-257 (1987). 7. Beck,J.R. &: Pauker, S.G. The Markov process in medical prognosis. Medical decision making, 3: 419-57 (1983). 8. Eddy, D.M. A computer-based model JM designing cancer control strategies. Bethesda, Maryland, National Cancer Institute, monograph No. 2 (1986). 9. Steensberg, J. Environmental decision making: the politics of disease prevention. Copenhagen, Almqvist & Wiksell Inter- national, 1989. 10. Harris,J.E. Environmental policy making: act now or wait for more information. In: National Research Council, Valuing 170 health risks, costs and benefits JM environmental decision making, Washington, D.C., National Academy Press, 1990. 11. Ozinoff, D. &: Boden, L.I. Truth and consequences: health agency responses to environmental health problems.Science, technowgy & human values, 12: 70-77 (1987). 12. McMicbael, A.J. Setting environmental exposure standards: current concepts and controversies. International journal of environmental health research, 1: 2-13 (1991). 13. Cook, R. Code of ethics for epidemiologists.Journal of clinical epidemiowgy. Supp. 1, 44: 135S-139S (1991). 14. Gordis, L. Ethical and professional issues in the changing practice of epidemiology. Journal of clinical epidemiowgy. Supp. 1, 44: 9S-13S (1991). 15. Sandman, P. Emerging communication responsibilities of epidemiologists. Journal of clinical epidemiowgy. Supp. 1, 44: 41S-50S (1991). 16. Hill, A.B. The environment and disease: association or causation? Proceedings of the Ruyal Society of Medicine, 58: 295- 300 (1965). Rapp. trimest. statist. sanit. mond., 48 (1995) WORLD HEALTH ORGANIZATION PUBLICATIONS Operation and Maintenance of Urban Water Supply and Sanitation Systems A Guide for Managers This book describes a systems approach to the opera- tion and maintenance of drinking-water and sanita- tion services in urban areas of developing countries. Addressed to managers and other personnel with decision-making responsibilities, the book responds to ample evidence that poor management has had the greatest single negative impact on the quality of water supply and sanitation services. Throughout the book, numerous lists, tables, and diagrams are used to help managers think through problems and find ways to solve them. The guide, which is intended to serve as a reference source and conceptual framework, covers virtually all the procedures, activities, projects, and areas of managerial responsibility, at different levels, needed to ensure that water supply and sanitation services function continuously, efficiently, and to their full capacity. Though most aspects of operation and main- tenance are covered, particular attention is given to procedures that can help control water losses. The book has sixteen chapters presented in five parts. The first part describes the use of the management systems approach to analyse the functions of drink- ing-water and sanitation agencies and to solve opera- tion and maintenance problems. Chapters explain what an operational system is, how it works, and what sub-systems it includes. Readers also learn how use of a systems approach to management facilitates detailed analysis of the agency, even in complex situations, without obscuring the overall picture. Part two, on management, provides a step-by-step account of the key responsibilities and functions involved in managing an agency's operation and maintenance activities. Specific functions are identi- fied for the different levels of senior, middle, and operational management. The third and most extensive part serves as a detailed guide to the planning and control of operation and maintenance procedures. While most attention is given to projects for controlling water loss in an agency's different systems, part three also contains chapters describing the objectives and component elements of programmes for controlling the produc- tion and quality of drinking-water, and for sewage collection, treatment, reuse, and disposal. In part four, on information systems, chapters cover the conceptual framework, methodology, and centres of decision for the development of management information systems and systems to support deci- sions. Of particular practical value is a five-page tabular presentation of performance indicators that can be used to assess the effectiveness of specific activities. The final part outlines actions to be carried out, and stages to be completed, in order to implement procedures described in the previous parts. Operation and Maintenance of Urban Water Supply and Sanitation Systems A Guide for Managers 1994, ix+ 102 pages (available in English; French in preparation) ISBN 92 4 154471 6 Sw.fr. 23.-/US $20.70 In developing countries: Sw.fr. 16.10 Orderno.1150416 WHO• DISTRIBUTION AND SALES• 1211 GENEVA 21 • SWITZERLAND WORLD HEALTH ORGANIZATION PUBLICATIONS Financial Management of Water Supply and Sanitation A Handbook This handbook describes a range of financial principles and methods for improving the management of water supply and sanitation services - whether large or small, urban or rural. Addressed to managers and other decision- makers, the book shows how financial mechanisms, such as cost recovery, cash raising, and cost containment, can be used to ensure that services are financially sustainable and able to meet users' needs. With this goal in mind, the book helps readers to think through all costs and responsibilities associated with each stage in a project's life span, and then to use this information to set objectives and calculate costs and benefits. Material in the handbook was tested in field activities in more than 20 countries and then further refined in numer- ous country and intercountry seminars involving over l OOO participants. Throughout the book, guidelines, warn- ings, and advice respond to common errors known to contribute to poor financial management and poor quality services. The book has thirteen chapters presented in two parts. Part one introduces some of the underlying principles for ascer- taining that all resources required for services are identi- fied and available. To help readers think through problems and be alert to potential errors, chapters explain the impor- tance of a two-way partnership between project executors and project beneficiaries, discuss ten essential elements of sustainability, describe the process of resource mobiliza- tion, and establish a financial model for meeting costs. Information ranges from a list of obstacles commonly encountered in developing countries, through tips on how to reduce costs and increase revenue, to the simple re- minder that water has a price that should always be paid for by consumers. Readers are also introduced to the concepts of resource coverage and liquidity maintenance as essen- tial components of any sustainable system. The second and most extensive part provides a practical guide to methods of cost recovery. Using numerous check- lists, charts, examples, and schedules for calculating pro- jected costs, chapters offer a step-by-step explanation of the financial and related activities required to achieve cost recovery at each stage in a project's life span, moving from planning and construction, through operation and mainte- nance, to eventual replacement. Separate chapters con- sider cost recovery in relation to the operating environ- ment, users' needs and expectations, technical options, and different methods for raising cash and ensuring that all necessary financial, physical, and human resources are mobilized. While emphasizing sustainability as the most desirable characteristic of any public utility, chapters also provide information on ways to assure that the basic needs of low-income groups are met. Of particular practical value are sections offering advice on how to collect relevant community data and establish a scheme for water tariffs in line with public health, financial, and economic criteria. The book also includes guidance on the design and implementation of appropriate reporting and accounting systems. Financial Management of Water Supply and Sanitation A Handbook 1994, x + 83 pages (available in English; French and Spanish in preparation) ISBN 92 4 154472 4 Sw.fr. 20.-/US $18.00 In developing countries: Sw.fr. 14.- 0rder no. 1150419 WHO• DISTRIBUTION AND SALES• 1211 GENEVA 21 • SWITZERLAND Publications of the World Health Organization 1995 WORLD HEAL TH FORUM An international journal of health development (Separate editions in English, French, Spanish, Arabic, Chinese and Russian) W<>rld lwil.th /arum is a quarterly journal for policy-makers, health planners, administrators, health educators, and public health workers of all kinds. It provides a medium for the presentation and discussion of new concepts in public health and new approaches to health problems, and it is devoted to the improvement of health through the promotion of health services covering the entire population and the undertaking of a wide variety of public health measures, whether or not they are supported by WHO. The Farum is the main organ ofWHO's Member States for the international exchange of health information and an active instrument in technical cooperation among developing countries. Subscription ( 4 numbers) ............................................... . Price per copy ............................................................... . Sw. fr. 75.- Sw. fr. 20.- BULLETIN OF THE WORLD HEALTH ORGANIZATION The Bulletin is the principal scientific organ of WHO: its role is to review progress in medical and related sciences ("update" articles) and bring to light new knowledge by publishing original papers on scientific research in Lhe laboratory and field. The Bulletin contains original articles in either English or French, with a summary in the other language. Bimonthly. Subscription (6 numbers) ............................................... . Price per copy ............................................................... . WEEKLY EPIDEMIOLOGICAL RECORD (Bilingual: English and French) Sw. fr. 176.- Sw. fr. 31.- Prepared for the guidance of health administrations and health authorities, the Wllkly epidemiological ree<>rd contains notifications made under the International Health, Regulations and information concerning their application. The &e<>rd also contains epidemiological information on communicable diseases of international importance. Annual subscription ......................................................... . Sw. fr. 209.- WORLD HEALTH STATISTICS QUARTERLY The W<>rld lwal.th statistics quaru:rly replaces (since 1978) the WMld lwal.th statistics report (published since 1967) and its forerunner theEpidemiological and vital statistics report (published since 1947). It deals with the detailed analysis of selected health topics of current interest. The Qµarterly contains articles in eithl!f French <>r English with a summary in both languages. Annual subscription ......................................................... . Price per copy ............................................................... . WORLD HEALTH STATISTICS ANNUAL (Bilingual: English and French) Sw. fr. llO.- Sw. fr. 31.- The forerunner of this series was the Annual epidemiological report of tlw League of Nations. It was followed by the Annual epidemiological and vital statistics issued by the World Health Organization. Latest publications: 1990. Vital statistics and life tables, international statistics on causes of disability, causes of death (single volume), 447 pages (out of stock) .................................................................... . 1991. Vital statistics and life tables, health and human development, causes of death (single volume), 371 pages ..................................................... . 1992. Vital statistics and life tables, implementation of the Global Strategy for Healh for All by the Year 2000 (single volume), 480 pages ....................... . 1993. Vital statistics and life tables, health and demographic data, and alternative sources for the collection of cause-of-death and vital events data in the absence of universal vital registration (single volume), 662 pages ................................. . 1994. Vital statistics and life tables, and worldwide health and demographic data (single volume), 460 pages ........................................................... . Sw. fr. 90.- Sw. fr. 100.- Sw. fr. 100.- Sw. fr. 100.- Sw. fr. 100.- Publications de /'Organisation mondiale de la Sante 1995 FORUM MONDIAL DE LA SANTE Revue internationale de developpement sanitaire (Editions separccs en fran~ais, anglais, espagnol, arabe, chinois et russe) Forum mondialde I.a santi est une revue trimestrielle destinee aux responsables des politiques sanitaires, aux planificateurs, administrateurs et educateurs sanitaires, enfin aux travailleurs de la sante publique de toutes categories. Tribune pour la presentation et la discussion de nouveaux concepts en sante publique et de nouvelles approches des problemes de sante. Forum se consacre a I 'amelioration de la sante par la promotion de services de sante couvrant la population tout entiere et d 'une vaste gamme de mesures de sante publique, qu' elles soienlou non soutenues par l'OMS. Il est le principal organe a la disposition des Etats Membres de l'OMS pour l'echange international d 'informations sanitaires en meme temps qu 'un instrument de cooperation technique entre pays en developpement. Abonnement (4 numeros) .............................................. . Lenumero .................................................................... . Fr. s. 75.- Fr. s. 20.- BULLETIN DE L'ORGANISATION MONDIALE DE LA SANTE Le Bulletin est le principal organe scientifique de l'OMS; il a pour role de passer en revue les progres des sciences medicates etapparentees (articles de la rubrique «Le point•) et de mettre en lumiere les connaissances nouvelles en presentanl dans des articles originaux les resultats de recherches scientifiques au laboratoire et sur le terrain. Le Bulletin contient des articles originaux en fran~ais ou en anglais, accompagnes d'un resume dans l'autre langue. Bimensuel. Abonnement (6 numeros) .............................................. . Lenumero .................................................................... . RELEVE EPIDEMIOLOGIQUE HEBDOMADAIRE (Bilingue: fran~ais et anglais) Fr. s. 176.- Sw. fr. 31.- Destine aux administrations sanitaires et aux services de sante, le &/eue ipidemiologique lwbdmnadaire contient les notifications exigees par le Reglement sanitaire international, ainsi que d'autres renseignemenLS concernanl !'application de ce reglement. Le &leue contient cgalement des informations epidemiologiques concernant les maladies transmissibles d'une importance internationale. Prix de l'abonnement annucl .......................................... Fr. s. 209.- RAPPORT TRIMESTRIEL DE STATISTIQUES SANITAIRES MONDIALES Le Rapport tnmestrielde statistiques mondia/.es, remplace ( depuis 1978) le Rapport de statistiques sanitaires mondial.es (publie depuis 1967) et son precurseur le Rapport epidhniologiquut demographique (publie depuis 194 7). Il presente des analyses detaillees sur des sujets spccifiques d'interct courant. Le Trimestriel presente des articles originaux en fran{ais ouen anglais, accompagnes d'un resume dans les deux langues. Prix de l'abonnement annuel ......................................... . Lenumero .................................................................... . Fr. s. HO.- Fr. s. 31.- ANNUAIRE DE STATISTIQUES SANITAIRES MONDIALES (Bilingue: fran~ais et anglais) Cet annuaire remplace les Statistiques epidbniologiques et demographiques annue/Jes publiees par ['Organisation mondiale de la Santi' et qui avaient, elles-memes, rem place le Rapport epidbniologique annu.el publie par la Societe des Nations. Publications recentes: 1990. Mouvement de la population et tables de survie, statistiqucs intemationales sur les causes d'incapacite, causes de deces (1 seul volume), 447 pages (epuise) ............................................... . 1991. Mouvementde la population et tables de survie, la sante et le developpement humain, causes de deces (1 seul volume), 371 pages ...................... . 1992. Mouvemenl de la population et tables de survie, mise en ceuvre de la Strategie mondiale de la sante pour tous d'ici l'an 2000 (1 seul volume), 480 pages .............................................................. . 1993. Mouvement de la population et tables de survie, donnees sanitaires et demographiques, methodes permettant de recueillir des donnees sur les causes de deces et les faits de l'etat civil en ['absence de systeme universe( d'enregistrement (I seul volume), 662 pages ................................. . 1994. Mouvement de la population et tables de survie, ainsi que donnees sanitaires et demographiques a l'echelle mondiale (1 seul volume), 460 pages Fr. s. 90.- Fr. s. 100.- Fr. s. 100.- Fr. s. 100.- Fr. s. 100.- Health and environment analysis and indicators for decision-making The Health and Environment Analysis for Decision-Making Project (HEADLAMP) aims to provide decision-makers, environmental health professionals and the community at large with valid and useful information on the health impact of environmental hazards at the local and national level. HEADLAMP combines methodologies in environmental epidemiology, human exposure assessment and other fields related to health and environment, in order to produce information which can be understood easily and serve as a basis for action. The articles in this issue of the World Health Statistics Quarterly describe the HEADLAMP project, its methods, the development and use of environmental health indicators, the decision-making process and the findings from field studies. Analyse et indicateurs sante et environnement pour la prise de decisions Le projet d'analyse sante et environnement pour la prise de decisions (HEADLAMP) est destine a fournir aux decideurs, aux specialistes de !'hygiene de l'environnement et a la communaute dans son ensemble des informations justes et utiles concernant l'impact sur la sante des risques lies a l'environnement aux niveaux local et national. Ce projet combine les methodes eco-epidemiologiques, !'evaluation de !'exposition de la population et d'autres disciplines liees a la sante et a l'environnement, en vue de fournir une information facile a comprendre qui servira de base a une action. Les articles de ce numero du Rapport trimestriel de statistiques sanitaires mondiales decrivent le projet HEADLAMP et les methodes auxquelles ii fait appel, expliquent comment mettre au point et utiliser des indicateurs de la salubrite de l'environnement, et presentent le processus de prise de decision et les resultats de plusieurs etudes de terrain.