Bacterial indicators of risk of diarrhoeal disease from drinking-water in the Philippines.

Bacterial indicators of risk of diarrhoeal disease from drinking-water in the Philippines C.L. Moe,1 M.D. Sobsey,2 G.P. Samsa,3 & V. Mesolo4 Inadequate measures of water quality have been used in many studies of the health effects associated with water supplies in developing countries. The present 1-year epidemiological-microbiological study evaluated four bacterial indicators of tropical drinking-water quality (faecal coliforms, Escherichia coli, enterococci and faecal streptococci) and their relationship to the prevalence of diarrhoeal disease in a population of 690 under-2-year-olds in Cebu, Philippines. E. coli and enterococci were better predictors than faecal coliforms of the risk of waterborne diarrhoeal disease. Methods to enumerate E. coli and enterococci were less subject to interference from the thermotolerant, non-faecal organisms that are indigenous to tropical waters. Little difference was observed between the illness rates of children drinking good quality water (< 1 E. coli per 100 ml) and those drinking moderately contaminated water (2-100 E. coli per 100 ml). Children drinking water with > 1000 E. coli per 100 ml had significantly higher rates of diarrhoeal disease than those drinking less contaminated water. This threshold effect suggests that in developing countries where the quality of drinking-water is good or moderate other transmission routes of diarrhoeal disease may be more important; however, grossly contaminated water is a major source of exposure to faecal contamination and diarrhoeal pathogens. Introduction Seriously polluted water supplies are used by at least 1500 million people worldwide (16). Because of the magnitude of the health problems associated with water of inadequate quality and quantity, substantial efforts have focused on how to evaluate and maximize the health benefits derived from improved water supplies. In many developing countries the high incidence of waterborne disease and wide- spread use of untreated and often highly polluted water sources make the accurate assessment of faecal contamination of water particularly impor- tant. The indicator bacteria commonly used to evalu- ate water quality in temperate climates may not be appropriate for the tropics, where water sources are I Graduate Research Assistant, Department of Environmental Sciences and Engineering, School of Public Health, University of North Carolina, Chapel Hill, NC, USA. Current position: Visiting Fellow, Viral Gastroenteritis Unit, Division of Viral Diseases (G04), Centers for Disease Control, 1600 Clifton Road, Atlanta, GA 30333, USA. Requests for reprints should be sent to the latter address. 2 Professor, Department of Environmental Sciences and Engi- neering, School of Public Health, University of North Carolina, Chapel Hill, NC, USA. 3 Assistant Professor, Division of Biometry and Informatics, Department of Community and Family Medicine, Duke Univer- sity, Medical Center, Durham, NC, USA, and Senior Research Scientist, Center for Health Services Research in Primary Care, Veterans Administration Medical Center, Durham, NC, USA. 4 Chairman, Microbiology Department, Cebu Institute of Medi- cine, Cebu, Philippines. Reprint No. 5176 typically warm (21-31 °C) and have high nutrient levels. These characteristics influence the survival and extra-enteral regrowth of indicator bacteria and encourage the growth of non-faecal, thermo- tolerant indigenous microorganisms. The latter can cause heavy background growth or give false posi- tive reactions and thereby interfere with enumera- tion of the indicator bacteria. Several investigators have reported difficulties in using total and faecal coliforms to measure the sanitary quality of tropical waters, including the ability of the bacteria to multiply, to survive for long periods, and to occur in high numbers in the absence of any identifiable source of faecal pollution (21). Current WHO bacteriological guidelines (38) for drinking-water recommend zero faecal coliforms per 100 ml of water. However, some investigators hold that this is not feasible for untreated, unpiped water sources used in developing countries (16, 39). There is therefore a need for epidemiological-micro- biological studies in tropical developing countries to define measures of water quality that accurately reflect the potential for disease transmission and to formulate realistic water guidelines applicable to the special problems and needs of such countries (14, 21, 39). Previous epidemiological field studies that at- tempted to relate water supply and diarrhoeal disease have been reviewed by McJunkin (28), Blum & Feachem (5), and Esrey et al. (13), inter alia. In an analysis of data pooled from nine separate studies, Esrey et al. concluded that improvements in water quality alone caused a median reduction of 16% in diarrhoea morbidity (range, 0-90%) and Bulletin of the World Health Organization, 69(3): 305-317 (1991) © World Health Organization 1991 305 C.L. Moe et al. that improvements in both water quality and avai- lability resulted in a median reduction of 37% (range, 0-82%) (13). Among the common method- ological problems observed by McJunkin (28) was the inadequate measurement of water quality. Most of 32 field studies of water supply and diarrhoeal disease he reviewed appeared not to have measured bacteriological water quality; instead, exposure to "igood" or "poor" water quality was estimated by the type of water source or level of service. Only 10 of the studies reported the bacteriological quality of the water; and of these, seven measured total coliforms (29-32, 34, 35, 37), one measured faecal coliforms (22), and two measured faecal coliforms and faecal streptococci (3, 15, 33). The present study used an epidemiological approach to evaluate four bacterial indicators of tropical drinking-water quality (faecal coliforms, Escherichia coli, enterococci, and faecal strepto- cocci) by examining the relationship between the concentrations of these bacteria in drinking-water and the prevalence of diarrhoea in a population that consumed this water. Faecal coliforms and faecal streptococci were selected as standard indicator bacteria, while E. coli and enterococci were chosen because of their successful use in other studies, because of evidence that they are relatively faecal- specific, and because of new selective methods for their enumeration (8, 39). Methods General description of the study A one-year study of the above-mentioned indicators of tropical drinking-water quality was carried out as part of two larger investigations on infant feeding practices and on the effect of water and sanitation on infant health in metropolitan Cebu, Philippines. Because the study shared the infrastructure and resources of these larger investigations, their design limited the type and frequency of the data collected to 2-monthly intervals. One of the larger studies examined the effects of water quality and use, excreta disposal, waste- water disposal, and solid-waste disposal practices on diarrhoeal disease morbidity among 690 infants. These infants were selected from nearly 2800 aged 0-24 months who were being followed in a 3-year prospective study of the relationship between infant feeding practices and infant morbidity and mortal- ity.8 In developing countries, under-2-year-olds suf- a Conducted jointly by the Carolina Population Center, University of North Carolina, the Office of Population Studies, University of San Carlos, and the Nutrition Center of the Philippines. fer the highest incidence of diarrhoeal disease (36), and are a sensitive population group for studying the effects of water quality on this disease. Data were collected in a series of interviews of mothers or caregivers. A baseline questionnaire that was administered before the birth of the child was followed by an interview several days after the baby's birth and then by interviews every 2 months until the child was 2 years of age. All questionnaires were administered by trained field-workers and then edited. The prevalence of diarrhoea morbidity in the study population was determined at the 2-monthly interviews, and rectal swabs were collected from all children who had diarrhoea at the time of the interview. Most of the cases of diarrhoea detected at home visits were relatively mild. All cases identified by the mother as diarrhoea were included, and no formal case definition was used. This approach has been used previously (19, 36), and is considered satisfactory. Of the 363 cases identified in the study, 82% involved the passage of three or more stools in 24 hours, which is similar to definitions of diarrhoea used in several other studies (36). The mean reported duration of the illness was about 4 days. The diarrhoea etiologies were determined and have been described elsewhere.' Study site and population Metropolitan Cebu has a population of slightly more than 1 million, and includes Cebu city, the second largest city in the Philippines. A total of 33 communities (barangays) were randomly selected to include a mixture of urban and rural, coastal and mountainous environments, and a considerable spread of social, economic, and health conditions. Preliminary data indicate a birth rate of about 23 per 1000 and an infant mortality rate of about 54 per 1000 (Office of Population Studies, unpublished reports, 1987). Diarrhoea, pneumonia, and malnu- trition are the leading causes of infant mortality, with diarrhoea being involved in approximately 17% of deaths during the first 2 years of life (Office of Population Studies, unpublished reports, 1987). The 690 infants were chosen randomly from all live births that occurred from May 1983 to April 1984, in the 33 selected barangays, excluding multi- ple births. Their age distribution in January 1985 at the beginning of the study is shown in Table 1; 53% were male and 74% lived in urban barangays. b Moe, C.L. An evaluation of four bacterial indicators of risk of diarrheal disease from tropical drinking-waters in the Philippines. Doctoral dissertation. The University of North Carolina at Chapel Hill, 1989. 306 WHO Bulletin OMS. Vol 69 1991. Bacterial indicators of diarrhoea in the Philippines Table 1: Age distribution of study children, Cebu, Philip- pines, January 1985 Age No. of children (months) 9-10 204 (29.6)a 11-12 152 (22.0) 13-14 83 (12.0) 15-16 86 (12.5) 17-18 96 (13.9) 19-20 69 (10.0) Total 690 (100.0) a Figures in parentheses are percentages. Water sampling programme All water sources used by the study population were routinely monitored for bacteriological quality. The proportion of families using the various types of sources included in the study is summarized in Table 2. It was intended to sample different types of sources at a frequency based on the proportion of the study population using the source and the number of sources in each category. In practice, however, each of the boreholes, wells and springs was sampled approximately five times between January and December 1985. The city water supply was sampled approximately 138 times at various points in the distribution system. Samples of water were collected in sterile 500-ml polypropylene bottles, employing methods that took into account those of a typical user of the source. For example, taps or spouts were not sterilized before sample collection; however, a sterile bucket was used to collect water samples from open wells with no pumps. All samples were transported to the laboratory in insulated containers that were chilled with frozen refrigerant packs, refrigerated overnight, and analysed the next morn- ing. The water samples were examined for the four indicator bacteria using membrane filtration methods. For each indicator 1-ml, 10-ml, and 100-ml samples of water were passed through 47-mm diameter, 0.45 Am porosity cellulose filters.c Phos- phate-buffered saline (pH 7.5) was used as a diluent for the filtration of the 1-ml samples and as a rinse. Faecal coliforms were enumerated on mFC agard and faecal streptococci on KF streptococcus agar (J).c During the last 6 months of the study not all water samples were analysed for faecal strepto- cocci because of economic constraints and problems c GN-6, Metricel, Gelman Sciences, Ann Arbor, Ml, USA. d Difco Laboratories, Detroit, Ml, USA. with heavy background growth and atypical colo- nies. E. coli was enumerated on mTEC agar accord- ing to the method described by Dufour et al. (12), an approach that is particularly suitable for tropical waters since it singles out E. coli from the interfer- ing background growth. However, the method assumes that the majority of thermo-tolerant, non- E. coli coliforms are urease-positive, and thus can be easily differentiated; this may not be true for tropical waters. The technique has, nevertheless, been used successfully by Carillo et al. in Puerto Rico (9). Enterococci were enumerated on mE medium followed by testing for esculin hydrolysis on esculin -iron agar, as described by Levin et al. (24). This method has potential for the analysis of tropical waters because of its ability to differentiate entero- cocci (a more faecal-specific indicator group re- stricted to Streptococcus faecalis and S. faecium) from background growth. Statistical analysis of the data The statistical analysis examined whether there was a consistent and significant relationship between any of the indicator bacteria and the risk of diarrhoeal disease. Initially, estimates of the concentrations of these bacteria in each water sample were obtained from the number of colonies on the membrane filters. Their density per 100 ml was calculated from the filter colony counts of the three sample dilutions, as outlined below, assuming that the average density was independent of the volume of water filtered and that the number of colonies per plate followed a Poisson distribution (20). * If the number of colonies on one filter was in the countable range (approximately 10-100 colonies for faecal coliforms, E. coli, and faecal streptococci; and 10-80 for enterococci), the density was esti- mated from the number of colonies on that filter. Table 2: Types of drinking-water sources used by the study population, Cebu, Philippines No. of families No. of sources Springs 59 (8.6)8 34 Open dug wells 55 (8.0) 38 Wells with pumps 22 (3.2) 15 Boreholes 350 (50.7) 230 MCWDb 168 (24.3) 1 Other piped waterc 36 (5.2) 1 Total 690 (100.0) 319 a Figures in parentheses are percentages. b Municipal piped system; chlorinated groundwater. c Untreated piped system in one rural neighbourhood. WHO Bulletin OMS. Vol 69 1991. 307 C.L. Moe et al. * If more than one filter had counts in the count- able range or if all three filters had counts <10, the density was estimated using a maximum likelihood function in which the total number of colonies on these filters was divided by the total volume of water filtered (20). * If one or two filters had counts <10 and the other filter(s) had colonies too numerous to count, a maximum likelihood function was used to estimate the number of colonies on the latter filter(s) (20). * If for all three dilutions the colonies were too numerous to count, the concentration was assumed to be twice the maximum of the countable range of the highest dilution, i.e., 200 colonies per ml. Each estimate was given a reliability rating (A-D) based on the following criteria: the relation of the colony counts to the countable range; the amount of background growth; the progression of increased counts at higher volumes; and the colony characteristics (typical or atypical). The following proportions of the approximately 2000 water sam- ples in the study had high reliability ratings: E. coli (85%), enterococci (79%), faecal coliform (77%), and faecal streptococci (60%). Low ratings were assigned to estimates calculated from less than 10 colonies, filters with heavy background growth, or filters with atypical colonies. If a water source was sampled more than once in a 2-month period, the estimate with the best reliability rating was taken. Also, if in this period several measurements of a water source had the same reliability rating, the highest estimate was used, since it reflected the worst case potential exposure for the population using that source. Although this may bias the data towards higher estimates of pollution, such extremes are more likely to cause disease than minor day-to-day variations in water quality. In the USA there is evidence that unexpectedly heavy contamination of water supplies is responsible for the occurrence of many outbreaks of waterborne diseases (11). The data obtained in the study were analysed as a series of six cross-sectional studies, rather than as a follow-up study. Each cross-sectional analysis in- cluded information for all the study households that were interviewed and whose water source was sampled during a 2-month period. This analysis assumes that the occurrence of diarrhoea in each 2-month period is an independent event. However, each study child was followed for more than one 2-month period (depending on how often his or her water supply was sampled), and his or her risk of diarrhoea may have been influenced by diarrhoea morbidity in the previous 2-month period. This approach also assumes that the level of water pollution reflected by the concentration of indicator bacteria at the time of sampling is repre- sentative of the level that the study child was exposed to during a given 2-month period. For sources with highly variable water quality, this assumption has limited validity and may have misclassified the exposure of some study children. However, because this misclassification is likely to be independent of disease status it tends to bias the results conservatively by underestimating the effect of water quality on diarrhoea morbidity (23). In the analysis of the disease-exposure relation- ship, the outcome (diarrhoeal disease at the time of the longitudinal interview) was considered to be a dichotomous variable (presence or absence of dis- ease) for each study child at each 2-monthly inter- view. The exposure of interest was the concentration of indicator bacteria in the drinking-water sources used by the study families. This exposure was considered as follows: - as a continuous variable ranging from <1 per 100 ml to >20 000 per 100 ml; - as a categorical variable with initially five cate- gories (<1, 2-10, 11-100, 101-1000, and >1000 per 100 ml); and - as a categorical variable with two categories (<1000 and >1000 per 100 ml based on the results of the preliminary stratified analysis). When exposure was considered as a continuous variable, estimates of the indicator bacteria level per 100 ml were log1o-transformed to reduce skewness. Consideration of the risk factors for diarrhoea is complicated by its multiple transmission routes and by underlying determinants that affect behaviour and health. In our analysis the effects of age, gender, urban or rural environment, and season on diarrhoeal disease were examined. Any potential confounding by the type of water source was analysed by stratifying for this. These results have been described elsewhere." Results Water quality Heavy background growth on membrane filters was a frequent problem in the analysis of faecal coliform and faecal streptococci that influenced the reliability of the data. The greater proportion of reliable estimates of E. coli and enterococci concentrations reflects fewer problems with background growth and atypical colonies. The mTEC and mE methods for * See footnote b, p. 306. 308 WHO Bulletin OMS. Vol 69 1991. Bacterial indicators of diarrhoea in the Philippines E. coli and enterococci, respectively, were advan- tageous because they allowed a clearer separation of the target organism. Random indole tests for pre- sumptive E. coli on the mTEC plates were usually positive, indicating that the method was reasonably specific. The temperature of 37 water samples measured in the field ranged from 24 °C to 30 °C (mean, about 27 °C). Table 3 shows the distribution of the densities of indicator bacteria by type of water source for those samples that were used in the disease-ex- posure analysis, i.e., one sample per water source per 2-month period. All the indicators exhibited the same general distribution pattern in terms of the overall quality of different types of water sources. For example, more samples (33-40%) of spring- water were in the category 101-1000 indicators per 100 ml than in any other density category for all four indicator bacteria. Boreholes, which ranged in depth from about 12 m to 37 m, were generally better protected sources and usually provided water of relatively good quality. More borehole samples (38-77%) had <1 indicator organism per 100 ml for all four indicators. A greater proportion of borehole sam- ples had higher densities of enterococci and faecal streptococci than of faecal coliforms and E. coli; this may reflect the longer survivability in water of enterococci and some species of faecal streptococci (27). Compared with open dug wells, dug wells with pumps had a greater proportion of samples in the two lowest indicator density categories (<1 and 2-10 indicators per 100 ml). At least 65% of water samples from open dug wells had faecal coliform, enterococci, and faecal streptococci densities of >100 indicators per 100 ml. E. coli densities were somewhat lower. The high contamination level of the majority of water samples from wells is not surprising in view of the geology of the study area, the construction of the wells, and local sanitation practices. Most of the wells were relatively shallow (<5 m deep) and either unprotected or only poorly protected. Parts of Cebu island consist of porous marine limestone, which results in rapid groundwater flow rates. The siting of pit latrines and communal bathing and laundry activities near water supplies creates conditions with a high potential for groundwater contamination. Springs tended to be less contaminated than open dug wells. For all four indicator bacteria, more than 50% of water samples from springs had >100 indicators per 100 ml. Most springs were unpro- tected from surface run-off. The small, untreated, piped system that served one rural barangay did not provide high quality water. All samples had at least 11-100 indicators per 100 ml and several had >1000 indicators per 100 ml. Also, the municipal piped system (MCWD) did not deliver water of consistently high quality. Table 4 shows the distribution of all samples taken from this system, according to the density of indicator bacteria. Frequent fluctuations in pressure and leaky pipes resulted in contamination in areas where water pipes lay in open gutters. The majority of samples had <1 indicator per 100 ml for faecal coliforms, E. coli, and enterococci; however, depending on the microorganism, 4-13% of the samples had >100 indicators per 100 ml. The MCWD system was sampled in several city barangays in each 2-month period. Families who reported using the system, either through a standpipe or a direct connection in their home, were assumed to have been exposed to the indicator density measured in MCWD samples from their barangay. Table 3: Percentage of water samples in the indicator density categories 0-4, according to water source type Faecal coliformsa Escherichia colia Enterococci Faecal streptococci Type of water source nb 0 1 2 3 4 nb 0 1 2 3 4 nb 0 1 2 3 4 nb 0 1 2 3 4 Springs 123 12 9 25 33 21 123 13 14 21 33 20 123 5 7 30 40 18 75 19 8 23 35 16 Open dug 131 2 7 26 44 21 129 3 14 40 30 13 127 0 6 25 43 26 75 16 8 11 40 25 wells Wells with 52 14 17 25 31 14 52 19 21 21 31 8 49 8 16 37 22 16 31 32 3 36 13 16 pumps Boreholes 751 74 1 1 5 4 6 748 77 1 1 5 3 4 738 65 20 8 4 3 485 38 32 18 7 5 Other piped 5 0 0 20 20 60 5 0 0 40 20 40 5 0 0 20 40 40 2 0 0 50 0 50 water a 0= one indicator per 100 ml; 1 = 2-10 indicators per 100 ml; 2 =11-100 indicators per 100 ml; 3 =101-1000 indicators per 100 ml; 4 = > 1000 indicators per 100 ml. b n = number of water samples WHO Bulletin OMS. Vol 69 1991. 309 C.L. Moe et al. Table 4: Distribution of water samples from the municipal piped system (MCWD), by indicator density % of samples with indicator density per 100 ml of: No. of Indicator samples < 1 2-10 11-100 101-1000 > 1000 Faecal coliforms 138 73 12 3 8 5 Escherichia coli 130 80 8 4 6 2 Enterococci 138 69 20 7 4 1 Faecal streptococci 74 42 30 24 1 3 Correlations between bacterial indicators The correlation coefficients between the log1o-con- centrations of the bacterial indicators in all the water supplies studied are shown in Table 5 for each 2- month period and for all time periods combined. All four indicators correlated reasonably well with each other. Not surprisingly, the correlation was strongest between faecal coliforms and E. coli (a subgroup of faecal coliforms) (r = 0.85). Entero- cocci (a subgroup of faecal streptococci) were also well correlated with faecal streptococci (r = 0.72). The correlations between all the indicator bacteria decreased in May and June, but this did not coincide with any substantial change in rainfall or tem- perature. Prevalence of diarrhoea No significant seasonal differences in the prevalence of diarrhoea were observed (P = 0.124) (Table 6). Age-specific seasonal analysis also showed no signi- ficant differences. This may reflect the lack of extreme seasonal differences in rainfall in Cebu or suggests that the dominant transmission routes of diarrhoeal disease in this study were not very sensitive to the effects of rainfall. An analysis of the prevalence of diarrhoea by age among 10-12-month-olds, 13-18-month-olds, and 19-24-month-olds revealed no significant differ- ences. However, children aged about 14 months or younger had slightly higher diarrhoea rates than those aged 19 months or older. Higher rates of diarrhoea may have been associated with weaning, but we did not investigate this. There were no significant differences in the prevalence of diarrhoea between males and females. Disease-exposure relationship Table 7 shows the distribution of diarrhoea cases over five exposure categories of the density of the indicator bacteria. The lowest exposure category (<1 indicator per 100 ml water) meets WHO guide- lines for faecal coliforms in drinking-water. Because no significant seasonal differences in illness rates were observed, Table 7 shows combined data for diarrhoea cases and children at risk for all six study periods. The pattern of disease rates for the first four exposure levels was similar for all four indicators but somewhat greater disease rates were associated with the highest category (>1000 indicators per 100 ml). Table 5: Correlations between log,,-indicator densities per 100 ml, by 2-month interval, and for all study intervals combined Correlation coefficient (r) for the period: Combined Jan.-Feb. Mar.-Apr. May-June July-Aug. Sept.-Oct. Nov.-Dec. intervals Faecal coliforms versus: Escherichia coli 0.89 0.85 0.76 0.92 0.81 0.92 0.85 Enterococci 0.81 0.74 0.44 0.84 0.64 0.70 0.68 Faecal streptococci 0.48 0.70 0.50 0.65 0.59 - a 0.58 E. coli versus: Enterococci 0.77 0.80 0.48 0.85 0.75 0.76 0.74 Faecal streptococci 0.46 0.74 0.54 0.71 0.62 -a 0.63 Enterococci versus: Faecal streptococci 0.58 0.82 0.60 0.85 0.72 _a 0.72 a No faecal streptococci analyses were performed in Nov.-Dec. 310 WHO Bulletin OMS. Vol 69 1991. Bacterial indicators of diarrhoea in the Philippines Table 6: Prevalence of diarrhoea among the study children by 2-month periods, Cebu island, Philippines Jan.- Mar.- May- July- Sept.- Nov.- Feb. Apri. June Aug. Oct. Dec. No. of cases of 40 62 65 49 44 17 diarrhoea No. of children 555 646 651 577 480 324 interviewed % prevalence 7.2 9.6 10.0 8.5 9.2 5.2 These results suggest there was an exposure thresh- old effect on disease rates for all four indicators. To further analysis this effect, we defined two exposure categories: low (91000 indicators per 100 ml); and high (>1000 indicators per 100 ml). Even though no significant time effect was observed, the Mantel-Haenszel x2 test for association was repeated, treating time as a stratification factor. This analysis tests for overall association and gives a corresponding overall P-value and estimate of over- all effect (risk ratio). Table 8 shows the prevalence of diarrhoea associated with low and high exposure for all four indicators during the entire study period, and summarizes the results of these analyses. All four indicators showed a positive relation- ship between exposure and disease. For E. coli and enterococci the overall P-values when time was controlled (E. coli, P = 0.02; enterococci, P = 0.03) were significant at the 0.05 level. For faecal strepto- cocci the overall P-value (0.08) was not significant at the 0.05 level, and the 95% confidence interval (CI) of the risk ratio (0.95-2.28) includes unity. Faecal coliforms had the smallest overall risk ratio and were not significant predictors of diarrhoeal disease at the 0.05 level. When the density of E. coli was taken as a predictor, the relative risk of diarrhoeal illness for a child in the high exposure group was 1.52, i.e., 52% greater than that for a child in the low exposure group. The 95% CI of the risk ratio (1.06-2.18) does not include unity. For enterococci exposure levels, children in the high group had a relative risk of 1.56 for diarrhoeal illness (95% CI: 1.04-2.33), i.e., 56% greater than that of children in the low category. The results of the Breslow-Day test for interac- tion (23), i.e., the homogeneity of risk ratios over time, showed that for all four indicators the relation- ship between exposure and disease was relatively consistent over all six periods. Stratified analysis of the relationship between water quality and risk of "highly credible diar- rhoea", i.e., those cases from which a diarrhoea pathogen was isolated from the rectal swab (34% of all diarrhoea cases) indicated that faecal coliforms and faecal streptococci were again not significant predictors of risk of diarrhoeal disease (Table 8). E. coli were marginally significant when the data were combined over time (P = 0.06), but not when time was included as a stratification variable (P = 0.12). Enterococci were highly significant predictors of illness both in the combined analysis (P = 0.005) and when time was controlled (P = 0.01). For enterococci, the overall risk ratio was 2.13, i.e., children in the high exposure group had a 113% greater risk of diarrhoeal disease than those in the low exposure group. Logistic regression analysis, as described by Kleinbaum et al. (23), was used to examine the disease-exposure relationship after controlling for the effects of age, gender, and urban/rural environ- ment. Although age and gender were not statisti- cally significant confounders, controlling for them increased the precision of the overall estimate of effect-the risk odds ratio. The urban/rural environ- ment was controlled because of evidence of mild confounding. Two logistic models were examined for each indicator microorganism. In the first model, the logarithm of the odds of having diarrhoea was Table 7: Stratified analysis of disease-exposure relationship for five exposure levels of the indicator bacteria for all time periods combined Indicator density No. of diarrhoea cases/ per 100 ml population at risk Faecal coliforms: < 1 106/1139 (9.3)a 2-10 16/204 (7.8) 11-100 16/196 (8.2) 101-1000 26/279 (9.3) > 1000 36/296 (12.2) Escherichia coli: < 1 104/1125 (9.2) 2-10 13/196 (6.6) 11-100 20/235 (8.5) 101-1000 25/248 (10.1) > 1000 33/220 (15.0) Enterococci: < 1 86/1001 (8.6) 2-10 37/339 (10.9) 11-100 26/281 (9.3) 101-1000 26/297 (8.8) > 1000 25/175 (14.3) Faecal streptococci: < 1 32/373 (8.6) 2-10 41/407 (10.1) 11-100 27/285 (9.5) 101-1000 13/141 (9.2) > 1000 22/149 (14.8) a Figures in parentheses are percentages. WHO Bulletin OMS. Vol 69 1991. 311 C.L. Moe et al. Table 8: Results of a stratified analysis of the disease-exposure and highly credible disease-exposure relationship using two exposure levels Disease exposure Highly credible disease exposure Faecal Escherichia Faecal Faecal Escherichia Faecal coliforms coli Enterococci streptococci coliforms cofi Enterococci streptococci Low exposurea No. of diarrhoea cases 164 162 175 113 63 64 64 44 Population at risk 1818 1804 1918 1206 1717 1706 1807 1137 Diarrhoea prevalence (%) 9.0 9.0 9.1 9.4 3.7 3.8 3.5 3.9 High exposure" No. of diarrhoea cases 36 33 25 22 14 13 13 8 Population at risk 296 220 175 149 274 200 163 135 Diarrhoea prevalence (%) 12.2 15.0 14.3 14.8 5.1 6.5 8.0 5.9 No. of peoplec 2114 2024 2093 1355 1991 1906 1970 1272 P-value (Mantel-Haenszel x2 test; 0.087 0.004 0.026 0.038 0.062 0.005 0.254 combined over time) 0.251 Overall P-value (Cochran-Mantel-Haenszel 0.285 0.023 0.032 0.084 0.448 0.117 0.01 0.354 X2 test; controlling for time) Overall risk ratiod 1.22 1.52 1.56 1.47 1.26 1.61 2.13 1.44 (0.85-1.74) (1.06-2.18) (1.04-2.33) (0.95-2.28) (0.69-2.30) (0.89-2.91) (1.20-3.80) (0.68-3.07) P-value (Breslow-Day test) 0.441 0.667 0.436 0.885 0.685 0.891 0.265 0.750 a <1000 indicators per 100 ml. b> 1000 indicators per 1 00 ml. cNo. of people who were interviewed and had their water source sampled during the study period. dFigures in parentheses are 95% confidence intervals. regressed against a continuous exposure variable (logl of the indicator density per 100 ml). In the second model, a dichotomous exposure variable (1 for >1000 indicators per 100 ml; 0 for <1000 indicators per 100 ml) was used. Both models used data combined from all six periods. The estimated model parameters for all four indicator bacteria are shown in Table 9. A better fit to the data was obtained with the categorical model since it reflects the threshold effect of exposure on the risk of diarrhoeal disease. For this model the odds ratios for being exposed to >1000 indicators per 100 ml were 1.49, 1.92, 1.94, and 1.81, respect- ively, for faecal coliforms, E. coli, enterococci, and faecal streptococci. E. coli and enterococci were strongly significant predictors (P < 0.01) and faecal streptococci were also significant predictors (P < 0.05) of disease risk. Faecal coliforms were marginally significant (P = 0.053). However, the 95% confidence intervals of the odds ratios for all four indicators overlap. The results of the logistic regression analysis support the conclusions drawn from the stratified analysis (Table 8). The odds ratios from the logistic regression analysis are similar to the risk ratios calculated in the stratified analysis and the 95% CI of these estimates overlap. The odds ratios were used to predict the probability of a typical member of the study popula- tion (an 18-month-old female living in an urban barangay) having diarrhoea (Table 9). The results indicate that drinking-water with >1000 E. coli per 100 ml produced a 15% chance of having diarrhoea on any given day, compared with a 9% chance for drinking less polluted water. Consumption of the more contaminated water was associated with a 1.9 times greater probability of having diarrhoeal dis- ease than that of drinking-water with <1000 E. coli per 100 ml. Discussion Relationship between water quality and diarrhoeal disease Our results have important implications for asses- sing and attempting to control waterborne diar- rhoeal illness in tropical developing countries. The prevalence of diarrhoea among children who drank good water (<1 E. coli per 100 ml) was about the same as that among children exposed to moderately contaminated water (2-100 E. coli per 100 ml). When water quality is good or moderate, WHO Bulletin OMS. Vol 69 1991.312 Bacterial Indicators of diarrhoea in the Philippines Table 9: Estimated parameters for logistic regression models using continuous or categorical exposure variablesa Faecal coliforms Escherichia coli Enterococci Faecal streptococci (n = 2114) (n = 2024) (n = 2093) (n = 1355) Continuous Categorical Continuous Categorical Continuous Categorical Continuous Categorical Intercept -2.18 -2.12 -2.29 -2.15 -2.13 -2.09 -2.45 -2.32 Slope 0.10 0.40 0.16 0.65 0.13 0.66 0.17 0.60 Standard error of slope 0.05 0.20 0.06 0.21 0.07 0.25 0.07 0.26 P-value 0.060 0.053 0.004 0.002 0.063 0.007 0.021 0.020 Odds ratio 1.10 1.49 1.18 1.92 1.14 1.94 1.19 1.81 (1.00-1.23)b (1.00-2.22) (1.05-1.31) (1.27-2.91) (0.99-1.31) (1.20-3.16) (1.03-1.37) (1.10-3.00) Probability of diarrhoea in 24 hoursc Low exposured 0.08 0.08 0.08 0.09 0.08 0.09 0.08 0.09 High exposure" 0.12 0.12 0.15 0.15 0.13 0.16 0.14 0.15 ' Data were combined over time, controlling for age, gender, and urban/rural environment. Figures in parentheses are 95% confidence intervals. c Calculated for an 18-month-old female in an urban barangay. d - 1000 indicators per 100 ml. @ > 1000 indicators per 100 ml. these findings suggest that other transmission routes of diarrhoeal disease are more important. This threshold effect of indicator density on risk of diarrhoeal illness tends to support Feachem's proposal for more flexible water quality guidelines for developing countries (14). Use of moderately contaminated water containing up to 1000 E. coli per 100 ml did not appear to cause an increased risk of diarrhoeal disease, and may be acceptable in areas where water treatment is not feasible and better quality sources are not accessible. Probably several transmission routes for diar- rhoeal disease were important in this study environ- ment. The households of the study population were typical of those in many developing countries in that few had running water and that water usually was stored in large containers. Excreta disposal facilities (generally pit latrines, water-sealed toilets, and flush toilets) were usually shared by large numbers of people. Also, many households did not have access to any such facilities and used open fields, river- banks, or seashores. Relatively poor personal hygiene and food preparation and storage practices were not uncom- mon in the study population, and may have con- tributed to disease transmission. Very few house- holds had refrigerators, and food was often cooked once a day and then eaten at various times thereafter with little or no reheating. Analysis of a common infant weaning food (rice pudding) left at room temperature revealed that it contained very high levels of bacteria by aerobic plate count. Most households in both urban and rural barangays had domestic animals (dogs, cats, poul- try, and pigs) living in close proximity in the yard or under the house. Direct or indirect contact with animal faecal material may be common and could result in transmission of diarrhoeal diseases with animal reservoirs, e.g., salmonella and campylo- bacter. The results of the larger studies that included the present study confirm the importance of several transmission routes of diarrhoeal pathogens in the first year of life (10). The methodology used in the present study permits only the investigation of a unidirectional relationship, i.e., it assumes that water quality can affect health but not vice versa. In the larger studies the possibility that the health status of a child is not only determined by water quality but that water use choices may be affected by the health status of the child was also investigated. The results indicate that among the urban popula- tion poor water quality, poor excreta disposal practices, poor food hygiene, and high community density were all significantly associated with an increased likelihood of diarrhoea in the first year of life. Breast-feeding appeared to exert a strong protective effect against diarrhoea, perhaps by decreasing exposure to foodborne pathogens and by providing immunological benefits. In our study, the type of water source used by each study family was examined to check possible confounding by socioeconomic status or geo- graphical location. The type of water source had no significant effect on the risk of diarrhoeal disease after controlling for the indicator density. f However, the density of E. coli was still a significant I See footnote b, p. 306. WHO Bulletin OMS. Vol 69 1991. 313 C.L. Moe et al. predictor of diarrhoeal disease after controlling for the type of water source. Type of water source may not be a good surrogate for socioeconomic status in this study population, since the municipal water supply was widely used regardless of socioeconomic status. Children in rural areas were more likely to be exposed to contaminated water, but those in the urban environment were generally at greater risk of developing diarrhoea than rural children for all indicator exposure categories. f Our results show that when the water supply was grossly contaminated (> 1000 E. coli per 100 ml), the rate of diarrhoeal disease was signifi- cantly higher, and it appears that in this situation water becomes a major source of exposure to faecal contamination and diarrhoea pathogens. The use of such highly contaminated water to prepare food may also increase foodborne transmission of diarrhoeal illness. In the Gambia, microbiological examination of weaning foods in various stages of preparation has confirmed that highly polluted water (1700-11 000 E. coli per 100 ml) was a major source of faecal contamination, which was not completely eradicated by cooking (4). Evidence for a threshold effect of water quality on diarrhoeal disease may explain some of the ambiguous results of previous studies of water quality and diarrhoeal disease in developing coun- tries. It is generally assumed that provision of a safe water supply results in a corresponding reduction in waterborne disease. However, in some developing countries the installation of improved water supplies in villages did not produce a consequent measurable decrease in diarrhoeal disease morbidity (35). Briscoe has proposed that the removal of a waterborne transmission route might not result in a substantial reduction in the rate of diarrhoeal disease because of the effects of other routes, and that improvements in water quality without concurr- ent improvements in sanitation and excreta disposal practices are not sufficient to significantly reduce the incidence of diarrhoea (6). This is supported by the findings of one study in the Philippines on the health impact of an improved water supply (piped water system) in two cities. Improved water quality had a positive impact on health, as indicated by the incidence of diarrhoea and the nutritional status of children aged 0-4 years in one city, but only among upper-income households, which already had ade- quate sanitary facilities and sound hygiene practices. Installation of an improved water source alone was therefore not sufficient to produce improved health (25). Also, two recent case-control studies of water supply, sanitation, and diarrhoea in the Philippines (2) and in Malawi9 suggest that improved water supply in itself is not enough to reduce the incidence of diarrhoeal disease in children under 5 years of age; however, the combined effect of improved water and sanitation was associated with a reduction in diarrhoea morbidity. Our results suggest that if a population is consuming moderately contaminated water (perhaps up to 100 E. coli per 100 ml), the introduction of an improved water supply that meets WHO guidelines (0 faecal coliforms per 100 ml) may not result in an appreciable reduction in diarrhoeal illness rates. However, if a population is exposed to a grossly contaminated water supply (>1000 E. coli per 100 ml), modest improvements in water quality to reduce E. coli levels by tenfold or more may significantly reduce diarrhoeal disease rates. In developing countries it may therefore not be greatly beneficial to health or cost-effective to use limited financial resources to provide high quality water supplies in the absence of other improvements in hygiene and sanitation. Instead, low-cost improve- ments to existing water supplies should be made to eliminate grossly contaminated water sources, and consideration should be given to improving sanita- tion and excreta disposal facilities to help eliminate other environmental transmission routes of diar- rhoeal diseases. Comparison of the indicator bacteria Although the densities of all four bacterial indicators showed a positive relationship with the prevalence of diarrhoea, because of laboratory methodology some were more useful than others. In accord with the findings of Cabelli et al. on recreational waters (8), we found that E. coli and enterococci appear to be better predictors than faecal coliforms of the risk of waterborne gastrointestinal illness. These results further support the contention that E. coli and enterococci are more faecal-specific indicators than faecal coliforms and possibly faecal streptococci (8, 24). E. coli and enterococci may correlate signifi- cantly with the rates of diarrhoeal disease because the methods used to enumerate them are more specific and less subject to interference from back- ground growth. Although faecal streptococci were significant predictors of disease in some of the analyses, substantial problems with background growth and atypical colonies detract from their value as indicator bacteria for tropical waters. Our results support the use of E. coli and enterococci as indicators of the quality of tropical 9 Young, B. & Briscoe, J. Water and health in rural Malawi: aspects of the performance, utilization and health impact of the Malawi self-help rural water supply project. USAID report, 1986. 314 WHO Bulletin OMS. Vol 691991. Bacterial indicators of diarrhoea in the Philippines freshwaters. The observed ability of E. coli to predict the risk of waterborne illness contrasts somewhat with the observations of Hazen that E. coli may be indigenous to tropical aquatic environ- ments (21). However, if low levels of E. coli are indigenous to tropical waters, this may explain why only high densities of E. coli, where perhaps the majority are from faecal contamination, caused an increased risk of diarrhoea. It should be noted that the threshold effect of indicator density on the prevalence of diarrhoea was observed for all four indicator bacteria. Low levels of any of these indicators might therefore reflect indigenous populations in the environment and not faecal contamination. Nonfaecal sources of faecal coliforms, some enterococci (S. faecalis var. liquefa- ciens), and faecal streptococci have been reported (17, 18, 21). The possibility that these indicators have nonfaecal sources should be further investig- ated in tropical regions where human and animal faecal contamination are absent. In tropical countries the use of conventional indicators such as total or faecal coliforms is more likely to result in misclassification of some water sources as contaminated, in instances where high indicator levels may not reflect faecal contamination and a risk of waterborne illness. In such situations the effect of improved water quality on diarrhoeal diseases may be seriously underestimated. This may partly explain why several studies that relied on total or faecal coliforms to evaluate water quality found no differences in diarrhoeal disease rates between those exposed to "good" versus "poor" quality water (7,15, 25, 26, 29, 32, 35). Our results demon- strate the importance of choosing an appropriate indicator of water quality in investigations of the effect of water quality on diarrhoea morbidity. Acknowledgements We thank Dr W. Flieger, Dr H. VanEngelen, Dr P. Echever- ria, Dr C. Haas, J. Vanderslice, E. Walag, F. Walag, A. Lim, Dr B. Popkin, and Dr J. Briscoe for their collaboration and assistance. We are also grateful to T. Bale, M. Tiu, A. Quijada, R. Quijada, and V. Ligutom for technical assis- tance. This work was partly supported by the Ford Founda- tion (grant no. 840-0544). Resume Indicateurs bacteriens du risque de maladie diarrheique due a l'eau de boisson aux Philip- pines Dans de nombreux pays en developpement, l'inci- dence elevee des maladies transmises par l'eau et l'utilisation tres repandue de sources d'eau non trai- tees et souvent fortement polluees rendent l'6valua- tion pr6cise de la contamination f6cale de l'eau particulierement importante. II est n6cessaire de dis- poser a la fois de mesures de la qualite de l'eau qui permettent de predire le risque de transmission de maladies, et de directives r6alistes concernant la qualite de l'eau qui soient applicables aux problemes particuliers des pays en developpement. De nombreuses etudes ont examine quelle etait la relation entre l'approvisionnement en eau et les maladies diarrh6iques; cependant, la plupart de ces 6tudes ne mesuraient pas la qualite bacteriologique de l'eau ou utilisaient des indicateurs bacteriens qui ne constituaient peut-etre pas une mesure appro- pri6e du risque pour les eaux tropicales. Les sources d'eau dans les pays tropicaux sont caract6risees par des temperatures 6lev6es et des taux 6lev6s de nutriments, conditions qui influencent la survie et la multiplication extra-enterique des bacteries indica- trices et favorisent la croissance de bact6ries thermo- tolerantes non f6cales, qui peuvent modifier les r6sul- tats de I'analyse. Nous avons utilise une approche 6pid6miologi- que pour evaluer quatre indicateurs bacteriens de la qualite d'eaux de boisson tropicales (coliformes f6c- aux, Escherichia coli, ent6rocoques et streptocoques fecaux) en examinant les relations entre leur concen- tration dans l'eau de boisson et la prevalence de la diarrh6e dans la population consommant cette eau. Pendant une periode d'un an, des echantillons ont ete r6gulierement recueillis dans 319 sources d'eau de boisson (sources, puits, forages et canalisations) utilisees par une population d'6tude de 690 enfants (ages de 6 a 24 mois) et analyses pour recherche des quatre indicateurs bact6riens. La pr6valence de la diarrhee a ete mesuree tous les deux mois, et des 6couvillonnages rectaux, pratiqu6s chez les cas de diarrhees et chez des enfants sains temoins, ont e analyses pour savoir quelles 6taient les etiologies de la diarrhee et les taux d'infection asymptomatique. Une m6thode avec analyse stratifiee et r6gression logistique a e employee pour examiner la relation entre la qualit6 de l'eau de boisson et la prevalence de la diarrh6e. Peu de diff6rences ont ete observees entre les taux de maladie chez les enfants qui avaient bu de l'eau de bonne qualite (definie suivant les normes de l'OMS) et ceux qui avaient bu de l'eau moderement contaminee. Cependant, les enfants exposes a de l'eau fortement contamin6e (>1000 E. coli par 100 ml) avaient un risque de diarrhee superieur de 52% a ce qu'il etait avec de l'eau moins contaminee. Cet effet de seuil de la qualite de l'eau sur les maladies diarrheiques laisse a penser que lorsque la WHO Bulletin OMS. Vol 69 1991. 315 C.L. Moe et al. qualite de l'eau est bonne ou moyenne dans les pays en developpement, d'autres voies de transmis- sion des maladies diarrheiques peuvent jouer un plus grand r6le. Cependant, un approvisionnement en eau fortement contaminee est dans tous les cas une source majeure d'exposition a la contamination feca- le et aux agents pathogenes de la diarrh6e. Ces resultats confirment l'importance de choisir un indicateur approprie pour evaluer le risque de maladie diarrh6ique associee a la consommation d'eau contamin6e par des pathogenes f6caux. Bien que la concentration des quatre indicateurs bacte- riens ait fait preuve d'une relation positive avec la pr6valence de la diarrh6e, E. cofi et les ent6roco- ques 6taient de meilleurs pr6dicteurs du risque de maladie transmise par l'eau. Les m6thodes de nume- ration de E. cofi et des enterocoques donnaient de meilleurs resultats dans des conditions tropicales qu'avec les coliformes et les streptocoques fecaux. Ces observations permettent de penser que dans les pays en developpement, des ameliorations meme modestes apportees aux approvisionnements d'eau fortement contaminee peuvent sensiblement reduire la prevalence des maladies diarrheiques dans la population. Quand les ressources economi- ques sont limitees, des ameliorations peu couteuses doivent etre apportees aux approvisionnements d'eau pour eliminer les sources fortement contami- nees, et il faut s'attacher a ameliorer l'hygiene et l'6vacuation des excr6ments pour parvenir a eliminer d'autres sources de transmission presentes dans 'environnement. References 1. Standard methods for the examination of water and wastewater, 14th edition. Washington, DC, American Public Health Association, 1980. 2. Baltazar, J. et al. Can the case-control method be used to assess the impact of water supply and sanita- tion on diarrhoea? A study in the Philippines. Bulletin of the World Health Organization, 66: 627-635 (1988). 3. Barrell, R.A.E. & Rowland, M.G.M. The relationship between rainfall and well water pollution in a West African (Gambian) village. Journal of hygiene, 83: 143-150 (1979). 4. Barrell, R.A.E. & Rowland, M.G.M. Commercial milk products and indigenous weaning foods in a rural West African environment: a bacteriological perspective. Joumal of hygiene, 84: 191-202 (1980). 5. Blum, D. & Feachem, R.G. Measuring the impact of water supply and sanitation investments on diarrheal diseases: problems of methodology. International journal of epidemiology, 12: 357-365 (1983). 6. Briscoe, J. Intervention studies and the definition of dominant transmission routes. American journal of epi- demiology, 120: 449-455 (1984). 7. Bruch, H.A. et al. Studies of diarrhoeal disease in Central America. V. Environmental factors in the origin and transmission of acute diarrheal disease in four Guatemalan villages. American journal of tropical medi- cine and hygiene, 12: 567-579 (1963). 8. Cabelli, V.J. et al. A marine recreational water quality criterion consistent with indicator concepts and risk analysis. Joumal of the Water Pollution Control Federa- tion, 55:1306-1314 (1983). 9. Carrillo, M. et al. Survival and enumeration of the fecal indicators Bifidobacterium adolescentis and Escherichia coli in a tropical rainforest watershed. Applied and environmental microbiology, 50: 468-476 (1985). 10. Cebu Study Team. Underlying and proximate determi- nants of child health: the Cebu Longitudinal Health and Nutrition Study. American journal of epidemiology, 133: 185-201 (1991). 11. Craun, G.F. Waterborne diseases in the United States. Boca Raton, FL, CRC Press, 1986. 12. Dufour, A.P. et al. Membrane filter method for enumer- ating Escherichia coli. Applied and environmental micro- biology, 41:1152-1158 (1981). 13. Esrey, S.A. et al. Interventions for the control of diarrhoeal diseases among young children: improving water supplies and excreta disposal facilities. Bulletin of the World Health Organization, 63: 757-772 (1985). 14. Feachem, R.G. Water supplies for low-income commu- nities: resource allocation, planning and design for a crisis situation. In: Feachem, R. et al., ed. Water, wastes and health in hot climates. New York, Wiley, 1977, pp. 75-95. 15. Feachem, R.G. et al. Water health and development: an interdisciplinary evaluation. London, Tri-Med Books, 1978. 16. Feachem, R.G. Bacterial standards for drinking water quality in developing countries. Lancet, 2: 255-256 (1980). 17. Feachem, R.G. et al. Sanitation and disease: health as- pects of excreta and wastewater management. Washing- ton, DC, World Bank, 1983, pp. 206-242. 18. Fujioka, R.S. et al. Naturally occurring coliforms and enterococci in freshwaters of Hawaii. Toxicity assess- ment, 3: 613-630 (1988). 19. Guerrant, R.L. et al. Prospective study of diarrheal illness in northeastern Brazil: patterns of disease, nutri- tional impact, etiologies and risk factors. Journal of infectious diseases, 148: 986-997 (1983). 20. Haas, C. & Heller, B. Averaging of TNTC counts. Applied and environmental microbiology, 54: 2069- 2072 (1988). 21. Hazen, T.C. Fecal coliforms as indicators in tropical waters: a review. Toxicity assessment, 3: 461-477 (1988). 22. Hebert, J. Effects of water quality and water quantity on nutritional status: findings from a south Indian com- munity. Bulletin of the World Health Organization, 63: 143-155 (1985). 23. Kleinbaum, D.G. et al. Epidemiologic research: princi- ples and quantitative methods. Belmont, CA, Lifetime Learning Publications, 1982. 24. Levin, M.A. et al. Membrane filter technique for enumeration of enterococci in marine waters. Applied microbiology, 30: 66-71 (1975). 316 WHO Bulletin OMS. Vol 691991. Bacterial indicators of diarrhoea in the Philippines 25. Levine, R.J. et al. Failure of sanitary wells to protect against cholera and other diarrhoeas in Bangladesh. Lancet, 2: 86-89 (1976). 26. Magnanl, R. et al. Evaluation of the provincial water project in the Philippines: final report. Washington, DC, International Statistical Programs Center, Bureau of the Census, U.S. Department of Commerce, 1984. 27. McFeters, G.A. et al. Comparative survival of indicator bacteria and enteric pathogens in well water. Applied microbiology, 27: 823-829 (1974). 28. McJunkin, F.E. Water and human health. Washington, DC, U.S. Agency for International Development, 1982. 29. Moore, H.A. Diarrheal disease studies in Costa Rica. IV. The influence of sanitation upon the prevalence of intestinal infection and diarrheal disease. American joumal of eidemiology, 82: 162-184 (1965). 30. Petersen, N.J. & Hines, V.D. The relation of summer- time gastrointestinal illness to the sanitary quality of the water supplies in six Rocky Mountain communities. American journal of hygiene, 71: 314-320 (1960). 31. Pournadeali, E. & Tayback, M. Potable water and village health: is primary prevention affordable? Public health reports, 95: 291-294 (1980). 32. Rajasekaran, P. et al. Impact of water supply on the incidence of diarrhoea and shigellosis among children in rural communities in Madurai. Indian journal of medical research, 66: 189-199 (1977). 33. Rowland, M.G.M. Epidemiology of childhood diarrhea in the Gambia. In: Chem, L.C. & Scrimshaw, N.S., ed. Diarrhea and malnutrition. New York, Plenum Press, 1983, pp. 87-98. 34. Schleissmann, D.J. et al. Relation of environmental factors to the occurrence of enteric diseases in areas of eastern Kentucky. Public Health Monograph No. 54, 1958 (Public Health Services Publication No. 591); also published in Public health reports, Vol. 73 (1 1), November 1958. 35. Shiffman, M.A. et al. Field studies on water, sanitation and health education in relation to health status in Central America. Progress in water technology, 11: 143-150 (1978). 36. Snyder, J.D. & Merson, M.H. The magnitude of the global problem of acute diarrhoeal disease: a review of active surveillance data. Bulletin of the World Health Organization, 60: 605-613 (1982). 37. Trivedi, B.K. et al. Bacteriological water quality and incidence of waterborne diseases in a rural population. Indian joumal of medical science, 25: 795-801 (1971). 38. Guidelines for drinking-water quality. Vol. 1: Recom- mendations. Geneva, World Health Organization, 1984, pp. 1-39. 39. Wright, R.C. A medium for rapid enumeration of E. coli in the presence of other faecal coliforms in tropical waters. Journal of hygiene, 88: 265-274 (1982). WHO Bulletin OMS. Vol 69 1991. 317