Effects of water quality and water quantity on nutritional status: findings from a south Indian community

Bulletin ofthe World Health Organization, 63 (1): 143 - 155 (1985) © World Health Organization 1985 Effects of water quality and water quantity on nutritional status: findings from a south Indian community JAMES R. HEBERT I Quantitative assessments of the relative effects on health of various aspects of water supply are virtually absentfrom the literature. Despite the lack of information, resources are being allocated throughout the developing world, for projects related to water and sanitation. The present study was designed specifically to overcome many of the methodo- logical problems that other researchers have faced. Data were collected concerning the nutritional status of 627 children in three urban communities in South India. Information was also collected on water quality, water quantity, household sanitation, socioeconomic conditions, and housing. A statistical technique is presented that allows for controlling potential confoundingfactors in the analyses. The results, in general, indicate that atyoung ages (i.e., under 3 years old) water quality is relatively more important as a determinant of nutritional status, while at older ages water quantity is relatively more important. With the designation of the 1980s as the United Nations Water and Sanitation Decade (1), water and sanitation improvements have become a priority in most developing countries. There remain, however, many unanswered questions concerning the relative effects of the various factors that influence sani- tation. Geographical location, seasonal and political factors, water distribution patterns, standards of living, and actual water usage behaviours must be taken into consideration when evaluating the relative importance of the quality versus the quantity of a water supply. The object of the present study is to evaluate the relative effects of the quantity or avail- ability of a water supply and its quality on the nutritional and health status of children. Several reviews in the literature discuss these problems in terms of the relative effects of the various characteristics of a water supply (2, 3 ).' Three under- lying issues, however, have made it virtually impossible to draw definite conclusions, even within the context of particular study situations. The first of these is the control of confounding factors, both in the design of studies and in the analyses (4-6). The ' Nutrition Programme Officer, UNICEF, Kampala, Uganda. Address for reprints: 10 Crescent St., Somerville, MA 02145, USA. a Report of a WHO Scientific Working Group on Environmental Health and Diarrhoeal Disease Prevention, Kuala Lumpur, 1979 (WHO unpublished document, DDC/80.5, 1980). second is concerned with selection of an estimator of health status that is not prone to severe recall bias and different interpretations (6, 7). The last involves measurement of all the sanitation-related variables of potential interest, including those related to water usage. The characteristics of a water supply are nearly always associated with the factors that affect health status (8).b These characteristics are also related to one another (9, 10), so it is important to describe them accurately so as to assess which aspect of water supply is responsible for a given effect. Since the character- istics of a water supply are themselves alleged to be affectors of health status, the issue of confounding must be considered (11). If confounding is ignored, as it usually is (2), then the results of such a study are doubtful. To compensate for the difficulty of proper and consistent field reporting of morbidity, there has been increased interest in using anthropometry to estimate a child's health status (12, 13).C The anthropometric measurements of children are easily monitored and reliable, if done correctly, and they can easily be b ISELY, R. B. Relationships among improved water supply and sanitation, diarrhoeal morbidity, and nutritional status. Unpublished report of a Working Meeting on 17 December 1982, Washington, DC. c RAHAMAN, M. M. Research protocolfor water and sanitation intervention project, Teknaf, Bangladesh. Unpublished report of the International Center for Diarrhoeal Disease, Dhaka, 1980. 4511 -143- J. R. HEBERT compared with reference values that have been well described and are available to a wide audience. There have been two main objections to the use of anthropometric indices to define health and/or nutri- tional status. One is that the various ethnic groups have different genetic potential to grow so that use of comparators from groups with different genetic make-up may be inappropriate (14, 15). The tendency of children from well-nourished, affluent groups in the less developed countries and of immigrants to the developed countries to grow to the level of US children, however, argues strongly against the importance of genetic factors (16, 17). Also, the refer- ence values may be used, as they are here, within a genetically homogeneous group for purposes of comparison only. The second objection to the use of anthropometric indices is that growth per se may not be of critical importance as a health outcome. Human growth is the product of genetic, metabolic, endocrine, and environmental factors. It is an extremely complex phenomenon for which only a few causal linkages are known with any certainty. While it is recognized that large size is not necessarily a desirable condition, numerous studies show that children who are ill or have a poor diet over short periods of time may be light for their height or length (18, 19). Poor diet may make a child prone to illness (20, 21) whereas an ill child can suffer nutritionally through food wastage (22, 23), metabolic losses (24), or anorexia (25). Insufficient nutrients or diversion of nutrients due to illness can, in turn, cause deviations of growth from what would otherwise be expected. Children who have many acute episodes or protracted periods of illness or food deprivation may be stunted (i.e., have low height or length for their age) (26, 27). Stunting, however, may indicate either a status of equilibrium in which a child may be quite healthy despite being short for his age, or ongoing stunting which is associated with current nutritional or health prob- lems. The present study was designed to test the relative effects of water quality and water quantity on human anthropometric status. In view of the multiple inter- correlations of parameters important in determining health, it is imperative to account for many variables simultaneously. There has therefore been an explicit aim to control for more than one covariate at a time. MATERIALS AND METHODS The study was conducted in Madras, the capital of Tamil Nadu state in South India (Fig. 1), which is located on the Bay of Bengal, approximately 12.80 NORTH COOUM RIVER w 0 ,n mm z C) I- SCALE: l--1km Fig. 1: Map of Madras to show location of the three communities. north of the equator, and is the fourth largest city in India. Situated on a broad coastal plain, the Madras area receives over three-quarters of its precipitation during the north-east monsoon, from October to January. This is also the coolest time of the year, with temperatures averaging 22-24 OC. Many communities in the city are caste homo- geneous, reflecting a near uniformity in food choice, hygienic behaviour, and occupation. To minimize the variability of behaviours related to eating and washing, it was decided to limit sampling to a single caste. In the three communities chosen (Fig. 2 and 3), over 95% of the people were from the Pattinavar caste. Most of the heads of the households were engaged in fishing, the traditional occupation of the caste; relatively little migration was thought to occur because the people occupied as fishermen had the advantage of an intimate knowledge of the local coast and markets. Two of the areas were composed of tenement housing which had replaced the traditional stick-and- mud huts. These buildings, which were about ten years old, were three or four storeys high and made of concrete. Each unit contained approximately 46.5 m2 of living area. The third community was adjacent to the first (Fig. 2). It was composed mainly of 144 EFFECT OF WATER QUALITY AND QUANTITY ON NUTRITIONAL STATUS to increase the number of taps in the distribution system. The new organization departed from this policy and undertook to increase the number of services by about 10% per year (T.S. Daivamani, personal communication, 1978) without any increase in the storage or pumping capacity to keep pace. Communities located far from the distribution centre, such as the study communities, were therefore at a distinct disadvantage, especially in the year 1978-79 which was very dry. The primary source of cooking- and drinking-water MARINA for the three communities was the piped water system of the Madras Metropolitan Water Supply and BEACH Sewerage Board. Water was provided from valveless taps in undrained subterranean access pits which functioned for approximately 3 hours each morning. Owing to the scarcity of water, there was always someone collecting water at the taps whenever the water was running. An alternative source of water was from local shallow wells. Access to these dug wells could be gained either directly by bucket and line, or the water could be pumped to overhead tanks which were provided in the tenement buildings. Pumping, Fig. 2: Map of Communities 1 and 3. = water mains * = water tap ¢ = on-site well 4= public latrine traditional huts built of mud and sticks, although there were a few multistorey homes of finished rock and plaster. The water supply Madras has significant variation in water quality and quantity. During the short north-east monsoon season, water tends to be abundant. During the rest of the year, the city's shallow reservoirs, which supply most of the water for the system, lose approximately half their total storage volume as a result of evaporation alone (28). The maximum of 220 million litres of water that can flow through the system daily during the wet season would allow for 70 litres per person per day. This assumes the water to be equally available during all seasons and to all the inhabitants of the city. However, many areas are severely under- served, especially during the dry season. The water supply system for the city of Madras became the responsibility of the Madras Metropolitan Water Supply and Sewerage Board (MMWSSB) in 1978, the year that this study began. Previously, the water supply had been the responsibility of the Madras City Corporation whose policy had been not SCALE: #-I 20m Fig. 3: Map of Community 2. - = water mains * =water tap NORTH NORTH 11 MARINA i BEACH ¢ = on-site well 145 146 J. R. HEBERT however, was erratic, although it was more frequent in community 2 than in community 1. Water from the wells was used mainly for washing and flushing latrines. It was generally brackish, and therefore was not used for cooking and drinking. The one well that was an exception was located in community 2. The water from that well (B-I in Fig. 3) was considered by the residents to be of good enough quality to be used for cooking and drinking. This well was a source of water for the distribution system, through a small pumping station, and was also accessible by bucket and line. Timing of the study The study was conducted longitudinally from November 1978 to January 1980. Water quality data were collected every two weeks throughout this period. Children aged 6 years and under from randomly selected households were followed from the post-monsoon measurement of January 1979 to a mid-monsoon period ending in December 1979. Selection of the children Census data were unavailable for the three com- munities. The exact number of households in the communities, however, was known and an estimate of the average number of children under 6 years of age per household was available from local agencies. The tenement populations were easily sampled, as each apartment and building was numbered; 191 random numbers were chosen and the households in the apart- ments corresponding to those numbers were enrolled in the study. The traditionally housed community 3 was smaller, consisting of 245 households, and it was not numbered in a way that enabled proper random sampling. All the households there were included in the study. A total of 662 children were eligible to be enrolled in the study. Of these, 627 were actually enrolled and measured at least once; 219 from community 1, 200 from community 2, and 208 from community 3. Data collection forms A questionnaire was prepared and presented orally in the local language by native speakers. Information was collected on the demography of the communities and the standard of living of the household; the ages, health histories, and educational and occupational status of the household members; and their socioeconomic status and behaviours relating to food and hygiene. This questionnaire was based on one used previously by Faigenblumd in Guatemala and a d FAIGENBLUM, J. M. The level of living in two rural Guatemalan communities and its relationship to health. PhD dissertation, University of North Carolina, Chapel Hill, 1978. "level-of-living" scale devised by Belcher (29) for cross-cultural comparisons. The questionnaire was pre-tested on ten households. Anthropometric techniques of the field staff were standardized in early December 1978. A validity check of the survey was carried out on a random subset of the participating households after the first measurements were made. Over 95% of the responses on the validity check agreed with the responses given on the original questionnaire. Personnel The field support staff consisted of two medical social workers, one part-time assistant, and four full- time assistants. The social workers were recruited from the School of Social Work at the University of Madras. They were fluent in both Tamil, the language spoken in the communities, and English, the language in which the questionnaire answers were recorded. The social workers administered the questionnaire orally, as most of the members of the communities were illiterate. They were also responsible for measuring the children and updating the responses to items on the questionnaire at the time of subsequent anthropometric measurements. The five field assistants were responsible for assisting the social workers in making the anthro- pometric measurements, verifying the accuracy of the answers to items on the questionnaire, ascertaining the location of the water tap that the household was actually using, and helping to estimate the quantity of water stored by the household. Water quality data Sampling for water quality was carried out approximately once every two weeks by the author. The first samples were taken in November 1978 and the last ones in early January 1980. Sterile sample bottles were filled directly from the flowing taps with the exception of water from well B-1 (see Fig. 3), where the sample was drawn from the dug well using a bottle with a line attached. The multiple tube fermentation technique was chosen as the assay for monitoring contamination of the water sources (30). The faecal coliform procedure was done as the second stage, or the confirmatory test, for all positive tubes in the first stage of the test. The concentration of faecal organisms was derived from the most probable number (MPN) table based on the statistical probability of finding the concentration of micro- organisms given a number of positive tubes. The natural logarithms of the coliform concentrations (MPN) were taken, and the averages of the Igg concentrations were derived for the interval between EFFECT OF WATER QUALITY AND QUANTITY ON NUTRITIONAL STATUS the beginning of the study and the first measurement and for the intervals between measurements for each pair of consecutive child measurements available in the study. Water quantity data Water quantity data were recorded at the beginning of the study and were updated each time the children were measured. Since cooking/drinking-water and domestic water for washing were from two distinct supplies, information concerning both types of water was recorded. The volume of cooking- and drinking- water was readily estimated by actually measuring the household storage containers and questioning the head female of the household as to the proportion of capacity used daily. The mean volume of water per person per day was used in our analyses. When the study was originally planned, it was hoped that the interval estimates of water quality and water quantity would correctly reflect the long-term quality and quantity estimates. Changes in the water supply system, however, led us to believe that this was not the case. Therefore another term, the estimate of cooking/drinking-water quantity at the beginning of the study, was added to the height-for-age model. The quantity of water for washing, which could not be measured as accurately, was graded on a scale from 1 to 3, where "1" indicated no direct access (i.e., through an overhead tank) and no separate storage facilities for such water in the household, "2" indicated no direct access but with a separate storage supply for such water, and "3" indicated a ready supply of water for washing from an overhead tank or, rarely, from a well. Other variables In addition to water quantity, other variables that could change during the intervals between anthropo- metric measurements were recorded. These included changes in withholding food during the two weeks preceding the anthropometric measurement and breast-feeding status, Information was also collected and used in analyses on the children's bathing and defecation habits, the family's socioeconomic status (indicated by the mother's and father's educational level, family income, and food cooking equipment), household sanitation (indicated by clothes washing, and the parents' washing and defecation behaviours), quality of housing (type of construction of walls and floors, water storage facilities, and presence of a latrine), food-related variables (food expenditure/ person/day, facilities for food storage and Table 1. Distribution of continuous variables used in analyses Variable Value Frequency (%) Mean S.D. Mother's education none 62.2 2.06 3.08 (years) 1-4 12.5 5-8 19.8 >8 5.5 Father's education none 30.8 4.10 3.45 (years) 1-4 18.2 5-8 39.5 >8 11.6 Family income < 22.25 23.6 37.07 21.99 (rupees/person/month) 22.25-31.24 24.8 31.25-44.99 26.1 >45 25.5 Food expense <0.86 25.2 1.09 0.30 (rupees/person/day) 0.86-1.00 30.8 1.01-1.25 22.0 > 1.25 22.1 Drinking-water use' < 6.66 19.9 10.84 6.57 (litres/person/day) 6.66-9.99 28.4 10.00-13.33 31.0 > 13.33 20.7 Water quality' < 1.33 23.7 2.97 1.69 (mean log coliform 1.33-2.99 26.2 organisms/100 ml) 3.00-4.32 24.2 (geometric mean= 19.49 > 4.32 25.9 a This variable is used in the second stage of analysis. b This represents the antilog of the overall mean log coliform concentration. 147 148 J. R. HEBERT Table 2. Distribution of other variables used in analyses Variable Frequency (%) Variable Frequency (%) Cooking equipment: Child's bathing: Open fire/ground level Mud hearth/knee level Concrete hearth/knee level Charcoal/kerosene stove Floor construction: Earthen Plaster over brick Concrete Roof construction: Palm leaves Tin Concrete Water storage in overhead tank: None Some Food preservation: None Some Father's occupation: Fisherman Not fisherman Proximity to sea: 25-150 metres 151-300 metres > 300 metres Plumbing wastepipes: Not present Present 31.5 19.0 38.5 11.0 14.9 33.7 51.4 17.7 14.9 67.4 59.2 40.8 89.5 10.5 53.8 46.2 36.2 36.3 27.5 16.8 83.2 Away from home At home Child's defecation: Near home Away from home Use of latrine Mother's defecation: No latrine Use of latrine Father's defecation: No latrine Use of latrine Washing of clothes: Away from home Athome Washing-water availability:' No separate supply Separate supply/hand carry Overhead tank Percentage by age category 0-18 months 19-36 months > 36 months Breast-feeding: No Yes Withholding food: No Yes 30.9 69.1 18.9 81.1 86.9 13.1 25.4 74.6 a This variable is used in the second stage of the analyses. preservation), proximity to the sea, and the father's occupation. A complete list of the variables used in the analyses and their distributional characteristics are shown in Table I for continuous variables, and Table 2 for other variables. Anthropometric measurements and age determination Age was determined by direct questioning. In cases where it was not known, reference was made to the astrological chart and the birth date was determined accurately to within a day in most cases. All measurements were performed in the household by the medical social workers. Weight was measured using Salter scales which could be hung from hooks in the ceilings of the tenements. Paediatric slings were also provided. The scales were adjusted to zero twice daily and were recalibrated once a week. Weights were recorded to the nearest 100 grams. Children were weighed without clothing or, for ambulatory children, with a light cotton shirt and pants only. Height or length was measured according to the guidelines outlined by Jelliffe (31). 50.1 49.9 25.7 37.4 36.9 9.0 91.0 57.5 42.5 49.9 50.1 39.9 23.0 37.1 99.2 0.8 *24.9 75.1 EFFECT OF WATER QUALITY AND QUANTITY ON NUTRITIONAL STATUS 149 Measurements were begun in late January 1979 and the final measurements were made in December 1979. The average interval between the first and second measurements was 5 months and between the second and third was 3j months. For each set of measurements, the weight of a child was compared with the median weight of children of the same height or length in the NCHS and WHO reference series.e f The percentage weight-for-height (or length) was found by dividing the actual weight by the reference median weight and multiplying the quotient by 100. Height-for-age and weight-for-age were derived by similar methods. STATISTICAL METHODS Each child could contribute up to three measure- ments depending on the number of times he or she was measured. Since the mean age of termination of breast-feeding in this population was found to be 18 months, the data were stratified by age so that children under 18 months old were in one age group, those aged 18-36 months in a second group, and those over 36 months old in the third age group. The four water-related variables used in the models were the average log coliform counts, availability of water for washing (on the scale described above), quantity of cooking/drinking-water in the interval preceding a child's measurement, and quantity of cooking/drinking-water at the beginning of the study. The analyses were carried out in two stages. In the first stage a linear predictor score was computed after each anthropometric index was regressed on all the non-water-related summary variables (socioecono- mic, housing, household sanitation, and food- related), along with the child's defecation and bathing habits, breast-feeding, withholding of food, proximity to the ocean, and the father's occupation (see Annex for a complete description of the analysis technique). The linear predictor score accounts for the combined effect of all the non-water-related variables used in the model. The second stage of the analyses consisted of regressing the anthropometric indices on this linear predictor score plus all of the water-related variables, as denoted in Tables 1 and 2. The regression coefficients of the water-related variables then represent the effects of those variables after controlling for the linear predictor score. All ' NCHS growth charts, 1976. Rockville, MD, National Center for Health Statistics, 1976. f Reference data for the height and weight of children. In: Measurement of nutritional impact. Unpublished document WHO/FAP/79.1, 1979, Annex 3, pp. 53-85. coefficients and corresponding P values are based on Type III sums of squares (32), i.e., the analysis accounts for the test variable with all the other predictors in the model. RESULTS Completeness offollow-up In order to use all the information available for the intervals it was necessary to accumulate the anthropo- metric indices for all three measurement times. Since there may have been different losses to follow-up in the three communities, the children who were lost in any measurement interval were compared with those who provided a full set of measurements. For all the indices, the children who were not measured a second or third time were similar to children who were measured all three times. Their age distributions were also very similar. The weight-for- height index showed the greatest difference. Even for this index, the test for difference between proportions of children in the tails of the distributions was insignificant. Nor were the means different (92.75±9.89 vs 95.54±8.55). Comparison of the children's anthropometric indices by community revealed no significant differences at any of the three measurement times. The age distributions were also the same. Regression analyses In Tables 3 to 5 the mean value of each variable is listed along with its regression coefficient. In order to give the reader a better understanding of how the variable affects the predicted value of the anthropo- metric index, the products of the regression coefficient with the 25th and 75th percentile values are also shown. The influence of the variable in the model can be estimated by comparing these fitted values. Results of the weight-for-height model for all three age groups (0-18 months, 19-36 months, and over 36 months) are shown in Table 3. Except for the average log coliform score, the predictor variables are associated with positive coefficients. In the youngest age group the linear predictor score, which estimates the combined effect of the twenty-one variables listed in the Methods section, has an insignificant regression coefficient (P= 0.21). Neither of the water quantity variables is significant. The average coliform concen- tration, however, is significant at P= 0.05, and the difference between the fitted values, .,X(75) and fiX(2s), shows more of an impact on the predicted value of weight-for-height than the other water- related variables (- 3.23 versus 1.20 for washing- J. R. HEBERT Table 3. Model of weight-for-height measurements Age and variable Mean value B P value BX(25)' BX(75)b 0-18 months: Intercept 95.77 Average coliform count 2.92 - 1.09 0.03 - 1.46 - 4.69 Washing-water quantity 2.02 +0.60 0.50 +0.60 +1.80 Cooking-water quantity 10.73 + 0.15 0.30 + 1.00 + 2.00 Linear predictor score -3.30 +0.55 0.21 - 2.67 -0.94 Model R2 = 0.06 n=218 19-36 months: Intercept 106.80 Average coliform count 3.13 -0.53 0.05 -0.71 -2.40 Washing-water quantity 1.94 +0.24 0.65 +0.24 +0.72 Cooking-water quantity 11.33 +0.07 0.32 +0.51 +0.93 Unear predictor score -11.74 +1.01 0.0001 - 13.35 - 9.92 Model R2=0.10 n= 348 Over 36 months: Intercept 93.42 Average coliform count 2.91 -0.13 0.39 -0.17 -0.56 Washing-water quantity 1.9i +0.25 0.37 +0.25 +0.75 Cooking-water quantity 10.66 + 0.12 0.004 + 0.80 + 1.60 Linear predictor score -0.14 + 1.04 0.0001 - 5.05 - 1.78 Model R2 = 0.03 n = 847 a X(l2s) represents the product of the estimate in the regression model times the twenty-fifth percentile value of the variable. b BX(75) represents the product of the estimate in the regression model times the seventy-fifth percentile value of the variable. Table 4. Model of weight-for-age measurements Age and variable Mean value B P value BX(25)a BX(75)b 0-18 months: Intercept 74.63 Average coliform count 2.92 - 1.06 0.04 - 1.42 - 4.84 .Washing-water quantity 2.02 +1.04 0.26 +1.04 + 3.12 Cooking-water quantity 10.73 +0.17 0.26 +1.13 + 2.27 Linear predictor score 5.73 +0.68 0.02 +2.47 - 5.11 Model R2 = 0.08 n=220 19-36 months: Intercept 81.88 Average coliform count 3.13 -0.63 0.06 -0.84 - 2.71 Washing-water quantity 1.94 +0.26 0.70 +0.26 +0.78 Cooking-water quantity 11.33 + 0.09 - 0.29 + 0.65 +1.20 Unear predictor score - 2.88 +1.03 0.0001 -4.92 -0.54 Model R2=0.13 n=351 Over 36 months: Intercept 71.06 Average coliform count 2.91 -0.31 0.14 -0.41 - 1.33 Washing-water quantity 1.97 +1.37 0.0003 + 1.37 + 4.11 Cooking-water quantity 10.66 + 0.20 0.0003 + 1.33 + 2.67 Linear predictor score 1.76 +1.39 0.0007 +1.86 + 3.07 Model R2= 0.04 n=850 BX(25) represents the product of the estimate in the regression model times the twenty-fifth percentile value of the variable. b HX(75) represents-the product of the estimate in the regression model times the seventy-fifth percentile value of the variable. 150 EFFECT OF WATER QUALITY AND QUANTITY ON NUTRITIONAL STATUS water quantity and 1.00 for cooking-water quantity). In the 18-36-month age group the linear predictor score is highly significant (P=0.0001). The water- related variables, however, exhibit the same pattern that they did for the younger children; the coliform count is relatively more important than quantity of water; the difference between the fitted valuesf8X(75) andflX(2s) is - 1.69 for average coliforms versus 0.48 for washing-water quantity and 0.42 for cooking- water quantity. For the children in the oldest age group, cooking- and drinking-water volume is the only water-related variable that is significant (P < 0.005). The difference between its fitted values, AX(75) and fOX(25), is also larger (0.80 versus 0.50 for washing-water quantity and - 0.39 for average coliform concentration). The linear predictor score persists as a significant variable in the model. Table 4, which presents the weight-for-age model in a form analogous to the weight-for-age model of Table 3, shows a very similar pattern. The signs of the coefficients all behave as in the previous model. Average coliform concentration is the only water- related variable which comes close to being significant in the two youngest age groups (P= 0.04 andP= 0.06, respectively). Again, it is the variable with the largest impact in terms of the difference in fitted.fX(7s) and fiX(25) values (- 3.42 and - 1.87, respectively, in the two youngest age groups). The linear predictor score is significant at all ages. In the oldest age group, however, both washing-water quantity and cooking/ drinking-water volume are highly significant (P= 0.0003) while the coliform count is not. The difference between the fitted values is also larger for each of the water quantity variables (2.74 for washing-water quantity and 1.34 for cooking-water quantity) than for average coliform concentration (-0.92). Table 5 shows the results for the height-for-age model. A fifth variable was added to the model, the original volume of water used per person. This variable was added because it was thought that height, which measures long-term nutritional status (26), would be best predicted by variables that indicate exposures over a long period of time. The original volume of water was felt to be a better estimate of long-term usage than estimates from later periods. For children under 36 months old, only the Table 5. Model of height-for-age measurements Age and variable Mean value B P value SX2s)' BX(75)b 0-18 months: Intercept 88.00 Average coliform count 2.92 -0.11 0.60 -0.15 -0.47 Washing-water quantity 2.02 +0.36 0.41 +0.36 + 1.08 Cooking-water quantity 10.73 +0.005 0.95 +0.03 +0.07 Original water quantity 12.93 +0.03 0.59 +0.20 +0.43 Linear predictor score 5.25 +0.75 0.002 +3.03 +4.97 Model RI = 0.06 n=218 19-36 months: Intercept 84.25 Average coliform count 3.13 -0.12 0.48 -0.16 -0.54 Washing-water quantity 1.94 + 0.08 0.83 + 0.08 + 0.24 Cooking-water quantity 11.33 +0.04 0.35 +0.29 +0.53 Original water quantity 13.76 -0.02 0.17 -0.13 -0.30 Linear predictor score 4.93 +1.05 0.0001 + 3.77 + 6.47 Model R2=0.1 1 n=350 Over 36 months: Intercept 85.77 Average coliform count 2.91 -0.09 0.43 -0.12 -0.39 Washing-water quantity 1.97 + 0.86 0.0003 + 0.86 + 2.58 Cooking-water quantity 10.66 +0.02 0.54 +0.13 +0.27 Original water quantity 12.91 +0.07 0.006 +0.47 +1.05 Linear predictor score 1.43 + 1.19 0.004 + 1.31 + 1.91 Model R' = 0.04 n=853 a BX(25, represents the product of the estimate in the regression model times the twenty-fifth percentile value of the variable. b ?X(7S) represents the product of the estimate in the regression model times the seventy-fifth percentile value of the variable. ISI J. R. HEBERT linear predictor score is significant. For no variable other than the linear predictor score is the difference between the fitted values .8X(75) and OX(25) greater than 0.75 and the average difference is approximately 0.28. For children over the age of 36 months, the quantity of washing-water and the average volume of cooking/drinking-water, as reported at the beginning of the study, are significant (P= 0.0003 and P = 0.006). The differences between the fitted values of these variables also have the greatest impact on the predicted values of height-for-age (1.72 for washing- water quantity and 0.58 versus - 0.27 for average coliform concentration and 0.14 for cooking/ drinking-water quantity). All coefficients behaved as in the previous models with the exception of the original volume estimate for the second age group, which was negative. DISCUSSION For children aged 3 years or older the findings of this study are consistent with the notion that water quantity is a relatively stronger determinant of child health than is water quality. For younger children, however, these data support the opposite view. That is, for the children under 3 years of age water quality tends to be a stronger predictor of nutritional status than either washing- or cooking/drinking-water volume or availability. The fact that the log coliform concentration does better in predicting weight-related indices than the water quantity variables may indicate several things. First, there may be a real age-specific difference in terms of effect. This would be easier to detect using labile weight-based indices in young children for whom the interval between measurements is a relatively larger proportion of their lives. Second, young children at the time of weaning tend to be very prone to intestinal diseases and diarrhoea (33). Water quality could be an important factor during this period of vulnerability and this fact could account for the significant regression coefficient. A third explanation might be that water quality is an important factor all along; the lack of significance in the over 3 year olds may reflect the failure of the interval quality measurement to account for exposure to water contamination over much longer periods of time. Statistical considerations Consistent with the experience of others studying environmental determinants of health status (34, 35), these models had low R2 values. The inability to explain the variability in the dependent variables may have two causes. One reason may be that the homogeneity of these groups made detection of real differences for a range of determinants more difficult than it would have been in a very heterogeneous group. A second possible reason for the low R2 has to do with the "noise" in such systems. Many of the real determinants of nutritional status are extremely difficult to measure with the accuracy of properly performed anthropometric measurements. The use of interval measurements that are much shorter than the length of a child's lifetime compounds this problem. The signs of the regression coefficients, with the exception of the coefficient for original water quantity in the height-for-age model (19-36 months), behaved as one would expect. Increased contamin- ation of water reduced the size of the predicted value of the anthropometric index while high scores for the quantity variables and the linear predictor score increased the predicted value of the index. The consistency of these findings show through the "noise" and indicate that, despite the low R2 value, the model may be appropriate. Extensive examination of the residuals also indicated that the assumptions of the general linear model were not violated. The linear predictor scores, which are composed of nineteen variables, tend to behave very differently in the three age categories of the weight-for-height model. Among the children who are not breast- feeding the negative effects of living far from the ocean, living in tenement buildings, and having food withheld in the period preceding measurement weigh heavily in forcing the value of the X(i) to be negative (see Table 2). The model of weight-for-age (see Table 3) shows a negative X(25) and X(75) value only for the children in the period just following the age at which half the children have terminated breast-feeding. The other age categories are associated with positive X(i) values. Strengths and weaknesses The appropriateness of using this study group was, in one sense, validated by its remarkable homogeneity for factors typically considered to be important determinants of children's nutritional status. Values of the water quality and quantity parameters were also uncorrelated with these other variables. The study group provided a certain amount of inherent "control" for variables which are usually con- founded with sanitation. The technique of fitting the model in two stages provided an additional way to account for confounding factors while making it possible to focus on the variables of interest in the analyses. 152 EFFECT OF WATER QUALITY AND QUANTITY ON NUTRITIONAL STATUS One weakness of the study is that the children were followed for only eleven months. To estimate the effects of changing factors on the nutritional status of preschool-aged children much older than 11 months of age is, perhaps, expecting too much. For height- for-age this is even more a problem than it is for weight-based indices. Another, more general, problem is that there are many other true deter- minants of nutritional status which are difficult to identify and usually very difficult to measure. An important consideration in any longitudinal study is loss to follow-up. While any loss is regret- table, it was shown that the children who left the study were essentially the same as those who stayed on. The amount of out-migration during the time of the study was surprisingly high. This may be, in part, due to the general instability that is quickly becoming charac- teristic of the Tamil villages of the Coromandel coast (36). The water availability problem may also have contributed to this out-migration. Policy implications At present, it is the convention that, when there is a choice, water quality is sacrificed for water quantity. Yet, it is known that the most vulnerable group consists of children below the age of 3 years. In absolute numbers and with regard to age-specific mortality rates, this is typically the target group of greatest interest. If the findings of this study are validated in future work, then an important policy issue is raised. It may be necessary to reconsider the relative merits of improving water quality or water quantity. It is hoped that the techniques and findings described here will motivate other researchers to look at the issue of assessing the health effects of particular aspects of sanitation in other populations. ACKNOWLEDGEMENTS This work was supported by the International Foundation for the Promotion of Nutrition Education and Nutrition Research, Zurich, Switzerland; OXFAM, Oxford, England; University of California at Berkeley (Department of International Education); and the Fund for Teaching and Research, Department of Nutrition, School of Public Health, Harvard University, Boston, MS, USA. RESUMIt INFLUENCE DE LA QUANTIT, ET DE LA QUALITt DE L'EAU SUR L'ITAT NUTRITIONNEL: OBSERVATIONS FAITES DANS UNE COLLECTIVITt DU SUD DE L'INDE Devant le manque de donn6es scientifiques permettant d'6valuer l'influence relative de la qualite et de la quantit6 (ou disponibilite) de l'eau d'approvisionnement sur la sante des enfants, on a ete amen6 a mettre sur pied une 6tude sur ce point dans l'Inde du Sud. Les trois collectivit6s retenues sont composees pour l'essentiel (a plus de 95%) de membres de la caste Pattinavar et sont implant6es sur les bords du Golfe du Bengale, dans les limites de la ville de Madras. Pour la plupart, les chefs de famille travaillaient comme pecheurs, la principale profession exercee par leur caste. Pour eviter les problemes que pose l'emploi d'indicateurs de morbidite, par exemple l'incidence des maladies diar- rheiques, on a choisi comme parametres des indicateurs anthropometriques de l'etat nutritionnel. Les hypotheses intervenant dans l'emploi de ces indicateurs ont e longuement etudiees. Les enfants ont e peses et mesures trois fois, la premiere apres la mousson de 1978-79 et les deux autres au cours de la saison seche puis de la mousson suivantes. La qualite de l'eau, estim&e d'apres sa teneur en coli- formes fecaux, a e mesuree en moyenne une fois tous les 15 jours. Le volume d'eau utilise pour la boisson et la cuisine a e determine au debut de l'etude, une mise a jour des chiffres etant effectuee chaque fois que l'enfant etait mesur6. Faute de pouvoir la mesurer avec la meme pr6cision, on a estim6 la quantit6 d'eau utilis6e ou disponible pour se laver et pour faire la lessive d'apres un bareme allant de 1 a 3. Parmi les autres variables dont on sait qu'elles influent sur la sant6 ou l'6tat nutritionnel de l'enfant, bon nombre ont et6 mesur6es et utilis&es comme determinants dans les analyses. Certaines des variables importantes ont ete condens&es sous forme d'un score r6capitulatif decrivant la situation socio-6conomique de la famille, les installations sanitaires A sa disposition, son logement et diverses variables en rapport avec l'alimentation. On a egalement note, et utilise pour les analyses, les pratiques en matiere d'alimen- tation maternelle, de constitution de stocks alimentaires, d'hygiene individuelle, de defecation, ainsi que la proximite du logement par rapport a la mer et la profession du chef de famille. L'analyse statistique a comporte deux stades. Dans le premier, on a etudie la regression de l'indicateur anthrometrique de l'etat nutritionnel en fonction de tous les 153 154 J. R. HEBERT parametres sans rapport avec l'eau. On a ainsi pu obtenir une variable recapitulative qualifiee de score predictif lineaire. Dans un second stade, on a etudie la regression de l'indicateur anthropometrique en fonction du score predictif lineaire et de toutes les variables en rapport avec l'eau. Ces analyses ont montre que, pour les enfants de moins de 36 mois appartenant a la population consid6r&e, la qualite de l'eau conditionne la valeur des indicateurs anthropometriques fondes sur le poids davantage que la quantite d'eau stockee pour la boisson et la cuisine ou que la quantite d'eau disponible pour se laver et faire la lessive. Dans le cas des enfants plus ages, la quantite d'eau disponible pour se laver et faire la lessive et la quantite d'eau utilis6e pour la boisson et la cuisine influent de faqon relativement importante sur les mensurations de l'enfant, contrairement a la qualite de 1'eau. Les resultats de cette etude, la premiere oui l'on ait essaye de reduire l'influence des facteurs parasites, posent un probleme de politique generale puisque, jusqu'ici, on etait convenu de sacrifier la qualite a la quantite en matiere d'approvisonnement en eau, quand ce choix etait inevitable. Or, on sait que le groupe le plus vulnerable chez les enfants est celui des "moins de trois ans". Si les observations faites ici se revelent plus gen6rales, il faudra peut-etre remettre en question l'inter& respectif d'une amelioration de la qualite ou d'une amelioration de la quantite de 1'eau distribuee aux populations. REFERENCES 1. International Drinking Water Supply and Sanitation Decade. New York, 1980. Report of the Secretary General (A/35/367)-United Nations. 2. BLUM, D. & FEACHEM, R. G. Measuring the impact of water supply and sanitation investments on diarrhoeal diseases: problems of methodology. International journal of epidemiology, 12: 357-365 (1983). 3. SAUNDERS, R. J. & WARFORD, J.J. Village water supply: economics and policy in the developing world. Baltimore, Johns Hopkins University Press, 1976. 4. STEWART, W. H. ET AL. The relationship of certain environmental factors to the prevalence of Shigella infection. American journal of tropical medicine and hygiene, 4: 718-724 (1955). 5. BRADLEY, D. J. & EMURWON, P. Predicting the epide- miological effects of changing water sources. East African medical journal, 45: 284-291 (1968). 6. WATKINS, C. J. ET AL. Doctor diagnosis and maternal recall of lower respiratory illness. Internationaljournal of epidemiology, 11: 62-66 (1982). 7. KOOPMAN, J. S. ET AL. Food, sanitation and socio- economic determinants of child growth in Colombia. American journal ofpublic health, 71: 31-37 (1981). 8. World Health Statistics Annual 1980-81. Iqfectious diseases: cases. Geneva. World Health Organization, 1981. 9. Manual ofstandards ofqualityfor drinking water sup- plies. New Delhi, Indian Council of Medical Research, 1975 (Special report series No. 44). 10. US ENVIRONMENTAL PROTECTION AGENCY. Interim primary drinking water standards. Federal Register, 40(51): 11990 (1975). 11. ANDERSON, S. ET AL. Statistical methods for compar- ative studies. New York, John Wiley and Sons, 1980. 12. WASH. Relating Improvements in Water Supply and Sanitation to Nutritional Status. Arlington, VA, 1982. WASH Technical Report Series 16. 13. CHEN, L. C. ET AL. The use of anthropometry for nutri- tional surveillance in mortality control programs. American journal of clinical nutrition, 34: 2596-2599 (1981). 14. GOLDSTEIN, H. & TANNER, J. C. Child growth standards. Lancet, 2: 35 (1980). 15. MARTORELL, R. & KLEIN, R. E. Food supplemen- tation and growth rates in preschool children. Nutrition reports international, 21: 447-454 (1980). 16. GRAITHCER, P. L. & GENTRY, E. M. Measuring chil- dren: one reference for all. Lancet, 2: 297-299 (1981). 17. HABICHT, J. ET AL. Height and weight standards for preschool children: how relevant are ethnic differences in growth potential? Lancet, 1: 611-615 (1974). 18. HARRIES, A. D. ET AL. Malnutrition in inflammatory bowel disease: an anthropometric study. Human nutrition: clinical nutrition, 36c: 307-313 (1982). 19. GRAITCHER, P. L. ET AL. Anthropometric indicators of nutrition status and morbidity. Journal of tropical pediatrics, 27: 292-298 (1981). 20. JAMES, J. W. Longitudinal study of the morbidity of diarrhoeal and respiratory infections in malnourished children. American journal of clinical nutrition, 25: 690-694 (1972). 21. TOMKINS, A. Nutritional status and severity of diar- rhoea among preschool children in rural Nigeria. Lancet, 1: 860-862 (1981). 22. TRIPATHY, K. ET AL. Effects of ascaris infection on human nutrition. American journal of tropical medicine and hygiene, 20: 212-218 (1971). 23. JELLIFFE, E. F. P. & JELLIFFE, D. B. Ascariasis and malnutrition: a worm's eye view. American journal of clinical nutrition, 34: 1976-1977 (1981). 24. DAVIDSqON, S. ET AL. Human nutrition and dietetics. Edinburgh, Churchill Livingstone, 1979. 25. MARTORELL, R. ET AL. The impact of ordinary illnesses on the dietary intakes of malnourished children. American journal of clinical nutrition, 33: 345-350 (1980). 26. WATERLOW, J. C. Classification and definition of protein-energy malnutrition: epidemiology and ap- proaches to control. In: Beaton, G. H. & Bengoa, J. M., ed., Nutrition in preventive medicine. Geneva. World Health Organization, 1976, Annex 5, pp. 530- 554. 27. WHO Technical Report Series No. 477, 1971 (Joint FAO/WHO Expert Committee on Nutrition: eighth report). EFFECT OF WATER QUALITY AND QUANTITY ON NUTRITIONAL STATUS 155 28. The Madras water system. Madras, Corporation of the City of Madras, 1955. 29. BELCHER, J. C. A cross-cultural level of living scale. Rural sociology, 37: 203-219 (1972). 30. Standard methods for the examination of water and wastewater. New York, The American Water Works Association and the American Public Health Associ- ation, 1973. 31. JELLIFFE, D. B. The assessment ofthe nutritional status of the community. Geneva, World Health Organization, 1966 (Monograph Series No. 53). 32. SAS user's guide, 1982 edition: Statistics. Cary, NC, SAS Institute Inc., 1982. 33. SNYDER, J. D. & MERSON, M. H. The magnitude of the global problem of acute diarrhoeal disease: a review of active surveillance data. Bulletin ofthe World Health Organization, 60: 605-613 (1982). 34. SCHIFFMAN, M. A. ET AL. Field studies on water, sani- tation and health education in relation to health status in Central America. Progress in water technology, 11: 143-150 (1979). 35. BEATON, G. H. & GHASSEMI, H. Supplementary feeding programs for young children in developing countries. American journal of clinical nutrition, 35: 864-916 (1982). 36. BLAKE, B. A. Cultural adaptation and technological change among Madras fishing populations. In: Estellie Smith, M., ed., Those who livefrom the sea: a study in maritime anthropology. St. Paul, West Publishing Co., 1977, pp. 97-1 10. Annex THE LINEAR PREDICTOR SCORE Here, the outcome variable, Y, the anthropometric index, is assumed to be linearly related to the water supply variables, X= (XI.....Xp), and to the house- hold and behavioural variables, Z= (Z I.....Zq). The true model may be represented as follows: P q Y=cx+ BiXi+ ejZj+error (1) With regard to testing hypotheses concerning the water-related variables, the other variables can be regarded as "nuisance" variables. Therefore the effect of the household and behavioural variables is summarized by means of a linear predictor score, ZO. This is done by constructing a model, Y=ro + eOjzj (2) and letting q ZO= E ejz; (3) The final model, which now includes the water- related variables, is given by: Y= a + fx3ixi + eozo (4) eO is the coefficient that allows correlation of the linear predictor score with Xi. If ZO were uncorrelated with Xi, then e0 would equal 1. This approach has the advantage of summarizing the "nuisance" variables into a single score, of reducing the number of variables in the final model, of taking into account the variables known to influence child growth, and of focusing the attention on the water-related variables of interest. It has been established, by a simulation technique, that the full model (equation 1) and the final model (equation 4) yield virtually identical values for the predicted value of the coefficients, fli, and their standard deviations, when the truefi values are 0 (Dr Martin Larson, personal communication, 1983). That is, if water supply variables do not affect anthropometric variables, the use of linear predictor scores will not change the type 1 error rates, i.e., the probability of claiming an effect of water supply variables when none is actually present. Furthermore, this holds true whether or not the X and Z variables are correlated. If the true ft values are non-zero, using equation 4 instead of equation 1 still yields nearly identical estimates of fti though their standard deviations may be increased. With the large sample sizes in this study, the effect on significance levels of the f terms is not likely to be important.