Risk Factors for Childhood Diarrhea Incidence: Dynamic Analysis of a Longitudinal Study : Epidemiology

Secondary Logo

Journal Logo

Original Article

Risk Factors for Childhood Diarrhea Incidence

Dynamic Analysis of a Longitudinal Study

Genser, Bernd; Strina, Agostino; Teles, Carlos A.; Prado, Matildes S.; Barreto, Mauricio L.

Author Information
Epidemiology 17(6):p 658-667, November 2006. | DOI: 10.1097/01.ede.0000239728.75215.86
  • Free


Diarrhea continues to be an important health problem in developing countries, especially among preschool children. Several longitudinal studies1–9 have been conducted in developing countries to identify risk factors for childhood diarrhea. Socioeconomic factors, household sanitary conditions, neighborhood basic sanitation infrastructure, and child-related or care-related variables (eg, hygiene behavior, anthropometric nutritional status, breast-feeding, or intestinal parasitic infections) have been identified, among others, as the major diarrhea determinants. Unfortunately, most studies on diarrhea incidence have neglected important dynamic features of the longitudinal design such as variation of diarrhea incidence during the follow-up period, the impact of time-varying variables, and autoregressive or intrasubject correlation among the repeated diarrhea episodes.2,10 A few studies have used longitudinal analysis techniques to address dynamic date features such as intrasubject correlation among the repeated disease episodes11,12 or to model diarrhea incidence by a time-to-event approach that accounts for the impact of time-varying variables.2,13 However, most authors of diarrhea studies do not properly address hierarchical interrelationships among the potential risk factors. A single multivariate model is not sufficient to explain the causality of disease occurrence when there are factors that do not cause morbidity directly (distal determinants), but instead act through interrelated proximate determinants (intermediate variables).14

We present the results of a longitudinal epidemiologic study conducted in Brazil with the aim to investigate risk factors for childhood diarrhea. In contrast with other studies, we applied an advanced analysis approach combining time-to-event modeling to address several dynamic features of the longitudinal design with a hierarchical effect-decomposition strategy to quantify the impact of risk factors grouped in different blocks.


Study Design and Population

This community-based longitudinal study was conducted in Salvador, a large developing urban center in northeastern Brazil with a population of 2.5 million. Households were recruited from 23 sentinel areas without public sewage system selected from 30 areas chosen to represent the full range of socioeconomic and environmental conditions in a wider study on the health impact of an extensive sanitation program.15,16 The sample has been described in detail elsewhere.17 Households with children under 3 years of age were selected at random from the full list of households in each sentinel area with one child in the eligible age range (0–36 months) randomly enrolled in each household. The study was conducted from October 2000 until January 2002 enrolling 1233 children over the first 10 months. Because diarrhea is a relatively rare event, we selected 902 children with a minimum follow-up time of 90 days for analysis.

Study Variables

Diarrhea data were collected by twice-weekly home visits carried out by 15 fieldworkers.4 During each visit, the field worker questioned the mother (or other caretaker) about the number and consistency of bowel movements; symptoms such as fever, vomiting, and blood in stool; and any treatments (such as rehydration, medical care, or use of medication) over the preceding 3 to 4 days. A day with diarrhea was defined as the occurrence of 3 or more loose stools or any number of loose stools containing blood in a 24-hour period starting when the child wakes in the morning.18 Episodes of diarrheal illness were defined using a minimum 3-day diarrhea-free gap to mark the beginning of a new episode.18,19 The field workers were blinded to these definitions of a “day with diarrhea” and “episode.”

To deal with the numerous explanatory variables, we defined a conceptual framework shown in Figure 1. The framework, similar to one suggested by other authors,14 takes into account interrelationships among risk factors by grouping them into 5 blocks. In addition, this framework assumes that diarrhea risk is affected by age and sex as well as past diarrhea episodes (autoregressive effect). Age was also considered as a potential effect modifier.

Conceptual framework of risk factors for diarrhea incidence in developing countries; time-varying variables are indicated by (t).

Information on socioeconomic status (block 1), neighborhood and household sanitation (block 2), and prenatal examination and birth weight (block 3) was collected at the time of recruitment into the study using a precoded questionnaire. Hygiene behavior (block 4) was collected by field workers based on their observation of hygienic and unhygienic behaviors of the child or the child's caretaker (details given elsewhere4). A composite hygiene behavior score was calculated for each child, and children were grouped into 3 categories: those for whom the observed behaviors were mainly hygienic, those for whom hygienic and unhygienic behaviors were observed with roughly equal frequency, and those for whom unhygienic behaviors were more commonly observed.

Stool samples were collected once during the follow-up (between May and July 2001) and examined for presence of intestinal parasites (block 5). The mother or caretaker was visited at home and given a container that was collected the next morning, placed on ice, and taken for immediate analysis. A single stool sample was examined (using the Kato-Katz method) for the presence and number of helminth eggs (Ascaris lumbricoides and Trichuris trichiura) and simple sedimentation for Giardia lambia cysts.20 Anthropometric measurements were carried out 3 times for each child (at baseline and after 6 and 12 months). Height-for-age z-scores were calculated using the EPINUT program (Epi Info 6.0; CDC, Atlanta, GA).

Statistical Analysis

We applied a time-to-event approach for statistical analysis taking the time until the beginning of a new diarrhea episode as the outcome variable. Time-on-study was chosen as the analysis time scale; other time scales (age and time since last diarrhea episode) were considered as time-varying covariates. The model we applied was an extended Cox proportional hazards model for multiple events, a special case of the Andersen-Gill counting process model.10,21,22 The approach assumes a common baseline hazard among the repeated events and uses a robust resampling variance estimation technique to adjust for within-child correlation.23 Days without morbidity information were excluded from the person-time assuming a noninformative missing pattern (“missing at random”).24 Association analyses (contingency tables, χ2 tests, correspondence analysis) were conducted to identify highly correlated variables that we then aggregated into index variables (socioeconomic status, water supply, and garbage disposal). Bivariate analyses were conducted by calculating hazard ratios (HRs) and 95% naive and robust25 (ie, adjusted for intracluster correlation) confidence intervals (CIs) across strata defined by potential risk factors. In bivariate analysis, missing values of the potential risk factors were treated as a separate category. Subgroup analyses in 3 age groups were conducted to examine effect modification by age.

For multivariate analysis, we applied an effect-decomposition strategy, similar to an approach suggested by other authors.14 The idea of this strategy is to fit a sequence of multivariate extended Cox regression models by including step-by-step blocks of potential risk factors according to a predefined hierarchy. For each block, we used a backward elimination procedure to identify significant risk factors (P < 0.10). By comparing the risk estimates obtained with the different models, we were able to examine the pathways by which the risk factors act. Following our conceptual framework (Fig. 1), we fitted 6 extended Cox regression models. Model A (block 1 only) sought to estimate the overall effect of socioeconomic status. Models B and C additionally included significant variables from blocks 2 and 3 and sought to estimate the effect of socioeconomic status not mediated through blocks 2 and 3. Model D (variables of block 1, 2, and 3) sought to estimate the effects of socioeconomic status not mediated through blocks 2 and 3 as well as the overall effect of variables in block 2. Model E (variables of blocks 1, 2, 3, and 4) sought to estimate the effect of socioeconomic status not mediated by blocks 2, 3, and 4, the effect of block 2 not mediated by block 4, and the overall effect of variables of block 3 and 4. Finally, model F (variables of blocks 1, 2, 3, 4, and 5) sought to estimate the effects of socioeconomic status not mediated through blocks 2, 3, 4, and 5; the effect of blocks 2 and 3 not mediated by blocks 4 and 5; the effect of block 4 not mediated by block 5; and the overall effect of the variables in block 5. All models were adjusted for age, sex, and time since last diarrhea episode. We fitted separate models that additionally included interaction terms to test for overall effect modification by age. Finally, we fitted a model including the number of previous episodes as a covariate for a subpopulation of 235 children with at least 4 diarrhea episodes. In all multivariate models, missing values for explanatory variables were imputed with the modal value. All statistical analyses were carried out using the statistical software package STATA (version 8.2; Stata Corp., College Station, TX).


Informed consent to participate in the study was obtained from all study households. Ethical approval for the study was given by the ethics review board of the Federal University of Bahia.


The study population included 902 children who were followed for a total of 285,424 child-days. The mean age (± standard deviation [SD]) at baseline was 19.5 (± 9.9) months, and the median follow-up time was 351 days (range = 90–426 days). Three fourth (77% [n = 693]) of the children had at least one diarrhea episode during the observation period. In total, we observed 2397 diarrhea episodes (6245 diarrhea days) with a mean (± SD) duration of 2.4 (± 1.6) days during 279,179 child-days at risk, resulting in a crude overall diarrhea incidence rate of 3.1 episodes/child-year (95% CI = 3.0–3.3). Diarrhea risk varied substantially among the children (from 0–26 episodes per child-year); this between-subject variation also affected the variance estimates and produced a slightly wider robust 95% CI (2.9–3.4).

Diarrhea risk was affected by several time-varying variables. First, the rate varied substantially with age; the incidence rate (adjusted for time-on-study) peaked at 5.9 episodes/child-year (5.1–6.8) in children aged 7 to 12 months and decreased with increasing age to 0.9 episodes/child-year in children aged 43 to 48 months (0.6–1.3) (Fig. 2).

Smoothed incidence rates over age bands of 6 months, adjusted for “time-on-study,” including a robust 95% confidence band, in 902 children (2397 diarrhea episodes, 279,179 person days at risk), aged 0 to 36 months at baseline from Salvador, Brazil, 2000–2002.

We also observed a substantial decline of incidence rate with time-on-study that was only partially explained by child's age (Fig. 3). The age-adjusted rate decreased markedly from 7.6 episodes/child-year (7.0–8.3) during the first 3 study months to 1.1 episodes/child-year (0.9–1.4) after 14 months. Finally, we found evidence of an autoregressive effect of past diarrhea episodes. The rate decreased substantially with increasing time since last episode, from 6.2 episodes/child-year (5.9–6.7) 2 weeks after the last episode, to 3.2 episodes/child-year (2.8–3.7) 7 to 8 weeks afterward, and to 1.9 episodes/child-year (1.5–2.3) 19 weeks afterward (Fig. 4). Using an autoregressive extended Cox regression model to adjust for time since last episode, age, and “time-on-study,” diarrhea incidence was reduced by 7% (6–8%) for every added week since the last episode. By contrast, an autoregressive model applied to children with 4 or more diarrheal episodes showed that the diarrhea rate was independent of the number of previous diarrhea episodes, (data not shown).

Smoothed crude and age-adjusted incidence rates as a function of “time-on-study” (intervals of 3 months), including robust 95% confidence bands, in 902 children (2397 diarrhea episodes, 279,179 person days at risk), aged 0 to 36 months at baseline, from Salvador, Brazil, 2000–2002.
Smoothed incidence rates, including a robust 95% confidence band, adjusted for age and “time-on study,” as a function of “time since last episode” (intervals of 2 weeks), in 902 children (2397 diarrhea episodes, 279,179 person days at risk), aged 0 to 36 months at baseline, from Salvador, Brazil, 2000–2002.

Table 1 presents the results of bivariate analysis of several factors possibly associated with diarrhea incidence. We identified potential risk factors in all blocks of the conceptual framework. Among the strongest explanatory factors were variables in the sanitation condition block (block 2) such as an unsatisfactory garbage disposal and absence of a toilet in the household (HR = 1.73 and 1.70, respectively) followed by poor socioeconomic status (HR = 1.58) (block 1). Other factors were prenatal examination (HR = 1.50 for no examination), intestinal parasitic infections (HR = 1.32 for one or more infections), specific infection by G. lambia (HR = 1.53), and hygiene behavior (HR = 1.28 for poor hygiene).

Associations of Risk Factors With Diarrhea (902 children, 2397 diarrhea episodes, 279,179 person-days at risk), Aged 0 to 36 Mo at Baseline, Salvador, Brazil, 2000–2002

The bivariate analyses stratified by age group (0–12, 13–36, and 37 months or more) showed evidence of effect modification by age for 6 potential risk factors (Table 2). Except for socioeconomic status, for which the strongest effect was observed in children older than 36 months, the strongest associations were identified in children age 13 to 36 months.

Subgroup Analyses Stratified by Age

Table 3 summarizes the results of the effect-decomposition approach consisting of 6 extended Cox regression models. We found notable overall effects for socioeconomic conditions, presence of open sewage nearby, toilet in household, flooding of habitation, prenatal examination, and intestinal parasitic infections (especially G. lambia). Hygiene behavior and height-for-age score showed no overall effect. Comparing effect estimators from the different models, the HR for socioeconomic status decreased markedly after adjusting for blocks 2 and 3. For example, comparing the poorest versus the richest group, the overall HR was 1.69, whereas after adjusting for block 2 the HR was 1.33, and after adjusting for block 3, it was 1.52.

Results of the Multivariate Extended Cox Regression Models

By contrast, the effect measures of neighborhood and household sanitary conditions (block 2 variables, model D) did not change substantially after adjusting for hygiene behavior (model E) and intestinal parasitic infections (model F). HRs for open sewage nearby, no toilet in habitation, and flooding of habitation were similar in all 3 models. The effect measures of child and care-related variables (no prenatal examination and low height-for-age score; block 3 variables, model E) also did not change after adjusting for intestinal parasitic infections.

The multivariate models with the interaction terms (extensions of model A, D, and E; data not shown) confirmed the hypothesis of effect modification by age for some of the risk factors identified with stratified bivariate analysis (Table 2). By extending model A, we found significant interactions for socioeconomic status (P = 0.021, global Wald test) with the poorest group having a HR of 2.3 times higher for children age more than 36 months than for children aged 0 to 12 months (95% CI = 1.2–4.3). Furthermore, by including an interaction term in model D, we found effect modification for presence of toilet (P = 0.016) and flooding of habitation (P = 0.002) and, by extending model E, interaction for height-for-age score (P = 0.003). In children aged 13 to 16 months compared with children aged 0 to 12 months, the HR for no toilet in household was 2.3 times higher (1.2–4.6), and the HRs of flooding of habitation and low height-for-age score were 1.4 (1.2–1.7) and 1.3 (1.1–1.5) times higher, respectively.


We investigated risk factors for childhood diarrhea using a time-to-event approach to address both dynamic features of the longitudinal design and hierarchical interrelationships among the potential risk factors. Age and an autoregressive effect of time since previous diarrhea episodes were identified by our dynamic models as the major determinants of the decline of diarrhea rate over time that we observed in this longitudinal study. Our study also showed that diarrhea incidence was strongly associated with risk factors located in different blocks. Moreover, by applying a hierarchical effect-decomposition approach based on a predefined conceptual framework, we showed that the effect of poor socioeconomic conditions on diarrhea incidence was mediated mostly by inadequate sanitation conditions with only a small further mediation attributable to child- or care-related variables and intestinal parasitic infections. By contrast, poor sanitation conditions and child and care-related variables apparently act directly on diarrhea risk.

By applying a time-to-event approach, we detected a number of dynamic features in this longitudinal study. First, diarrhea risk adjusted for time-on-study varied substantially with age, showing a peak in children aged 7 to 12 months and a dramatic decrease with increasing age thereafter. This is similar to findings from other studies.2,5 Also, we identified effect modification by age with the strongest risk factor associations found mostly in children aged 13 to 36 months. This issue has been addressed in few previous studies. A study conducted in Guinea Bissau reported significant interaction with age for several risk factors explained by breast-feeding practices.5,26 Another recently published study conducted in a poor Peruvian community also reported interaction of age with water supply and sanitation level.2 The lower relative risks that we observed in children aged 0 to 12 months could be explained either by the protective effect of breast-feeding or by the fact that very young children spend more time indoors and therefore have reduced exposure to contamination near the household. However, the mean duration of exclusive breast-feeding in our study was only 2.1 months (SD = 2.4), so the latter explanation is more likely. The reduced effect of neighborhood sanitation factors observed in children older than 36 months could be due to the fact that hygienic behavior usually improves in older children. Interestingly, we observed the highest effect of socioeconomic status in older children, a finding that might indicate that socioeconomic status is a proxy for factors other than sanitation conditions in this age group, (eg, education level, diet, or access to health facilities).

We also observed a substantial decline of incidence over the course of the study that was only partially explained by age. Although this pattern has been reported in other longitudinal studies, the reasons for this decline are unclear. A study from Brazil has attributed the decrease to observed improvements in nutritional status.27 A more plausible explanation for our cohort, among whom the prevalence of malnutrition did not change during the study (data not shown), might be an effect of repeated domiciliary visits by study personnel (“Hawthorne effect”).28,29 Also, part of the decline might be explained by the autoregressive effect of past diarrhea episodes. Because a diarrhea episode in the past increased the child's diarrhea risk, even a small “Hawthorne effect” would have increasing impact over time. This shows the importance of carefully adjusting the temporal factors to correctly estimate incidence as a function of age adjustment. Other studies have used simple analysis approaches to show that recent diarrhea episodes (ie, those occurring in the previous 14 days) increased the risk of occurrence of a new episode.5,30,31 We explored this by modeling the decline of risk after an episode in an autoregressive Cox model; there was a 7% decrease in diarrhea risk predicted for every passing week since the last diarrheal episode. An explanation of this autoregressive effect might be that the cause of a diarrhea episode may still persist and cause another episode or that diarrhea episodes may alter the host defenses by damaging the intestinal epithelium. However, by examining the diarrhea rates in children with 4 or more episodes and comparing the rates of the first, the second, and the third diarrhea events, we were able to show that the number of previous episodes did not affect the diarrhea risk, thus favoring the first interpretation.

We confirmed poor socioeconomic, neighborhood, and domestic sanitary conditions (presence of an open sewage nearby, absence of toilet in household, flooding of habitation) as well as some child- and care-related exposure variables (absence of prenatal examination) and presence of intestinal parasites as being the major risk factors for diarrhea incidence. These have also been shown in previous studies.2,5,32,33 Intestinal parasitic infections showed associations with diarrhea incidence, a finding that has been discussed by other authors.34–37 By analyzing specific intestinal parasitic infections separately, we found an association with G. lambia, which can cause diarrhea.34,38,39 No associations were found for T. trichuris, which also can cause diarrhea, and for A. lumbricoides, which may be considered to be a proxy for poor neighborhood and domestic sanitation conditions (risk factors already addressed by our conceptual model).36,39,40 Also, we found an increased diarrhea risk in children whose mothers did not seek prenatal examination, a finding in line with other studies.14,41 Lack of prenatal care may represent a proxy for unfavorable risk factors (besides the child and care-related factors included in our conceptual model) such as limited access to antenatal care facilities.42

Finally, we were not able to show an effect of hygiene behavior in contrast with a previous study conducted in the same city.4 Stunting, which has been associated with diarrhea in other studies,5,30,43–46 was not a risk factor in our study.

In contrast to other diarrhea studies, we applied a dynamic analysis to address some typical characteristics of longitudinal data. One important issue is intrasubject correlation due to repeated diarrhea episodes. We used both robust variance estimation to adjust for repeated measures on the same child and an autoregressive model to account for serial correlation among sequential diarrhea episodes. Neglecting intrasubject correlation, as other authors have,47,48 might result in biased variance estimates (too-narrow confidence intervals of relative risks). Diarrhea risk often varies during a longitudinal study due to time-varying variables (eg, age). When this feature is not properly addressed, parameter and variance estimates may be biased. A study conducted in Guinea-Bissau5 showed that the diarrhea risk varied during the follow-up due to the effect of age and other time-varying variables, although authors assumed a constant rate during the follow-up.

The simple hierarchical effect-decomposition strategy used here is not without limitations. Under this approach, introduced by Victora et al,14 consistent estimates of direct (not mediated) effects can be obtained only when there is no confounding at the level of the intermediate variable.49 For example, to estimate the effect of socioeconomic status not mediated by sanitation conditions, we had to assume that there were no unobserved covariates associated with sanitation conditions and diarrhea risk. We think this assumption is realistic for socioeconomic status as well as the other intermediate blocks. Our conceptual framework considered a wide range of potential diarrhea determinants and grouped them in meaningful blocks, and it seems unlikely that other unobserved factors are associated with both an intermediate block (eg, sanitation conditions) and diarrhea incidence.

In conclusion, childhood diarrhea remains an important health concern in urban developing centers with diarrhea rates that peak at 6 episodes/child-year among children 7 to 12 months of age. The dynamic features identified in our study underline the necessity of advanced analytic approaches for evaluating risk factors of diarrhea or other acute diseases in longitudinal studies. As a matter of fact, observational studies in children pose special analysis problems because the risk of disease usually depends on age and other time-varying variables, which must be properly addressed in statistical analysis. Moreover, as our data showed, age can also act as an effect modifier for some risk factors. As a further main result, our study shows the necessity of conceptual frameworks for ordering the complex hierarchical interrelationships among risk factors of childhood diarrhea. Poor socioeconomic status, which we had initially considered the most distal diarrhea determinant, was found to be highly associated with diarrhea incidence with a large proportion of its effects mediated by inadequate household and neighborhood sanitation. Only a small part of the socioeconomic status effect was mediated by other unfavorable child- and care-related variables or the high prevalence of intestinal parasitic infections. The implication is that diarrhea rates in developing countries could be substantially decreased by interventions aimed to improve the sanitary and general living conditions of households.


The authors thank the field work team, especially their supervisor, J. C. Goes. The authors also thank Dr Sandy Cairncross for his useful comments on the manuscript and Craig Milroy for the linguistic corrections.


1. Lima AA, Guerrant RL. Persistent diarrhea in children: epidemiology, risk factors, pathophysiology, nutritional impact, and management. Epidemiol Rev. 1992;14:222–242.
2. Checkley W, Gilman RH, Black RE, et al. Effect of water and sanitation on childhood health in a poor Peruvian peri-urban community. Lancet. 2004;363:112–118.
3. Luby SP, Agboatwalla M, Painter J, et al. Effect of intensive handwashing promotion on childhood diarrhea in high-risk communities in Pakistan: a randomized controlled trial. JAMA. 2004;291:2547–2554.
4. Strina A, Cairncross S, Barreto ML, et al. Childhood diarrhea and observed hygiene behavior in Salvador, Brazil. Am J Epidemiol. 2003;157:1032–1038.
5. Molbak K, Jensen H, Ingholt L, et al. Risk factors for diarrheal disease incidence in early childhood: a community cohort study from Guinea-Bissau. Am J Epidemiol. 1997;146:273–282.
6. Manun'ebo MN, Haggerty PA, Kalengaie M, et al. Influence of demographic, socioeconomic and environmental variables on childhood diarrhoea in a rural area of Zaire. J Trop Med Hyg. 1994;97:31–38.
7. Mertens TE, Jaffar S, Fernando MA, et al. Excreta disposal behaviour and latrine ownership in relation to the risk of childhood diarrhoea in Sri Lanka. Int J Epidemiol. 1992;21:1157–1164.
8. el Samani FZ, Willett WC, Ware JH. Predictors of simple diarrhoea in children under 5 years—a study of a Sudanese rural community. Soc Sci Med. 1989;29:1065–1070.
9. Moraes LR, Cancio JA, Cairncross S, et al. Impact of drainage and sewerage on diarrhoea in poor urban areas in Salvador, Brazil. Trans R Soc Trop Med Hyg. 2003;97:153–158.
10. Kelly PJ, Lim LL. Survival analysis for recurrent event data: an application to childhood infectious diseases. Stat Med. 2000;19:13–33.
11. Colford JM Jr, Wade TJ, Sandhu SK, et al. A randomized, controlled trial of in-home drinking water intervention to reduce gastrointestinal illness. Am J Epidemiol. 2005;161:472–482.
12. Hellard ME, Sinclair MI, Forbes AB, et al. A randomized, blinded, controlled trial investigating the gastrointestinal health effects of drinking water quality. Environ Health Perspect. 2001;109:773–778.
13. Checkley W, Epstein LD, Gilman RH, et al. Effects of Cryptosporidium parvum infection in Peruvian children: growth faltering and subsequent catch-up growth. Am J Epidemiol. 1998;148:497–506.
14. Victora CG, Huttly SR, Fuchs SC, et al. The role of conceptual frameworks in epidemiological analysis: a hierarchical approach. Int J Epidemiol. 1997;26:224–227.
15. Barreto ML SA, Prado M, et al. Saneamento básico e saúde: fundamentos científicos para avaliação do impacto epidemiólogico do programa de saneamento ambiental da Bahia de Todos os Santos (Bahia Azul). In: Heller ML, Monteiro TC, et al, eds. Saneamento e Saúde nos Países em Desenvolvimento. Rio de Janeiro: CC&P Editores; 1997:7–35.
16. Strina A, Cairncross S, Prado MS, et al. Childhood diarrhoea symptoms, management and duration: observations from a longitudinal community study. Trans R Soc Trop Med Hyg. 2005;99:407–416.
17. Teixeira Md Mda G, Barreto ML, Costa Mda C, et al. Sentinel areas: a monitoring strategy in public health. Cad Saude Publica. 2002;18:1189–1195.
18. Baqui AH, Black RE, Yunus M, et al. Methodological issues in diarrhoeal diseases epidemiology: definition of diarrhoeal episodes. Int J Epidemiol. 1991;20:1057–1063.
19. Morris SS, Cousens SN, Lanata CF, et al. Diarrhoea—defining the episode. Int J Epidemiol. 1994;23:617–623.
20. Neva FA BH. Basic Clinical Parasitology, 6th ed. Norwalk; 1994.
21. Andersen PK GR. Cox's regression model for counting processes: a large sample study. Ann Stat. 1982;10:1100–1120.
22. DR C. Regression models and life-tables. J R Stat Soc. 1972;34:187–220.
23. Lin DY WL. The robust inference for the Cox proportional hazards model. J Am Stat Assoc. 1989;84:1074–1078.
24. Little RA, Rubin DB. Statistical Analysis With Missing Data, 2nd ed. Wiley; 2002.
25. Fawzi WW, Herrera MG, Willett WC, et al. Dietary vitamin A intake and the risk of mortality among children. Am J Clin Nutr. 1994;59:401–408.
26. Molbak K, Gottschau A, Aaby P, et al. Prolonged breast feeding, diarrhoeal disease, and survival of children in Guinea-Bissau. BMJ. 1994;308:1403–1406.
27. Schorling JB, Wanke CA, Schorling SK, et al. A prospective study of persistent diarrhea among children in an urban Brazilian slum. Patterns of occurrence and etiologic agents. Am J Epidemiol. 1990;132:144–156.
28. Haggerty PA, Muladi K, Kirkwood BR, et al. Community-based hygiene education to reduce diarrhoeal disease in rural Zaire: impact of the intervention on diarrhoeal morbidity. Int J Epidemiol. 1994;23:1050–1059.
29. Mayo E. The Human Problems of an Industrial Civilization. New York: Macmillan; 1933.
30. Baqui AH, Black RE, Sack RB, et al. Malnutrition, cell-mediated immune deficiency, and diarrhea: a community-based longitudinal study in rural Bangladeshi children. Am J Epidemiol. 1993;137:355–365.
31. Sazawal S, Bhan MK, Bhandari N, et al. Evidence for recent diarrhoeal morbidity as a risk factor for persistent diarrhoea: a case-control study. Int J Epidemiol. 1991;20:540–545.
32. Esrey SA, Feachem RG, Hughes JM. Interventions for the control of diarrhoeal diseases among young children: improving water supplies and excreta disposal facilities. Bull World Health Organ. 1985;63:757–772.
33. Esrey SA, Potash JB, Roberts L, et al. Effects of improved water supply and sanitation on ascariasis, diarrhoea, dracunculiasis, hookworm infection, schistosomiasis, and trachoma. Bull World Health Organ. 1991;69:609–621.
34. Reinthaler FF, Linck G, Klem G, et al. Intestinal parasites in children with diarrhea in El Salvador. Geogr Med. 1988;18:175–180.
35. Addy PA, Antepim G, Frimpong EH. Prevalence of pathogenic Escherichia coli and parasites in infants with diarrhoea in Kumasi, Ghana. East Afr Med J. 2004;81:353–357.
36. Smith H, Dekaminsky R, Niwas S, et al. Prevalence and intensity of infections of Ascaris lumbricoides and Trichuris trichiura and associated socio-demographic variables in four rural Honduran communities. Mem Inst Oswaldo Cruz. 2001;96:303–314.
37. Stanton B, Silimperi DR, Khatun K, et al. Parasitic, bacterial and viral pathogens isolated from diarrhoeal and routine stool specimens of urban Bangladeshi children. J Trop Med Hyg. 1989;92:46–55.
38. Genta RM. Diarrhea in helminthic infections. Clin Infect Dis. 1993;16(suppl 2):S122–129.
39. Hashmey R, Genta RM, White AC Jr. Parasites and diarrhea. I: protozoans and diarrhea. J Travel Med. 1997;4:17–31.
40. Gamboa MI, Basualdo JA, Kozubsky L, et al. Prevalence of intestinal parasitosis within three population groups in La Plata, Argentina. Eur J Epidemiol. 1998;14:55–61.
41. De Souza AC, Petersont KE, Cufino E, et al. Underlying and proximate determinants of diarrhoea-specific infant mortality rates among municipalities in the state of Ceara, north-east Brazil: an ecological study. J Biosoc Sci. 2001;33:227–244.
42. Emond A, Pollock J, Da Costa N, et al. The effectiveness of community-based interventions to improve maternal and infant health in the northeast of Brazil. Rev Panam Salud Publica. 2002;12:101–110.
43. Schorling JB, McAuliffe JF, de Souza MA, et al. Malnutrition is associated with increased diarrhoea incidence and duration among children in an urban Brazilian slum. Int J Epidemiol. 1990;19:728–735.
44. Tomkins A. Nutritional status and severity of diarrhoea among pre-school children in rural Nigeria. Lancet. 1981;1:860–862.
45. el Samani EF, Willett WC, Ware JH. Association of malnutrition and diarrhea in children aged under five years. A prospective follow-up study in a rural Sudanese community. Am J Epidemiol. 1988;128:93–105.
46. Sepulveda J, Willett W, Munoz A. Malnutrition and diarrhea. A longitudinal study among urban Mexican children. Am J Epidemiol. 1988;127:365–376.
47. Aziz KM, Hoque BA, Hasan KZ, et al. Reduction in diarrhoeal diseases in children in rural Bangladesh by environmental and behavioural modifications. Trans R Soc Trop Med Hyg. 1990;84:433–438.
48. Huttly SR, Hoque BA, Aziz KM, et al. Persistent diarrhoea in a rural area of Bangladesh: a community-based longitudinal study. Int J Epidemiol. 1989;18:964–969.
49. Robins JM, Greenland S. Identifiability and exchangeability for direct and indirect effects. Epidemiology. 1992;3:143–155.
Copyright © 2006 Wolters Kluwer Health, Inc. All rights reserved.