Ascaris lumbricoides Infection Following School-Based Deworming in Western Kenya: Assessing the Role of Pupils' School and Home Water, Sanitation, and Hygiene Exposures

Water, sanitation, and hygiene (WaSH) technologies and behaviors can prevent infection by soil-transmitted helminth species independently, but may also interact in complex ways. However, these interactions are poorly understood. The purpose of this study was to characterize how school and home WaSH exposures were associated with Ascaris lumbricoides infection and to identify relevant interactions between separate WaSH technologies and behaviors. A study was conducted among 4,404 children attending 51 primary schools in western Kenya. We used multivariable mixed effects logistic regression to characterize how various WaSH exposures were associated with A. lumbricoides infection after annual school-based deworming. Few WaSH behaviors and technologies were independently associated with A. lumbricoides infection. However, by considering relevant interdependencies between variables, important associations were elucidated. The association between handwashing and A. lumbricoides depended largely upon the pupils' access to an improved water source. Among pupils who had access to improved water sources, A. lumbricoides prevalence was lower for those who handwashed both at school and home compared with neither place (odds ratio: 0.38, 95% confidence interval: 0.18–0.83; P = 0.01). This study contributes to a further understanding of the impact of WaSH on A. lumbricoides infection and shows the importance of accounting for interactions between WaSH technologies and behaviors.

Because there were many water, sanitation, and hygiene (WaSH) technologies and behaviors of interest, we used a number of guiding criterion during the variable specification and model specification.
We first aggregated variables to the correct levels based on causal hypotheses of how they might affect the outcome. Some variables were collected at the pupil level, although they were intrinsically school-level variables, and were therefore aggregated to the school level.
We assessed the homogeneity of variables to see if they have enough variation to be included in the analyses. We also assessed that there were sufficient numbers in cells of categorical variables, and when we observed small cell counts, we considered the possibility of combining similar categories to resolve the problem. We had originally considered the possibility of herd protection from some variables, including school handwashing, school sanitation, and community sanitation. That is, we allow for the possibility that pupils' Ascaris lumbricoides infection may be affected through grouplevel adherence, even in the absence of individual-level adherence. When considering the school-level handwashing variable, in 44 of the 51 schools, fewer than 10% of the pupils reported always washing their hands after defecation. For sanitation variables, almost all pupils reported always using a latrine for defecation at both school and home. Because heterogeneity was poor, we were unable to assess herd protection as originally intended.
Although access to WaSH is important, we generally assumed that pupils' helminth infection could be affected only through the use of WaSH and not through access alone. For example, the presence of a handwashing station can only affect pupil health through handwashing.
Our primary exposures of interest were access to an improved water source, access to comprehensive sanitation (captured by several variables), and practice of handwashing, with separate variables for each of these primary exposures at both school and at home. We also had interest in several other WaSH technologies and behaviors. We attempted to control for a number of important confounders, and to include relevant interactions between variables. The inclusion of each variable was chosen a priori based on biological plausibility and on the previous literature. Sometimes separate variables measured similar constructs, and in Supplemental Table 1 we show correlations between these variables, and the reasons why we chose to include specific variables in our models. We also assessed collinearity of variables in the full model and eliminated terms that were collinear (measured by the presence of high condition indices with several high variance decomposition proportions). 1 Further details on each variable of interest are also discussed in the article.
It was initially unclear whether or not we should control for the prevalence of A. lumbricoides at baseline, as we hypothesized that it could act as either an intermediate variable or a confounding variable. We therefore used a datadriven approach, and ran two separate adjusted models, with and without the "baseline A. lumbricoides prevalence" variable, comparing the change in the coefficients from the model for each of the WaSH variables between the reduced and full model. The two models gave generally similar results (both the point estimates and variances), indicating minimal confounding due to this variable. Only the estimate for one variable-the % of ventilated improved pit latrines in school-was > 10% different from the fully adjusted estimate, but we believed this was more likely to be an intermediate. We had originally considered the possibility of group-level, or herd protection. However, handwashing was poor at most schools (see Table 3), so the variable lacked the necessary heterogeneity to be able to include in our multivariable models Always handwash with soap and water after defecating (same as above, but aggregated)  3 †We found similar results in unadjusted analyses that used the "last handwash" variables, instead of the "always handwash" variables. Pupils who reported washing their hands after they last defecated at school had an odds ratio (OR) of 0.75 (95% confidence interval [CI]: 0.54-1.06; P = 0.10), and for washing their hands after they last defecated at home the OR was 0.94 (95% CI: 0.69-1.28; P = 0.68). ‡The unadjusted analysis assessing the association between an improved water source at school and A. lumbricoides  *Models included handwashing × water interaction terms and controlled for all of the other water, sanitation, and hygiene (WaSH) variables and confounder variables. †At many schools, improved school water sources did not reliably supply water throughout the year, so here we constrained the definition of an improved school water source to also require water reliability.