Estimating the effects of water supply and sanitation on risk of diarrheal disease.

摘要

Intervention trials (ITs) have been used to estimate the reduction in diarrheal disease associated with improved water, sanitation, and hygiene. Issues associated with inadequate randomization and blinding procedures, publication bias, participation bias, and generalizability, however, affect the validity of these studies. Here, I present methods utilizing cross-sectional data from 48 countries representing 10 global regions that illustrate how cross-sectional data can be used in conjunction with marginal structural model (MSM) estimators to estimate the causal effects that improved water supply and sanitation facilities would have on reducing diarrheal disease in children. Additionally, I employ Monte Carlo simulations to identify the amount of experimental treatment assignment (ETA) bias and confounding bias associated with the IPTW estimator for countries in South America. Causal estimates of reductions in diarrheal disease in African children of 4%, 11%, and 18% were attributable to basic improvements to water supply, sanitation, and combined interventions, respectively. These estimates are notably smaller than those obtained from previous ITs. Contrary to previous IT results, these findings suggest that implementation of water supply and sanitation interventions simultaneously is more beneficial than separate implementations of each type of intervention. Based on causal DR population-level estimates, basic improvements implemented on a population-wide basis would be most effective if conducted in the Africa D, Africa E, Eastern Mediterranean D, and America D world regions. Results from Monte Carlo simulations suggest that confounding bias was approximately 10 to 1000 times greater than bias associated with ETA violations in the America D region, suggesting that ETA bias is negligible compared to confounding bias in this region. Cross-sectional data used in conjunction with MSM techniques offer an inexpensive way for investigators to obtain causal estimates and may potentially be more valid than those obtained from ITs in the context of diarrheal disease.