Estimating the causal effects of water supply and sanitation improvements on diarrheal disease using cross-sectional data

Currently, developing countries have few tools to decide on how best to allocate resources towards interventions that will lower disease burden. At a global level the World Health Organization has developed a method to estimate disease burden associated with poor water, sanitation, and hygiene levels primarily based on results from a small number of randomized control trials (RCTs) conducted throughout the world. Recently developed statistical methods known as marginal structural models (MSMs) allow researchers to estimate causal effects, and therefore estimate disease burden, from cross-sectional data. Compared to RCTs, cross-sectional data is useful when generalizing to large populations and is often more cost effective. Using country specific data from the Demographic Health Surveys (DHS) we demonstrate how causal effects associated with improvements to water supply and sanitation facilities may be estimated using cross-sectional data from 27 African countries obtained between 1995 and 2003 and the MSM procedure known as G-computation. Results indicate that populations with no basic services would experience reductions in diarrhea of 4% (95% CI: -1%, 13%) when water supply is improved, 11% (95% CI: 7%, 17%) when sanitation is improved, and 18% (95% CI: 14%, 24%) when both are improved. There can be no replacement for data obtained from internally valid RCTs, however, with recent developments in statistical methodology, we propose that cross-sectional data be considered as a valuable complement to RCTs, especially when used to make decisions regarding national and regional-level health policy.