Exploring the relationship between access to water, sanitation and hygiene and soil-transmitted helminth infection: a...

Soil-transmitted helminths (STH) - a class of parasites that affect billions of people - can be mitigated using mass drug administration, though reinfection following treatment occurs within a few months. Improvements to water, sanitation and hygiene (WASH) likely provide sustained benefit, but few rigorous studies have evaluated the specific WASH components most influential in reducing infection. There is a need for alternative analytic approaches to help identify, characterize and further refine the WASH components that are most important to STH reinfection. Traditional epidemiological approaches are not well-suited for assessing the complex and highly correlated relationships commonly seen in WASH.

We introduce two recursive partitioning approaches: classification and regression trees (C&RT) and conditional inference trees (CIT), which can be used to identify complex interactions between WASH indicators and identify sub-populations that may be susceptible to STH reinfection. We illustrate the advantages and disadvantages of these approaches utilizing school- and household-level WASH indicators gathered as part of a school-based randomized control trial in Kenya that measured STH reinfection of pupils 10 months following deworming treatment.

C&RT and CIT analyses resulted in strikingly different decision trees. C&RT may be the preferred approach if interest lies in using WASH indicators to classify individuals or communities as STH infected or uninfected, whereas CIT is most appropriate for identifying WASH indicators that may be causally associated with STH infection. Both tools are well-suited for identifying complex interactions among WASH indicators.

C&RT and CIT are two analytic approaches that may offer valuable insight regarding the identification, selection and refinement of WASH indicators and their interactions with regards to STH control programs; however, they represent solutions to two distinct research questions and careful consideration should be made before deciding which approach is most appropriate.