A Comparison of Causal Mediation Methods of Observational and Experimental Global Health Data

Mediation analysis aims to assess the mechanisms of action of exposures, such as medical or public health interventions. Understanding why interventions work can help us to effectively adapt and target our interventions. The DGHI Research Design & Analysis Core (RDAC) has seen an increase in requests to support research that seeks to estimate mediated pathways in global health. There are two modern methods for conducting a mediation analysis: a regression-based approach using a potential outcomes framework; and a structural equation modeling framework. Each method has its own strengths and limitations; however, a comparison of the performance of each method has not been thoroughly assessed. The primary objective of the proposed research is to utilize available data to evaluate the strengths and limitations of both approaches to estimate mediated pathways. Data proposed for this study come from a recently completed clinical trial of a psychosocial intervention in Tanzania. These longitudinal data display characteristics that make it ideally suited for this comparative research. The study will employ statistical packages in Stata and R software and compare the ability of each method to estimate direct and indirect effects under various assumptions about the data structure and relationships among variables. In addition to developing expertise for the principal investigator, the study will produce two peer-reviewed manuscripts and training materials for relevant RDAC personnel. Knowledge from this study will contribute to two funded NIH studies, one submitted R01, one planned R34, one planned R01, and likely numerous future DGHI grant applications.