Keywords: between-study Heterogeneity, distributed data analysis, distributed research network, heterogeneous treatment effect, pragmatic trial
Although many pragmatic trials have aimed to identify heterogeneous treatment effects (HTE), it is well known that the needed sample size to identify an interaction is substantially larger than the sample size needed to identify a main effect due to the increased uncertainty in interaction effect estimates and the need to account for multiple testing. Since most sample size calculations are conducted for identifying main effects only, investigations of interactions are usually underpowered and are generally considered to be only hypothesis generating. More efficient methods for quantifying HTE are needed to facilitate the use of healthcare system-based research networks to generate the evidence to support precision medicine. I will present a two-stage hypothesis testing procedure, which first filters the covariates by the heterogeneity in their marginal effects and then tests for interactions. I will also describe a distributed data analysis procedure to facilitate the use of this method in networks with limited data sharing capabilities.