Primary Submission Category: Generalizability/Transportability
Examining subgroup-specific treatment effects in multi-source data: source-specific inference and transportability to an external population
Authors: Guanbo Wang, Alex Levis, Issa Dahabreh,
Presenting Author: Guanbo Wang*
One major challenge in estimating effect heterogeneity is that the sample size of the data used is typically not enough to capture how effects vary according to the effect modifiers precisely. Therefore, there is interest in synthesizing evidence across multi-source data (e.g., multi-center trials, meta-analyses of randomized trials, pooled analyses of observational cohorts) to improve the precision of estimators of heterogeneous treatment efficacy. Furthermore, when combining information from multi-source data, the samples typically do not represent a common target population of substantive interest. This raises the question of how to combine information from multi-source data in a way that is interpretable in the context of some meaningful target population of interest while using evidence across multi-source data to improve efficiency. We develop and evaluate methods for using multi-source data to estimate subgroup treatment effects in an external target population or the populations underlying the data sources. We propose a doubly robust estimator that, under mild conditions, is non-parametrically efficient and allows for nuisance functions to be estimated using machine learning methods. We illustrate the methods in meta-analyses of randomized trials for schizophrenia and bipolar disorder.