Clark, Justin M

Abstract Recent work in causally interpretable meta-analysis (CIMA) has bridged the gap between traditional meta-analysis and causal inference. Whereas traditional meta-analysis results generally do not apply to any well-defined population, CIMA approaches specify a target population to which meta-analytic treatment effect estimates are transported. Although theoretically attractive, these approaches currently have some practical limitations. Most assume that all studies in the meta-analysis have individual participant data (IPD), which is rare in practice because most trials share only aggregate data. We propose a method to perform CIMA using a combination of aggregate data and IPD. This method borrows information from studies with IPD to augment the aggregate data and create aggregate-matched synthetic IPD (AMSIPD), which can be used readily in the existing CIMA framework. By allowing use of both aggregate data and IPD, the method opens CIMA to more applications and can avoid biases arising from using only studies with IPD. We present a case study and simulations showing the AMSIPD approach is promising and merits further investigation as an advancement of CIMA.