The meaning of confounding adjustment in the presence of multiple versions of treatment: an application to organ transplantation

Abstract

Causal inference for treatments with many versions requires a careful specification of the versions of treatment. Specifically, the existence of multiple relevant versions of treatment has implications for the selection of confounders. To illustrate this, we estimate the effect of organ transplantation using grafts from donors who died due to anoxic drug overdose, on recipient graft survival in the US. We describe how explicitly outlining the target trial (i.e. the hypothetical randomized trial which would answer the causal question of interest) to be emulated by an observational study analysis helps conceptualize treatment versions, guides selection of appropriate adjustment variables, and helps clarify the settings in which causal effects of compound treatments will be of value to decision-makers.