Keywords: Estimands, intercurrent events, causal effects, causality conditions
We present an overview of different approaches in defining treatment effects in randomized clinical trials with intercurrent events. The most well-known is Neyman-Rubin potential outcomes framework. Others are Dawid’s decision-theoretic approach, Pearl’s graphical modeling approach, structural equation modeling, mediation analysis. We will focus on the stochastic theory of causality that clearly defines causal effects in terms of probability theory, conditional expectations, and filtration that reflects the temporal order of intercurrent events. An important feature of this alternative theory is the causality conditions that imply unbiasedness of effect estimators and have empirically testable implications. They can be used for covariate selection and study design.