Variable importance measures for heterogeneous causal effects
share
The conditional average treatment effect (CATE) of a binary exposure on an outcome is often studied on the basis that personalised treatment decisions lead to better clinical outcomes. The CATE, however, may be a complicated function of several covariates representing patient characteristics. As such, clinicians rely on researchers to summarise insights related to treatment effect heterogeneity. Clinical research usually focuses on the ATE, which averages the CATE across all patient characteristics, or considers the treatment effect in patient strata, though strata are rarely chosen systematically. Here we present new nonparametric treatment effect variable importance measures (TE-VIMs). These may guide clinicians as to which patient characteristics are important to consider when making treatment decisions, and help researchers stratify patient populations for further study. TE-VIMs extend recent regression-VIMs, viewed as nonparametric analogues to ANOVA statistics. Moreover, TE-VIMs are not tied to a particular model, thus are amenable to data-adaptive (machine learning) estimation of the CATE, itself an active area of research. Estimators for the proposed statistics are derived from their efficient influence curves and these are illustrated through a simulation study and an applied example.
READ FULL TEXT