Primary Submission Category: Heterogeneous Treatment Effects
A nonparametric Gail-Simon test and estimand for qualitative effect heterogeneity
Authors: Aaron Hudson, Mats J. Stensrud, Oliver Dukes, Ricardo Brioschi,
Presenting Author: Aaron Hudson*
Qualitative heterogeneity or effect modification, occur when treatment is beneficial for certain sub-groups and harmful for others. This specific type of heterogeneity is of clinical interest when treatment decisions will be tailored to individual characteristics. The problem of testing for qualitative heterogeneity has been well-studied when the comparison is made between finite subgroups; for example, Gail and Simon (1985) proposed a likelihood ratio test in the context of discrete covariates. However, the problem is more challenging when the potential effect modifiers are continuous, and one wishes to infer the conditional average treatment effect under a nonparametric model. In this talk, we propose a class of nonparametric tests for qualitative heterogeneity as a natural extension of the Gail-Simon test. Compared with some recent approaches, our proposal can incorporate a variety of structured assumptions on the conditional average treatment effect, extends to moderate/high-dimensional covariates and does not require sample splitting. The utility of the proposal is borne out in simulation studies and a re-analysis of a recent clinical trial.