Third Year Candidacy Presentation: Selective Inference for Interaction Trees

In precision medicine, subgroup identification is crucial for designing personalized treatments by uncovering heterogeneous treatment effects. Tree-based methods provide an interpretable and flexible framework for detecting treatment–covariate interactions. This presentation begins with a brief overview of our prior work extending interaction trees to longitudinal settings by integrating mixed models for repeated measures (MMRM), enabling subgroup discovery with repeated outcomes.

A key limitation of existing tree-based subgroup identification methods is the lack of valid statistical inference following data-driven subgroup selection. To address this issue, recent developments in post-selection inference are discussed, with a focus on the Tree Value framework, which provides valid statistical inference for regression trees after model selection. Building on this framework, the extension of post-selection inference methods to interaction trees is presented, along with preliminary results focusing on inference for the first split.

Advisors: Jimin Ding, Lei Liu

This presentation has been moved to Zoom due to inclement weather:

https://wustl.zoom.us/j/99064737856?pwd=bf6IvVJE9aDZ393cdWUAS9Cn96CSZX.1