Estimating Heterogeneous Treatment Effects for Interval and Right-Censored Time-to-Event Responses

Date of Award

Spring 2025

Project Type

Dissertation

Program or Major

Statistics

Degree Name

Doctor of Philosophy

First Advisor

Qi Zhang

Second Advisor

Michelle Capozzoli

Third Advisor

Pei Geng

Abstract

The estimation of heterogeneous treatment effects (HTE) has powerful utility in biological processes, such as determining the efficacy of a drug. Performing designed random experiments allows for estimation of HTEs with a causal interpretation. In many cases these experiments can be costly and time consuming and thus one may need to rely on observational studies to answer a research question. The implications of substituting observational studies for designed experiments is two-fold: causal effects from a designed experiment now have a correlational interpretation and accumulation of data can become unreliable especially in time-to-event follow-up data. Recent Machine Learning methods such as random forests have provided causal effect estimation for observational studies with continuous, binary, or right-censored outcomes. This thesis extends these data-driven methods for the case when an event is known to occur only between two known time points, known as interval-censored data. Utilizing the random forest method, we introduce an influence function that targets the response for data partitioning and provides novel forest-based weights that are used in the construction of non-parametric survival curves for causal effect estimation.

This document is currently not available here.

Share

COinS