Novel Statistical Methods for Restricted Mean Survival Time and Patient Preference Augmented Dynamic Treatment Regimes in Observational Studies

Abstract

In this dissertation, we develop three new statistical methods and estimating procedures in survival analysis with restricted mean survival time and in evaluating the optimal treatment decision rules by involving patient preference.

Restricted mean survival time (RMST) is a clinically interpretable and meaningful survival metric defined as the patient's mean survival time up to a pre-specified time horizon of interest, denoted as L. No existing RMST regression method allows for the covariate effects to be expressed as functions over time, which is a considerable limitation in light of the many hazard regression models that do accommodate such effects. To address this void in the literature, in the first project of my dissertation, we propose an inference framework for directly modeling RMST as a continuous function of L. We apply our method to kidney transplant data obtained from the Scientific Registry of Transplant Recipients (SRTR).

The second and third projects of my dissertation consider personalized treatment decision strategies in the management of chronic diseases, such as end stage renal disease, which typically consists of sequential and adaptive treatment decision making. This can be formulated through a dynamic treatment regime (DTR) framework, where the goal is to tailor treatment to each individual given their medical history in order to maximize a desirable health outcome. We develop a new method, Augmented Patient Preference incorporated Reinforcement Learning (APP-RL), to incorporate a patient's latent preferences through data augmentation into a tree-based reinforcement learning method to estimate optimal dynamic treatment regimes for multi-stage, multi-treatment settings. For each patient at each stage, we derive their posterior distribution of preferences given responses to a questionnaire, and then subsequently weight multiple outcomes with the estimated preferences to identify the optimal stage-wise personalized decision. APP-RL is robust, efficient, and leads to interpretable DTR estimation.

We further extend the APP-RL ideas into the survival setting with censored data in the last project. We investigate a two-stage treatment setting where patients have to decide between quality of life and survival restricted at maximal follow-up. We successfully develop a method that incorporates the latent patient preference into a weighted utility function that balances between quality of life and survival time, in a Q-learning model framework. We further propose a corresponding m-out-of-n Bootstrap procedure to accurately make statistical inferences and construct confidence intervals on the effects of tailoring variables, whose values can guide the personalized treatment strategies.