Yichen Yan

Title: Statistical Methods for Treatment Decision Making Across Comprehensive Clinical Evidence Scenarios
Date: May 6th, 2026
Time: 1:30pm
Location: Zoom
Supervised by: Haolun Shi and Tianyu Guan

Abstract: Clinical treatment development and evaluation require statistical methods that support decision making under complex outcomes, model uncertainty, and limited data availability. This dissertation develops a unified methodological framework across three stages of the clinical evidence pipeline. The first part proposes a generalized phase I/II dose optimization design for multi-categorical and multi-graded toxicity and efficacy outcomes. By converting clinically graded outcomes into quasi-probability measures and utility-based desirability scores, the proposed Bayesian interval design supports adaptive dose escalation and de-escalation while accommodating non-monotonic dose–efficacy relationships. It improves identification of the optimal biological dose. The second part extends the decision target from a single dose amount to a complete dosing regimen within a model-informed drug development framework. Recognizing that treatment regimens are inherently multidimensional and that the true pharmacokinetic pharmacodynamic (PK/PD) structure is rarely known, this work integrates virtual patient simulation, Bayesian model averaging, and Pareto-based multi-objective optimization. It identifies structurally robust regimens by jointly evaluating efficacy, safety, total drug usage, and treatment duration under potential discontinuation. The third part proposes a Target Aggregate Data Adjustment (TADA) method for transportability analysis when only summary-level data are available from the target population. For time-to-event outcomes subject to informative censoring, TADA combines method-of-moments participation weights with inverse probability of censoring weights in a two-stage weighting framework, enabling causal treatment effect estimation despite restricted data access. In all, this dissertation advances a coherent statistical perspective on treatment decision making. Supported by case studies, it contributes to a more realistic, robust, and clinically relevant framework for modern biostatistical inference and decision support in drug development.

Keywords: Dose optimization; Causal inference; Transportability analysis; Aggregate-level data; Pharmacokinetic/pharmacodynamic modeling; Multi-objective optimization