Dissertation

A Unified Testing Framework in Preclinical Tumor Growth Studies Using Prioritized Endpoints

Longhurst, Colin A

2026

Overview

Longitudinal tumor growth studies serve an essential role in preclinical therapeutic evaluation, acting as precursors to human clinical trials. Despite the prevalence of these experiments, there is little consensus on how best to analyze the resulting data, largely due to under-emphasized data challenges such as non-linearity, censoring, and correlated errors. The overarching goal of this dissertation is to introduce prioritized, composite estimators into the preclinical tumor-growth literature, demonstrate their practical utility, and provide several new statistical derivations that address open methodological questions. Chapter 1 considers a two-arm experiment in which each animal has a single tumor under study. Leveraging common design features, a rank-based test statistic is developed from prioritized composite scores, which combine survival outcomes and longitudinal tumor measurements. The resulting non-parametric test is robust to several of the motivating data challenges, yields a simple and interpretable estimand, and can serve as a unified approach for analyzing tumor growth data. In Chapter 2, we consider the case of a two-arm experiment where each animal may have several tumors under study. Here, we extend our testing framework using a time-dependent win ratio kernel and derive a cluster robust variance estimator for the win ratio by finding its influence functions, using the general theory of statistical functionals. Chapter 3 introduces the proportional odds model as a regression framework for assessing synergy in multi-arm experiments under an independent-data design. The chapter explores theoretical connections between the model and well-known rank-based tests, extends transformations that link regression coefficients to interpretable probability indices, and derives a semiparametric estimating equation based on conditional-likelihood principles. Chapter 4 then considers clustered multi-arm experiments. The time-dependent win-ratio kernel from Chapter 2 is embedded in the proportional win-fractions model, and an asymptotically valid cluster-robust variance estimator is derived for use with clustered tumor-growth designs. Throughout the dissertation, interpretation and implementation of the proposed methods are illustrated using several HPV(+) head and neck squamous cell carcinoma xenograft studies. Accompanying software implementations are demonstrated on these same data sets, with an emphasis on generic functions whose use extends well beyond preclinical applications. Extensive simulation studies evaluate operating characteristics such as power, type I error, bias, and compare the proposed methods with existing approaches. Taken together, the methods developed here provide a simple, interpretable, and data-robust framework for statistically assessing therapeutic benefit in tumor growth studies.

• Publisher: ProQuest Dissertations & Theses
• Subject: Biomedical engineering / Biostatistics / Oncology
• ISBN: 9798273346383