Mathematical Modeling of Tumor Response Dynamics to Predict Progression-Free Survival in Patients with Recurrent High-Grade Glioma

Graduation Year

2024

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Biology (Cell Biology, Microbiology, Molecular Biology)

Major Professor

Heiko Enderling, Ph.D.

Committee Member

Joel Brown, Ph.D.

Committee Member

Solmaz Sahebjam, Ph.D.

Committee Member

Hsiang-Husan Michael Yu, Ph.D.

Committee Member

Dung-Tsa Chen, Ph.D.

Committee Member

Adam MacLean, Ph.D.

Keywords

individual dynamic predictions, joint modeling, nonlinear mixed effects modeling, survival analysis, tumor growth inhibition

Abstract

In this dissertation, I aim to develop a mathematical model describing tumor volume response dynamics to perform individual dynamic predictions of progression-free survival (PFS) on patients with recurrent high-grade glioma (rHGG).

Patients with rHGG have a dismal prognosis with median overall survival (OS) of <12 months and median PFS of <7 months. However, there is a wide heterogeneity in treatment responses. Therefore, to aid clinicians with making decisions to alter therapeutic protocol, I would like to predict patient-specific PFS.

To perform individual dynamic predictions, I employ the Claret tumor growth inhibition (TGI) model. I further develop this model by coupling it with two different survival models. Inter-patient heterogeneity is also taken into account through mixed effects, including covariate effects. Model PFS predictions were evaluated using receiver operating characteristic (ROC) curve analysis as well as Brier score (BS).

The developed mathematical model was able to recapitulate population-level as well as patient-specific tumor response dynamics and PFS. I found sex, study, age, prior bevacizumab failure, and number of recurrences to be significant covariates effecting tumor response dynamics of PFS. In the final iteration of the model, I found that the joint model that incorporated tumor response dynamics into risk of progression was superior to a Cox proportional hazard model typical of survival analyses of clinical data in performing individual dynamic predictions for time horizons longer than the next observation.

In this dissertation, I have motivated the consideration of individual dynamic predictions as a clinical endpoint. To be clinical translatable, these results will need to be prospectively validated on an external cohort. A further development of the sample size analysis performed in this dissertation can then be used to assist in designing such a clinical study.

This document is currently not available here.

Share

COinS