Modeling Tumor Growth Using Partly Conditional Survival Models: A Case Study in Colorectal Cancer.

Area of Science:

• Oncology
• Biostatistics
• Medical Imaging Analysis

Background:

• Traditional survival models often rely on untestable assumptions for longitudinal tumor measurements.
• Flexible modeling is needed to accurately assess the relationship between tumor dynamics and patient survival.

Purpose of the Study:

• To introduce and demonstrate the application of the partly conditional (PC) survival model for analyzing longitudinal tumor measurements and overall survival.
• To evaluate the association between tumor burden changes and survival in a phase III colorectal cancer trial.

Main Methods:

• Utilized PC survival modeling, a semiparametric approach, on longitudinal volumetric computed tomography data from 1,025 patients in the VELOUR trial.
• Modeled overall survival as the outcome, with covariates including baseline tumor burden, changes in tumor burden, and treatment (aflibercept plus chemotherapy).
• Investigated both unstratified and time-stratified PC models without assuming tumor growth distribution.

Main Results:

• The change in tumor burden was significantly associated with overall survival (HR, 1.04; 95% CI, 1.02-1.05; P < .001), suggesting aflibercept impacts survival partly by altering tumor growth.
• Baseline tumor size remained prognostic for survival even after accounting for changes over time (HR, 1.02; 95% CI, 1.01-1.02; P < .001).

Conclusions:

• The PC survival model provides a flexible method for characterizing associations between longitudinal tumor burden and survival time.
• This approach can be applied to various longitudinal data types and integrated into ongoing clinical trials to incorporate accumulating disease assessment information.