A Data-Calibrated Gompertz Tumor Growth Model for Predicting Chemoradiation Response In Murine Tumors.
Abstract
Purpose
Mechanistic tumor growth models provide a framework for linking observed tumor kinetics to underlying biological processes; however, their application to real preclinical datasets is limited by incomplete longitudinal measurements and uncertainty in model parameters. We developed a data-driven mechanistic model predicting tumor growth trajectories, treatment response, and outcome metrics using caliper-based measurements in murine tumor models undergoing chemoradiation. This framework enabled in silico predictions of preclinical experimental outcomes and treatment variability.
Methods
Murine oral tumors were treated with cisplatin-based chemoradiation on a fixed schedule. Bidirectional caliper measurements were collected thrice weekly from induction until endpoint. Tumor volumes were computed using a cylindrical approximation, and induction-to-treatment growth rates were estimated using tumor volume at treatment normalized by time since induction. These measurements were used to construct sampling distributions of model inputs, including initial tumor volume, induction-to-treatment growth rate, and survival thresholds derived from endpoint measurements. A mechanistic Gompertz-based tumor growth model incorporating chemotherapy and radiotherapy effects was implemented. Bayesian optimization was used to estimate model parameters governing radiation sensitivity, chemotherapy sensitivity, and carrying capacity by comparing distributions of simulated and measured tumor volumes at a fixed post-treatment timepoint.
Results
This method integrated heterogeneous experimental measurements into a unified modeling framework across 123 mice. Bayesian optimization converged to stable parameter estimates while accounting for probabilistic overall survival and cure events representing biological variability. Simulated tumor trajectories reproduced the observed distribution of post-treatment tumor volumes, and estimated parameters fell within biologically plausible ranges consistent with expected treatment response behavior.
Conclusion
Our data demonstrates a practical approach for calibrating mechanistic tumor growth models to real preclinical data using limited longitudinal measurements. By combining simplified growth assumptions with Bayesian parameter estimation, the framework provides quantitative insight into tumor kinetics and treatment response, supporting in silico preclinical studies and robust experimental design in radiation oncology.