In silico Clinical Trials Via Mathematical Modeling Predict Response to Chemoradiation Therapy for Head and Neck Cancer
Abstract
Purpose
To develop an in silico clinical trial framework based on mathematical modeling of tumor response to chemoradiation therapy as a prognostic indication of patient outcomes.
Methods
We propose to model the change in tumor volume in response to chemoradiation therapy using an ordinary-differential-equation (ODE) model with Gompertz growth dynamics, a linear-quadratic (LQ) model of radiation effects, and a log-cell kill model of chemotherapy action. The model was parameterized using pre-treatment and intra-treatment (20 Gy) 18F-FDG-PET/CT data from 87 patients with head and neck squamous cell carcinoma enrolled in a prospective clinical trial. Population variability in tumor growth and volume was captured using kernel density estimation and Latin hypercube sampling. Treatment sensitivity parameters were estimated via Bayesian optimization by comparing simulated and observed intra-treatment tumor volume distributions. We validated our model by comparing simulated and observed progression-free survival (PFS) using a probabilistic tumor control condition and a clinically informed tumor recurrence criterion. Predictions were evaluated using Kaplan-Meier analysis and uncertainty due to limited trial sample size was mitigated by bootstrapping. We further applied our model to investigate the impact of radiation dose modulation on PFS.
Results
Patient data comprised mainly oropharyngeal carcinomas (72%), of which the majority were HPV+ (87%). Median PFS was 40.7 months across the cohort. SUVmean was used to parameterize the growth rate (mean = 4.11, variance = 2.77). Optimization of treatment parameters yielded radiosensitivity and chemosensitivity parameters consistent with literature estimates (0.01 [1/Gy] and 0.004 [m2/mg] respectively). Kaplan-Meier survival analysis showed good agreement between simulated and observed PFS and subsequent modeling highlighted the benefit of increased radiation dose on PFS
Conclusion
Mathematical models of tumor volume dynamics parameterized using 18F-FDG-PET/CT imaging data generated predictions of tumor response to chemoradiation that captured observed patient outcomes, representing a shift towards precision medicine in radiation oncology.