Data from Modeling Tumor Evolutionary Dynamics to Predict Clinical Outcomes for Patients with Metastatic Colorectal Cancer: A Retrospective Analysis
crossref（2023）
Over 50% of colorectal cancer patients develop resistance after a transient response to therapy. Understanding tumor resistance from an evolutionary perspective leads to better predictions of treatment outcomes. The objectives of this study were to develop a computational framework to analyze tumor longitudinal measurements and recapitulate the individual evolutionary dynamics in metastatic colorectal cancer (mCRC) patients. A stochastic modeling framework was developed to depict the whole spectrum of tumor evolution prior to diagnosis and during and after therapy. The evolutionary model was optimized using a nonlinear mixed effect (NLME) method based on the longitudinal measurements of liver metastatic lesions from 599 mCRC patients. The deterministic limits in the NLME model were applied to optimize the stochastic model for each patient. Cox proportional hazards models coupled with the least absolute shrinkage and selection operator (LASSO) algorithm were applied to predict patients' progression-free survival (PFS) and overall survival (OS). The stochastic evolutionary model well described the longitudinal profiles of tumor sizes. The evolutionary parameters optimized for each patient indicated substantial interpatient variability. The number of resistant subclones at diagnosis was found to be a significant predictor to survival, and the hazard ratios with 95% CI were 1.09 (0.79–1.49) and 1.54 (1.01–2.34) for patients with three or more resistant subclones. Coupled with several patient characteristics, evolutionary parameters strongly predict patients' PFS and OS. A stochastic computational framework was successfully developed to recapitulate individual patient evolutionary dynamics, which could predict clinical survival outcomes in mCRC patients.
Significance:A data analysis framework depicts the individual evolutionary dynamics of mCRC patients and can be generalized to project patient survival outcomes.
