PCN212 PERSONALIZED DISCRETE EVENT SIMULATION OF PROGRESSION-FREE SURVIVAL AND OVERALL SURVIVAL FOR FIRST-LINE DOUBLET CHEMOTHERAPY WITH OR WITHOUT BEVACIZUMAB IN METASTATIC COLORECTAL CANCER

Determining the optimal treatment pathway for metastatic colorectal cancer (mCRC) patients is challenging given the many possible treatment combinations and sequences. Simulation models combining patient and disease characteristics to estimate effectiveness of treatment sequencing strategies have potential to guide treatment decisions. As a first step to a comprehensive sequencing model, we simulated progression-free survival (PFS) and overall survival (OS) for first-line doublet chemotherapy with or without bevacizumab. Parametric survival models and a logistic regression model, predicting event-type, were developed based on registry data of mCRC 867 patients to populate a discrete event simulation (DES). An exhaustive variable selection procedure was performed, considering clinical relevance and statistical performance. Models’ discrimination and calibration were assessed using bootstrapping to correct for optimism. For DES, predicted and observed medians and Kaplan-Meier plots were compared and probabilistic sensitivity analysis was performed. Models showed reasonable discrimination and good calibration. A C-statistic of 0.66 and Brier-score of 0.09 were observed for the logistic regression model. For the survival models, C-statistics were 0.65, 0.62 and 0.62 (PFS), and 0.70, 0.68 and 0.67 (OS), at 0.5, 1, and 2 years respectively. Modified Hosmer-Lemeshow statistics showed good calibration, except for short-term predictions. Simulated medians and Kaplan-Meier plots matched those observed well. Exploratory analyses estimated that cohort-level median PFS (95% confidence interval) may be further improved from 265 days (248, 280) to 288 days (270, 307) by targeting a different treatment for 219 (25%) patients. This simulation model is the first in mCRC to reflect patient heterogeneity and estimate population-level impact of treatment sequencing strategies. It is made publicly available together with an interactive data visualization tool. After further expanding the simulation model, insights in key drivers of health economic outcomes can be obtained and sequencing strategies can be identified that optimize clinical outcomes given resource constraints.