Comparative simulation study on calibrating MISCAN-colon using ABC-SMC with adaptive multi-dimensional tolerance updating

This study aimed to investigate the performance of approximate Bayesian computation as a method for calibrating stochastic micro-simulation models and compare it to an algorithm used in current practice. We implemented Approximate Bayesian Computation Sequential Monte Carlo (ABC-SMC) with adaptive multi-dimensional tolerance updating and used this to estimate two colorectal cancer stool test sensitivities. As a baseline algorithm for comparison we implemented Nelder-Mead calibration extended for stochastic models. The model calibrated in this study is MISCAN-colon, developed in collaboration between Memorial Sloan-Kettering Cancer Center and Erasmus Medical Center. Calibration data was simulated with this same model. Our implementation of ABC-SMC turned out to improve the accuracy, efficiency and consistency of calibration significantly compared to Nelder-Mead.