Calibration and Validation of the Microsimulation Model in Cancer of the Bowel (MiMiC-Bowel), an Individual Patient Simulation Model for Investigation of the Cost-effectiveness of Personalised Screening and Surveillance Strategies

The Microsimulation Model in Cancer of the Bowel (MiMiC-Bowel) incorporates individual cancer risk to inform colorectal cancer (CRC) screening decisions in England. This work reports calibration and cross-validation of the MiMiC-Bowel.

Due to the complex and computationally intensive nature of the model a step-wise calibration approach was taken utilising manual and automated algorithmic fitting. Natural history disease model parameters were estimated via calibration to several data targets including English and adjusted German data relating to previously unscreened persons. The model was cross-validated to four international models. Incompatibility in calibration data was assessed by analysing CRC incidence, convergence feasibility, and predictions of screening test sensitivity. Data incompatibility was addressed by giving less weight in fitting the parameters to target data with lower reliability and applicability, and adjusting data sets to reflect differences between localities. The MiMiC-Bowel predicted 60% sensitivity of flexible sigmoidoscopy for CRC and 59% for high-risk adenoma. MiMiC-Bowel’s predictions of CRC cases that arise from potentially detectable adenomas and impact of perfect polypectomy on risk reduction were within ranges reported by the US models, but the predictions of the sojourn and dwell time were not (29.1 and 5.3 years vs 8-24 and 2-4 years respectively). CRC risk increased less rapidly by age in MiMiC-Bowel than German data suggests, which is supported by population data on pre-screening CRC incidence.