Description and Validation of the Colorectal Cancer and Adenoma Incidence & Mortality (CRC-AIM) Microsimulation Model

Abstract

BackgroundMicrosimulation models of colorectal cancer (CRC) have helped inform national screening guidelines and health policy decision-making. However, detailed descriptions of particular underlying assumptions are not published, limiting access to robust platforms for exploratory analyses. We describe the development and validation of the Colorectal Cancer and Adenoma Incidence and Mortality (CRC-AIM) microsimulation model, a robust model built to facilitate collaborative simulation studies on disease progression and early detection through screening interventions.DesignWe used the Cancer Intervention and Surveillance Modeling Network (CISNET) CRC models, specifically CRC-SPIN, as a foundation for CRC-AIM’s formulas and parameters. In addition, we developed novel submodels and recalibrated various parameters to address gaps and discrepancies in publicly available information. Along with evaluating the natural history and screening detection outcomes from CRC-AIM, we determined the impact of using different life tables (cohort versus period) on natural history outcomes.ResultsCRC-AIM demonstrated substantial cross-model validity when comparing multiple natural history and screening outputs and probability curves to those from CISNET models, particularly CRC-SPIN. Additionally, using period life tables, CRC-AIM’s cumulative probability of developing CRC from ages 40 to 100 (7.1%) lies within the range of the CISNET models (6.7% to 7.2%). Using cohort tables, that probability increases to 8.0%. One notable difference is that, regardless of life table used, the cumulative probability of dying from CRC (3.2% for period; 3.8% for cohort) is slightly higher in CRC-AIM than the CISNET models (2.7% to 2.8%), due to CRC-AIM’s different methodology for determining survival. Additionally, there is substantial overlap (e.g. 94-95% overall agreement for strategies on and off the efficient frontier for stool-based strategies) across multiple screening overlay outputs between CRC-AIM and the CISNET models, especially CRC-SPIN.ConclusionsWe developed and validated a robust CRC microsimulation model, CRC-AIM, and demonstrate the influence of life table choice on downstream outputs. We further describe CRC-AIM’s parameters and include complete component tables to enhance transparency and encourage collaboration.

Impact of screening and follow‐up colonoscopy adenoma sensitivity on colorectal cancer screening outcomes in the CRC‐AIM microsimulation model

“…The outcomes of CRC‐AIM have been qualitatively and quantitatively validated against the CISNET models 14 . Full details of the model have been described elsewhere 14 .…”

“…The outcomes of CRC‐AIM have been qualitatively and quantitatively validated against the CISNET models 14 . Full details of the model have been described elsewhere 14 . As with all CRC microsimulation models, CRC‐AIM has natural history and screening components (additional details in Supplemental Methods).…”

“…A previous study showed that endoscopists who were aware of a positive mt‐sDNA screening result (unblinded) had a longer colonoscopy withdrawal time, found significantly more adenomas, and had a higher ADR compared with those endoscopists who were not aware of a positive test (blinded) 13 . To more accurately model the differential clinical outcomes between screening and follow‐up colonoscopy exams, we used the validated Colorectal Cancer and Adenoma Incidence and Mortality Microsimulation Model (CRC‐AIM) 14 to simulate the impact on estimated outcomes, including CRC incidence, CRC mortality, and life‐years gained (LYG), as well as adenoma miss rate (AMR) and ADR when assuming different adenoma sensitivities between screening and combined follow‐up and surveillance colonoscopies.…”

Impact of Screening and Follow-up Colonoscopy Adenoma Sensitivity on Colorectal Cancer Screening Outcomes in the CRC-AIM Microsimulation Model

“…14 Full details of the model have been described elsewhere. 14 As with all CRC microsimulation models, CRC-AIM has natural history and screening components (additional details in Supplemental Methods). The natural history component models the progression from adenoma development, to preclinical cancer, to symptomatic CRC in unscreened patients.…”

“…A previous study showed that endoscopists who were aware of a positive mt-sDNA screening result (unblinded) had a longer colonoscopy withdrawal time, found significantly more adenomas, and had a higher ADR compared with those endoscopists who were not aware of a positive test (blinded). 13 To more accurately model the differential clinical outcomes between screening and follow-up colonoscopy exams, we used the validated Colorectal Cancer and Adenoma Incidence and Mortality Microsimulation Model (CRC-AIM) 14 to simulate the impact on estimated outcomes, including CRC incidence, CRC mortality, and life-years gained (LYG), as well as adenoma miss rate (AMR) and ADR when assuming different adenoma sensitivity between screening and combined follow-up and surveillance colonoscopies.…”