TL;DR
This paper introduces CMOST, a novel, open-source microsimulation framework for modeling colorectal cancer (CRC) natural history and screening strategies. By providing a transparent and adaptable tool, CMOST aims to facilitate cost-effectiveness analyses for diverse healthcare scenarios and screening interventions, addressing a critical need for robust computational models in CRC prevention.
What Problem Does This Paper Address?
Colorectal cancer (CRC) remains a significant cause of mortality, and while screening strategies like colonoscopy can reduce this, optimizing these strategies is challenging. Randomized controlled trials for CRC prevention are often impractical due to the need for large populations and long follow-up periods. Existing computational microsimulation models that could address these questions are typically not publicly available, hindering independent verification, adaptation, and further development. This creates a gap in accessible tools for designing and evaluating new screening interventions and tailoring them to local conditions.
Methods
The authors developed CMOST, a microsimulation model implemented in MATLAB. It simulates the natural history of CRC, including adenoma initiation, progression through six distinct stages, and transition to cancer. The model also incorporates the effects of screening interventions like colonoscopy and rectosigmoidoscopy, accounting for detection probabilities, complications, and polypectomy. A key methodological contribution is the implementation of an automated parameter calibration system that allows the model to be fitted to epidemiological benchmarks for adenoma prevalence and CRC incidence. The model also allows for the exploration of different "adenoma dwell times" (the time from adenoma initiation to cancer), offering three parameterized versions (CMOST8, CMOST13, CMOST19) to assess the impact of this uncertain parameter.
Datasets
The model was calibrated using published epidemiological data for adenoma prevalence and CRC incidence. Specifically, it utilized age- and sex-specific adenoma prevalence rates and data from the Surveillance Epidemiology and End Results (SEER) database for colon and rectum cancer incidence and mortality rates (2005-2009). The model's predictions were validated against data from a large randomized rectosigmoidoscopy screening trial that was not used for calibration.
Key Insights and Ideas
- Open-Source is Crucial: The lack of publicly available microsimulation models is a significant barrier. CMOST's open-source nature is a major step towards transparency, reproducibility, and collaborative advancement in CRC modeling.
- Adenoma Dwell Time Matters, But Robustness Exists: While the exact adenoma dwell time is unknown and a critical parameter, the study demonstrates that CMOST can produce similar predictions for key outcomes (incidence, mortality, cost-effectiveness) across a reasonable range of dwell times (8, 13, and 19 years). This suggests a degree of robustness in the model's conclusions despite this uncertainty.
- Optimizing Screening is Complex: The optimal timing for a single screening colonoscopy varies depending on the outcome of interest (incidence reduction, mortality reduction, or life years gained), highlighting the trade-offs involved in screening strategy design.
- Cost-Effectiveness is Achievable: Even with a 3% annual discount rate, single screening colonoscopies are predicted to be highly cost-effective, with optimal costs per life year gained well below the commonly accepted threshold of $100,000 USD.
Key Results
- CMOST's predictions for CRC incidence and mortality reduction were highly similar to those of a large randomized rectosigmoidoscopy screening trial, validating the model.
- The optimal age for a single screening colonoscopy to maximize incidence reduction was around 58 years, while for maximizing discounted life years gained, it was around 49-50 years.
- The most cost-effective single screening colonoscopy was found to be at age 59, costing $17,211 discounted USD per life year gained (for the CMOST13 version).
- CMOST's predictions for incidence and mortality reduction were comparable to other established microsimulation models like MISCAN, CRC-SPIN, and SimCRC.
- The model demonstrated that screening costs for colonoscopy remained below the $100,000 USD per life year gained threshold, indicating cost-effectiveness.
Limitations and Future Work
The authors acknowledge several limitations:
- Unknown Natural History Parameters: Key parameters, particularly the adenoma dwell time, remain uncertain. While CMOST offers flexibility by allowing different dwell times, the precise value is unknown.
- Serrated Adenoma Pathway Not Explicitly Modeled: The model does not explicitly account for the serrated adenoma pathway, which may have different progression characteristics. However, it does include a "direct cancer" pathway to partially address cancers that may arise without detectable adenomas.
- No Benign Polyps: The model does not account for benign non-adenomatous polyps.
- Calibration to Contemporary Data: CMOST was calibrated to contemporary CRC incidence data, unlike some older models that used data from 1977. This difference can explain some discrepancies in predictions.
Future work could involve incorporating the serrated adenoma pathway more explicitly, further refining parameter calibration with new data, and adapting the model for use in diverse epidemiological settings, such as developing countries. The open-source nature of CMOST is expected to facilitate these advancements.
Summary updated