Development and validation of a model to predict outcomes of colon cancer surveillance.

ABSTRACT

PURPOSE: Clinical trials suggest that intensive surveillance of colon cancer (CC) survivors to detect recurrence increases curative-intent treatment, although any survival benefit of surveillance as currently practiced appears modest. Realizing the potential of surveillance will require tools for identifying patients likely to benefit and for optimizing testing regimens. We describe and validate a model for predicting outcomes for any schedule of surveillance in CC survivors with specified age and cancer stage. METHODS: A Markov process parameterized based on individual-level clinical trial data generates natural history events for simulated patients. A utilization submodel simulates surveillance and diagnostic testing. We validate the model against outcomes from the follow-up after colorectal surgery (FACS) trial. RESULTS: Prevalidation sensitivity analysis showed no parameter influencing curative-intent treatment by > 5.0% or overall five-year survival (OS5) by > 1.5%. In validation, the proportion of recurring subjects predicted to receive curative-intent treatment fell within FACS 95% CI for carcinoembryonic antigen (CEA)-intensive, computed tomography (CT)-intensive, and combined CEA+CT regimens, but not for a minimum surveillance regimen, where the model overestimated recurrence and curative treatment. The observed OS5 fell within 95% prediction intervals for all regimens. CONCLUSION: The model performed well in predicting curative surgery for three of four FACS arms. It performed well in predicting OS5 for all arms.