Presentation (CTI)

OBJECTIVES: Allowing for fast cancer progression is crucial for evaluating screening strategies, as its omission can bias outcomes and overestimate cost-effectiveness. While Markov models are common, capturing rapid progression typically requires short cycle lengths, which can be computationally intensive. We aimed to demonstrate the importance of accounting for fast cancer progression in screening models across diseases with differing natural histories.
METHODS: In R, we developed three hybrid individual-level cancer screening models with differing progression speeds: bladder (fastest), kidney and prostate (slowest). The models used annual cycles to sample cancer onset and sampled time-to-progression from a calibrated Weibull distribution to define the stage distribution. Analyses were performed on a simulated population of 500,000 individuals. In cancers diagnosed before death, we calculated the sojourn time and estimated the proportion of cancers progressing through multiple stages within one year to illustrate the limitations of conventional Markov models using annual cycles.
RESULTS: For bladder cancer, the mean sojourn time was 3.96 years with 0.46% of cancers progressing from Stage II to IV within a year and 0.05% progressed through all stages within one year. In kidney cancer, with a sojourn time of 4.71 years, 2.86% of cancers progressed from Stage II to IV within a year, though none progressed through all stages within one year. In prostate cancer, despite a long sojourn time of 10.70 years, 3.49% progressed from Stage II to IV within a year and 0.13% through all stages within one year.
CONCLUSIONS: Rapid progression through multiple cancer stages within one year may occur even in slow-progressing cancers such as prostate. To avoid overestimating cost-effectiveness, cancer models should account for this possibility, either through discrete event simulation or the integration of time-to-event components.