Effects of Screening for Colorectal Cancer: Development and Validation of a Multistate Markov Model

Abstract

Simulation models are a powerful tool to overcome gaps of evidence needed to inform medical decision making. Here, we present development and application of a multistate Markov model to simulate effects of colorectal cancer (CRC) screening, along with a thorough assessment of the model's ability to reproduce real-life outcomes. Firstly, we provide a comprehensive documentation of the model development, structure and assumptions. Secondly, to assess the model's external validity, we compared model-derived cumulative incidence and prevalences of colorectal neoplasms to (1) results from KolosSal, a study in German screening colonoscopy participants, (2) registry-based estimates of CRC incidence in Germany, and (3) outcome patterns of randomized sigmoidoscopy screening studies. We found that (1) more than 90% of model-predicted neoplasm prevalences were within the 95% confidence intervals of the observed prevalences in the KolosSal study; (2) the 15-year cumulative CRC incidences estimated by simulations for the German population deviated by 0.0-0.2 percent units in men and 0.0-0.3 percent units in women when compared to corresponding registry-derived estimates; and (3) the time course of cumulative CRC incidence and mortality in the modelled intervention group and control group closely resembles the time course reported from sigmoidoscopy screening trials. Summarized, our model adequately predicted colorectal neoplasm prevalences and incidences in a German population for up to 25 years, with estimated patterns of the effect of screening colonoscopy resembling those seen in registry data and real-world studies. This suggests that the model represents a valid tool to assess the comparative effectiveness of strategies for CRC screening.

Strongly Divergent Impact of Adherence Patterns on Efficacy of Colorectal Cancer Screening: The Need to Refine Adherence Statistics

“…19 The model's external validity (its ability to reproduce real-life outcomes) was assessed by comparing model-derived cumulative incidences and prevalences to results from KolosSal, a German cohort study, registry-based estimates of CRC incidence in Germany, and outcome patterns of randomized sigmoidoscopy screening studies. 20 The model source code, developed in the statistical software R (version 3.6.3), is available in Supplementary Appendix 1b.…”

“…Furthermore, the model was subjected to a thorough assessment of its external validity, and was found to adequately predict colorectal neoplasm prevalences and incidences in a German population, with estimated patterns of the effect of screening colonoscopy resembling those seen in registry data and real-world studies. 20 Major limitations concern model simplifying assumptions and uncertainties related to input parameters where evidence was limited, for instance with regard to transition rates for age groups 50-54 years and 80+ years, true screening test performance characteristics in Germany and potential differences between sexes in this respect. In addition, the chosen approach of a Markov-type simulation model is very well suited to quantify average effects on the level of populations, but less so for individuals potentially at substantially lower or higher risk of CRC.…”

“…Specific strengths and limitations of the model per se have been described previously. 10,20,40 Briefly, a major strength is the use of input parameters derived specifically from the German general population using the world's largest screening colonoscopy registry. Furthermore, the model was subjected to a thorough assessment of its external validity, and was found to adequately predict colorectal neoplasm prevalences and incidences in a German population, with estimated patterns of the effect of screening colonoscopy resembling those seen in registry data and real-world studies.…”