Abstract
Background: To develop a novel simulation model for ductal carcinoma in situ (DCIS), fully validate it, and provide new estimates for DCIS in the setting of population-based biennial screening.
Methods: A micro-simulation Markov model for DCIS (SimDCIS) was developed. Input parameters were independently derived from literature and transition parameters were age- and grade-dependent. The model was applied to the Dutch biennial screening program. SimDCIS was internally, cross, and externally validated by comparison of the model output to data from the Netherlands Cancer Registry, a previously published modelling study on the United Kingdom (UK) Frequency Trial, and the UK screening program, respectively. Univariate and probabilistic sensitivity analyses were performed to estimate uncertainty. DCIS regression, progression to invasive breast cancer (IBC), clinical detection, and screen-detection were estimated in the Dutch screening setting.
Results: SimDCIS excellently matched observed data in internal, external, and cross validation. The model was most sensitive to changes in DCIS onset probability, and the maximum variation in the screen-detection rate was 11%. In the Dutch screening setting, DCIS regression, progression to IBC, clinical detection, and screen-detection was estimated at 7% (0-14%), 19% (15-24%), 7% (0-14%), and 63% (58-68%), respectively. Grade distribution was 20% grade 1, 38% grade 2, and 42% grade 3.
Conclusion: SimDCIS provides strong predictive accuracy across validation methods and is particularly sensitive to changes in DCIS onset probability. Most DCIS will be found through screening, of which less than 50% of DCIS will be grade 3, less than 1 in 10 will regress, and only 1 out of 5 DCIS will progress to IBC in the setting of biennial screening.