How cervical screening interferes with cervical cancer and cervical cancer death is a complex matter. Therefore, also the models we develop to help optimizing screening interventions, are not simple. Nor is it simple to compare the models. Especially where the models differ in structure, parameters act differently and cannot be directly compared. This most often is the case for the parts in de models where unobservable processes are described: the natural history of disease and how screening and subsequent treatment interfere with that history. Any tools to clarify similarities and differences between models are welcome.
Cancer incident and death cases after screening participation have 3 possible origins: 1) the cancer-precursor was not present yet at the time of the screening, 2) the precursor was missed by the screening, or 3) treatment of the screen-detected precursor was not successful in preventing further cancer and cancer death. In models, we can discern these cases and in CISNET, we developed the MCLIR method to do so. It is based on comparing model output of specially designed - partly hypothetical - screening scenarios. We considered 3 of the CISNET cervical models, as they were developed independently before CISNET started, for the USA, Australia and the Netherlands. We compared them for cytology screening effectiveness after a one time screening, and compared the contribution of the three reasons for screening failure.
Overall, the patterns in incidence and mortality reduction after screening were quite similar between the models. This was particularly true for the role of the duration of cancer precursors. The models differed somewhat in the ability of cytology to detect these precursors. The models differed most in how successful treatment of screen-detected conditions was in preventing further morbidity and mortality.