Modelling issues in cancer screening

Abstract

The two main goals of modelling cancer screening are data analysis and evaluation. In data analysis, analytical-numerical statistical models are used to test hypotheses about preclinical disease, the screening test, and the association between early detection and risk of dying from the cancer. Evaluation in cancer screening is supported by model-based prediction of screening effects and cost-effectiveness. Simulation models are suitable for these tasks, and can also be used to identify efficient age-ranges and intervals between screening tests. Striking differences exist between screening models for cervical cancer and breast cancer, which are the two cancer types for which screening is common practice. The two main problems in cervical cancer screening are the proportion of progressive and regressive among screen-detected lesions, and the impact of screening on incidence and mortality. In breast cancer, regression is not (yet) a big issue, and the relationship between screening and mortality reduction has been demonstrated in randomized controlled trials (at least for women older than 50 years). The weakest link in current breast cancer models is the association between earliness of detection and improvement in prognosis. The modelling outcomes and their usefulness are decisively influenced by the data sets that were used in quantifying the model, and the subclassifications of the data that were considered. New or pending modelling issues include HPV-based screening in cervical cancer, screening models for colorectal cancer, the use of surrogate outcome measures and model-based meta-analysis of screening trials.

An analysis of the efficacy of serial screening for familial nasopharyngeal carcinoma based on Markov chain models

“…Transition probabilities were derived between successive phases applying the Markov chain model [8,9]. This model is progressive and non-reversible.…”

The MISCAN-COLON Simulation Model for the Evaluation of Colorectal Cancer Screening

“…The incomplete knowledge of the natural history of the disease makes it difficult to estimate the effectiveness of screening strategies. As has been shown for other cancers, mathematical models can be useful to test hypotheses about the incidence and natural history of disease and screening characteristics and subsequently to make predictions of the (cost-) effectiveness of screening strategies (5). Both Eddy (6) and Wagner et al (7) have constructed models to estimate the potential benefits and costs of several CRC screening strategies, including FOBT, FSIG, BE, and CSCPY screening.…”

Using observational data to estimate an upper bound on the reduction in cancer mortality due to periodic screening

“…in the absence of screening) conditional on no death from competing risk. Previous approaches [1-3] used data from refusers, substituting m R for m A . However this requires a strong unreasonable assumption as well as data from refusers, which is often not available.…”