Article
Systematic evaluation of methodological features in decision-analytic prostate cancer screening models
Search Medline for
Authors
Published: | September 13, 2012 |
---|
Outline
Text
Background: Prostate cancer remains the second most frequent cancer diagnosis for men, and its incidence is increasing. In spite of results from randomized trials, still disagreement exists whether prostate specific antigen (PSA) screening for prostate cancer causes more benefit than harms. Aspects of overdetection and overdiagnosis due to screening are discussed as well. Therefore, decision-analytic modelling plays an important role in decision making regarding prostate cancer screening.
Objectives: To identify the different methodological features used in published screening models for prostate cancer and to give an insight into their similarities and their differences in order to derive recommendations for future screening models.
Methods: We performed a systematic literature search for decision-analytic PSA-screening models in Embase and Pubmed. Methodological features of the models were extracted and compared. Besides common model features such as model type, simulation technique, time horizon and model outcomes, we focused on more specific aspects of cancer screening models. These were how incidence, cancer mortality, test performance, benefit and harms of treatment, overdiagnosis and overtreatment were modelled, including which data were used. A particular point of interest was whether the models applied a stage-shift or cure approach to model the effect of screening.
Results: The majority of included models were economic models, followed by benefit-risk, explanatory and other models. Model types used were decision trees, Markov models, discrete-event simulations and mathematical equation models. Screening strategies varied widely among the studies, differing in screening intervals, parallel screening tests with Digital Rectal Exam (DRE) or Transrectal Ultrasound (TRUS) and Prostate specific antigen-cutoff points. The common diagnostic confirmation was done by biopsy. Only a few models explicitly modelled cancerogenesis, specific cancer treatment, or overdetection. A minority considered quality of life, allowing to evaluate benefits and harms of screening with the same metric. Stage-shift and cure approaches were both used to model the effect of screening.
Conclusions: Examined models built a very heterogeneous group of decision-analytic screening models. The two main methodological approaches were stage-shift and cure models. As results from these approaches differed, further research should focus on such discrepancies.