An evaluation of prostate cancer screening using a decision analytic model

As early prostate cancer is often asymptomatic, the disease is often not diagnosed until it has reached an advanced, incurable stage. However, if diagnosed when it is still confined to the prostate, prostate caner is potentially curable. At present, there are no completed prospective evaluations or other scientific evidence to suggest that prostate cancer mortality is reduced or that quality of life is increased either by curative treatment or by screening programmes. However, the potential effects of a screening programme can be modelled using decision analytic computer software. The aim of this project was to develop a model for comparing the expected quality-adjusted life expectancy for a group of men subjected to a programme of screening for prostate cancer to that of a control group with conventional case-finding, given limited empirical data on the effectiveness of both strategies. Due to limitations of existing data, the analysis was developed during the early stages of evaluating a prostate cancer screening programme. We wish to illustrate how an analytic tool such as this model can be used in the future. The model was functionally separated into two parts. The first part, the observation submodel, covers the first 6 years of the programme and classifies the screened population into different quality of life states, based on observed outcomes. The second part, the simulation submodel, predicts quality adjusted life-years for healthy participants and for prostate cancer patients who receiving curative treatment or expectant management. Data for the model were obtained from a pilot programme on screening for prostate cancer in which a randomly selected population sample has been screened by digital rectal examination in Norrköping Sweden in 1987 and 1990. The outcome evaluated in the model was quality-adjusted life expectancy for individuals in the cohort invited to the screening programme, compared to results for population controls. While the preliminary results identify some health benefits associated with screening, more accurate empirical data for a number of key variables could improve the evaluation. One feature of this model is that it combines solid, empirical data from the observation submodel with simulated results. When better empirical data on the actual consequences of different strategies are available, they can easily be analysed by using this model