Methods We present a Markov model of population screening for abdominal aortic aneurysms, with parameter estimates taken from a large randomised trial of screening. This model will enable estimation of the long-term cost-effectiveness of screening, but must first be validated and consideration given to structural and parameter uncertainty. To assess internal validity, predictions from the model are compared with the observed data in the trial at four years of follow-up. Comparisons with the trial are made in terms of costs and outcomes, in addition to numbers and timing of key events. The degree of complexity of the model is addressed by investigating the full model alongside possible simplified structures. Probabilistic sensitivity analysis of the model at four years is undertaken to assess the influence of parameter uncertainty.

Results The full model proposed includes delays to receiving surgery, incidental aneurysm detection, and non-attendance at screening. Not all parameters could be estimated directly from the trial data. This model validates well with the observed data from the trial in terms of key events such as mortality and surgical interventions, both in total and cumulatively over time. Censoring in the trial is accommodated in order to make these comparisons. Simplified model structures do not validate so well, and are unlikely to be reliable for long-term extrapolation. Probabilistic sensitivity analysis suggests that increased model complexity results in increased imprecision of cost-effectiveness results.

Conclusions Using this model as an exemplar, we conclude that internal validation of cost-effectiveness models should include three components: comparison of observed and predicted total numbers of key events, numbers of events over time, and mean costs and life-years. Censoring and half-cycle adjustment may need to be incorporated, as well as the delays that often occur in the practical implementation of medical interventions.