An Empirical Comparison of an Individual Sampling Model and Markov Model for Screening of Atrial Fibrillation: Final Results
Posted: 12 Jun 2007
Date Written: 2007
Background: There are a number of modelling approaches available to health economists when faced with a decision problem. One of the main areas of uncertainty is at what level of structural complexity a patient level simulation should be the approach of choice rather than a Markov cohort simulation. A decision problem with the potential for structural complexity is screening for atrial fibrillation (AF), where patient characteristics strongly influence patient pathways. AF is an abnormal heartbeat rhythm, most prevalent in patients aged 65 and over, whose incidence increases with age and individuals with the condition are five times more likely to have a stroke. The condition can be diagnosed with an ECG, and subsequently treated, however there is currently no formal screening programme in the UK.
Objective: The primary objective was to compare the processes and outputs of two alternative types of model for the decision problem of screening for AF, and to determine the most appropriate method for this case study.
Methods: Data from published sources and a randomised controlled trial of AF screening was used to populate two decision analytic models addressing the long-term cost-effectiveness of screening for AF: a Markov cohort model and an individual sampling model (ISM). Initially, a comprehensive ISM was constructed to address the policy question. Following this, a Markov model was constructed to mirror the ISM. Base case and probabilistic sensitivity analyses were conducted. The models were compared using a number of criteria including structure, analytical input and results.
Results: The nature of the Markov model prevented the inclusion of some of the features of the ISM. Therefore, the ISM was amended so that the two model types could be directly compared in terms of their results. Although the output of these two models was broadly similar, a number of issues highlighted the differences between the model processes and output. The ISM was able to capture the full complexity of the decision problem by considering appropriate patient attributes such as AF status and stroke history. A key feature of the ISM was the inherent flexibility of the model. Therefore a number of features were incorporated in the ISM with greater ease, and in some cases were not achievable in the Markov model without the structural complexity becoming impractical. In both models, probabilistic sensitivity analysis (PSA) was feasible. However, the first-order uncertainty component of the ISM, representing realistic patient variability, resulted in greater uncertainty around the cost-effectiveness result. Only when the number of individuals the ISM runs for became very large did the result approach that produced by the Markov model, which considered an (unrealistic) infinite population.
Conclusions: In this clinical context, the ISM demonstrated greater flexibility and richness of structure by incorporating patient attributes in a simple and intuitive way compared with the Markov model where a large number of health states were required. When faced with a decision problem, the choice of model type is critical and the complexity of the decision problem needs to be considered carefully.
Keywords: decision modelling, patient level simulation, economic evaluation
Suggested Citation: Suggested Citation