Purpose: Despite recognized limitations, Markov cohort models remain the standard modeling method adopted in economic evaluations. Although discrete event simulation (DES) circumvents many of these limitations, it has not yet gained much popularity. This study compares the techniques' ability to address a common research question.

Methods: A cohort Markov model (SHAPE programmed in Microsoft© Excel) and an individual patient DES (RIVER in Arena©) evaluating long-term outcomes and the cost-utility of rimonabant in overweight or obese individuals were compared. Common data sources and disease concepts were used. In addition to a comparison of predictions, the models were evaluated in terms of data requirements, file size, computational burden, development time and breadth of results.

Results: SHAPE is 8.1MB while the Arena© RIVER file is 3.6 MB. Analyses with SHAPE take a few seconds, whereas simulation of 1,000 patients takes 30-45 seconds with RIVER. Core programming time was similar for both models, but testing and validation time were roughly 50% longer for RIVER. Additional time was required to develop inputs to RIVER because of the need to create multivariate time- and patient-specific equations for risk factors changes – SHAPE only requires mean values. Outcomes are in the same decision range for both models, but are more favorable in RIVER because retaining information on individuals' medical histories allows the contribution of time spent with improved risk factors (e.g., lipid levels) to be accounted for even after risk factors change upon discontinuation of treatment. Diabetes complications are also more precisely modeled. RIVER provides more detailed results, including distributions of changes in risk factors, proportions of patients attaining risk factor goals, and number of admissions for complications by location of care. Subgroups are immediately specifiable within RIVER, while for SHAPE, such analyses require updating inputs for each subgroup and running patient profiles separately.

Conclusions: The DES required additional analysis time to set up inputs, additional validation time and longer run times. These differences are not substantial in light of the additional outcomes and precision allowed. The increased precision provided by DES, and the simplifications required in the Markov model, led to better cost-utility results with the simulation. Although in this case, both results are in the same range, differences in modeling technique may have an impact on policy decisions in borderline situations.