A Comparison of Markov and Discrete-Time Microsimulation Approaches: Simulating the Avoidance of Alcohol-Attributable Harmful Events from Reduction of Alcohol Consumption Through Treatment of Alcohol Dependence

Background and Objective When modelling the pathophysiology of a disease, it is important to select a modelling approach that can adequately replicate its course. The objective of this paper was to compare the outcomes obtained by the Markov and discrete-time microsimulation modelling approaches using nalmefene clinical trial data. Methods Markov and microsimulation modelling approaches assessing alcohol dependence treatment with psychosocial support with or without nalmefene were compared in terms of the modelled evolution of patients’ alcohol consumption and the resulting occurrence of alcohol-attributable harmful events over 1 year. Results Comparison of the proportion of the modelled population at different levels of alcohol consumption over time revealed systematic differences arising from the different modelling techniques: a lower number of patients reaching abstinence, a higher number of patients at higher drinking levels, and, overall, a smoother evolution of alcohol consumption in the microsimulation. Reasons are discussed in the paper. While the models produced similar occurrences of alcohol-attributable harmful events as a whole, distinct results for the individual events were observed, explained by the specific pathophysiology of occurrence of these events and how their implementation was adapted to fit the limitations of the compared modelling approaches; however, these differences were only statistically significant for one of the eight events. Conclusions For a general public health or health economic assessment of alcohol use disorders, it is possible to achieve similar results with the compared approaches. To assess a patients’ disease course, taking into consideration alcohol-attributable harmful events, the microsimulation approach may provide more precise results. However, further external validation of the models is needed and this additional precision may be outweighed by the greater computational burden of a microsimulation approach.