Comparative evaluation of methods approximating drug prescription durations in claims data: modeling, simulation, and application to real data.

Purpose The purpose of this study was to compare the predictive accuracy of different methods suggested for approximation of drug prescription durations in claims data. Methods We expanded a well-established modeling and simulation framework to compare approximated drug prescription durations with ‘true’ (i.e., simulated) durations. Real claims data of persons aged ≥65 years insured by the German nationwide ‘Statutory Health Insurance Fund’ AOK between 2010 and 2012 provided empiric input parameters that were completed with missing information on actual dosing patterns from an observational cohort. The distinct approximation methods were based on crude measures (one tablet a day), population-averaged measures (defined daily doses), or individually-derived measures (longitudinal coverage approximation of the applied dose, COV). As a proof-of-principle, we assessed the methods' performance to predict the well-characterized bleeding risks of anticoagulant, antiplatelet, and/or non-steroidal anti-inflammatory drugs. Results When applied to modeling and simulation data sets, the closest, least biased, and thus most accurate approximation was observed using the COV approximation. In a real-data example, rather similar results to an external reference were obtained for all methods. However, some of the differences between methods were meaningful, and the most reasonable and consistent results were obtained with the COV approach. Conclusion Based on theoretically most accurate approximations and practically reasonable estimates, the individual COV approach was preferable over the population-averaged defined daily dose technique, although the latter might be justified in certain situations. Advantages of the COV approach are expected to be even bigger for drug therapies with particularly large dosing heterogeneity. Copyright © 2016 John Wiley & Sons, Ltd.