gms | German Medical Science

22. Jahrestagung der Gesellschaft für Arzneimittelanwendungsforschung und Arzneimittelepidemiologie (GAA)

Gesellschaft für Arzneimittelanwendungsforschung und Arzneimittelepidemiologie

03.12. - 04.12.2015, Dresden

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

Meeting Abstract

  • author presenting/speaker Andreas Meid - University of Heidelberg, Department of Clinical Pharmacology and Pharmacoepidemiology, Heidelberg, Germany
  • Renate Quinzler - University of Heidelberg, Department of Clinical Pharmacology and Pharmacoepidemiology, Heidelberg, Germany
  • Jürgen-Bernhard Adler - Wissenschaftliches Institut der AOK (WIdO), Berlin, Germany
  • Dirk Heider - Hamburg Center for Health Economics, Hamburg, Germany
  • Hans-Helmut König - Hamburg Center for Health Economics, Hamburg, Germany
  • corresponding author Walter-Emil Haefeli - University of Heidelberg, Department of Clinical Pharmacology and Pharmacoepidemiology, Heidelberg, Germany

Gesellschaft für Arzneimittelanwendungsforschung und Arzneimittelepidemiologie e.V. (GAA). 22. Jahrestagung der Gesellschaft für Arzneimittelanwendungsforschung und Arzneimittelepidemiologie. Dresden, 03.-04.12.2015. Düsseldorf: German Medical Science GMS Publishing House; 2015. Doc15gaa04

doi: 10.3205/15gaa04, urn:nbn:de:0183-15gaa040

Veröffentlicht: 9. Dezember 2015

© 2015 Meid et al.
Dieser Artikel ist ein Open-Access-Artikel und steht unter den Lizenzbedingungen der Creative Commons Attribution 4.0 License (Namensnennung). Lizenz-Angaben siehe



Background: In older people, serious adverse drug reactions are a frequent cause of hospital admissions. Many of them are caused by drug-drug interactions (DDIs) that modify the effect of simultaneously or successively administered drugs [1]. Therefore, the evaluation of DDIs requires accurate determination of the (co-)medication, treatment durations, and temporal overlap [2]. However, claims data from the German health care system do not offer explicit information on the days of supply that could be used to determine drug prescription durations and thus drug exposure. Consequently, accurate definition of concomitant usage requires adequate approximation of the prescription durations [2]. Among the different options to achieve this goal, no comparative evaluations have been carried out so far.

Our objective was to determine the most appropriate method to approximate prescription durations of the potentially interacting drug classes of non-steroidal anti-inflammatory drugs (NSAIDS) and antithrombotic drugs. This goal was pursued by means of modeling and simulation (M&S) to obtain ‘true’ durations in simulated claims data sets. Assessing the temporal relationship between concomitant drug use and clinical events in an exemplary application, we further investigated the impact of different approximation methods on clinical outcomes.

Materials and Methods: Although actual dosing patterns, adherence, and thus true durations are unknown in claims data, M&S can incorporate such information derived from well-characterized cohorts and use such data on dosing regimens as simulation input to generate representative claims data sets [3]. We determined the agreement of approximation methods with simulated ‘true’ durations by means of the performance measures of mean durations, mean relative bias, and mean absolute error. Methods for approximation of prescription durations can be based on package sizes [4] (i.e. one tablet a day, OTAD), the defined daily dose (DDD) [5] defined as the assumed average maintenance dose per day in a drug’s main indication [6], or the drug coverage (COV) based on an averaged fraction of prescribed dose units obtained from the longitudinal prescription history (or a standard dose in case of a single prescription) [7].

Analyses of clinical outcomes were based on claims data of the “Statutory Health Insurance Fund” AOK (nationwide). We identified patients eligible for antithrombotic medication and longitudinally followed them up for 24 months for hospitalization due to bleeding events. We assessed the impact of drug exposure to antithrombotic and NSAID treatment by time-dependent Cox proportional hazards models because drug treatments may be transient and may change during the study period. Statistical analyses were performed using the R software/environment version 3.1.0.

Results: Incorporating the inter-individual dosing variability of older people into the simulation of claims data sets allowed us to compare the approximated prescription durations to simulated ‘true’ durations. Overall, the longitudinal approach of coverage approximation (COV) yielded the closest and thus best results followed by proposed durations based on DDDs. This was evident in all performance measures.

We then investigated the potential impact of different approximation strategies by applying them to a real claims data set. Cox regression estimates differed only marginally between distinct approximation methods and were close to an external reference demonstrating a substantially elevated bleeding risk of the concomitant usage of both drug classes [7].

Conclusion: When investigating a heterogeneous group of drugs modifying the effect of another co-administered drug, longitudinally averaging the coverage was the best approach to approximate drug prescription duration and thus exposure. In an exemplary application, we could corroborate existing evidence on an elevated bleeding risk upon concomitant administration of NSAIDs and antithrombotic drugs. Generally, our results encourage researchers to conduct various sensitivity analyses when investigating claims data, not only regarding study designs and outcome definitions, but also concerning methods to determine drug exposure.

This study was supported by grants 01GY1329B and 01GY1320B from the German Ministry of Education and Research (BMBF)


Marengoni A, Pasina L, Concoreggi C, Martini G, Brognoli F, Nobili A, et al. Understanding adverse drug reactions in older adults through drug-drug interactions. Eur J Intern Med. 2014 Nov;25(9):843-6.
Tobi H, Faber A, van den Berg PB, Drane JW, de Jong-van den Berg LT. Studying co-medication patterns: the impact of definitions. Pharmacoepidemiol Drug Saf. 2007 Apr;16(4):405-11.
Murray RE, Ryan PB, Reisinger SJ. Design and validation of a data simulation model for longitudinal healthcare data. AMIA Annu Symp Proc. 2011;2011:1176-85.
Mangiapane S, Busse R. Prescription prevalence and continuing medication use for secondary prevention after myocardial infarction: the reality of care revealed by claims data analysis. Dtsch Arztebl Int. 2011 Dec;108(50):856-62
World Health Organisation Collaborating Centre for Drug Statistics Methodology. ATC classification index with DDDs. 2014. Available from: Externer Link
Abbas S, Ihle P, Harder S, Schubert I. Risk of hyperkalemia and combined use of spironolactone and long-term ACE inhibitor/angiotensin receptor blocker therapy in heart failure using real-life data: a population- and insurance-based cohort. Pharmacoepidemiol Drug Saf. 2015 Apr;24(4):406-13.
Schjerning Olsen AM, Gislason GH, McGettigan P, Fosbol E, Sorensen R, Hansen ML, et al. Association of NSAID use with risk of bleeding and cardiovascular events in patients receiving antithrombotic therapy after myocardial infarction. JAMA. 2015 Feb 24;313(8):805-14.