gms | German Medical Science

MAINZ//2011: 56. GMDS-Jahrestagung und 6. DGEpi-Jahrestagung

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V.
Deutsche Gesellschaft für Epidemiologie e. V.

26. - 29.09.2011 in Mainz

How to construct a control group: a comparison of different matching approaches

Meeting Abstract

Suche in Medline nach

  • Robert Krohn - AQUA-Institut, GmbH, Göttingen

Mainz//2011. 56. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds), 6. Jahrestagung der Deutschen Gesellschaft für Epidemiologie (DGEpi). Mainz, 26.-29.09.2011. Düsseldorf: German Medical Science GMS Publishing House; 2011. Doc11gmds069

DOI: 10.3205/11gmds069, URN: urn:nbn:de:0183-11gmds0699

Veröffentlicht: 20. September 2011

© 2011 Krohn.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Gliederung

Text

Background: Since 2004 German statutory health care funds are obliged to offer their insured a general practitioner (GP) centered health care contract. The Idea of these contracts is basically to strengthen the GP in his role as a gatekeeper for the health care system to achieve better quality in medical care and cost effectiveness. Requirement for the evaluation of such programs is the choice of an appropriate control group for a self selected study population. In this context, comparison of different matching methods and their evaluation is of great importance.

Methods: Our dataset consists of routine data provided by statutory health insurance funds for the years from 2004 to 2008. To evaluate the effectiveness of an enrolment model it is essential to build up a control group. The researcher has to cope with heterogeneity and self selection. Randomization is unfeasible due to the data characteristics, originated by the voluntarily design of enrolment models. Within two different evaluation projects we applied different approaches to construct a control group, inter alia propensity score matching. Comparison of the results based on different quality measures (T-Test, Kruskal Wallis Test and standardized mean difference (Cohen’s D)) to explore if there are disparities between matching methods for different variables.

Results: The different quality measures indicated that there is no gold standard for matching approaches. Each approach has its own advantages and disadvantages the researcher has to choose between. However, direct covariate matching is of limited use in practical studies where many variables must be accounted for. The use of propensity score matching is encouraged for such studies.

Conclusions: Matching approaches are the method of choice to find a control for a self selected study population. The right choice is depending on the data set, the mathematical background of the researcher and the complexity of the algorithm.


References

1.
Ariel Linden JL. Using Propensity Scores to Construct Comparable Control Groups for Disease Management Program Evaluation. Disease Management Health Outcomes. 2005;12(2):107-15.
2.
Austin PC. A critical appraisal of propensity-score matching in the medical. STATISTICS IN MEDICINE. 2008;27:2037-249.
3.
Riens B, Broge B, Kaufmann-Kolle P, Pöhlmann B, Grün B, Ose D, Szecsenyi J. Creation of a control group by matched pairs with GKV routine data for the evaluation of enrollment models. Gesundheitswesen. 2010;72(6):363-70.
4.
Rosenbaum PR, Rubin DB. The Central Role of the Propensity Score in Observational Studies for Causal Effects. Biometrika. 1983;70(1):41-55.
5.
Til Stürmer MJ. A review of the application of propensity score methods yielded increasing use, advantages in specific settings, but not substantially different estimates compared with conventional multivariable methods. Journal of Clinical Epidemiology. 2006;59(5):437-47.