[Creation of a control group by matched pairs with GKV routine data for the evaluation of enrollment models]

Gesundheitswesen. 2010 Jun;72(6):363-70. doi: 10.1055/s-0030-1249687. Epub 2010 May 4.
[Article in German]


Background: Various study approaches can be considered for the investigation of the efficiency of enrollment models, like GP-centred health-care contract or disease management programmes. As an active and independent enrollment into care models is effected by the insured (self-selection), a randomisation cannot be applied. The matched pairs design - in which for every insured a control insured with comparable morbidity is selected - presents an alternative investigation method. A precondition is a model that describes appropriately the morbidity on the basis of available routine data.

Target: The aim of this study was to develop a procedure that selects comparable insured persons on the basis of routine data of the statutory health-care funds.

Methods: Apart from age, gender, care status, insured status, days of disability, region, health insurance and belonging to an enrollment model, also ambulant as well as stationary performance data for the year 2005 following the PCG/DCG procedure for morbidity-oriented matching design developed by Lamers and Vliet (2003) were applied. Thereby the consumption of certain medications prescribed is determining for the allocation of patients to pharmaceutical cost groups (PCG). Additionally a classification into diagnosis cost groups (DCG) according to stationary diagnoses was conducted.

Results: Within the scope of the enrollment models the formation of matched pairs following the PCG/DCG procedure represents an appropriate study design for the creation of a control group. In the first year of enrollment the insured of the interventional and those of the control group show a comparable morbidity. When applying 9 matching criteria a control insured person can be found for 87% of the enrolled individuals.

Discussion/conclusions: There are various and complex possibilities to define morbidity. Variable parameters within the presented matched pairs design are the number of used matching criteria as well as minimum drug consumption limit relevant for the classification in PCGs. Alternative models are possible for morbidity definition considering, apart from the stationary diagnosis, also the ambulant diagnosis. When taking into account a higher number of morbidity criteria, the matched pairs design is confronted with dimensionality issues. The propensity score matching is discussed as approach to solve this problem.

Publication types

  • English Abstract

MeSH terms

  • Computer Simulation
  • Data Interpretation, Statistical*
  • Germany
  • Health Services Accessibility / statistics & numerical data*
  • Insurance, Health / statistics & numerical data*
  • Models, Statistical*