gms | German Medical Science

Kongress Medizin und Gesellschaft 2007

17. bis 21.09.2007, Augsburg

Comparing Propensity Score based methods with traditional regression techniques in extreme situations using a random reallocation approach

Meeting Abstract

Suche in Medline nach

  • Susanne Stampf - University Medical Center Freiburg, Freiburg
  • Angelika Caputo - University Medical Center Freiburg, Freiburg
  • Claudia Schmoor - University Medical Center Freiburg, Freiburg

Kongress Medizin und Gesellschaft 2007. Augsburg, 17.-21.09.2007. Düsseldorf: German Medical Science GMS Publishing House; 2007. Doc07gmds401

Die elektronische Version dieses Artikels ist vollständig und ist verfügbar unter:

Veröffentlicht: 6. September 2007

© 2007 Stampf et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen ( Er darf vervielf&aauml;ltigt, verbreitet und &oauml;ffentlich zug&aauml;nglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.



Background: In observational studies the exposure effect estimated via regression techniques is often compared to estimations from a Propensity Score (PS) based approach, though these techniques differ in their way to use covariate information. Often both methods lead to similar results when both exposure and response depend on the same explanatory variables [1] . Different results typically occur if the sets of covariates for exposure and response are different or even disjoint.

Methods: We investigate the properties of the two approaches in extreme situations by means of so-called random reallocation (RARE) [2] of a real data set of patients undergoing coronary surgery. The aim of this prospective observational study was to analyze the impact of the degree of platelet inhibition (exposure status) on the 30-day clinical response to elective percutaneous catheter intervention (PCI) with Clopidogrel loaded stents [3]. Investigated response variables were "any severe complication after PCI" and Troponin concentration.

The aim of RARE is to create extreme situations characterized by conditional independencies. After selecting a binary/categorial variable A whose direct dependence structures should remain, the remaining variables are divided into subsets X and Y between which independence conditional on A should hold. One observation is rewritten as (X,Y,A). Within each category j of A the observation (X_i,Y_i) is split into X_i and Y_i and the set of Y_i is randomly reallocated to X_i, i=1,...,n_j. In contrast to pure simulation studies, the real dependence structure within subsets remains, and we obtain a complex and realistic data set where we can compare effect estimates.

Results: RARE offers an opportunity to investigate the behaviour of different estimation approaches in specific settings. We briefly describe the underlying idea of RARE and present the application to the data with the interpretation of results.


Stürmer T, et al. A review of the application of the propensity score methods yielded increasing use, advantages in specific settings, but not substantially different estimates compared to conventional multivariable models. Journal of Clinical Epidemiology. 2006;59(5):437-61.
Holländer N, Sauerbrei W, Schumacher M. Confidence intervals for the effect of a prognostic factor after selection of an 'optimal' cutpoint. Statistics in Medicine. 2004;23:1701-13.
Hochholzer W, et al. Impact of the Degree of Peri-Interventional platelet Inhibition After Loading With Clopidogrel on Early Clinical Outcome of Elective Coronoary Stent Placement. Journal of the American College of Cardiology. 2006;48(9):1742-50.