Propensity score adjustment with multilevel data: Setting your sites on decreasing selection bias

Michael E. Griswold, A. Russell Localio, Cynthia Mulrow

Research output: Contribution to journalEditorialpeer-review

72 Scopus citations

Abstract

Pooling approach is one of a myriad of issues in PS methods. Additional topics include choice of inclusion in outcome models (subclassification, matching, weighting, and doubly robust estimation [7]) and longitudinal PS estimation rather than only baseline scores as both Ray and colleagues and we did. We conclude that PS methods are valuable if imperfect tools for addressing some of the selection biases inherent in observational studies. Appropriately accounting for within-site imbalances associated with various prognostic factors or prescription patterns can help minimize the effect of selection bias on observed outcomes. The types of sensitivity analyses that we described may help diagnose deficiencies in PS models and results. Using such techniques, we found variation in PS estimates, balance achieved, and site-specific end point risks. However, consistency in the overall results for Ray and colleagues' analyses is comforting and offers a counterpoint to previous studies. Given multiple other studies with conflicting or uncertain results (8 -10), we believe that the safety of coprescription of PPIs and clopidogrel remains an unanswered question.

Original languageEnglish (US)
Pages (from-to)393-396
Number of pages4
JournalAnnals of internal medicine
Volume152
Issue number6
DOIs
StatePublished - Mar 16 2010
Externally publishedYes

ASJC Scopus subject areas

  • Internal Medicine

Fingerprint

Dive into the research topics of 'Propensity score adjustment with multilevel data: Setting your sites on decreasing selection bias'. Together they form a unique fingerprint.

Cite this