Tripartite Matching in Multi-Group Observational Studies

• Activity Number: 624
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Zhenzhen Xu*+ and John Kalbfleisch
• Companies: Abbott Laboratories and University of Michigan
• Keywords: Matching ; Observational Studies ; Tripartite Matching ; Propensity Score ; Euclidean Distance ; Multi-group study

Abstract

The method of propensity score matching has been widely used in observational studies to control for bias. Often in practice, observational studies may involve several different exposures of interest and the aim is to compare their exposure effects. For example, in assessing the effect of an exposure variable, some investigators use a second control group in an effort to detect the hidden biases in unobserved covariates. In other cases, several risk factors may be of interest, or there may be multiple levels of a given risk factor. In such cases, the propensity score matching method, which was designed to match only two treatment groups, needs to be extended.

In this paper, we propose two new algorithms for matching with three groups, the symmetric and asymmetric tripartite matching with matched triples, and compare the proposed methods with the incomplete block design with disjoint pairs introduced by Bo and Rosenbaum (2004, Journal of Computational and Graphical Statistics). The proposed methods can be extended to match more than three groups. We illustrate the new methods with data from observation studies and evaluate the performance of the methods with a simulation study.