Using the Stratification Score to Adjust U-Statistics to Account for Confounding Covariates

• Activity Number: 411
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Glen Satten*+ and Somnath Datta and Andrew Allen
• Companies: CDC and University of Louisville and Duke University
• Keywords: Stratification Score ; U-Statistic ; Rank-sum Test ; Confounding ; Case-control study ; Weighted analysis

Abstract

When comparing the distribution of a variable (such as exposure) between two groups (such as case and control participants in a case-control study), observed differences in exposure may be due to confounding covariates (variables that are causally related to both exposure and group membership). In the presence of confounding, a significant difference between exposures in two groups may be explained entirely by differences in the distribution of confounding covariates. To account for confounding, we use the stratification score, which plays a role in retrospective studies that is analogous to that played by the propensity score in prospective studies. Allen and Satten (AJE 2011) showed that reweighting the data using the stratification score allows meaningful comparison of the exposure distribution, in the presence of confounding, under the null hypothesis that exposure is unrelated to group membership. Here we use this approach to develop adjusted U-statistics that can test the equality of exposure distributions in two (or more) groups in the presence of confounding, concentrating on an adjusted version of the Rank-Sum test.