Supplementary data collection with case-cohort analysis to address potential confounding in a cohort study of thromboembolism in oral contraceptive initiators matched on claims-based propensity scores


  • Disclosure statement: this study, which was conducted by Ingenix i3 Drug Safety, was funded by a research contract from Berlex. The role of Berlex was limited to review of the manuscript. Drs Eng, Seeger, and Walker, and Ms Clifford, Ms Loughlin and Ms Mentor, Ms Loughlin and Ms Mentor were full-time employees of Ingenix, a subsidiary of UnitedHealth Group during the course of this study. Ingenix i3 Drug Safety retains final publication rights.



Residual confounding is a potential limitation of pharmacoepidemiologic studies, and in particular, studies based on administrative claims data that do not capture lifestyle and clinical confounders. We describe an application of the case-cohort design to assess residual confounding by thromboembolic risk factors (e.g., smoking and obesity) not captured in claims data in a claims-based cohort study of thromboembolism among matched oral contraceptive (OC) initiators.


This study was conducted using the Ingenix Research Data Mart, a database containing medical claims for approximately 12 million members of a large health plan of the United States. We randomly sampled 701 OC initiators from cohorts of ethinyl estradiol/drospirenone (n = 22 429) and other OC initiators (n = 44 858) identified in the years 2001–2004 and matched by propensity score in a claims-based cohort study. Supplementary data on risk factors not measured in the cohort study were collected from medical records for the sample. We estimated the risk ratio of thromboembolism adjusted for the supplementary variables using Cox regression modified for a case-cohort design, and compared it to the rate ratio from the cohort study.


The risk ratio adjusted for the supplementary variables was 0.90 (95 per cent (%) confidence interval (CI): 0.49, 1.68) which was similar to the rate ratio (0.92; 95%CI: 0.50, 1.63), indicating negligible confounding by the supplementary variables in the cohort study.


Case-cohort methods were used to assess residual confounding in a claims-based cohort study. This approach adds to a growing number of methods to evaluate residual confounding in cohort studies. Copyright © 2008 John Wiley & Sons, Ltd.