The Causal Effect of Postmenopausal Hormone Therapy on Coronary Heart Disease

• Activity Number: 99
• Type: Invited
• Date/Time: Monday, August 9, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Miguel A. Hernán*+ and James Robins
• Companies: Harvard School of Public Health and Harvard School of Public Health
• Keywords: causal inference ; observational studies ; survival analysis

Abstract

The Nurses' Health Study (NHS) found a lower risk of coronary heart disease among users of postmenopausal hormone therapy. The Women's Health Initiative (WHI), a randomized trial, reported the opposite finding. The following limitations of the NHS design may explain the discrepancy: (1) lack of comparability between women that did and did not initiate therapy, (2) lack of comparability between women who continued and discontinued therapy, and (3) inability to detect short-term effects of therapy. The NHS can be conceptualized as a randomized trial with unknown randomization probabilities. We describe a study protocol for the NHS trial and propose its re-analysis under the intention to treat principle, which is immune to bias due to explanation 2. We use different statistical models (Cox proportional hazards, accelerated failure time, risk ratio models) to reduce the possibility of bias due to idiosyncratic misspecification of any particular model. We propose a strategy to assess the influence of explanation 3. Finally, we describe analyses to assess the sensitivity of our causal estimates to bias from explanation 1.