Marginal Structural Models to Estimate the Causal Effect of Switching Highly Active Antiretroviral Therapy on the Survival of HIV-Positive Women.

• Activity Number: 279
• Type: Contributed
• Date/Time: Wednesday, August 14, 2002 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology*
• Author(s): Yolanda Barrón*+ and Stephen Cole and Kathryn Anastos and Stephen Gange+
• Affiliation(s): Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health and Montefiore Medical Center and Johns Hopkins Bloomberg School of Public Health
• Keywords: Causality ; Confounding ; Marginal Structural Models ; Cox Models ; Cohort Studies ; HIV/AIDS

Abstract

In prospective observational studies with time-varying exposures, standard Cox survival models may produce biased effect estimates in the presence of time-varying confounding affected by previous exposure. We used Robins' marginal structural models (MSMs), which are fit using inverse probability-of-treatment weights, to estimate the effect of changing therapy on the survival of participants from the Women's Interagency HIV Study who had initiated highly active antiretroviral therapy. We adjusted for time-varying markers of disease progression (e.g., CD4+ cell count, HIV RNA) and compared the results with standard Cox models. Preliminary analyses indicate the effect of discontinuing therapy on survival using standard Cox models (relative hazard [RH] = 2.59; 95% confidence interval [CI]: (1.63, 4.12)) was attenuated using MSMs (RH = 1.86; conservative 95% CI: (1.05, 3.28)). Additional analyses will show the joint causal effect of time-varying markers of disease progression and discontinuation of therapy on survival.