Causal Inference for Survival and Recurrent Events with Missing Exposure: Application to Treatment of HIV/TB Coinfection in Western Kenya

• Activity Number: 6
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ann W. Mwangi and Joseph Hogan*+ and Jane E. Carter and Rami Kantor and Abraham Siika
• Companies: Brown University and Brown University and Miriam Hospital and Miriam Hospital and Moi University School of Medicine/AMPATH
• Keywords: Causal inference ; HIV ; Observational data ; Informative censoring ; Survival analysis ; Confounding bias

Abstract

The optimal timing to initiate combination antiretroviral treatment (cART) is a major concern for patients who are co-infected with HIV and TB. Data from randomized trials are still limited. We use data on 8,810 HIV/TB co-infected patients in western Kenya to quantify the causal effect of cART initiation time on mortality and recurrence of opportunistic infection (OI).

Our cohort had drop out leading to censoring of both exposure and outcome. We categorized patients into 5 groups based on timing of cART initiation following TB treatment initiation: (0,2], (2,8], (8,16], (16,35] and >35 weeks.

We use marginal structural hazards models, fitted with IPW methods, to capture causal effects of cART initiation time on mortality and rate of OI recurrence. For those having missing exposure, we used the weight model to impute indicators of initiating cART at the remaining intervals. We also develop sensitivity analysis to assess the effect of key assumptions used in the imputations. Our method allows 763 individuals with missing treatment time to be included. The fitted model suggests that hazard for death and recurrence of OI are substantially reduced by initiating cART early.