Flexible Models for Estimating Optimal Treatment Initiation Time for Survival Endpoints: Application to Timing of CART Initiation in HIV/TB Co-Infection

• Activity Number: 122
• Type: Topic Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Liangyuan Hu* and Joseph Hogan
• Companies: Brown University and Brown University
• Keywords: Causal inference ; HIV/TB co-infection ; Missing data ; Survival analysis ; Optimal cART initiation time ; Inverse weighting

Abstract

Timing of combinational antiretroviral therapy (cART) initiation is important in HIV/Tuberculosis (TB) co-infection. Early initiation during TB treatment increases drug toxicity, the risk of inflammatory immune reconstitution, and cost burden; late initiation increases risk for morbidity and mortality associated with HIV/AIDS. Evidence from recent RCTs and observational studies generally supports early initiation. However, existing studies do not give specifics about optimal initiation time or precise recommendations for those with CD4>100. We use data from a large observational cohort to gain more detailed information about treatment effects in practical settings. We formulate a causal structural model that flexibly captures the joint effects of treatment initiation time and treatment duration using smoothing splines, and develop methods for fitting the model to observational data with complicated censoring patterns where both treatment and outcome are event times and subject to censoring. Our methods can generate survival curves corresponding to specific treatment times; and can separately characterize effects of timing and duration on treatment. We fit the model to data from 4903 individuals in a large HIV treatment program in Kenya, and use it to estimate optimal initiation times by CD4 subgroups. Our findings are consistent with RCTs but have "higher resolution" in the sense of generating CD4-specific rules that can be used to complement current treatment guidelines.