Evaluating Intervention and Mediator Effects on HIV/STI Risk Reduction When Mediators and Outcomes Are Measured Repeatedly

• Activity Number: 124
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Alisa Stephens*+ and John Jemmott III and Marshall Joffe
• Companies: and University of Pennsylvania and University of Pennsylvania
• Keywords: mediation ; controlled direct effects ; causal inference ; longitudinal data analysis

Abstract

Mediation analysis is a valuable tool for determining the mechanisms by which randomized interventions impact study outcomes. Underlying this type of analysis is the temporal ordering of the intervention, potential mediators, and outcomes, with interventions believed to affect mediators, and mediators in turn believed to impact desired outcomes. When mediators and outcomes are measured repeatedly, overlapping timing of mediator and outcome assessments complicates analysis. We use Structural Mean Models to define controlled direct effects of study interventions and mediator effects on outcomes when mediators and outcomes are measured longitudinally. Model parameters are estimated through a semiparametric G-estimation procedure. Our analysis utilizes all observed information on mediators and outcomes to estimate effects, which can increase the efficiency of mediation analyses. We use this approach to assess mediation of the effect of a sexual behavior risk reduction intervention on condom usage among adult men and compare results to standard approaches that consider a single mediator and outcome measurement.