Causal inference for disseminated effects of randomized interventions and components of packages of randomized interventions

Dr. Ashley Buchanan, who was a postdoctoral research fellow in Biostatistics supported, in part, by this award, and is now an Assistant Professor at the University of Rhode Island, and I are developing methods for the estimation of individual, disseminated and composite effects for interventions which may have substantial resonance within a community or network beyond individuals who directly receive the intervention. To motivate the research, we reanalyzed HPTN 037, a trial that randomized networks of injection drug users to a package of education and counseling to reduce HIV incidence and high-risk behaviors for HIV infection. We developed new statistical methods for point and interval estimation of individual, disseminated and composite effects for prevention studies with a network feature. We proved that the average effect is less than or equal to the composite effect. We developed two different modeling approaches for this problem, specifically, an aggregate and stratified model, and each make different assumptions about the effects of the covariates included in the model. Even when the networks are randomized to the intervention, the validity of some of these quantities of interest do not benefit from randomization, and we developed causal inference methods for estimating these quantities, with a novel application of marginal structural models for time-varying confounding. We have prepared two manuscripts and one is accepted at the American Journal of Epidemiology (Buchanan A., Vermund S., Friedman S., Spiegelman D., Assessing individual and disseminated effects in network-randomized studies. American Journal of Epidemiology. In Press 2018). The first is focused on estimating the individual and disseminated effects of the randomized intervention and the second is about the individual and disseminated package component effects. In this second paper, we are deriving an expression for the sandwich estimator of the variance, which may provide a gain in efficiency. We are also developing a simulation-based study to evaluate the performance of our estimators, including variance estimators. We developed a collaborative relationship with Dr. Samuel Friedman, who is a leading drug abuse and HIV researcher, at the National Development and Research Institutes. Dr. Samuel Friedman has been a collaborator on our manuscripts, ensuring our methodological work has a high impact in the field. He has a wealth of network data among injection drug users focusing on HIV prevention and has agreed to share data from three of his studies, namely the Social Factors and HIV Risk Study (SFHR), Networks, Norms, and HIV Risk among Youth Study (NNAHRAY), and Transmission Reduction Intervention Project (TRIP). We obtained URI IRB approval to perform secondary analyses in two of these studies and Dr. Friedman has shared the datasets.