The overarching goal of this proposal is to develop innovative statistical methods for designing more effective
HIV treatment and prevention interventions, along with more effective implementation strategies to deliver them.
Due to HIV secondary transmission and social influence of preventive behaviors, the intervention received by
one individual can have an effect (or spill over) on the risk for HIV infection, risk behaviors, retention in care and
treatment adherence of other individuals. This mechanism, called ‘interference’ in causal inference, remains a
major challenge for the evaluation of HIV interventions.
Statistical methods accounting for interference are necessary for valid estimation of the individual effect of an
intervention and of the overall population effect, as well as for understanding the extent to which social context
plays a role though spillover effects and how it can be leveraged. This proposal will develop innovative methods to
1) disentangle individual and spillover effects of time-varying package intervention components in cluster random-
ized trials with interference and non-compliance to the assigned components; 2) in network-based and cluster
randomized studies, correct for bias due to misspecification of the interference sets, that is, the sets of individuals
whose treatment affects the outcome of others. 3) identify individuals who are more likely to influence their peers
to adopt behavioral changes and evaluate the improved effectiveness of strategies that target these individuals.
This project will define novel causal estimands for the causal questions of interest, and extend marginal structural
modeling methodology to adjust for confounding and spillover and to evaluate hypothesized strategies leveraging
component-specific effects and influence heterogeneity. Bias correction for mismeasured transmission and social
influence networks will be based on a main study/validation study approach comparing the use of phylogenetic-
based clusters, social and sexual networks, and spatially-based networks and clusters. User-friendly software
implementing the proposed methods will be developed and made publicly available to facilitate their uptake. To
further facilitate dissemination, short courses about the new methods and software will be offered.
The development of statistical methods will be motivated and applied to two large cluster randomized trials in
Botswana (BCPP) and South Africa (TasP) and three network-based peer education studies (HPTN 037, CHAT,
STEP), providing new insights into effective combinations of HIV interventions at the individual and community
level and into novel strategies to leverage spillover and strengthen the impact of these interventions. The methods
will be broadly applicable to many public health interventions for HIV and other infectious diseases.
Our proposal fits within the mission of the National Institute of Mental Health - Division of AIDS Research, as
well as that of a recent NOSI (NOT-AI-21-054), to advance the development and testing of behavioral and biomed-
ical interventions by incorporating social context and behavioral science to understand and leverage transmission
networks and social influence, as well as their social determinants.