I propose to implement an innovative mixed methods study using machine learning techniques for text analytics and predictive modeling in combination with advanced social network analysis to provide valuable information about the online communicative and social contexts that contribute to HIV prevention and care engagement among young Black men who have sex with men (YBMSM). The study aims to answer the following questions: (1) How are the semantic features of social media communication (i.e., key terms and concepts) among YBMSM related to their HIV prevention and care engagement?; (2) How are structural features of observed social media relationships among YBMSM related to their prevention and care engagement?; and (3) Can social media use patterns predict future prevention and care engagement? To pursue these questions, I will draw on data collected March 2016-March 2019 from YBMSM participants in an ongoing network HIV prevention intervention (N=423). Prevention and care engagement behaviors include retention in care (HIV/PrEP/Primary), STI/HIV testing, and condom use. Participant social media data include: (1) Facebook posts, (2) Facebook friendships, and (3) Facebook group memberships. K99 Phase: Research conducted during the K99 Phase will use machine learning techniques for textual analysis, semantic network characterization, and regression models to identify the semantic features of participants' communication and their association with HIV prevention and care engagement. I will receive mentored training in 5 key areas: (1) knowledge of issues relevant to HIV prevention, care and treatment for MSM; (2) machine learning techniques for text analytics and predictive modeling; (3) advanced stochastic network modeling, including exponential random graph models (ERGMs); and (4) design and implementation of social media based interventions for health promotion; and (5) professional development. R00 Phase: For the R00 I will employ one-mode and two- mode ERGMs to determine how prevention and care engagement behaviors affect the structure of observed social media relationships. I will then develop a predictive model for prevention and care engagement on the basis of individuals' social media use patterns. This K99/R00 award mechanism will be critical to my success in achieving long-term success as an expert in the social and communicative dynamics of HIV prevention and care engagement in high-risk populations. It will also pave the way to an R01 application ? based on a longitudinal design ? of YMSMs' social and communication engagement in a more diverse array of online social networking sites, including both general purpose and dating platforms. The goal of this project will be to utilize that information to establish classes of YMSM on the basis of their heterogeneous patterns of social media use to better understand their synergistic effects on HIV prevention and care engagement practices over time. This is an important next step in understanding how online social environments impact prevention engagement and will help identify causal relationships that may be amenable to network intervention.
This proposal is an innovative mix methods design that utilizes machine learning techniques for text analytics and predictive modeling in combination with advanced social network analysis to provide valuable information about the online communicative and social contexts that contribute to HIV prevention and care engagement among young Black men who have sex with men (YBMSM). The proposed research will sharpen our understanding of the degree to which online social environments support protective or risky behavioral norms and, consequently the degree to which they reveal viable points for intervention. It will also demonstrate how social media data can help front-line staff profile clients in near real-time and predict future HIV-related outcomes like retention failure, thereby serving as an indicator to guide decisions on screening, treatment, and other service provisions.
