Abstract: Using Computational Methods to Assess the Implementation of a Peer-to-Peer HIV Prevention Intervention for Adolescent Men Who Have Sex with Men (Society for Prevention Research 27th Annual Meeting)

546 Using Computational Methods to Assess the Implementation of a Peer-to-Peer HIV Prevention Intervention for Adolescent Men Who Have Sex with Men

Friday, May 31, 2019
Seacliff C (Hyatt Regency San Francisco)
* noted as presenting author
Carlos Gallo, PhD, Research Assistant Professor, Northwestern University, Chicago, IL
Cady Berkel, PhD, Associate Research Professor, Arizona State University, Tempe, AZ
Kevin Moran, MS, Data Scientist, Northwestern University, Chicago, IL
C. Hendricks Brown, PhD, Professor, Northwestern University, Chicago, IL
Brian S. Mustanski, PhD, Professor, Northwestern University, Chicago, IL
Introduction: Adolescent men who have sex with men (AMSM) account for 79% of HIV diagnoses among young people. Text messaging-based interventions overcome many barriers in reaching high risk HIV populations by reducing stigma and geographical distance. Further, the nature of the intervention allows for the use of innovative computational methods in evaluating the delivery and participant responsiveness to the program. In the Guy2Guy (G2G) randomized controlled trial, the mHealth intervention sent prevention scripted messages related to safer sex and provided an interactive platform that matched AMSM into dyads for peer support and skill practice. We hypothesized that linguistic similarity of the texters would increase their responsiveness to the intervention, and in turn promote safer sex behaviors.

Methods: We measured the linguistic style of text messages exchanged by AMSM during a 5-week period (N=132 AMSM). Dyads were matched on age, geographical location, and sexual experience, but not linguistic similarity. We automatically extracted linguistic features, such as emotional content, part of speech tags, and pronouns. A regression model was trained to predict engagement and satisfaction based on dyad’s linguistic alignment.

Results: Linguistic alignment was positively correlated with the number of messages exchanged in the peer support platform (p<0.002). We also found that texters who composed messages with higher linguistic alignment and higher emotional content used condoms during sex and had more HIV tests relative to those whose messages had less linguistic alignment and emotional content.

Conclusions: Computational linguistics methods are efficient, scalable, and non-obtrusive in monitoring the implementation of text-based interventions. It reduces the need to hire and train personnel dedicated to measuring key features of the mHealth interventions delivery and implementation. Further, because the analysis of linguistic style can reveal the linguistic inter-personal differences, it can be used to tailor the content and timing of text-messages, and to optimize the peer-to-peer matching, based on participants’ linguistic profile.