Recruitment Source and Participant Retention in a Digital HIV Prevention Intervention for Young Black and Latino Men and Transgender Women: Secondary Analysis of the HealthMpowerment 2.0 Randomized Controlled Trial

Background: Young Black and Latino men who have sex with men and transgender women who have sex with men (YBLMT) experience disproportionate HIV-related health disparities in the United States. Digital health interventions offer scalable HIV prevention and support services for these populations. However, recruitment strategies may influence both sample demographics and participant retention, which is critical for intervention effectiveness. Objective: This study examined whether recruitment source was associated with retention in a mobile health randomized controlled trial and assessed demographic differences across recruitment platforms. Methods: Data were drawn from the HealthMpowerment (HMP) 2.0 randomized controlled trial (N=750; July 2020 to September 2022). Participants aged 15 to 29 years were recruited through 4 sources: social media advertisements (n=202), partner-seeking apps (n=394), prior study contacts (n=118), and other sources (eg, community outreach, peer referrals; n=36). Retention was defined as completing at least 3 of 4 follow-up surveys over 12 months, equivalent to completing at least 80% (n=4) of all 5 study measurement occasions, consistent with the United States Preventive Services Task Force (USPSTF) criteria for cohort study follow-up. Chi-square and Fisher exact tests assessed unadjusted retention differences across recruitment sources. Multivariable logistic regression adjusting for age, HIV status, race, ethnicity, and gender identity estimated adjusted odds ratios (aOR) for the association between recruitment source and retention. A sensitivity analysis requiring completion of all 4 follow-up surveys was also conducted. A predictive logistic regression model incorporating Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance further explored demographic and recruitment predictors of retention. Results: <.001). Participants recruited via social media (184/202, 91.1%) and prior study contacts (103/118, 87.3%) had higher retention than those recruited via partner-seeking apps (288/394, 73.1%) and other sources (23/36, 63.9%). After adjusting for demographics, social media (aOR 2.80, 95% CI 1.51-5.18) and study contacts (aOR 2.11, 95% CI 1.14-3.92) remained significantly associated with higher retention compared to partner-seeking apps. HIV-positive status (aOR 0.64) and gender-diverse identity (aOR 0.38) were independently associated with lower retention. The sensitivity analysis yielded directionally consistent findings, with the social media advantage remaining significant. The SMOTE-adjusted model improved recall for low-retention participants from 6% (2/33) to 64% (21/33) but reduced overall accuracy from 77.3% (116/150) to 64.7% (97/150). Conclusions: Recruitment source was associated with both participant characteristics and long-term retention in this digital HIV prevention intervention. Social media and prior research networks yielded higher retention, whereas partner-seeking apps reached populations at higher HIV risk but were associated with lower retention. Targeted retention strategies that account for both recruitment platform and participant characteristics (including HIV status and gender identity) may help optimize engagement and sample diversity in future digital health interventions targeting YBLMT.