Optimizing machine learning-based identification of sexual health influencers for HIV self-testing distribution among men who have sex with men in China: a secondary analysis of a quasi-experimental trial.

BACKGROUND

Secondary distribution of HIV self-testing can expand testing among men who have sex with men. A parent quasi-experimental trial found that sexual health influencers (SHIs) identified by a machine-learning model achieved greater peer uptake than those identified by an empirical scale, but the factors driving distribution effectiveness remain unclear.

METHODS

We conducted a secondary analysis of the parent trial (ChiCTR2000039632) in five provinces in southern China. Participants were randomized to either the empirical scale or machine-learning identification arm, and SHIs were then selected within each arm. The current analysis included 196 SHIs with complete follow up. Effective distribution was defined as motivating at least one peer to test. Predictor variables captured sociodemographic, behavioral and network characteristics. Multivariable Poisson LASSO models with multiple imputation and cross-validation estimated adjusted incidence rate ratios (aIRRs).

RESULTS

Among 196 SHIs (mean age 28 years), 46.9% achieved effective distribution. A total of 286 unique peers returned verified HIV self-testing results, of whom 118 (41.3%) were first-time testers. Peer HIV testing was associated with stronger awareness of peers' HIV status (aIRR 1.31; 95% CI 1.16-1.48) and more active sharing of HIV information (aIRR 1.28; 95% CI 1.10-1.50). SHIs who had been accompanied during their own HIV test recruited more newly tested peers (aIRR 1.94; 95% CI 1.12-3.37).

CONCLUSIONS

Our data have implications for secondary distribution of HIV self-test kits and suggest the need for capacity strengthening related to peer engagement in HIV self-testing.