Machine Learning Identifies Sexual Behavior Subgroups Among Men Who Have Sex with Men in Switzerland
In a recent study, Salazar-Vizcaya et al. used a machine-learning technique to depict distinct longitudinal sexual behaviour patterns among men who have sex with men (MSM). They gathered data from MSM seeking counselling at 12 voluntary counselling and testing centres in Switzerland. The authors used an agglomerative hierarchical clustering algorithm to recognise subgroups of sexual behavioural status trajectories. A behavioural status at a given time point reflected the latest self-reported record including 5 main variables (online-dating, number of anal intercourse partners, condomless anal intercourse with non-steady partners [nsCAI], group sex and partnership status.
In the final analysis, they included 2349 men with 11269 counselling sessions between January 2017 and May 2019. Over the study period, overall frequencies of condomless anal intercourse, group sex, online-dating and having had more than 5 anal intercourse partners in the previous 12 months increased, whereas partnership status remained stable. The top 6 levels of the agglomerative hierarchical clustering pattern defined 10 clusters, of which 6 fulfilled heuristic subgroup criteria (A, B, C, D.1, D.2). Members of subgroup A (14% of the study population) reported high frequencies of sexual behaviours associated with sexually transmitted infection exposure (STI, i.e., nsCAI, group sex, >5 anal intercourse partners in previous 12 months), met partners mainly online and were single. Subgroup B (21%) included individuals reporting high frequencies of online dating and decreasing involvement in group sex and number of partners. Members of subgroup C (23%) reported decreasing rates of online-dating and high, increasing involvement in sexual behaviours associated with STI exposure. Subgroup D.1 (25% of whole studypopulation) had the lowest frequencies of online dating and higher frequencies of group sex, whereas subgroup D.2 (10%) was characterized by mostly meeting their partners online, but reported the lowest frequencies of group sex and having had >5 anal intercourse partners in the previous 12 months. Interestingly, data from the first visit predicted sexual behaviour trends with an accuracy between 64% (subgroup D2) and 86% (subgroup A).
Overall, the study depicted heterogenous longitudinal sexual behaviour patterns among MSM and showed that first visit records were – in general – a good subgroup predictor. The results of this study may help frame sexual health advice to address specific sexual behaviour subpopulations among MSM.