Creating HIV risk profiles for men in South Africa: a latent class approach using cross‐sectional survey data

Abstract Introduction Engaging at‐risk men in HIV prevention programs and services is a current priority, yet there are few effective ways to identify which men are at highest risk or how to best reach them. In this study we generated multi‐factor profiles of HIV acquisition/transmission risk for men in Durban, South Africa, to help inform targeted programming and service delivery. Methods Data come from surveys with 947 men ages 20 to 40 conducted in two informal settlements from May to September 2017. Using latent class analysis (LCA), which detects a small set of underlying groups based on multiple dimensions, we identified classes based on nine HIV risk factors and socio‐demographic characteristics. We then compared HIV service use between the classes. Results We identified four latent classes, with good model fit statistics. The older high‐risk class (20% of the sample; mean age 36) were more likely married/cohabiting and employed, with multiple sexual partners, substantial age‐disparity with partners (eight years younger on‐average), transactional relationships (including more resource‐intensive forms like paying for partner’s rent), and hazardous drinking. The younger high‐risk class (24%; mean age 27) were likely unmarried and employed, with the highest probability of multiple partners in the last year (including 42% with 5+ partners), transactional relationships (less resource‐intensive, e.g., clothes/transportation), hazardous drinking, and inequitable gender views. The younger moderate‐risk class (36%; mean age 23) were most likely unmarried, unemployed technical college/university students/graduates. They had a relatively high probability of multiple partners and transactional relationships (less resource‐intensive), and moderate hazardous drinking. Finally, the older low‐risk class (20%; mean age 29) were more likely married/cohabiting, employed, and highly gender‐equitable, with few partners and limited transactional relationships. Circumcision (status) was higher among the younger moderate‐risk class than either high‐risk class ( p  < 0.001). HIV testing and treatment literacy score were suboptimal and did not differ across classes. Conclusions Distinct HIV risk profiles among men were identified. Interventions should focus on reaching the highest‐risk profiles who, despite their elevated risk, were less or no more likely than the lower‐risk to use HIV services. By enabling a more synergistic understanding of subgroups, LCA has potential to enable more strategic, data‐driven programming and evaluation.