Previous work identified risk factors for new HIV infections in sub-Saharan African populations but patterns of association are not consistent across studies. Different risk factor definitions and low power may explain some inconsistencies. Statistical power has not previously been estimated in these risk factor analyses. We harmonised population-based longitudinal data from general population studies in 6 sub-Saharan African countries, partners in the Network for Analysing Longitudinal Population-based HIV/AIDS data on Africa, to assess risk factors for new HIV infections. Potential risk factors were identified from the literature and a modified version of the proximate determinants framework.
Individual level data covering 2005 to the end of follow up (2012-2016) were obtained for each study. Data were arranged for survival analysis with first HIV negative test as the start of observation and HIV seroconversion as the failure event. Indviduals were censored at death, out migration and end of follow up. 70 imputations of seroconversion date were used to overcome interval censoring.
Time-varying risk factors were: residence, residential mobility, time since first sex, marital status, numbers of partners in lifetime and last year, acquisition of new partners, types and combinations of partnerships, male circumcision, condom use and age gaps between partners. Piecewise exponential regression models were fitted separately by study for men and women aged 15-24 and 25-49. Crude hazard ratios were compared between studies. We estimated the statistical power to detect each association. Study- and sex- and age-specific multivariate models were fitted and consistency of risk factors evaluated. Where warranted, the pooled effects of risk factors are estimated.
99097 people contributed 351457 person years (203266 from women). There were 5274 seroconversions (3711 among women). Figure 1 shows the crude hazard ratio for HIV infection by selected risk factors. Most consistent findings across studies were that new & multiple partners and being formerly married increased risk whilst being circumcised decreased risk. Condom use was protective among people who had higher risk partnerships.
Effect size and strength of evidence varied across studies and age groups and for each risk factor. Whilst lack of statistical power explains some heterogeneity there are likely to be real differences in the importance of some risk factors between populations.