International STD Research & Reviews

Background and Objectives: Despite significant strides being made in HIV prevention and treatment, adolescent girls and young women (AGYW) continue to be disproportionately affected by HIV infection globally. Many of the current studies on HIV prevalence globally have focused on linear or Gaussian-based models, which do not represent the bounded, skewed, nonlinear, and hierarchical nature of HIV prevalence data, and which may result in misleading inference and a lack of understanding of the dynamics of the HIV epidemic. The objective of this study was to describe long-term temporal trends and spatial variability in HIV prevalence among women aged 15-24 years at the population level with a sophisticated multilevel nonlinear modelling approach.

Method: A country-year panel dataset of 269 countries and territories in the World Development Indicators database (1990–2024) was analysed. HIV prevalence in the female population aged 15-24 years was the outcome variable. To model bounded prevalence proportions, multilevel beta mixed-effects regression was used, including natural cubic splines to model nonlinear temporal trends, random intercepts and slopes for country-specific heterogeneity, and repeated measurements. Model performance was evaluated by the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), likelihood ratio tests, and simulation-based residual diagnostics.

Results: The spline beta mixed-effects model with random intercepts and random slopes was the best model fit, with values of AIC = −60,918.34 and BIC = −60,858.28, respectively, significantly better than the linear model (ΔAIC = 3,657.51). Nonlinear temporal effects were highly significant (χ² = 3663.51, p < 0.001) and showed a sharp increase in the epidemics in the 1990s followed by a gradual decrease since 2000. Each spline component was significantly associated (p < 0.001) with HIV trajectories, which were complex, nonlinear, and significantly different. There was evidence of significant variation and clusters of high burden of disease across countries, especially in sub-Saharan Africa. The overall model performance was satisfactory according to the diagnostic assessments, and there was no sign of any problems regarding dispersion, influential outliers or zero-inflation.

Conclusion: This study proposes a comprehensive hierarchical beta spline modelling framework that overcomes several important challenges in the currently available literature on HIV prevalence studies by being able to model cross-country heterogeneity, longitudinal dependence, nonlinear epidemic evolution and bounded outcomes. This information echoes the longstanding disparities of HIV burden between men and women and within young women. To advance towards the end of the HIV epidemic and SDGs 3 and 5, targeted adolescent-focused HIV prevention programmes need to be strengthened, sexual and reproductive health services need to be made more accessible and gender-responsive interventions need to be implemented in high-burden settings.