Objectives: Improve models for estimating HIV epidemic trends in sub-Saharan Africa (SSA).
Design: Mathematical epidemic model fit to national HIV survey and ANC sentinel surveillance (ANC-SS) data.
Methods: We modified EPP to incorporate age and sex stratification (EPP-ASM) to more accurately capture the shifting demographics of maturing HIV epidemics. Secondly, we developed a new functional form for the HIV transmission rate, termed 'r-hybrid', which combines a four-parameter logistic function for the initial epidemic growth, peak, and decline followed by a first-order random walk for recent trends after epidemic stabilization. We fitted the r-hybrid model along with previously developed r-spline and r-trend models to HIV prevalence data from household surveys and ANC-SS in 177 regions in 34 SSA countries. We used leave-one-out cross validation with household survey HIV prevalence to compare model predictions.
Results: The r-hybrid and r-spline models typically provided similar HIV prevalence trends, but sometimes qualitatively different assessments of recent incidence trends because of different structural assumptions about the HIV transmission rate. The r-hybrid model had the lowest average continuous ranked probability score, indicating the best model predictions. Coverage of 95% posterior predictive intervals was 91.5% for the r-hybrid model, versus 87.2 and 85.5% for r-spline and r-trend, respectively.
Conclusion: The EPP-ASM and r-hybrid models improve consistency of EPP and Spectrum, improve the epidemiological assumptions underpinning recent HIV incidence estimates, and improve estimates and short-term projections of HIV prevalence trends. Countries that use general population survey and ANC-SS data to estimate HIV epidemic trends should consider using these tools.