Validating HIV Viral Suppression Threshold Adjustments for Comparable Estimates Using Data From Nationally Representative Household Surveys in Sub-Saharan Africa.

Abstract

Introduction: To enable comparable global assessments of viral load suppression (VLS) among people living with HIV on antiretroviral therapy, the Joint United Nations Programme on HIV/AIDS applies a model to adjust VLS estimates reported at different thresholds to a common viral load ≤1000 copies/mL definition. We assessed performance of the current reverse Weibull model and alternatives using survey data from sub-Saharan Africa.

Methods: Using data from 21 Population-based HIV Impact Assessment surveys in 16 sub-Saharan African countries (2015-2022), we assessed 6 models (Weibull, reverse Weibull, Pareto, Fréchet, gamma, and lognormal) in adjusting VLS reported at viral load <50, <200, <400 copies/mL to ≤1000. We compared predictions using parameters from Johnson et al and recalibrated using Population-based HIV Impact Assessment surveys, assessing whether new shape parameters improved adjustments and varied by sex and age.

Results: In adjustments from all thresholds, the Weibull model had the lowest prediction errors (root-mean-squared error for <200 to ≤1000: Weibull: 1.9%; reverse Weibull: 3.1%; Pareto: 2.5%). Prediction errors for reverse Weibull and Pareto models were higher in subgroups with low VLS compared with Weibull. Across 21 surveys, in adjustments from <200 to ≤1000, reverse Weibull overestimated VLS by 2.3%, compared with 1.5% by Weibull and Pareto. The Fréchet, gamma, and lognormal models performed similarly to Weibull. Shape parameter estimates for the Weibull and reverse Weibull were slightly higher after recalibration and varied by sex and age.

Conclusions: The Weibull, Fréchet, gamma, and lognormal models provided more reliable VLS adjustments across thresholds than the previously recommended reverse Weibull model, avoiding inflated VLS estimates that could obscure gaps in HIV treatment programs and underestimate HIV transmission risks.