Predictors of Virologic Failure Among Adults with Low-Level Viremia: A Retrospective Cohort Study in Oromia, Ethiopia

Low-level viremia (LLV), defined as an HIV-1 RNA level of 50 to 999 copies/mL, is a common but clinically challenging state among adults receiving antiretroviral therapy (ART). Although LLV does not meet the conventional threshold for virologic failure, it may identify individuals at increased risk of subsequent treatment failure. We evaluated predictors of progression from LLV to virologic failure and characterized the time-dependent probability of progression in a large Ethiopian ART cohort. Methods We conducted a retrospective cohort study among adults with at least one LLV episode documented between October 2023 and June 2025 across 105 health facilities in Oromia, Ethiopia. The primary outcome was progression from LLV to virologic failure, defined as an HIV-1 RNA level of ≥1000 copies/mL, within 18 months. Predictors were evaluated using generalized estimating equations (GEE) with patient-level clustering. A risk stratification model was derived using L1-penalized logistic regression with a patient-level train/test split and validated using grouped cross-validation. Time to virologic failure was evaluated using Kaplan-Meier methods, with a sensitivity analysis based on alternative censoring definitions. Results Among 6713 LLV observations contributed by 6167 unique patients, 189 virologic failure events occurred over 18 months of follow-up. The cumulative incidence of virologic failure was 0.76% at 6 months, 2.22% at 12 months, and 3.02% at 18 months at the patient level. In multivariable GEE analysis, receipt of second- or third-line ART was strongly associated with increased odds of virologic failure (adjusted odds ratio [aOR], 3.93; 95% CI, 2.85-5.43; P < .001), whereas multi-month dispensing (MMD) of ≥90 days was associated with lower odds of virologic failure (aOR, 0.54; 95% CI, 0.38-0.79; P = .001). Increasing age was independently associated with a modest reduction in the odds of virologic failure (aOR per year, 0.99; 95% CI, 0.98-1.00; P = .046). Sex, adherence status, prior visit frequency, prior viral load documentation, and prior viral load magnitude were not independently associated with virologic failure after adjustment. The risk-stratification model demonstrated moderate discrimination (area under the receiver operating characteristic curve [AUC-ROC], 0.79; precision-recall area under the curve [PR-AUC], 0.09) but poor calibration, with a mean predicted probability of 38.9% compared with an observed event prevalence of 2.23%. Kaplan-Meier analysis estimated a cumulative probability of virologic failure of 1.27% at 6 months, 8.10% at 12 months, and 34.51% (95% CI, 26.31%-44.40%) at 18 months in the primary cohort; estimates were similar in the sensitivity cohort (35.14%; 95% CI, 26.95%-44.95%).