Early adherence predicts treatment outcomes and mortality in people living with HIV and multidrug-resistant tuberculosis

Purpose: Treatment success for RR/MDR-TB remains suboptimal, particularly among people living with HIV. Electronic dose monitors support adherence, but strategies to act on these data are limited. We hypothesized that early adherence to bedaquiline predicts treatment outcomes and could guide individualized interventions.

Methods: We prospectively enrolled adults with RR/MDR-TB and HIV initiating bedaquiline-containing regimens in KwaZulu-Natal, South Africa. Adherence was measured using real-time electronic dose monitors. Weekly adherence data through week 24 trained time-aware XGBoost models to predict end-of-treatment outcomes, and latent class growth analysis (LCGA) at the earliest predictive time point identified distinct adherence trajectories.

Results: Among 282 participants, model performance improved over time, achieving good discrimination by week 4 (AUC > 0.80), predictive of subsequent adherence. LCGA using 4-week data identified three adherence trajectories: high, moderate-stable, and early-declining. Favorable outcomes were lowest in the early-declining group (55% vs. 77% and 76%), which also had the highest mortality (36% vs. 12% and 12%). Cox models confirmed a higher risk of unfavorable outcomes among early decliners.

Conclusion: Adherence in the first four weeks of bedaquiline therapy provides a powerful, early signal of treatment outcomes. Real-time adherence monitoring could enable risk stratification and tailored interventions to improve RR/MDR-TB and HIV care.