Socio-ecological vulnerability subgroups predict adherence and treatment outcomes in multidrug-resistant tuberculosis and HIV: a latent class analysis

Background: Despite the availability of effective all-oral regimens, treatment outcomes for rifampicin-resistant or multidrug-resistant tuberculosis (RR/MDR-TB) remain suboptimal, particularly in high HIV-burden settings. Factors across the socio-ecological spectrum strongly influence medication adherence and outcomes, yet approaches to identifying high-risk groups remain limited. We aim to identify clinically meaningful subgroups of individuals with RR/MDR-TB and HIV based on baseline socio-ecological vulnerabilities and examine whether subgroup membership is predictive of medication adherence and treatment outcomes.

Methods: Eligible participants were adults with confirmed RR/MDR-TB and HIV, initiating a bedaquiline-containing regimen and on antiretroviral therapy (ART), who were prospectively enrolled as part of the PRAXIS study (November 2016 to April 2020) in KwaZulu-Natal, South Africa. Adherence to bedaquiline and ART was measured using a cellular-enabled electronic dose monitor (EDM). We used a latent class analysis (LCA), a method for grouping individuals based on shared characteristics, to identify classes based on shared socio-ecological characteristics. A random forest model was developed and internally validated to predict class assignment. Cox proportional hazard models examined the association between class membership and treatment outcomes.

Results: Among 370 participants (median age, 36 years, 54% female), LCA identified three distinct socio-ecological classes. Class 1, characterized by severe socio-economic vulnerability, had the lowest treatment success rate (67%), while Class 3, characterized by relative socio-economic stability, had the highest (86%). The random forest model using only four socio-ecological variables achieves 91.9% classification accuracy, with marital status, gender, BMI, and functional status as the strongest predictors of class membership. Adjusted analysis demonstrated class membership independently predicted outcomes, even after accounting for bedaquiline adherence, indicating that structural vulnerabilities confer risk beyond adherence alone.

Conclusion: In this study, socio-ecological classes were associated with significantly different treatment outcomes, independent of bedaquiline adherence. Structural vulnerability independently predicted poor outcomes in RR/MDR-TB and HIV, supporting integration of socio-ecological risk stratification into differentiated care models.

Clinical trial: Not applicable. In a prospective study, we identified socio-ecological subgroups among individuals with RR/MDR-TB and HIV that predict treatment outcomes independent of medication adherence, highlighting the potential of predictive modeling to inform risk stratification and guide patient-centered interventions.