Characterizing Treatment Adherence Trajectories in the endTB Multisite Cohort of Drug-Resistant Tuberculosis Patients: An Application of Group-Based Trajectory Modeling

Background In tuberculosis (TB) care, adherence is often assessed using a simple 80% threshold, which may overlook meaningful patterns. We analyzed adherence trajectories among individuals treated for rifampicin- or multidrug-resistant TB (RR/MDR-TB) in the endTB observational study to identify more informative patterns. Methods We applied a joint latent class mixed model to classify adherence trajectories and assess their relationship with treatment outcomes. Model performance was compared to common classification methods (eg 80% adherence threshold) using Kendall's τb and area under the receiver operating curve for predicting unsuccessful outcomes. Results Among 1787 individuals, we identified 4 adherence patterns: "consistently high" (72.5%), "high to low" (14.3%), "low to high" (7.3%), and "consistently low" (5.9%). Compared to the "consistently high" group, those in "high to low" (hazard ratio [HR] = 23.2; 95% confidence interval [CI]: 15.7-24.3) and "consistently low" (HR = 43.2; 95% CI: 26.2-71.5) groups had significantly higher risk of unsuccessful outcomes, while the "low to high" group did not (HR = 0.7; 95% CI: .1-3.8). Our trajectory model more accurately predicted outcomes than common classification methods (P Conclusions Group-based trajectory modeling provides more nuanced insights into adherence patterns than conventional classification methods. Our findings demonstrate that patients with RR/MDR-TB who exhibited initial poor adherence followed by subsequent improvement achieved clinical outcomes comparable to those with consistently high adherence throughout treatment. This finding challenges the prevailing assumption that sustained high adherence is necessary for treatment success, suggesting that adherence patterns, rather than overall adherence rates, may be more predictive of clinical outcomes in the management of RR/MDR-TB.