Introduction

High levels of treatment adherence are critical for achieving optimal treatment outcomes among patients with tuberculosis (TB), especially for drug-resistant TB (DR TB). Current tools for identifying high-risk non-adherence are insufficient. Here, we apply trajectory analysis to characterize adherence behavior early in DR TB treatment and assess whether these patterns predict treatment outcomes.

Methods

We conducted a retrospective analysis of Philippines DR TB patients treated between 2013 and 2016. To identify unique patterns of adherence, we performed group-based trajectory modelling on adherence to the first 12 weeks of treatment. We estimated the association of adherence trajectory group with six-month and final treatment outcomes using univariable and multivariable logistic regression. We also estimated and compared the predictive accuracy of adherence trajectory group and a binary adherence threshold for treatment outcomes.

Results

Of 596 patients, 302 (50.7%) had multidrug resistant TB, 11 (1.8%) extremely drug-resistant (XDR) TB, and 283 (47.5%) pre-XDR TB. We identified three distinct adherence trajectories during the first 12 weeks of treatment: a high adherence group (n = 483), a moderate adherence group (n = 93) and a low adherence group (n = 20). Similar patterns were identified at 4 and 8 weeks. Being in the 12-week moderate or low adherence group was associated with unfavorable six-month (adjusted OR [aOR] 3.42, 95% CI 1.90-6.12) and final (aOR 2.71, 95% 1.73-4.30) treatment outcomes. Adherence trajectory group performed similarly to a binary threshold classification for the prediction of final treatment outcomes (65.9% vs. 65.4% correctly classified), but was more accurate for prediction of six-month treatment outcomes (79.4% vs. 60.0% correctly classified).

Conclusions

Adherence patterns are strongly predictive of DR TB treatment outcomes. Trajectory-based analyses represent an exciting avenue of research into TB patient adherence behavior seeking to inform interventions which rapidly identify and support patients with high-risk adherence patterns.