Global Health
Adherence to multidrug resistant-tuberculosis treatment in a prospective study of people on treatment: a latent class analysis Sarah Brumfield* Sarah Brumfield Pawandeep Kaur Sonali Sarkar Valeria C Rolla Ashavaid Tester Padmapriyadarsini Chandrasekaran Amrose Pradeep Sanjay Gaikwad Afranio Kritski Fernanda CQ Mello Marina Figueiredo Camilla Rodrigues Timothy Sterling C Robert Horsburgh
Multidrug resistant-tuberculosis (MDR-TB) is a threat to TB elimination; successful treatment is important to interrupt transmission. However, treatment adherence by MDR-TB patients is difficult due to high toxicity, limited medication availability, and stigma associated with accessing treatment. Poor adherence can lead to adverse outcomes, including prolonged disease, recurrence, and death. To identify potential causal effects of predictors on treatment outcomes, accurate and complete measurement of adherence is important when studying MDR-TB.
We measured treatment adherence in the Predictors of Resistance Emergence Evaluation in MDR-TB patients on treatment (PREEMPT) study, a prospective cohort of people on treatment for MDR-TB. Adherence was collected using: 1) a self-reported questionnaire, 2) a visual analog scale (VAS), and 3) review of directly observed therapy (DOT) records. Follow-up visits took place every four weeks from baseline until week 24, and another visit took place at week 36. Due to the COVID-19 pandemic, VAS and especially DOT were difficult to collect in 2020-2021. For each of the three measures, we calculated binary variables indicating whether the participant reported taking at least 80% of their medications in the prior 30 days. These were used to conduct a latent class analysis assuming 2 underlying classes of adherence.
We enrolled 328 participants at 5 sites in India and 3 sites in Brazil between January 2019-December 2023. These participants completed 2,199 follow-up visits. Of these, 198 (9%) visits were missing self-reported adherence, 403 (18%) were missing VAS, and 957 (44%) were missing DOT information. Our latent class analysis predicted that 79% of participants were in class 1 (‘adherers’) and 21% were in class 2 (‘non-adherers’).
Using a latent class analysis, we were able to summarize participants’ adherence into a single variable that can be used in subsequent analyses.
