Prediction models for adverse drug reactions during tuberculosis treatment in Brazil

Abstract

AbstractBackgroundTuberculosis (TB) treatment-related adverse drug reactions (TB-ADR) can negatively affect adherence and treatment success rates.MethodsWe developed two prediction models for TB-ADR. We included drug-susceptible pulmonary TB participants who initiated standard TB therapy. TB-ADR were determined by physician-assigned attributions of causality, and described according to affected organ system, timing, and grade. Potential predictors of TB-ADR included concomitant medication (CM) use, HIV-status, glycated hemoglobin (HbA1c), age, body mass index (BMI), sex, substance use, and TB drug metabolism variables (e.g.,NAT2acetylator profile). Bootstrapped backwards selection was used to develop the models. Cox proportional hazards regression was used to evaluate TB-ADR risk.ResultsThere were 156 TB-ADR among 102 (11%) of the 945 participants included. Most TB-ADR were hepatic (n=82;53%), grade 2 (n=121;78%), and occurred inNAT2slow acetylators (n=62;61%). The main prediction model included CM use, HbA1c, alcohol-use, HIV-infection, BMI, and age. The alternative model included the same variables, except replaced BMI withNAT2. Both models had good performance and fit. CM use and HIV-infection increased TB-ADR risk.ConclusionsThe model with only clinical variables and that withNAT2were highly predictive of TB-ADR. TheNAT2model provides rationale to evaluate isoniazid dose adjustment and ADR risk.