Despite successful completion of therapy for drug-susceptible pulmonary tuberculosis (PTB), a subset of individuals experience an unfavorable treatment outcome of tuberculosis (TB) recurrence. India accounts for 26% of the global burden of TB and few data on predictors of TB recurrence exist, especially with a focus on India. A TB recurrence prediction model could enable clinicians to identify patients at risk for recurrence during antituberculosis therapy (ATT) and may be used to alter patient care strategies, such as enhanced monitoring post treatment for high-risk individuals.
We conducted a retrospective analysis leveraging 3 NIH and Indian government funded observational TB cohorts in India (TB-DM, c-TRIUMPh, and eDOTS) designed to assess risk factors associated with unfavorable TB treatment outcomes. Adults newly diagnosed with PTB were enrolled and initiated on ATT, with 18-24 months of follow-up. A priori we selected 8 clinical and laboratory candidate predictors for recurrence based on previously published prediction models and expert opinion. We randomly selected 80% of the dataset and used multivariable logistic regression and bootstrapped backwards selection (repetitions=1000) to identify the best predictors of TB recurrence. We measured model accuracy by receiver operating characteristic (ROC) curve and area under the curve (AUC). We tested model fit by Akaike information criterion (AIC) and reliability by Kappa statistic. We internally validated the model with the remaining 20% of the dataset, generating the model accuracy, sensitivity, and specificity.
Among 1164 adults diagnosed with PTB who completed ATT and achieved cure, 95 (8%) subsequently experienced recurrence. The most important predictors of TB recurrence were female sex, low body mass index (BMI), ever smoker history, month 2 (M2) smear positivity, and M2 culture positivity (Table 1). The model exhibited a c-statistic of 0.68 (95% CI 0.52-0.84) and a Kappa statistic of 0.23. Internal validation revealed an accuracy of 90% (95% CI 0.86-0.93), a sensitivity of 0.94 and a specificity of 0.31.
Our prediction modeling revealed female sex, low BMI, ever smoker history, M2 smear positivity, and M2 culture positivity as the most important predictors to discriminate TB recurrence from sustained cure.