A point-of-care prediction tool for recurrent tuberculosis [TBN 092]

A study aimed to develop and validate a simple prediction model for recurrent tuberculosis (TB) that could be used at treatment completion by outreach workers to identify TB survivors at highest risk for recurrence in resource-limited, high-burden settings. The analysis used data from the TB Aftermath noninferiority trial conducted in Maharashtra, India. Participants were enrolled between January 2021 and October 2023, and follow-up data through October 2024 were analyzed.

The TB Aftermath trial enrolled 1,076 adults (≥18 years) who had completed or been cured of TB according to India's National TB Elimination Programme (NTEP) definitions. Individuals with pulmonary and/or extrapulmonary TB were eligible regardless of treatment regimen. Participants entered the study within 60 days of treatment completion and were randomized to phone-based or home-based symptom screening at 6 and 12 months after treatment. All participants also received an 18-month home visit. For this analysis, investigators included participants with at least 12 months of follow-up or those who experienced an outcome. Individuals lost to follow-up within 12 months and those who died without documented TB recurrence were excluded. The primary outcome was recurrent TB diagnosed within 18 months of treatment completion, including both microbiologically confirmed and clinically confirmed recurrences. Candidate predictors were evaluated using exploratory analyses and LASSO regression, followed by model development and validation using separate training (60%) and validation (40%) datasets.

Among 1,033 eligible TB survivors, 85 (8.2%) experienced recurrent TB within 18 months. Of these recurrences, 52 (61%) were microbiologically confirmed and 64 (75%) involved pulmonary disease. The median participant age was 35 years (IQR 27-48), 52% were male, and the median BMI was 19.9 kg/m² (IQR 17.6-23.2). Nine predictors were shortlisted: sex, household income, biomass fuel exposure, BMI, unhealthy alcohol use (AUDIT), smoking history, peak expiratory flow (PEF), disease site, and history of more than one TB episode. The highest-performing and most practical models contained five variables. Several five-item models demonstrated moderate discrimination, all including BMI and PEF. The selected model consisted of sex, household income, BMI, PEF, and history of more than one TB episode, achieving a cross-validated c-statistic of 0.69 (95% CI 0.56-0.83). Model calibration in the validation set was acceptable (Hosmer-Lemeshow P=.053; calibration intercept 0.03, 95% CI -0.03 to 0.09; slope 0.66, 95% CI 0.08-1.24). Performance did not differ significantly between early and late recurrence. Sensitivity analyses, including complete-case analyses and analyses restricted to microbiologically confirmed recurrence, identified the same five predictors and supported the robustness of the model.

The study concluded that a parsimonious five-item model using routinely obtainable characteristics can moderately predict TB recurrence after treatment completion among TB survivors in India. As a prediction-model development and validation study, the level of evidence is moderate. Limitations include modest discriminatory performance, relatively low recurrence event numbers, missing and poor-quality PEF measurements, and potential limitations in generalizability beyond similar high-burden public-sector settings. 

Source: Cox SR, Moe AH, Gupte AN, Kadam A, Valawalkar S, Gupte N, Lele G, Kendall EA, Baillie C, Barthwal MS, Kakrani A. A point-of-care prediction tool for recurrent tuberculosis. Clinical Infectious Diseases. 2025 Dec 15;81(6):e612-22.