Predicting tuberculosis progression in school contacts: novel host biomarkers for early risk assessment

The low positive predictive value of tuberculin skin tests and interferon-γ release assays often results in unnecessary prophylaxis. This study aimed to identify antigen-specific biomarkers with high accuracy for predicting progression to active tuberculosis (ATB). QuantiFERON supernatants from a school tuberculosis outbreak cohort were analyzed, tracking students over two years to identify ATB cases. We assessed 67 cytokines using the Luminex Multiplex Array kit and applied LASSO and multivariate logistic regression to select predictors. A nomogram was developed from the coefficients of top predictors. Model performance was evaluated by AUC, C-index, and AIC. The levels of FGFbasic, GM-CSF, MPIF-1/CCL23, as well as the combinations of ratios of FGFbasic/GM-CSF and FGFbasic/MPIF-1/CCL23 were significantly associated with the risk of ATB. AUC values for the prediction models based on individual cytokines ranged from 0.607 to 0.713, notably lower than those of the fixed models based on the logistic regression (0.932) and LASSO regression (0.939). The LASSO regression model exhibited the best predictive performance, with a higher sensitivity (0.858 vs. 0.818) and specificity (0.949 vs.0.923), lower AIC (36.323 vs. 38.232), and equivalent C-index (0.939) compared to the traditional logistic regression model. The biomarkers identified in this study offer valuable insights for developing a more precise tool to identify individuals at high risk for rapid progression to ATB disease, enabling targeted interventions. The combination of multiple immune indicators shows significant promise in improving diagnostic accuracy.