Peripheral cytokine distinguish recent from Mycobacterium tuberculosis infection and active tuberculosis

Background Recent Mycobacterium tuberculosis infection ( MTB ) carries a higher risk of progression to active tuberculosis (TB) than MTB infection, but the immune features that distinguish these stages remain unclear. Methods In this multicenter observational study in Jiangsu Province, China, we enrolled 228 participants with MTB infection ( n = 88), recent infection ( n = 44), or active TB ( n = 96). Twelve plasma cytokines and chemokines were quantified using a multiplex Luminex platform. Group differences were assessed using non-parametric tests with a two-stage false discovery rate (FDR) correction and age- and sex-adjusted residual models. Biomarker performance was evaluated by repeated Monte Carlo cross-validation using an ensemble feature-selection framework. Post-hoc power analyses were conducted to evaluate the robustness of subgroup comparisons. Results Age, but not sex, differed across groups. After age- and sex-adjusted comparisons, six markers differed across groups. While recent infection exhibited a trend toward higher IL-23 and IL-12 concentrations compared to remote infection, post-hoc analysis revealed these subgroup differences had small effect sizes and were statistically underpowered. In the combined infection-vs.-TB comparison, IL-8 and CCL1 were higher in active TB, whereas IL-23, IL-12, IL-10, and IL-1ra were lower. Among multivariable models, XGBoost performed best (mean AUC 0.866), followed by the ensemble model (0.865) and random forest (0.861). Decision curve analysis suggested potential clinical utility across exploratory threshold probabilities, although no clinically validated operating point was predefined. Conclusion Active TB exhibits a distinct peripheral cytokine pattern compared to MTB infection. Recent infection may represent an exploratory trend toward preserved Th1/Th17 signaling, though larger longitudinal cohorts are required for confirmation. Peripheral cytokine profiles integrated into multivariable models may provide complementary triage information, although external validation is required before clinical application.