Development and preliminary validation of a CD161 immunoscore-based screening method for latent tuberculosis infection in a Chinese training cohort

Current latent tuberculosis infection (LTBI) screening is limited by tuberculin skin test (TST) cross-reactivity, return-visit requirements, and interferon-gamma release assay (IGRA) costs. We developed a flow cytometry and deep learning CD161 immunoscore screening method, evaluating it across two phases ( N = 269) in Jiangxi, China. Phase 1 ( n = 110; 92 males, 18 females) used healthy controls ( n = 88) and clinically cured, post-treatment tuberculosis (TB) patients ( n = 22) for model development. This post-treatment cohort served as a proxy and was not equivalent to untreated LTBI. Phase 2 ( n = 159, all males) provided an independent, three-site proof-of-concept evaluation against TST, using IGRA as the reference standard. In phase 1, the optimal model achieved 100% sensitivity and 92% specificity. In phase 2, using a non-independently validated, post hoc Youden-selected cutoff (60.57), the immunoscore demonstrated 81.8% sensitivity, 80.4% specificity, and an area under the receiver operating characteristic curve (AUC) of 0.833. This performance was comparable to TST (90.9% sensitivity, 72.3% specificity, AUC = 0.816, and DeLong P = 0.701). The immunoscore showed significantly higher specificity than TST ( P = 0.049) but similar sensitivity ( P = 0.521), with an IDI of 0.077 ( P = 0.021) and good calibration ( P = 0.553). These exploratory, all-male results do not establish discrimination between untreated LTBI, post-treatment sensitization, or active TB, necessitating diverse validation.