Blood biomarkers show promise for tuberculosis detection and tracking

Tuberculosis remains a major global health challenge, and clinicians and researchers are eager for better tests that can diagnose disease early and monitor whether treatment is working. To bring clarity to a large but fragmented literature, Adrian R. Martineau and colleagues carried out a systematic review and meta-analysis of studies that tested blood for microbial traces of Mycobacterium tuberculosis (Mtb). They searched MEDLINE, EMBASE and Scopus for work published between 1 January, 1990 and 22 October, 2025, looking specifically for studies that sampled human blood to detect four classes of microbial biomarkers: cell-free Mtb DNA, cell-associated Mtb DNA, protein/peptide Mtb antigens and lipid/glycolipid Mtb antigens. The team identified 137 eligible articles in total. Of these, 109 contributed data to analyses that estimated how accurately each biomarker class could distinguish people with tuberculosis disease from those without, and 13 provided data that allowed assessment of how biomarker levels changed after starting antimicrobial therapy. This effort pooled results across diverse studies to give a broad picture of how these blood tests perform in practice.

To evaluate diagnostic accuracy the researchers used hierarchical summary receiver operating characteristic (HSROC) models and reported area under the HSROC curve (AUC), sensitivity and specificity for each analyte class. For cell-free Mtb DNA targets the pooled AUC was 0.87 (95% CI 0.84 to 0.89) with sensitivity 61.5% (51.0 to 71.0) and specificity 93.0% (88.1 to 96.1); these estimates came from 4,878 samples across 34 study/setting/biomarker combinations. For cell-associated Mtb DNA targets the AUC was 0.93 (0.90 to 0.95) with sensitivity 43.9% (29.4 to 59.4) and specificity 97.1% (94.5 to 98.5) based on 3,589 samples from 32 investigations. Protein/peptide targets showed AUC 0.94 (0.92 to 0.96) with sensitivity 78.9% (73.2 to 83.6) and specificity 92.9% (90.7 to 94.5) using 10,260 samples from 61 investigations. Lipid/glycolipid targets had AUC 0.96 (0.94 to 0.97) with sensitivity 68.6% (54.1 to 80.3) and specificity 97.0% (94.0 to 98.5) from 3,287 samples in 22 investigations. For longitudinal changes after antimicrobial therapy, pooled risk differences in proportions testing biomarker-positive were -0.44 (-0.89 to 0.01; 68 samples, 5 investigations) for cell-free Mtb DNA, -0.46 (-0.88 to -0.03; 89 samples, 5 investigations) for cell-associated Mtb DNA, and -0.24 (-0.75 to 0.28; 160 samples, 4 investigations) for protein/peptide antigens. No studies reported on responses of lipid/glycolipid antigens to therapy. Heterogeneity was moderate (I2 25-50%) for most studies.

The overall picture that emerges is cautiously optimistic but incomplete. Molecular and biochemical blood biomarkers appear capable of detecting tuberculosis disease with reasonably high overall accuracy — AUC values ranged from 0.87 to 0.96 across biomarker classes — and specificity was consistently higher than sensitivity, meaning these tests were better at ruling in disease than ruling it out. Importantly, cell-associated Mtb DNA showed a statistically significant decline after starting antimicrobial therapy, and similar but less certain trends were seen for cell-free Mtb DNA and protein/peptide antigens. However, the authors warn that most of the primary studies included were assessed as being at high risk of bias (98 of 109 studies in the bivariate analyses and 11 of 13 in the longitudinal analyses), which weakens confidence in the pooled findings. The review therefore calls for higher quality, well-designed studies to confirm the diagnostic and treatment-monitoring value of these blood biomarkers before they can be widely recommended. The study was funded by Barts Charity, The Medical College of Saint Bartholomew’s Hospital Trust and the Wellcome HARP Doctoral Fellowship Scheme.