New blood test combination predicts who will get active tuberculosis

Tuberculosis remains one of the world's major infectious threats, with about a third of people infected and a minority progressing from latent infection to active disease. Identifying which people will progress is critical because targeted preventive treatment could stop illness before it starts, but current tools do not reliably pick out the relatively small number who will become sick. In a study conducted among household contacts in Brazil, corresponding author Padmini Salgame and collaborators set out to compare and combine existing molecular risk tests to improve prediction. Rather than evaluating signatures separately, the team used a single laboratory platform to put 15 published gene signatures head-to-head. The investigators worked with RNA extracted from whole blood collected at baseline from people who had household exposure to tuberculosis, aiming to see which signatures — alone or in combination — best flagged those who went on to develop active TB. The study focused on improving specificity without losing sensitivity so that fewer people would receive unnecessary preventive treatment while still catching most who will progress.

To carry out the comparison, the researchers used the NanoString platform to profile expression of the 15 published gene signatures in whole blood RNA. They scored signature expression using two established scoring methods, GSVA and PLAGE, allowing direct comparison on the same technological platform. The team also applied machine learning to derive a new, parsimonious signature and explored whether combining this new signature with a published signature would improve predictive performance. They report that combining signatures enhanced specificity and that a particular combined signature met WHO TPP levels for a triage test. The combined signature achieved 90.9% sensitivity and 88% specificity, with a positive predictive value (PPV) of 0.24 and a negative predictive value (NPV) of 1. These results came from the household contact cohort in Brazil and reflect head-to-head testing of multiple published signatures on the NanoString platform using GSVA and PLAGE scoring.

The findings suggest a practical way to narrow down who should receive preventive therapy: by using a combined blood transcriptional signature, clinicians could better identify the minority of exposed people most likely to progress to active tuberculosis. Because the combined signature reached WHO TPP thresholds for a triage test, it shows promise for real-world screening programs that need both high sensitivity and improved specificity. A PPV of 0.24 means that about one in four people flagged by the test would go on to develop active disease, while an NPV of 1 indicates those testing negative were very unlikely to progress in this cohort. The study highlights the advantage of testing multiple published signatures on the same platform and using machine learning to produce a streamlined component that enhances overall performance. If validated in larger and varied populations, this combined signature could help target prophylactic treatment, reduce unnecessary treatment, and ultimately prevent TB morbidity and mortality.