New data tool sorts tuberculosis by disease severity

Many current approaches to describing tuberculosis (TB) at diagnosis boil complex disease into a simple yes-or-no label — for example, “advanced” versus “minimal or early disease.” But TB does not always fit into two neat boxes: as people move through stages of illness we expect rising bacterial burden and mounting inflammation, and those changes map to growing severity and worse outcomes. To explore that range, Samantha Malatesta and colleagues created a new method called tuberculosis SeveriTy Assessment Tool for Informed Stratification (TB-STATIS). TB-STATIS is designed to identify multiple distinct clinical severity classes that can exist when patients first present to care, and to assign a predicted disease class to each person based on the measurements available. The idea is to move beyond a binary snapshot and to use the data already collected in clinics — from laboratory tests to images and symptoms — to paint a fuller picture of where someone sits on the spectrum of TB disease at the time of diagnosis.

TB-STATIS pulls together information from multiple data sources — for example smear microscopy, chest x-ray findings, and reported symptoms — and uses a data-driven modeling framework inspired by event-based modeling to infer a set of disease severity classes. In computer simulations the team tested the method under different conditions and found that TB-STATIS could recover the true set of classes across varying sample sizes, mixtures of data sources, and levels of uncertainty in the measurements. The researchers then applied TB-STATIS to two real-world data sets: an observational TB cohort in South Africa and data from a global phase 3 clinical trial that tested the non-inferiority of two 4-month regimens compared to the standard 6-month regimen for the treatment of TB. The disease classes produced by TB-STATIS correlated with culture conversion, a commonly used proxy for TB treatment response, suggesting the classes capture clinically meaningful differences in disease.

The work introduces a practical way to stratify patients by TB severity using information already collected in many care settings. By identifying multiple clinically meaningful strata rather than forcing a two-group split, TB-STATIS could help researchers and clinicians better describe who is likely to do well with standard care and who might need different monitoring or treatment approaches. Because the tool provides a predicted disease class for each individual based on observed data, it can be applied in observational studies and clinical trials to compare outcomes across more refined patient groups, and it may improve interpretation of trial results like those from the phase 3 study. Overall, this integrated, data-driven approach offers a clearer map of the spectrum of TB disease at presentation and a way to tie that map to measurable treatment response markers such as culture conversion.