Blood signatures that see tuberculosis recurrence coming, and cannot see failure

Design

Prospective multicentre cohort with nested case-control matching (about 1:3) and benchmarking against WHO target product profile criteria

Population

263 adults with recurrence-free cure matched to 33 with treatment failure, 24 who died and 9 with recurrence, across five Brazilian sites

Primary outcome

Prediction of death, treatment failure and recurrence by 22 blood transcriptomic signatures at baseline, two months and end of treatment

Effect

Recurrence prediction AUC 0.71-0.91 at baseline and two months, falling to 0.42-0.89 at end of treatment; 2/22 signatures met WHO criteria at baseline, 13/22 at two months, none at end of treatment; treatment failure prediction AUC below 0.70 throughout; several baseline signatures predicted death with AUC 0.80 or more

Tuberculosis treatment is monitored on sputum, which is slow, often unobtainable, and a poor guide to who will relapse after apparently successful treatment. This Brazilian study tested whether host blood gene expression could do better. Adults with culture-confirmed, drug-susceptible pulmonary tuberculosis at five sites gave whole-blood samples at baseline, at two months and at the end of treatment, and were followed for 24 months. Twenty-two previously published transcriptomic signatures were measured and benchmarked against the WHO target product profile.

The results split cleanly. Signatures measured at baseline and at two months predicted recurrence, with areas under the curve between 0.71 and 0.91; 13 of 22 met the WHO minimum criteria at two months against only 2 at baseline. Several baseline signatures predicted death during treatment or follow-up, with AUC of 0.80 or above. But prediction of microbiological treatment failure was poor at every timepoint (AUC below 0.70), and no signature met the benchmark at the end of treatment — the point at which a clinician most wants to know whether to stop.

The practical implication is about where such a test would sit: at two months, to decide whether to shorten treatment in an early responder or intensify it in someone at high risk — not at the end, as a test of cure.

• No test to order; these are research assays measured by microfluidic RT-qPCR
• The useful timepoint is two months, not end of treatment
• Note that these signatures predict recurrence and death but not culture-positive failure
• Recurrence outcomes rested on only nine cases; treat the upper AUC estimates cautiously
• Continue existing sputum-based monitoring; nothing here replaces it

Why it matters

It points at individualising tuberculosis treatment duration, which is the change that would matter most to patients on six months of therapy.

Don't overread it

Nine recurrence events underpin the recurrence analysis; these are exploratory benchmarks, not a validated test.

The statistics, in plain English

An area under the curve of 0.71 to 0.91 is a wide range across 22 different signatures, not a confidence interval for one: the best performed well and the worst barely better than chance. The recurrence analysis rested on nine recurrence cases matched against 263 cured participants, which is far too few for a stable estimate, and the highest AUCs in such a setting are usually the most over-optimistic. The failure-prediction result, consistently below 0.70, is the more reliable finding because it is a negative one.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.