OBJECTIVES: The Xpert MTB/RIF Ultra (Ultra) assay is widely used to diagnose pulmonary tuberculosis (TB), but 'very low' or 'trace' results may reflect either paucibacillary TB or TB-negative disease, complicating clinical decision-making. We evaluated host transcriptomic signatures to identify culture-confirmed paucibacillary TB among individuals with Ultra very low or trace results. METHODS: We performed whole blood targeted transcriptional profiling of 90 symptomatic adults from Uganda, Kenya, and South Africa with Ultra very low/trace sputum results. An 81-gene customized NanoString panel representing 13 published TB signatures was analyzed using machine learning to derive a novel four-gene host signature ("TRACE4") predicting MGIT and LJ culture positivity. Validation included individuals with other respiratory diseases (n=18) and North American-TB-negative controls (n=20). Data were randomly split 75/25 into training (n=67) and test (n=23) sets. Diagnostic performance was evaluated against WHO targets. RESULTS: TRACE4 outperformed all published signatures in both training (AUC 0.89) and test sets (AUC 0.88), achieving 82% specificity at 75% sensitivity. It also exceed reclassification based on prior TB history (specificity 0.58 (95% CI: 0.45-0.69); sensitivity 0.35 (95% CI: 0.15-0.61)). TRACE4 showed 100% specificity in non-TB controls. CONCLUSION: TRACE4 shows promise for identifying paucibacillary culture-positive pulmonary TB in this diagnostically challenging group.
Lopez et al. (Fri,) studied this question.