Abstract
BACKGROUND: The scale-up of tuberculosis preventative treatment is a global health priority yet constrained by our inability to identify which exposed contacts should be prioritised. We evaluated the diagnostic accuracy and clinical utility of new and established tests, including the first deployable host mRNA assay, done near point-of-care to diagnose incident tuberculosis in contacts in Mozambique, South Africa, and Uganda. METHODS: Contacts microbiologically-negative for tuberculosis or without symptoms were followed for 12 months. Xpert MTB-Host Response (Xpert MTB-HR), haemoglobin, and CRP was done by a minimally trained healthcare worker at M0 and M6 using finger-prick blood. Tuberculosis disease was identified at months 0, 6 and 12 by symptom- and chest X-ray-screening (microbiological testing if symptomatic or able to expectorate sputum or CXR-positive; CXR M12 only). Predictive accuracy and decision curve analyses were done, with microbiological confirmation serving as the primary reference standard. RESULTS: Among 3,031 contacts, 66 incident cases occurred. At a 75% sensitivity threshold (the minimum WHO-recommended sensitivity), Xpert MTB-HR and haemoglobin had specificities of 55% (95% confidence interval 53-56) and 25% (24-26), respectively. CRP's highest sensitivity with non-zero specificity was 52% (38-65) at 71% (69-72) specificity. At 75% specificity (the minimum WHO-recommended specificity), sensitivities were 61% (48-74), 37% (25-51) and 41% (28-55), respectively. Xpert MTB-HR and CRP in parallel (either positive) had highest sensitivity [76% (62-87)] and specificity of 59% (58-61), and the best clinical benefit for a number-willing-to-treat of 20-50. CONCLUSION: No tests individually met WHO sensitivity and specificity criteria but Xpert MTB-HR came closest. This work resets expectations for host RNA tests. Multi-marker strategies, including personalised risk scores, should be pursued.