PET–CT benchmarked detection and 5-year progression of asymptomatic tuberculosis: a longitudinal, prospective cohort study

BACKGROUND: F]-fluorodeoxyglucose PET-CT) and compared with chest x-ray computer-aided detection (CAD). METHODS: We enrolled a prospective longitudinal cohort in Khayelitsha, Cape Town, South Africa, of asymptomatic, HIV-uninfected contacts aged 18-65 years of patients with rifampicin-resistant tuberculosis, a tuberculosis high-risk group not eligible for chemoprophylaxis. Participants underwent baseline PET-CT, chest x-ray, phlebotomy, and intensive sputum collection, and were classified into four PET-CT lung categories: consistent with tuberculosis, inactive tuberculosis, other lesions, and normal. Chest x-ray was processed by three types of CAD software (CAD4TB [version 7.0], qXR [version 3.0.0], and Lunit [version 3.1.4.111]). Follow-up included symptom-agnostic tuberculosis screening (23-38 months) and provincial register review (≤74 months), and a subgroup had repeat PET-CT (5-15 months). Tuberculosis was defined as bacteriologically confirmed or clinically diagnosed. The primary outcome measures were hazard ratio (HR) for tuberculosis diagnosis and treatment by baseline PET-CT lung abnormality category with normal as the reference group, and diagnostic performance of chest x-ray CAD software using area under the receiver operator characteristic curve (AUC). FINDINGS: 250 asymptomatic adults were enrolled between March 3, 2015, and Oct 11, 2017, irrespective of tuberculosis history or previous infection, and followed up for 1107 person-years (median 4·7 years [IQR 4·0-5·1]). 18 (7%) participants were treated for tuberculosis (16 [89%] of 18 bacteriologically confirmed). Six of 18 participants were diagnosed at baseline (four requiring induced sputum culture) and 12 of 18 after a median of 32 months (IQR 12-35). By baseline PET-CT category, tuberculosis was diagnosed and treated in 12 (41%) of 29 participants with scans consistent with tuberculosis, two (7%) of 30 with scans consistent with inactive tuberculosis, two (2%) of 83 with scans showing other lesions, and two (2%) of 108 with scans showing normal lungs. Participants with baseline PET-CT scans consistent with tuberculosis had the highest risk of 5-year tuberculosis diagnosis (HR 28·54 [95% CI 6·37-127·81] compared with those with scans showing normal lungs, p