Accuracy of lung and abdominal ultrasound for tuberculosis diagnosis: a prospective cohort study from India

Background

Point-of care ultrasound is considered to hold promise in tuberculosis (TB) screening. However, most available abdominal ultrasound data focuses on HIV-infected cohorts and for lung ultrasound (LUS) data is very sparse. We aimed to determine accuracy of lung and abdominal ultrasound in a cohort of presumed TB in a tertiary care hospital in India.

Methods

Adult patients with presumed TB were enrolled prospectively and underwent a comprehensive ultrasound evaluation. Accuracy of individual and a predetermined combination of findings was determined against a TB reference standard (mycobacterial culture and PCR). Diagnostic potential of a multi-variable model combining clinical and ultrasound findings was explored using generalized mixed methods and random forest approach. (German trial registry DRKS00026636)

Findings

We included 541 participants of whom 102 (19%) had TB and 1% had HIV. “Focused assessment with sonography for HIV-associated tuberculosis” (FASH) showed moderate sensitivity (51%, 95%-CI 41-60) and specificity (70%, 95%-CI 66-74). Small consolidations on LUS showed high sensitivity (98%, 95%-CI 93-99), but were unspecific (14%, 95%-CI 11-18). Exploratory LUS variations showed higher specificity (e.g., large apical consolidations: sensitivity 22%, specificity 86%). Predictive modelling for ultrasound and clinical variables revealed an Area Under the Curve of 0.79 in the receiving operator curve.

Interpretation

Accuracy of ultrasound does not meet requirements of a stand-alone diagnostic or screening test. However, accuracy for some ultrasound findings is comparable with CXR. Additionally, ultrasound may aid disease severity assessment and microbiological sampling strategies. Research into alternative analyses (e.g., artificial intelligence) may enable wider applications.

Funding

Grant TTU 02.911, German Center for Infection Research (Deutsches Zentrum für Infektionsforschung).

Research in context

Before undertaking the study, LUS for TB had been assessed in a small number of studies limited with uncertain ultrasound characterization of TB-related findings with lack of adequate terminology and unclear specificity for TB. FASH-studies in HIV+ and few studies in HIV- have shown moderate sensitivity and specificity, but the study design and reference standards were not robust enough for generalizability.

Our ultrasound study of LUS and FASH has a prospective cohort from a TB-endemic setting (India), we recruited 541 participants, the largest such cohort. This study brings to the body of evidence novel findings, backed by a robust study design and using a comprehensive reference standard. We were able to describe accuracy in a predominantly HIV-negative cohort of patients with presumed TB disease and compared our index testing protocol with the CXR, the most commonly used imaging modality. The implications from our study were that no single finding or combination of findings on LUS reached target product profiles (TPP) proposed by WHO, suggesting LUS as triage or diagnostic tool is too non-specific. The FASH accuracy in our study was in line with previous data, this study provides for it a solid foundation. The accuracy of submodules of LUS and FASH reaches that of CXR. Roles for ultrasound in TB care may lie in aiding diagnosis, assessment of disease severity, guiding of microbiological sampling or therapy monitoring. Further studies should focus on the accuracy of LUS in people living with HIV and children, evaluating ultrasound as a part of a diagnostic algorithm and the use of artificial intelligence to improve the yield of TB-POCUS.