Study population
In this secondary analysis of cross-sectional data, 304 adult patients were serially enrolled from an outpatient clinic of a rural hospital in Andhra Pradesh during 2012–13. Hospital physicians referred patients with clinical signs and symptoms of TB. Sputum and blood samples were collected from all suspected cases of TB, per standard of care. Demographic information and clinical history (including vaccinations and comorbidities) were collected from patient interviews.
The study protocol received approval from the institutional review board at St. John’s Research Institute (Bangalore, Karnataka, India). All study participants provided informed consent prior to any data collection. All analyses occurred subsequent to data de-identification (Additional file 1).
Assessment of outcomes
Active and latent TB were assessed in sputum and blood samples by three techniques: Xpert (Cepheid, Sunnyvale, California, United States), AFB, and QFT-G (Cellestis Limited, Carnegie, Victoria, Australia). The presence of Mycobacterium tuberculosis (MTB) in sputum was evaluated by Xpert and AFB. For Xpert assays, sputum specimens were decontaminated with N-acetyl-L-cysteine-sodium hydroxide, and processed according to manufacturer instructions. As a fully automated real-time multiplex polymerase chain reaction assay that is based on the GeneXpert platform, Xpert also assessed rifampin resistance (via rpoB gene mutations). AFB was based on standard protocol for Ziehl-Neelsen staining with conventional light microscopy.
Additionally, QFT-G measured the interferon-gamma response triggered by incubation with two MTB antigens (early secretory antigenic target-6; culture filtrate protein-10). QFT-G is recommended as an aid for diagnosing TB infection, as it can serve as an alternative to tuberculin skin tests but does not distinguish between latent versus active TB. For the QFT-G assay, blood was initially collected in QFT-G tubes. Samples were centrifuged to obtain plasma, processed according to manufacturer protocol, and assessed by enzyme-linked immunosorbent assay.
Assessment of exposure and other covariates
Diabetes status was based on patient self-report. Human immunodeficiency virus (HIV) status was assessed by a rapid HIV test (TRI-DOT; J. Mitra & Co. Pvt. Ltd.; New Delhi, India), per standard Revised National Tuberculosis Control Programme guidelines for individuals with suspected TB.
Statistical analyses
Relative diagnostic performance was assessed by calculating sensitivity, specificity, predictive values, and AUC (by receiver operator characteristic [ROC] curve visualization). Xpert results were considered as the gold standard for comparing conventional TB diagnostics. Study participants were categorized as either previous or new cases, based on whether anti-TB treatment was previously received. Statistical analyses were conducted with SAS software (version 9.4; SAS Institute Inc., Cary, NC, USA); two-tailed P-values <0.05 were considered statistically significant.
Patient demographic and health characteristics were reported as means (standard deviations [SDs]) or percentages. Group differences were assessed by t-tests (continuous variables with normal distribution; Satterwaite method for unequal variance) or Mantel-Haenszel χ2-tests (categorical variables). Potential covariates were identified a priori, based on previous literature. Initially, associations between potential covariates were assessed by univariate log-binomial regression models; variables with p < 0.2 were included in the full model. Subsequently, only variables with p < 0.05 were retained in the final multivariate log-binomial regression model.