Accuracy of upper respiratory tract samples to diagnose Mycobacterium tuberculosis: a systematic review and meta-analysis

Summary Background Pulmonary tuberculosis due to Mycobacterium tuberculosis can be challenging to diagnose when sputum samples cannot be obtained, which is especially problematic in children and older people. We systematically appraised the performance characteristics and diagnostic accuracy of upper respiratory tract sampling for diagnosing active pulmonary tuberculosis. Methods In this systematic review and meta-analysis, we searched MEDLINE, Cinahl, Web of Science, Global Health, and Global Health Archive databases for studies published between database inception and Dec 6, 2022 that reported on the accuracy of upper respiratory tract sampling for tuberculosis diagnosis compared with microbiological testing of sputum or gastric aspirate reference standard. We included studies that evaluated the accuracy of upper respiratory tract sampling (laryngeal swabs, nasopharyngeal aspirate, oral swabs, saliva, mouth wash, nasal swabs, plaque samples, and nasopharyngeal swabs) to be tested for microbiological diagnosis of tuberculous (by culture and nucleic acid amplification tests) compared with a reference standard using either sputum or gastric lavage for a microbiological test. We included cohort, case-control, cross-sectional, and randomised controlled studies that recruited participants from any community or clinical setting. We excluded post-mortem studies. We used a random-effects meta-analysis with a bivariate hierarchical model to estimate pooled sensitivity, specificity, and diagnostics odds ratio (DOR; odds of a positive test with disease relative to without), stratified by sampling method. We assessed bias using QUADAS-2 criteria. This study is registered with PROSPERO (CRD42021262392). Findings We screened 10 159 titles for inclusion, reviewed 274 full texts, and included 71, comprising 119 test comparisons published between May 13, 1933, and Dec 19, 2022, in the systematic review (53 in the meta-analysis). For laryngeal swabs, pooled sensitivity was 57·8% (95% CI 50·5–65·0), specificity was 93·8% (88·4–96·8), and DOR was 20·7 (11·1–38·8). Nasopharyngeal aspirate sensitivity was 65·2% (52·0–76·4), specificity was 97·9% (96·0–99·0), and DOR was 91·0 (37·8–218·8). Oral swabs sensitivity was 56·7% (44·3–68·2), specificity was 91·3% (CI 81·0–96·3), and DOR was 13·8 (5·6–34·0). Substantial heterogeneity in diagnostic accuracy was found, probably due to differences in reference and index standards. Interpretation Upper respiratory tract sampling holds promise to expand access to tuberculosis diagnosis. Exploring historical methods using modern microbiological techniques might further increase options for alternative sample types. Prospective studies are needed to optimise accuracy and utility of sampling methods in clinical practice. Funding UK Medical Research Council, Wellcome, and UK Foreign, Commonwealth and Development Office.


Supplementary data tables and figures
Supplementary tables S1: Search strategy S2: Table to  S9: Nasopharyngeal reference standard meta-regression and table of results S10: Table of data from naso-pharyngeal aspirate reports and clinical diagnosis as a reference standard used in the meta-analysis. S11: Data table to show analysis of reports of oral swabs as sample type. S12: Oral swab reference standard meta-regression and table of results S13: Data table to show analysis of reports of other sample type data.
Supplementary Figures: SF1: Summary receiver operating curve of laryngeal sampling for active pulmonary tuberculosis, with random effects meta-analysis SF2: Summary receiver operating curve of nasopharyngeal aspirate for active pulmonary tuberculosis, with random effects meta-analysis SF3: Summary receiver operating curve of nasopharyngeal aspirate for active pulmonary tuberculosis when clinical diagnosis used as a reference standard, with random effects meta-analysis SF4: Summary receiver operating curve of oral swab for active pulmonary tuberculosis, with random effects meta-analysis SF5: Bias of individual studies presented via each domain question and overall rating. SF6: Bias of individual studies presented by domain. Table S1:

Search strategy for systematic review
Search terms ("Mycobacterium tuberculosis" OR "Mycobacterium tuberculosis complex" OR "tuberculosis" OR "TB" OR "consumption" OR "wasting" OR "thysis" OR "pthysis OR "Koch's Disease") AND ("oral swab" OR "oral" OR "laryngeal swab" OR "laryngeal" OR "tonsils" OR "saliva" OR "Waldemeyers ring" OR "pharynx" OR "pharyngeal") AND ("diagnosis" OR "diagnostics" OR "samples" OR "sampling") Plus database specific strategy terms Medline (1847-1950): Human Medline (1950current): Human, Clinical Key: Diagnosis Global Health: As above, remove consumption Global Health archive: As above, remove consumption Supplementary not meet the criteria for unconfirmed TB. Osorio 2020 (i) signs/symptoms: (a) persistent cough (>2 weeks), unremitting cough; (b) weight loss/failure to thrive; © persistent (>1 week), unexplained fever reported by guardian; (d) persistent, unexplained lethargy or reduced playfulness; (e) infants 0-60 days with additional signs and symptoms like neonatal pneumonia, unexplained hepatosplenomegaly or sepsis-like illness; (ii) findings on chest X-ray congruent with pulmonary TB (presence of lymphadenopathy and/or abnormalities consistent with TB as new infiltrates) and read by two blinded operators (clinician and TB expert); (iii) history of exposure to M. tuberculosis within the preceding 12 months; or (iv) response to antituberculosis treatment yet no acid-fast bacillus on the sputum smear or a negative Xpert MTB/RIF test. 17 No TB 28  Participants had to have a visible cervical lymph node mass measuring >1 cm × 1 cm and persisting for >1 month despite antibiotic therapy for at least 5 days or a parenchymal abnormality on chest radiograph in addition to at least one of the following symptoms: 1) cough or wheezing >4 weeks not resolved after treatment with antibiotics, with cough continuing for at least 2 weeks after starting antibiotics (for hospitalized children only, respiratory distress or diagnosis of severe pneumonia not responding to antibiotics after 5 days or any cough >4 weeks despite at least 5 days of antibiotics), 2) moderate or severe malnutrition (defined as weight-for-height Z score <-2 standard deviations (SD) or -3SD, respectively) not responding after 3 weeks of treatment for malnutrition, and 3) reported fever >7 days not responding after 5 days of antibiotics or antimalarials. Children were excluded from the study if they were currently on tuberculosis treatment or isoniazid preventive therapy (IPT) or had received treatment for tuberculosis in the past year or IPT in the last 6 months.
S Table S3: Full descriptive data including methods used available as excel spreadsheet.
https://osf.io/9nuvq     1 S12: Oral swab reference standard meta-regression A larger variability in reference standard was seen in oral swab studies. A meta-regression therefore was carried out focused on sample type over culture v. Xpert as the model would have been overfitted with a larger number of covariates. Sixteen were classified as using sputum and four as gastric lavage. Firstly a separate meta-analysis was run for each reference test (Table S11) , and there was statistical evidence (Chi2 = 6.16, 2 df, P = 0.046) of association of reference test type with sensitivity and/or specificity of oral swabs. Two further models were then analysed one comparing sensitivity with reference sample as a covariate (C) and one of specificity (D). These did not show a statistical significant difference in the modelled specificity or sensitivity (likelihood ratio comparing model B and C: Chi2 = 2.01, 2 df, P = 0.156, model B and D: Chi2 = 2.30, 2 DF P = 0.130 [Table S11]). Lastly a model allowing for unequal variances (E) was run which showed a significant effect on specificity of reference sample types for sputum (Fixed effect specificity sputum Pr 9.27 x 10^-10).