Data sources and search strategies
We searched MEDLINE, EMBASE, and the Cochrane Infectious Diseases Group Specialized Register. Our last search was carried out on July 16, 2019. Furthermore, we manually reviewed the bibliography of the included articles. The primary search terms were “Xpert,” “GeneXpert,” “Cepheid,” “MTB/RIF,” and “Tuberculosis.” The search methodology applied for each database is shown in Additional File 1. The bibliography was screened for full-length research articles in all languages. Moreover, we conducted a full-text review to select articles exclusively on pediatric EPTB.
The following were the inclusion criteria: (1) studies using Xpert as a diagnostic tool for detecting EPTB in comparison with a reference standard in each study, with all non-respiratory samples (i.e., lymph node aspirate or tissue, CSF, pleural fluid, etc.); (2) studies evaluating the diagnostic performance of Xpert; (3) studies providing pediatric (0–18 years) data. Studies were included regardless of HIV infection.
We excluded reviews, letters, editorials, expert opinions, animal experiments, and studies that only presented as an abstract. Studies that did not include separate pediatric data were also excluded. We attempted to include all types of EPTB samples; however, studies reporting the use of gastric lavage samples were excluded because they were intended for the diagnosis of pulmonary TB. Studies including samples from fewer than five patients and studies with no or insufficient data to construct a 2 × 2 contingency table to determine sensitivity and specificity were also excluded. If data were obtained in more than one article from the same author, the article with the most data was selected.
Two review authors (YS Seo and JK Ahn) independently assessed titles and abstracts in accordance with the inclusion and exclusion criteria, followed by a full-text review of the selected studies. Discrepancies regarding the inclusion of articles between the two authors were resolved by the third author (DS Kim).
Composite reference standard (CRS)
To compare the accuracy of Xpert, mycobacterial culturing or a CRS were used as reference standards herein. The CRS was defined by the authors of each study. Owing to paucibacillary characteristics of extrapulmonary TB, the clinical diagnosis of TB was also included. The CRS included histopathological, smearing, and clinical response analysis to treatment with anti-TB therapy along with culturing.
Qualitative assessment was performed using the Quality Assessment of Diagnostic Accuracy Studies‐2 (QUADAS‐2) tool . All eligible studies were evaluated on the basis of four domains: patient selection, index test, reference standard, and flow and timing. Each domain was assessed in terms of the risk of bias, and the first three domains were assessed in terms of concerns regarding applicability.
We determined the sensitivity and specificity of Xpert with 95% confidence intervals in comparison with culturing or the CRS. To assess heterogeneity among the studies, the chi-square test was performed. Heterogeneity was defined as a p-value of <0.10. In case of heterogeneity, different thresholds were considered to influence sensitivity and specificity. To assess the presence of a threshold effect, Spearman’s ρ correlation analysis was performed with ρ > 0.6 implying a threshold effect. Sensitivities and specificities of Xpert in each study were determined and subjected to the meta-analysis with a bivariate random-effects model. We plotted the summary receiver operating characteristic (sROC) curve with this model. R-package mada (version 3.5.1.) was used to generate forest plots and an ROC curve.