Accuracy of Xpert MTB/RIF assay for the diagnosis of tuberculous pleural effusion

Abstract Background Tuberculosis poses a severe threat to human health. At present, compared with the traditional diagnostic methods for tuberculosis pleural effusion, such as Löwenstein–Jensen culture, pleural biopsy, and Ziehl–Neelsen smear microscopy, Xpert MTB/RIF was regarded as an emerging technology for its efficiency. The Xpert MTB/RIF accuracy for tuberculous pleural effusion diagnosis was evaluated in this systematic study. Materials and methods We searched the relevant literature published before January 2021 in PubMed, Cochrane, EMBASE, and Web of Science databases. Utilizing Review Manager 5.3 software, the quality of the included literature was evaluated based on the Quality Assessment of Diagnostic Accuracy Studies criteria. Sensitivity, specificity, and the summary receiver operating characteristic curves were plotted and analyzed with Metadisc 1.40 software. We used Stata 12.0 software to evaluate the publication bias of this study. Results Eighteen articles were identified in total. The sensitivity of Xpert MTB/RIF in the pleural effusion was 0.24, and specificity was 1.00, respectively. The area under the summary receiver operating characteristic curve was 0.9737, which indicated that the overall accuracy of the Xpert MTB/RIF was high. In addition, based on the Deeks funnel plot, no publication bias of the study was found. Conclusion Xpert MTB/RIF is a rapid method with high specificity but relatively low sensitivity for detecting Mycobacterium tuberculosis in pleural effusion. Its less sensitivity made it difficult to be used clinically, but the high specificity suggests that it can be used as a specific diagnostic method for tuberculous pleural effusion.


| Inclusion criteria and exclusion criteria
Three researchers screened the retrieved literature in accordance with pre-defined inclusion and exclusion criteria. Each paper was proofread by two researchers independently. In the situation of disagreement, the third researcher would make the judgment and obtain the complete screening result finally.
The inclusion criteria were summarized as follows: (1) analysis of human specimens, (2) English version, (3) Xpert MTB/RIF was in comparison with another reference standard to test the accuracy of diagnosis for tuberculous pleural effusion, and (4) the data in the article are enough to create a four-cell table.
The exclusion criteria were summarized as follows: (1) nonhuman samples; (2) repeated publications, conference abstracts, letters, case reports, editorials, reviews, and meta-analyses; (3) lack of four-grid table data; and (4) the literature lacks a gold standard or Xpert MTB/RIF analysis.
Detailed flowcharts for inclusion and exclusion are shown in additional materials.

| Data collection
Data extraction and quality assessment of all literature were first completed independently by two researchers. Results were reviewed, and inconsistencies were discussed by the two researchers.
If an agreement cannot be reached, the third researcher will make an evaluation. Finally, a consensus will be reached based on the judgments of the three researchers, summarizing all the results.

| Data extraction
Three researchers extracted relevant data of the study articles, including the name of the first author, study design, country, year of publication, sample size, reference standard, and false positive (FP), true positive (TP), true negative (TN), and false negative (FN). After that, three researchers focused on the final extraction results and set up a feature table for the extracted data.

| Quality assessment standard
The Quality Assessment for Diagnostic Accuracy Studies (Q UADAS-2) 11 was used as a criterion to evaluate the quality of the included studies. Afterward, Review Manager (Version 5.3) software was applied to evaluate the diagnostic accuracy of Xpert MTB/RIF. The risk of bias for each study was evaluated using "yes," "unclear," and "no," according to the eleven criteria in the four parts of QUADAS-2 (patient selection, index test, standard gold method, flow, and time). Charting with the software, we analyzed the risk of bias and suitability issues, including patient selection, indicator trials, reference criteria, procedures, and timing.

| Statistical analysis
Sensitivity, specificity, positive-likelihood ratio (PLR), negativelikelihood ratio (NLR), and diagnostic odds ratio (DOR) were generated using Meta disc (version 1.40). We plotted and analyzed the summarized receiver operating characteristic (SROC) curves and calculated the area under the curve (AUC). Stata (version 12.0) software was used to draw Deeks funnel plots to assess whether there was bias in the literature. Finally, quality assessment was studied using the Review Manager (version 5.3) software.

| Search results
From the databases mentioned above, 125 relevant articles were identified, including 44 in PubMed, 7 in Cochrane Library, and 74 in the Web of Science. Of the 125 references, 51 were duplicates.
A total of 28 articles remained after reviewing the initial selection of titles and abstracts. Then, a further ten articles were excluded after the full-text screening of the remaining literature for the following reasons: 1 article was a meta-analysis, two lacked reference standard, five were unable to extract complete data, one was a non-English article, and another could not be found its full text.

| Characteristics of eligible studies
Data were extracted from the final 18 articles, and feature information, such as the author name and year of publication, is summarized in Table 1.

| Quality assessment
The quality of the 18 articles (Figures 1 and 2

| Data analysis
MetaDiSc was applied to analyze the fourfold

| Publication bias
In a meta-analysis, the Deeks funnel plot (Figure 9) generated by Stata 12.0 was used to test the data. The Egger test showed that the p-value of this study was 0.148 > 0.050, indicating that no publication bias was found in the study.

| DISCUSS ION
According to the data from World Health Organization in 2019, tuberculosis (TB) remains the leading cause of morbidity and death worldwide, with an annual number of deaths of over With the data gained, Xpert MTB/RIF in this study suggested high specificity and low possibility of misdiagnosis. However, its sensitivity was not high enough as a diagnostic method. We suggested that Xpert might be used in combination with other diagnostic methods.
However, the current research still had some limitations.
Firstly, we only retrieved and extracted data from the literature published in the four English databases, leading to a lack of comprehensiveness and bias. Secondly, our study only included articles from the beginning of the study through January 2021. In addition, we did not delve into the effects of other potential factors on the results. Finally, the reference standards in each literature were not wholly consistent, making the results have a certain probability of bias.
According to the data analysis of our study, we learned that the low sensitivity of Xpert might be caused by the low bacteria load of mycobacterium in the tuberculous pleural fluid. 15

ACK N OWLED G EM ENTS
We acknowledge TCGA and GEO database for providing their platforms and contributors for uploading their meaningful datasets.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflicts of interest.