Diagnostic Performance of Xpert MTB/RIF in Tuberculous Pleural Effusion: Systematic Review and Meta-analysis

Abstract

A systematic review investigating the role of Xpert MTB/RIF in the diagnosis of tuberculous pleural effusion (TPE) was conducted. The pooled sensitivities and specificities of Xpert MTB/RIF were 51.4% and 98.6%, respectively, with culture used as a reference standard and 22.7% and 99.8%, respectively, with a composite reference standard (CRS) used as the benchmark. Xpert MTB/RIF has low sensitivity but excellent specificity in the diagnosis of TPE.

TEXT

Tuberculous pleural effusion (TPE) is the second most common site of extrapulmonary tuberculosis (EPTB) (1). Currently, the best tool for diagnosing TPE is a thoracoscopic pleural biopsy, an invasive procedure, with sensitivity ranging from 93 to 100% (2–5). The most widely used diagnostic test for TPE is the pleural fluid adenosine deaminase (ADA) level, which has variable sensitivity ranging from 47 to 100% (6). The sensitivities of pleural fluid microscopy and culture in TPE are about 10% and 20%, respectively (5, 7, 8). The absence of a simple reference standard makes the treatment of TPE empirical (nonmicrobiological) in most circumstances.

Xpert MTB/RIF (GeneXpert) has not only high sensitivity and specificity in smear-positive pulmonary tuberculosis (TB) (98% and 98%, respectively) but also reasonable diagnostic performance in smear-negative pulmonary tuberculosis (67% and 99%, respectively) (9). However, a recent meta-analysis reported the pooled sensitivity and specificity of GeneXpert in TPE as 46.4% and 99.1%, respectively, compared with those of pleural fluid mycobacterial culture (10). Since then, several new studies evaluating GeneXpert in TPE have been published. Herein, we perform an updated meta-analysis on the role of Xpert MTB/RIF in the diagnosis of TPE.

The systematic review was conducted using the standard methodology for systematic reviews of diagnostic test accuracy (see the detailed methodology in the supplemental material) (11, 12). Institutional ethics committee approval was not required as this was a systematic review of published data.

Our search yielded 155 citations, of which 24 studies (2,486 patients) were included in the current analysis (see Fig. S1 in the supplemental material) (7, 8, 13–34). There were equal numbers of studies from the developed and the developing countries (see Tables S1 and S2 in the supplemental material) (8, 14, 16–22, 24, 32, 33). Only two studies included children (13, 15). Seven studies included people living with HIV infection (8, 17, 18, 20–22, 27). Twenty-one studies (2,167 patients) and 10 studies (937 patients) used a microbiological standard (mycobacterial culture) (7, 8, 13–16, 19, 20, 22–34) or a composite reference standard (CRS) (7, 14–18, 21, 22, 24, 28), respectively, to assess the diagnostic accuracy of GeneXpert (see Table S3). Seven studies performed Xpert MTB/RIF on frozen pleural fluid (7, 15, 19, 21, 23, 28, 30), while the remaining used fresh samples (see Table S4). Nine studies concentrated the pleural fluid sample by centrifugation before performance of GeneXpert (7, 8, 16–18, 20, 22, 24, 29). The overall study quality assessed by the QUADAS-2 tool showed a low risk of bias except for studies using a case-control design (see Fig. S2). The risk of bias for the index test domain was largely due to a lack of information on blinding.

The pooled sensitivity and specificity of Xpert MTB/RIF compared to those of pleural fluid mycobacterial culture were 51.4% (95% confidence interval [CI], 43.3 to 59.7) and 98.6% (95% CI, 97.1 to 99.6), respectively (Fig. 1). The pooled sensitivity and specificity of Xpert MTB/RIF versus those of a CRS were 22.7% (95% CI, 12.8 to 36.9) and 99.8% (95% CI, 97.2 to 99.9), respectively (Fig. 1). The area under the concentration-time curve (AUC) values for Xpert MTB/RIF were 0.843 and 0.721 for pleural fluid cultures and a CRS, respectively (see Fig. S3 in the supplemental material). There was no heterogeneity for the outcomes of sensitivity (I², 33.2) and specificity (I², 15.9) using culture as the reference standard. However, there was significant heterogeneity for the outcomes of sensitivity (I², 80.8) and specificity (I², 91.3) using a CRS. There was no evidence of publication bias (see Fig. S4).

FIG 1.

Forest plot of the sensitivity and specificity of Xpert MTB/RIF in tuberculous pleural effusion using a pleural fluid culture reference standard and a CRS. The sensitivity and specificity of each individual studies are represented by a square through which runs a horizontal line (95% confidence interval). TP, true positive; FP, false positive; FN, false negative; TN, true negative.

On subgroup analysis, the pooled sensitivity was significantly higher in studies that used concentrated pleural fluid samples (P = 0.004). The pooled sensitivity was also higher but not statistically significant in studies using fresh samples, in those published from countries with high TB burdens, and in smaller studies (see Table S5 in the supplemental material). On Bayesian analysis, Xpert MTB/RIF has a high positive predictive value in areas with high TB prevalence, while the negative predictive value was not affected by the TB prevalence (see Fig. S5 in the supplemental material).

The results of this meta-analysis suggest that Xpert MTB/RIF can detect TPE in 22.7% and 51.4% of patients, using a CRS and a pleural fluid culture as the reference standard, respectively. The true sensitivity of Xpert MTB/RIF is expected to lie somewhere between the two estimates when an ideal reference standard such as thoracoscopic biopsy specimens with histopathology is used (4, 5). The specificity of Xpert MTB/RIF was very high with either reference standard, making it an excellent “rule-in” test. Compared to the previous meta-analysis on TPE, our study identified 14 additional studies and almost twice as many patients (10). We also performed a subgroup analysis to determine factors affecting the yield of GeneXpert. Further, we used the bivariate random effect model for pooling estimates. Thus, the validity of our results is significantly higher than that of the previous meta-analyses.

Is there any role for GeneXpert in the diagnosis of TPE, given the poor sensitivity? In clinical practice, the initial test used in the diagnosis of TPE is the pleural fluid ADA level. If the ADA level is >70 U/liter, most patients receive antituberculosis treatment (ATT). In those with ADA levels between 40 to 70 U/liter, ATT is given if the pretest probability of TPE is high (age of <45 years, nonsmoker, straw-colored effusion, and high tuberculosis prevalence area). Patients with ADA levels of <40 U/liter are almost always subjected to thoracoscopic pleural biopsy (4). Obviously, GeneXpert cannot be used alone for the diagnosis of TPE, given its low sensitivity. However, unlike other tests such as pleural fluid interferon gamma release assays (IGRAs) (35), the specificity and thus the positive predictive value of GeneXpert in the diagnosis of TPE are high. In fact, among 1,000 patients with pleural effusions with pretest probabilities of TPE of 5% and 20%; the use of GeneXpert will enable detection of 26 and 103 patients, respectively (see Table S6 in the supplemental material). This would obviate the need for an invasive procedure such as pleural biopsy in these patients.

Our meta-analysis has a few limitations. Pleural fluid culture and a CRS are both poor reference standards for diagnosing TPE, and future studies should use pleural tissue histology as the reference standard. Most studies included in the current analysis lacked information on blinding. Also, sample collection and processing varied across the studies. We have studied the performance of GeneXpert as a standalone test, and its utility when combined with other investigations like pleural fluid ADA levels cannot be commented upon. With the current analysis, we cannot comment on the ability of Xpert MTB/RIF for detecting rifampin resistance, as such information was not provided by the studies included in our analysis. We included studies published in English only, so some other studies might have been missed. Finally, our results cannot be extrapolated to children or people living with HIV infection, as separate information was not available for these subgroups.

In conclusion, Xpert MTB/RIF has poor sensitivity for the diagnosis of TPE. However, given its high specificity, it can potentially obviate the need for an invasive procedure in at least one-fourth of patients with TPE.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.