Interferon Gamma Release Assays for Diagnosis of Pleural Tuberculosis: a Systematic Review and Meta-Analysis

Abstract

The role of interferon gamma release assays (IGRAs), although established for identifying latent tuberculosis, is still evolving in the diagnosis of active extrapulmonary tuberculosis. We systematically evaluated the diagnostic performance of blood- and pleural fluid-based IGRAs in tuberculous pleural effusion (TPE). We searched the PubMed and Embase databases for studies evaluating the use of commercially available IGRAs on blood and/or pleural fluid samples for diagnosing TPE. The quality of the studies included was assessed through the QUADAS-2 tool. The pooled estimates of sensitivity and specificity with 95% confidence intervals (95% CI) were generated using a bivariate random-effects model and examined using forest plots and hierarchical summary receiver operating characteristic (HSROC) curves. Indeterminate IGRA results were included for sensitivity calculations. Heterogeneity was explored through subgroup analysis and meta-regression based on prespecified covariates. We identified 19 studies assessing the T.SPOT.TB and/or QuantiFERON assays. There were 20 and 14 evaluations, respectively, of whole-blood and pleural fluid assays, involving 1,085 and 727 subjects, respectively. There was only one good-quality study, and five studies used nonstandard assay thresholds. The pooled sensitivity and specificity for the blood assays were 0.77 (95% CI, 0.71 to 0.83) and 0.71 (95% CI, 0.65 to 0.76), respectively. The pooled sensitivity and specificity for the pleural fluid assays were 0.72 (95% CI, 0.55 to 0.84) and 0.78 (95% CI, 0.65 to 0.87), respectively. There was considerable heterogeneity; however, multivariate meta-regression did not identify any covariate with significant influence. There was no publication bias for blood assays. We conclude that commercial IGRAs, performed either on whole-blood or pleural fluid samples, have poor diagnostic accuracy in patients suspected to have TPE.

INTRODUCTION

Tuberculosis (TB) is a common etiology of pleural effusion, especially in developing countries (1). A definitive microbiological diagnosis is achieved by tuberculous pleural effusion (TPE) in only a few patients, and the diagnostic accuracy of other pleural fluid investigations is suboptimal (1, 2). Estimation of adenosine deaminase or interferon gamma and detection of the mycobacterial genome in pleural fluid are used as diagnostic surrogates (2). Pleural biopsies may improve the diagnostic yield; however, these procedures are invasive, require expertise, and are not free from complications (3–5).

In recent years, interferon gamma release assays (IGRAs) have emerged as an immunodiagnostic tool to detect tuberculous infection. IGRAs quantify interferon gamma released by T-lymphocytes in response to stimulation by specific antigens encoded in region of difference 1 (RD1) of the Mycobacterium tuberculosis genome. An enzyme-linked immunosorbent spot (ELISpot) assay (T-SPOT.TB; Oxford Immunotec Limited, United Kingdom) and an enzyme-linked immunosorbent assay (ELISA) (QuantiFERON; Cellestis Limited, Australia) are commercially available. Neither can distinguish latent from active tuberculosis, and indeterminate assays are a significant problem. Despite these issues, IGRAs have been evaluated for diagnosing pulmonary and extrapulmonary tuberculosis, using blood or extrasanguinous samples (6–8).

The simplicity and noninvasive nature of IGRAs offer an attractive alternative approach for a quicker diagnosis of TPE. A meta-analysis of seven publications up to January 2010 concluded that IGRAs were not useful for diagnosing TPE (9). Since then, several new studies have been published. Herein we perform a systematic review and meta-analysis using a more clinically relevant algorithm of IGRA interpretation in patients with TPE.

MATERIALS AND METHODS

This review was conducted in accordance with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (10). Since the study was a systematic review and meta-analysis of published articles, patient consent or approval from the institutional ethics committee was not necessary.

Search strategy.

We searched the PubMed and Embase databases for articles published up to December 2014 in any language, using the following free text terms: (tuberculosis OR tubercular OR tuberculous OR TB OR mycobacterium OR mycobacterial) AND (pleura OR pleural OR pleuritis OR pleurisy) AND (interferon OR IFN OR interferon-gamma OR gamma-interferon OR interferon gamma assay OR interferon gamma release assay OR IGRA OR interferon release assay OR QuantiFERON OR T-SPOT OR ELISpot OR enzyme-linked immunosorbent spot OR T cell based assay OR T cell response). We also searched for additional studies from bibliographies of selected and previous review articles.

Study selection.

A primary review of titles and abstracts was done to screen articles potentially suitable for more detailed evaluation (Fig. 1). A secondary full-text review of manuscripts of all such potentially eligible articles was carried out by two reviewers (A.N.A. and R.A.) independently to identify studies suitable for inclusion. Any disagreements were resolved through consensus. We included studies meeting the following criteria: (i) at least 10 TPE patients; (ii) original data on the evaluation of diagnostic accuracy using a commercial assay based on RD1 antigens; and (iii) assays performed on pleural fluid and/or blood samples. A study was included only if it provided diagnostic accuracy figures or allowed calculation of sensitivity and specificity from observations reported as numerical data or dot plots. Immunological studies describing experimental data and case reports, conference abstracts, reviews, editorials, and letters not reporting original data in sufficient detail were excluded.

FIG 1.

Flow diagram for study selection. IGRA, interferon gamma release assay; TB, tuberculosis.

A diagnosis of TPE was based on any of the following criteria: (i) microbiological (demonstration of acid-fast bacilli on pleural fluid or biopsy smear or culture or PCR positivity for M. tuberculosis in pleural or other clinical specimens), (ii) pathological (pleural biopsy showing granulomatous inflammation with or without staining for mycobacteria), or (iii) clinical (response to antitubercular therapy).

For studies reporting data on more than one kind of assay, all investigations had to be performed on synchronously collected samples. For studies on overlapping groups of patients, only the study with the larger sample size was included. For studies reporting data on the same patients, the one providing greater methodological detail was included.

For studies conducting more than one IGRA on the same clinical samples, we considered results from each different assay as an independent evaluation. If two or more diagnostic thresholds were reported, we chose the one suggested by the manufacturers. When necessary, authors of some studies were contacted for supplemental information.

Data extraction and quality assessment.

Information from all studies was extracted independently by two reviewers (A.N.A. and R.A.). Articles in foreign languages were translated to English before evaluation (Google Translate). The following details were extracted: (i) the year of publication, study design, calendar period of study, and country where the study was performed; (ii) the age-band of the participants; (iii) the proportion of the patients seropositive for HIV infection; (iv) the type of IGRA used and the criteria for test interpretation; (v) the proportion of TPE patients diagnosed using microbiological and/or histopathological criteria (here termed “definite TB”); and (vi) the number of positive, negative, and indeterminate results for each group of patients.

The study quality was assessed through the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) tool, appropriately tailored to suit this review (11).

Data synthesis and analysis.

For each study, we calculated the sensitivity and specificity, with corresponding 95% confidence intervals (95% CI). We pooled the results using a bivariate random-effects model to generate summary estimates for the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR). Indeterminate results, wherever reported, were excluded before specificity was calculated and false negatives were considered in the calculation of sensitivity (6, 12). Separate analyses were carried out for assays conducted on whole blood and pleural fluid samples. The statistical software Stata (intercooled version 12.0; Stata Corp., College Station, TX) was used for analysis.

Heterogeneity was calculated using the I² statistic and defined as low, moderate, and high when I² was more than 25%, 50%, and 75%, respectively (13). Heterogeneity was further explored through sensitivity analysis stratified by predefined covariates. These included the IGRA test type, the national TB prevalence, the robustness of the reference standard (definite TB of >75% versus others), blinding (whether explicitly stated or not), reporting of indeterminate results, and the study sample size (>50 patients or not). Countries with an estimated TB prevalence of ≥0.1% were categorized as having high TB prevalence (14). A restricted maximum likelihood estimation meta-regression analysis was conducted to evaluate heterogeneity, using the natural logarithm of DOR as an independent variable. Publication bias was assessed through Deeks' funnel plot, with a value of P <0.10 for the slope coefficient indicating significant asymmetry suggestive of bias (15). Hierarchical summary receiver operating characteristic (HSROC) curves were generated to summarize global test performance and display variation in diagnostic accuracy among studies (16).

RESULTS

Study characteristics.

Our search yielded 606 citations, of which 21 studies finally met our selection criteria (Table 1) (17–37). Of the 21 articles, one each was published in the Chinese and Korean languages and the remaining were English. All studies were conducted on adult patients, and patients were from 12 countries; only one study was multicentric (37). Fifteen studies reported data from countries with a high TB prevalence (Table 1).

TABLE 1.

Summary characteristics of studies included in meta-analysis

Investigators, year (reference)	Country	TPE/non-TPE patients enrolleda	Age (yr)b		Gender (M:F)c		HIV status of TPE patients	IGRA methodd	Clinical sample	Test resultse
			TPE patients	Non-TPE patients	TPE patients	Non-TPE patients				TPE patients			Non-TPE patients
										TP	FN	I	TN	FP	I
Liu et al., 2014 (17)	China	59 (59)/52	39 ± 19	57 ± 15	41:18	29:23	All negative	T-SPOT.TB	Fluid	54	5	0	47	5	0
Zhang et al., 2014 (18)	China	45 (NRf)/49	NR	NR	NR	NR	All negative	T-SPOT.TB	Fluid	40	5	0	42	75	47
Keng et al., 2013 (19)	Taiwan	31 (30)/57	63.9 ± 19.1	63.7 ± 15.4	23:8	38:19	All negative	T-SPOT.TB	Fluid	12	2	17	21	17	19
Liu et al., 2013 (20)	China	55 (28)/43	39 (25–59)	57 (47–67)	18:37	18:25	All negative	T-SPOT.TB	Blood	51	4	0	27	16	0
									Fluid	52	2	1	39	3	1
Cirak et al., 2012 (21)	Turkey	23 (23)/77	41.3 ± 19.0	NR	20:3	60:17	NR	QFT-G	Blood	10	0	13	42	23	12
Eldin et al., 2012 (22)	Egypt	20 (20)/18	48.0 ± 14.7		26:12		NR	QFT-GIT	Blood	14	6	0	14	4	0
									Fluid	12	8	0	15	3	0
Gao et al., 2012 (23)	China	58 (17)/20	49 (12–91)	50 (12–85)	45:13	13:7	NR	QFT-GIT	Blood	54	4	0	18	2	0
Kang et al., 2012 (24)	Korea	21 (18)/15	39.9 ± 16.0	63.3 ± 17.5	15:6	6:9	All negative	QFT-G	Blood	10	5	6	5	5	5
									Fluid	4	0	17	7	6	2
								T-SPOT.TB	Blood	18	2	1	8	6	1
									Fluid	15	0	6	3	8	4
Ates et al., 2011 (25)	Turkey	43 (17)/29	32.9 ± 20.7	53.3 ± 19.4	47:25		NR	QFT-GIT	Blood	30	12	1	15	10	4
									Fluid	21	4	18	23	3	3
Chung et al., 2011 (26)	Korea	54 (40)/43	43 (19–92)	69 (20–94)	37:17	33:10	All negative	QFT-GIT	Blood	40	12	2	22	14	7
Losi et al., 2011 (27)	Italy	18 (15)/30	32 ± 16	71.5 ± 15	10:8	22:8	NR	QFT-GIT	Blood	14	4	0	19	11	0
									Fluid	15	3	0	16	14	0
Katiyar et al., 2010 (28)	India	52 (38)/50	47.5 (NR)		72:30		All negative	QFT-GIT	Blood	47	5	0	43	7	0
Dheda et al., 2009 (29)	South Africa	48 (48)/19	37.0 ± 14.4	55.3 ± 15.7	23:25	9:10	22 positive	QFT-GIT	Blood	26	4	8	9	4	5
									Fluid	22	19	1	12	3	3
								T-SPOT.TB	Blood	30	6	3	9	7	1
									Fluid	34	6	4	9	6	2
Lee et al., 2009 (30)	Korea	14 (NR)/12	NR	NR	NR	NR	All negative	QFT-G	Blood	12	2	0	7	4	1
Lee et al., 2009 (31)	Taiwan	19 (15)/21	60.5	65.5	15:4	12:9	All negative	T-SPOT.TB	Blood	14	4	1	19	2	0
									Fluid	18	1	0	18	3	0
Liao et al., 2009 (32)	Taiwan	19 (11)/13	NR	NR	NR	NR	NR	T-SPOT.TB	Blood	14	5	0	11	2	0
Baba et al., 2008 (33)	Norway	28 (12)/6	39 (20–70)		22:12		23 positive	QFT-GIT	Blood	17	1	6	5	0	0
									Fluid	12	1	14	3	0	2
Chegou et al., 2008 (34)	South Africa	30 (16)/32	35 ± 13	NR	18:12	17:15	9/14 positive	QFT-GIT	Blood	16	6	0	12	5	0
									Fluid	13	10	0	13	2	0
Nishimura et al., 2008 (35)	Japan	10 (8)/13	NR	NR	NR	NR	All negative	QFT-G	Blood	6	4	0	12	1	0
Ariga et al., 2007 (36)	Japan	28 (28)/47	60.5 ± 22.0	72.1 ± 10.8	24:4	40:7	All negative	QFT	Blood	21	6	0	33	14	0
									Fluid	27	1	0	46	1	0
Losi et al., 2007 (37)	Italy, Germany, and Netherlands	20 (10)/21	34 (27–52)	71 (44–80)	13:7	15:6	NR	T-SPOT.TB	Blood	18	2	0	14	7	0
									Fluid	19	1	0	16	5	0

^aFigures in parentheses are the numbers of patients with definite tuberculosis (diagnosis confirmed by microbiological and/or histopathological investigations). TPE, tuberculous pleural effusion.

^bFigures are means ± SD or medians with corresponding interquartile ranges.

^cM, male; F, female.

^dIGRA, interferon gamma release assay; QFT, QuantiFERON; QFT-G, QuantiFERON TB Gold; QFT-GIT, QuantiFERON TB Gold (in-tube).

^eTP, true positive; FN, false negative; I, indeterminate; TN, true negative; FP, false positive.

^fNR, not reported in study.

Ten studies examined both blood and pleural fluid samples. Of these, pleural fluid data were not further analyzed in one study describing only nine TPE patients (32). Among the 13 studies providing information on HIV status, a substantial proportion of TPE patients were HIV seropositive in only 3 (Table 1) (29, 33, 34). Nine studies used T-SPOT.TB assays, while 14 used various generations of QuantiFERON assays (Table 1). Two studies reported results with both assays, and each assay was evaluated independently (24, 29). Hence, there were 20 evaluations of whole-blood assays (from 18 studies) involving 597 TPE patients (inclusive of 41 indeterminate results) and 488 patients with other effusions (excluding 36 indeterminate results). In addition, there were 16 evaluations of pleural fluid assays (from 14 studies) involving 516 TPE patients (inclusive of 78 indeterminate results) and 416 patients with other effusions (excluding 36 indeterminate results).

Quality of studies included.

All studies prospectively enrolled patients suspected to have TB. In five studies, investigators were blinded to the clinical details at the time of the laboratory assay (18, 29, 33–36). Investigators were aware of the clinical details in one study, whereas other studies did not report on blinding (32). Manufacturer-specified thresholds were used to categorize results in all but five studies. These five studies either used a modified threshold or derived one from receiver operating characteristic curves and hence did not report indeterminate results (20, 22, 28, 34, 36). Two studies excluded patients with indeterminate results prior to analysis (18, 27). All studies except one provided details on the reference standards used to diagnose TPE (30). In five studies, every TPE patient had definite TB (17, 21, 22, 29, 36). In five other studies, >75% TPE patients had a similar definite diagnosis (Table 1) (19, 24, 27, 31, 35).

The overall study quality, assessed by the QUADAS-2 tool, showed a low risk of bias for one study only (Fig. 2) (35). The risk of bias for the index test domain largely resulted from a lack of information on blinding in 13 studies. Four studies also showed a high risk of applicability concerns (Fig. 2).

FIG 2.

Summary of methodological quality of studies according to the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) tool. See references 17 to 37 for details.

IGRAs on whole-blood samples.

The sensitivities varied from 0.43 to 0.93 (Fig. 3), with a pooled estimate of 0.77 (95% CI, 0.71 to 0.83). The specificities varied from 0.50 to 1.00, with a pooled estimate of 0.71 (95% CI, 0.65 to 0.76). Pooled estimates for PLR, NLR, and DOR were 2.68 (95% CI, 2.15 to 3.33), 0.32 (95% CI, 0.24 to 0.43), and 8.39 (95% CI, 5.21 to 13.51), respectively. Figure 4 shows the corresponding HSROC curve. The summary sensitivity/specificity point is away from the desired upper left corner, and the location and shape of curve suggest that the test was a suboptimal discriminator.

FIG 3.

Forest plot of studies evaluating sensitivity and specificity of interferon gamma release assays for the diagnosis of tuberculous pleural effusion using whole-blood (top panel) or pleural fluid (bottom panel) samples. Open and filled squares represent sensitivity/specificity estimates from studies done using T.SPOT.TB and various generations of QuantiFERON assays, respectively. See references 17 to 37 for details.

FIG 4.

Hierarchical summary receiver operating characteristic (HSROC) curves summarizing the performance of the interferon gamma release assays for diagnosis of tuberculous pleural effusion using whole-blood (top) and pleural fluid (bottom) samples. Each individual study is represented by an open circle, whose size is proportional to the inverse standard error of sensitivity and specificity. The filled square represents the summary estimate of the test accuracy, with the surrounding dashed zone outline denoting the 95% confidence region around this estimate.

There was moderate heterogeneity between studies (I² of 66.43% and 40.27%, respectively, for sensitivity and specificity). On subgroup analysis, the studies using T-SPOT.TB, those having >75% of patients with definite TB, and those in which indeterminate results were either not reported or excluded were associated with lesser heterogeneity in sensitivity (Table 2). Similarly, smaller studies, those having >75% of TPE patients with definite TB and those from low-prevalence countries were associated with lesser heterogeneity in specificity (Table 2). However, the magnitude of change between the summary estimates of sensitivity and specificity in each subgroup was small. Multivariate meta-regression revealed that no factor significantly influenced heterogeneity (Table 3).

TABLE 2.

Subgroup analysis for exploration of factors influencing heterogeneity

Parameter	Category (no. of studies)a	Pooled sensitivity (95% CIb)	I2 (%)	Pooled specificity (95% CI)	I2 (%)	Pooled diagnostic odds ratio (95% CI)
Whole-blood assays
Country	High TB prevalence (13)	0.80 (0.73–0.86)	67.31	0.71 (0.63–0.78)	46.73	10.15 (5.31–19.40)
	Low TB prevalence (7)	0.70 (0.59–0.79)	51.20	0.69 (0.61–0.76)	28.46	5.07 (2.74–9.35)
Definite TBc	>75% of TPE patients (11)	0.70 (0.63–0.77)	45.99	0.68 (0.62–0.73)	29.11	4.96 (3.19–7.72)
	<75% of TPE patients (9)	0.84 (0.77–0.90)	64.01	0.75 (0.65–0.82)	48.11	15.81 (7.54–33.17)
Type of assay	QuantiFERON (14)	0.74 (0.66–0.81)	68.26	0.72 (0.65–0.78)	43.32	7.40 (3.99–13.73)
	T-SPOT.TB (6)	0.83 (0.75–0.90)	38.12	0.71 (0.59–0.80)	44.02	12.18 (6.82–21.77)
Total study sample	>50 patients (9)	0.80 (0.69–0.87)	41.09	0.70 (0.62–0.77)	50.41	9.12 (4.11–20.24)
	≤50 patients (11)	0.74 (0.66–0.80)	29.12	0.74 (0.65–0.81)	35.38	7.99 (4.74–13.47)
Indeterminate results	Specifically reported (15)	0.76 (0.67–0.82)	71.67	0.69 (0.62–0.76)	41.12	7.07 (4.04–12.37)
	Excluded/not reported (5)	0.82 (0.72–0.88)	28.43	0.75 (0.66–0.83)	34.49	13.49 (6.09–29.89)
Pleural fluid assays
Country	High TB prevalence (10)	0.74 (0.55–0.87)	95.90	0.76 (0.63–0.86)	46.61	5.43 (1.20–24.62)
	Low TB prevalence (6)	0.77 (0.52–0.91)	86.87	0.86 (0.67–0.95)	80.87	20.45 (4.24–98.57)
Definite TB	>75% of TPE patients (10)	0.74 (0.54–0.88)	83.07	0.74 (0.57–0.86)	83.47	6.43 (1.13–36.49)
	<75% of TPE patients (6)	0.78 (0.54–0.91)	90.38	0.87 (0.80–0.92)	0.00	21.93 (5.37–89.44)
Type of assay	QuantiFERON (8)	0.60 (0.40–0.77)	82.02	0.83 (0.66–0.92)	77.12	7.15 (1.69–30.28)
	T-SPOT.TB (8)	0.85 (0.72–0.92)	89.08	0.78 (0.64–0.88)	84.29	13.65 (2.66–82.33)
Total study sample	>50 patients (9)	0.78 (0.60–0.90)	90.48	0.85 (0.74–0.92)	45.39	14.67 (2.58–83.43)
	≤50 patients (7)	0.72 (0.46–0.89)	87.05	0.68 (0.50–0.81)	68.01	5.46 (1.24–24.09)
Indeterminate results	Specifically reported (10)	0.66 (0.47–0.80)	87.38	0.71 (0.56–0.83)	76.72	3.17 (1.19–8.42)
	Excluded/not reported (6)	0.87 (0.70–0.95)	85.71	0.88 (0.81–0.93)	43.95	56.30 (9.43–336.06)

^aTB, tuberculosis; TPE, tuberculous pleural effusion.

^bCI, confidence interval.

^cDefinite TB: microbiological or pathological confirmation of the diagnosis.

TABLE 3.

Multivariate meta-regression to evaluate factors associated with interferon gamma release assay positivity in tuberculous pleural effusion

Covariate	t	P>|t|	RDOR (95% CI)a
Whole-blood assays
High-TBb-prevalence country	0.82	0.43	1.55 (0.49–4.92)
Microbiological/pathological confirmation in >75% patients	−2.13	0.05	0.34 (0.12–1.01)
QuantiFERON assay (vs T-SPOT.TB)	−0.79	0.44	0.63 (0.18–2.18)
Total sample size of >50 patients	−0.03	0.98	0.99 (0.33–2.91)
Indeterminate results excluded/not reported	0.96	0.35	1.72 (0.51–5.78)
Pleural fluid assays
High-TB-prevalence country	−1.48	0.17	0.14 (0.00–2.70)
Microbiological/pathological confirmation in >75% patients	0.41	0.69	1.64 (0.11–24.34)
QuantiFERON assay (vs T-SPOT.TB)	−1.50	0.17	0.18 (0.01–2.28)
Total sample size of >50 patients	1.37	0.20	4.56 (0.39–53.91)
Indeterminate results excluded/not reported	1.80	0.10	8.24 (0.60–523.17)

^aCI, confidence interval; RDOR, relative diagnostic odds ratio.

^bTB, tuberculosis.

A Deeks' funnel plot asymmetry test revealed slope coefficients of −6.31 (P = 0.471) for the studies included. This suggested data symmetry and hence a lack of publication bias (Fig. 5).

FIG 5.

A Deeks' funnel plot assessment test for evaluation of the potential publication bias for interferon gamma release assays on whole-blood (left) and pleural fluid (right) samples. The plot for whole blood shows the symmetric distribution of the log of diagnostic odds ratios against the inverse root of effective sample sizes (ESS), indicating the absence of any publication bias. However, the plot for pleural fluid is suggestive of publication bias.

IGRAs on pleural fluid samples.

The sensitivities varied from 0.19 to 0.96 (Fig. 3), with a pooled estimate of 0.75 (95% CI, 0.60 to 0.86). The specificities varied from 0.27 to 1.00, with a pooled estimate of 0.79 (95% CI, 0.69 to 0.87). The pooled estimates for PLR, NLR, and DOR were 3.65 (95% CI, 2.12 to 6.28), 0.31 (95% CI, 0.17 to 0.56), and 11.74 (95% CI, 4.01 to 34.39), respectively. The corresponding HSROC curve (Fig. 4) showed a wider scatter of individual study points, and the summary diagnostic estimates were largely similar to those for whole-blood assays.

There was considerable heterogeneity between the studies (I² of 88.9% and 79.6%, respectively, for sensitivity and specificity). On subgroup analysis, none of the parameters studied accounted for heterogeneity in sensitivity. When studies in which >75% of TPE patients had definite TB were removed, the remaining six studies showed homogeneity (I² of 0%) in the specificity estimates (Table 2). Multivariate meta-regression revealed that no factor significantly influenced heterogeneity (Table 3).

A Deeks' funnel plot asymmetry test revealed slope coefficients of −35.64 (P = 0.0.039) for the studies included. This suggested data asymmetry and hence a publication bias (Fig. 5).

DISCUSSION

The results of this meta-analysis, similar to those of a previous analysis, suggest a limited role for whole blood or pleural fluid IGRA in the diagnosis of TPE. However, our meta-analysis is an improvement over the previous one in several aspects (9). Our search strategy was broader as we queried two databases without linguistic restrictions. We identified 15 additional studies for review, including two publications overlooked by the previous meta-analysis (32, 35). Compared to the earlier meta-analysis, we summarized blood IGRA data for more than three times as many patients and pleural fluid IGRA data for more than twice as many patients. We also categorized indeterminate results in TPE patients as false negatives. This reflects the real-life clinical decision-making scenario, where any “nonpositive” report is indicative of the absence of disease, more appropriately. This approach was adopted by a recent study, as well as by previous meta-analyses (6, 12, 19). The inclusion of indeterminate results in assessing sensitivity and the large sample size are major strengths of this analysis.

Pooled sensitivity and specificity estimates were rather low for both blood and pleural fluid IGRAs. These values suggest that the IGRA is a poor discriminator as a single test among patients suspected to have TPE. Based on these pooled estimates, a Bayesian analysis suggests that the increment in the probability of having TPE in a given patient is small if the IGRA result for either blood or pleural fluid samples is positive (Fig. 6). Similarly, negative results do not reduce the probability of TPE enough to use the IGRA as a “rule-out”' investigation (Fig. 6). In fact, the diagnostic accuracy of the IGRA is even inferior to that of pleural fluid adenosine deaminase, a test commonly employed for diagnosing TPE (38). Several factors may explain this poor performance. False positives due to latent TB in non-TPE patients may have compromised specificity, especially in patients from high-burden settings. While a diagnostic cutoff is available for whole-blood assays, a similar threshold is not defined for pleural fluid. The variable dilutions of pleural fluid T cells might have resulted in more negative or indeterminate results with the QuantiFERON assays. High background pleural fluid interferon gamma levels in TPE patients may also contribute to indeterminate or false-negative results. Overall, the results of this meta-analysis mirror the findings of poor diagnostic accuracy of IGRAs in the diagnosis of other forms of active extrapulmonary tuberculosis (6–8). In fact, the World Health Organization strongly advises against using an IGRA as a diagnostic marker for active tuberculosis in low- and middle-income countries (39). However, experts continue to advocate the use of IGRA as an adjunct to other standard investigations in patients with suspicious, difficult-to-diagnose tuberculosis in low-burden countries (40).

FIG 6.

Bayesian conditional probability plots for interferon gamma release assays (IGRAs) on whole-blood (solid curves) and pleural fluid (dashed curves) samples. The curves depict the estimated posttest probability of tuberculous pleural effusion in a patient, given a pretest probability of disease and a positive or negative IGRA result, using the pooled estimates of IGRA sensitivity and specificity derived during the meta-analysis.

Our meta-analysis has certain limitations. The quality of several studies was poor, mainly due to the lack of a description of blinding. Several studies included small numbers of patients, although we tried to overcome this by including only those studies evaluating at least 10 TPE patients. Most studies also used rather imperfect reference standards for diagnosing TPE, although it is not always possible to have microbiological/pathological confirmation in a routine clinical setting. There was substantial heterogeneity between the studies included. We investigated the causes of the statistical heterogeneity through a stratified analysis; however, no potential factor contributed significantly to heterogeneity on the multivariate meta-regression modeling. We studied the performance of the IGRA as a standalone test and cannot comment on its additive value when combined with other laboratory investigations.

In conclusion, the results of this meta-analysis suggest that commercial IGRAs (whole blood or pleural fluid) have poor diagnostic accuracy in patients with suspected TPE. We strongly suggest that an IGRA should not be used as a biomarker for diagnosing TPE.