Unstimulated Pleural Fluid Interferon Gamma for Diagnosis of Tuberculous Pleural Effusion: a Systematic Review and Meta-analysis

Unstimulated interferon gamma may be a useful pleural fluid biomarker in the diagnosis of tuberculous pleural effusion (TPE). However, the exact threshold of pleural fluid interferon gamma and its accuracy during routine clinical decision-making is not clear.

ABSTRACT

Unstimulated interferon gamma may be a useful pleural fluid biomarker in the diagnosis of tuberculous pleural effusion (TPE). However, the exact threshold of pleural fluid interferon gamma and its accuracy during routine clinical decision-making is not clear. We assessed the performance of pleural fluid interferon gamma in diagnosing TPE and tried to identify a useful assay threshold. We queried the PubMed and Embase databases for publications indexed until May 2020 that provided both sensitivity and specificity data on unstimulated pleural fluid interferon gamma for diagnosis of TPE. A bivariate random effects model was employed to compute summary estimates for diagnostic accuracy parameters, both overall as well as at threshold ranges of <2, 2 to 5, and >5 IU/ml. We retrieved 2,048 citations, of which 67 publications (7,153 patients) were assessed in our review. The summary estimates for sensitivity, specificity, and diagnostic odds ratio were 0.93 (95% confidence interval [CI], 0.91 to 0.95), 0.96 (95% CI, 0.94 to 0.97), and 310.72 (95% CI, 185.24 to 521.18), respectively. Increasing interferon gamma thresholds did not translate into any substantial change in diagnostic performance; however, eight studies using thresholds of >5 IU/ml showed poorer diagnostic accuracy estimates than other studies with lower thresholds. None of the prespecified subgroup variables significantly influenced relative diagnostic odds ratios in a multivariate meta-regression model. All publications demonstrated a high risk of bias. Unstimulated pleural fluid interferon gamma level provides excellent accuracy for diagnosing TPE and has the potential of becoming a first-line test for this purpose.

INTRODUCTION

In several developing nations, tuberculous pleural effusion (TPE) is the most frequent etiology for lymphocytic exudative pleural effusions and one of the more common manifestations of extrapulmonary tuberculosis (TB) (1). Although pleural fluid is a simple and relatively convenient sample for the diagnostic evaluation of suspected TPE, microbiologic confirmation remains uncommon despite the availability of newer molecular tools (2). Even though a novel cartridge-based nucleic amplification test (GeneXpert) has improved diagnostic yield in paucibacillary clinical specimens, its performance in pleural fluid remains suboptimal. A recent meta-analysis reported that GeneXpert can correctly identify 51.4% of culture-proven TPE and only 22.7% of TPE diagnosed through less stringent composite clinical criteria (3). Several pleural fluid biomarkers have been suggested to improve TPE diagnosis. Pleural fluid adenosine deaminase (ADA) has been the most extensively studied among these. Pleural fluid ADA levels are considerably higher among patients with TPE than other effusions, and a recent meta-analysis has confirmed a relatively good overall diagnostic accuracy (4).

The entry of mycobacterial antigen into the pleural space eventually triggers a CD4⁺ T-cell-mediated delayed hypersensitivity reaction. These activated lymphocytes get compartmentalized within the pleural cavity and release interferon gamma, interleukin-12, and other related cytokines that mediate and propagate the strong T-helper type 1 (Th1) response aimed at granuloma formation and containment of mycobacterial infection (5, 6). Interferon gamma release assays measure interferon gamma released from T-lymphocytes secondary to stimulation by specific antigens. Initially believed to be a good immunodiagnostic surrogate to identify TB, the diagnostic performance and accuracy of commercially available interferon gamma release assays is rather poor in diagnosing TPE (7). The levels of unstimulated pleural fluid interferon gamma have also been noted to be higher in TPE patients. Two previous meta-analyses have summarized the diagnostic characteristics of pleural fluid interferon gamma. The first, conducted on 13 studies published until 2000, yielded a summary sensitivity and specificity estimate of 0.96, which was marginally better than that noted for ADA (0.93) (8). The second reviewed 22 studies published until 2006 and reported summary sensitivity of 0.89 and specificity of 0.97 (9). Several additional studies have been published in this field since then. Moreover, previous meta-analyses have provided no clarity on optimal methodology or a clinically useful diagnostic threshold for identifying TPE. Both the need to clarify these issues and the availability of many recent publications underscore the need for conducting a fresh analysis (10). Here, we assess the diagnostic performance of unstimulated pleural fluid interferon gamma for diagnosing TPE in patients of pleural effusion through an updated systematic review and meta-analysis.

METHODS

Our review was performed per guidelines from the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (11, 12). Since our methodology involved only a systematic review and meta-analysis of already indexed publications, an approval from our Institutional Ethics Committee was not necessary. The protocol for this systematic review was registered with the PROSPERO database (CRD42017058843).

We searched Embase and PubMed databases since inception for articles indexed up to 31 May 2020 using the following search terms: (Tuberculous OR Tubercular OR Tuberculosis OR Mycobacterium OR Mycobacterial OR TB) AND (Pleural OR Pleura OR Pleurisy OR Pleuritis OR Nonrespiratory OR Nonrespiratory OR Extrapulmonary OR Extrapulmonary) AND (Interferon OR Gamma-interferon OR Interferon gamma OR IFN). We also identified additional studies from the bibliography of selected articles as well as our personal records. No linguistic restrictions were applied.

We screened the titles and abstracts of citations obtained during our literature search for published articles potentially suitable for a more detailed evaluation. We excluded animal experiments, articles not primarily related to TPE, studies not assessing pleural fluid interferon or assessing only interferon gamma release assays, case reports, conference abstracts, editorials and reviews, and letters to editor not presenting original data on diagnostic accuracy. Two reviewers (A. N. Aggarwal and R. Agarwal) independently read the full texts of all remaining studies to identify those appropriate for further analysis. Any disagreements were settled by consensus. Finally, we identified original studies providing numerical information regarding both sensitivity and specificity of unstimulated pleural fluid interferon gamma for diagnosing TPE as well as those where numerical or graphical data were available to compute both measures. We also excluded biochemical, experimental, and descriptive studies. Among studies presenting results on overlapping data sets of subjects, we selected only the publication reporting the largest patient sample. Wherever information was reported on two or more diagnostic thresholds, we picked the one associated with the largest sum of specificity and sensitivity. We selected studies that diagnosed TPE using either one or more of the following diagnostic standards: (i) microbiologic (presence of acid-fast bacilli or positivity for M. tuberculosis on culture or nucleic acid amplification tests, in pleural fluid, pleural biopsy, or any other clinical specimen), (ii) histopathologic (pleural biopsy specimen demonstrating granulomatous inflammation), and (iii) clinical (adequate resolution of effusion after empirical antitubercular therapy). We adopted this composite definition of reference standard, since a single well-performing investigation is not available for confirmation of TPE; this leads to enrollment of many subjects with only a probable diagnosis of TPE in a research setting.

Two assessors (A. N. Aggarwal and R. Agarwal) extracted data independently from all chosen articles with respect to study design, year of publication, countries where studies were carried out, etiology of nontuberculous pleural effusions, HIV status, technique of interferon gamma assay and its threshold, the proportion of patients having confirmed diagnosis of TPE using microbiologic or pathological criteria (referred to here as having definite TB), and the number of positive and negative assay results for each category of subjects. The QUADAS-2 (quality assessment for studies of diagnostic accuracy) tool was utilized for describing and summarizing the quality of each study (13).

We calculated the sensitivity and specificity figures for each study together with their 95% confidence intervals (95% CI). We decided to pool our data and compute summary sensitivity and specificity estimates through a bivariate random effects model, as we expected significant heterogeneity and variability at study level. We constructed an HSROC (hierarchical summary receiver operating characteristic) plot for summarizing overall test performance and accuracy and to graphically identify the differences in diagnostic accuracy across the included studies (14). We also computed summary positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) estimates. In general, an NLR below 0.1 and PLR exceeding 10 indicates the diagnostic test’s ability to exclude or confirm, respectively, the disorder being evaluated (15).

Heterogeneity was expressed using the I² statistic. We judged heterogeneity to be high if the I² value exceeded 0.75 (16). We investigated potential causes of heterogeneity by using predefined covariates to stratify our data. These subgroups included the year of publication (up to 2010 or later), prospective study design or otherwise, national TB burden, prevalence of TB among all study subjects (>0.5 or less), study sample size (total of >100 patients or lesser), robustness of the TPE reference standard (composite clinical criteria or definite TB), nature of nontuberculous pleural effusions (whether transudative pleural fluids are included or not), method of assaying interferon gamma (enzyme-linked immunosorbent assay [ELISA], radioimmunoassay, or others), diagnostic threshold of assay (below 2 IU/ml, 2 to 5 IU/ml, or above 5 IU/ml), and blinding in the study. The 30 countries with high burden of TB were identified as currently suggested by the World Health Organization (17). Interferon gamma threshold subgroups were arbitrarily categorized to approximately reflect the 25th and 75th percentiles of assay thresholds from the studies evaluated in a previous meta-analysis (9). Different units for reporting test threshold were adjusted by considering equivalence between 1 IU and 50 pg of interferon gamma (National Institute for Biological Standards and Control, UK) (18). Restricted maximum likelihood estimation analysis was used to perform metaregression to additionally evaluate heterogeneity using a natural logarithm of the DOR as the independent variable. Deek’s funnel plot was drawn to graphically review publication bias, with a slope coefficient with a P value of <0.10 indicative of substantial asymmetry suggesting bias (19). The quality of evidence (confidence in the estimated effects) regarding diagnostic accuracy (both overall and for different threshold categories) was assessed and summarized through the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach (20).

Data analysis was performed through the statistical package Stata (Intercooled version 12.0; Stata Corp., College Station, TX). We added 0.5 to all cells having zero values before performing any computations involving logarithmic or logit transformations.

RESULTS

We identified 2,046 publications through our search of the two electronic databases. We also included two more studies retrieved through other sources. Following an initial review of titles and abstracts, we assessed 151 full-text articles in detail (Fig. 1). Sixty-seven publications were finally selected for data synthesis (see Table S1 in the supplemental material). All except eight (11.9%) studies were published in the English language. The details regarding individual studies are summarized in Tables S2 and S3. These 67 publications provide data on 2,657 TPE patients and 4,496 patients having pleural effusion from another etiology, with an overall TPE prevalence of 37.1%. Thirty-one (46.3%) studies were reported from countries currently designated high burden by the WHO. All except six (9.0%) publications had a prospective study design. Only 11 (16.4%) studies ensured blinding. Only 20 (29.9%) studies provided information on inclusion/exclusion of HIV-seropositive subjects. Less than half of all studies (32, 47.8%) included patients exclusively with exudative pleural effusion. Thirty-seven (55.2%) studies utilized a definite TB reference standard for diagnosing TPE, while others (30; 44.8%) employed various composite clinical criteria. Most studies (56; 83.6%) used ELISA to estimate interferon gamma levels; however, the diagnostic thresholds differed widely from 0.01 to 23.41 IU/ml (Table S3). Six (9.0%) studies reported diagnostic accuracy data at more than one diagnostic threshold. Overall, we evaluated 30 (44.8%) studies that used an interferon gamma diagnostic threshold of <2 IU/ml or equivalent and another 29 (43.3%) using 2 to 5 IU/ml or equivalent as the cutoff. Only eight (11.9%) publications included results from interferon gamma thresholds exceeding 5 IU/ml.

FIG 1.

Study selection process.

We observed a high risk of bias for all of the included publications on performing assessment of study quality using the QUADAS-2 tool (Fig. S1). The bias primarily resulted in the index test domain, as investigators did not use prespecified thresholds for interpreting interferon gamma assay results and/or failed to provide information on study blinding. Sixteen (23.9%) studies additionally had applicability concerns in the patient selection domain.

The diagnostic accuracy estimates derived from different publications are provided in Table S4. The sensitivity varied between 0.61 and 1.00 in different studies (Fig. 2), with a summary sensitivity of 0.93 (95% CI, 0.91 to 0.95) (Table 1). The specificity ranged between 0.68 and 1.00, with summary specificity of 0.96 (95% CI, 0.94 to 0.97). The summary NLR and PLR estimates were 0.07 (95% CI, 0.06 to 0.10) and 22.80 (95% CI 16.61 to 31.29), respectively (Fig. S2). The summary DOR was 310.72 (95% CI, 185.24 to 521.18). On inspection of the HSROC curve, the point representing summary sensitivity/specificity was located near the preferred upper left corner of the graph (Fig. 3). Shape and location for the HSROC curve indicated that the unstimulated pleural fluid interferon gamma assay was overall a good discriminator.

FIG 2.

Coupled forest plot of 67 studies reporting on diagnostic accuracy of unstimulated pleural fluid interferon gamma for tuberculous pleural effusion. Individual study estimates are depicted by solid squares, and the horizontal lines correspond to their 95% confidence limits. Summary sensitivity and specificity estimates are indicated by vertical dashed lines.

TABLE 1.

Evaluation of factors affecting diagnostic accuracy of unstimulated pleural fluid interferon gamma assay^a

Factor	Group (no. of studies)	Summary sensitivity (95% CI)	I2, %	Summary specificity (95% CI)	I2, %
Overall	All studies (67)	0.93 (0.91–0.95)	79.24	0.96 (0.94–0.97)	79.25
Publication yr	Up to 2010 (38)	0.94 (0.91–0.96)	78.66	0.97 (0.95–0.98)	67.31
After 2010 (29)	0.91 (0.87–0.94)	78.21	0.94 (0.91–0.96)	78.62
Design of study	Prospective (61)	0.93 (0.91–0.95)	79.70	0.96 (0.94–0.97)	79.51
Not prospective or not specified (6)	0.93 (0.77–0.98)	78.04	0.98 (0.82–1.00)	80.49
Burden of TB in country	High (31)	0.91 (0.88–0.93)	71.17	0.95 (0.93–0.96)	59.84
Not high (36)	0.95 (0.92–0.97)	85.40	0.97 (0.95–0.98)	87.77
Prevalence of TB in the study population	>50% (24)	0.92 (0.87–0.95)	83.98	0.94 (0.90–0.97)	68.28
≤50% (43)	0.93 (0.91–0.95)	73.55	0.96 (0.95–0.98)	82.91
Effusion characteristics	Only exudates (33)	0.92 (0.88–0.94)	74.69	0.95 (0.93–0.97)	79.84
Transudates also or not specified (34)	0.94 (0.91–0.97)	82.93	0.96 (0.95–0.98)	77.94
Assay technique	ELISA (56)	0.93 (0.90–0.95)	78.76	0.96 (0.95–0.97)	79.18
Radioimmunoassay (6)	0.97 (0.90–0.99)	89.82	0.97 (0.90–0.99)	89.82
Other (5)	0.91 (0.86–0.94)	21.06	0.90 (0.87–0.93)	29.21
Study sample size	>100 subjects (18)	0.94 (0.91–0.96)	78.01	0.96 (0.93–0.97)	83.03
≤100 subjects (49)	0.92 (0.89–0.94)	78.15	0.96 (0.94–0.98)	79.97
Blinding in the study	Done (11)	0.96 (0.93–0.98)	45.85	0.97 (0.96–0.98)	45.51
Not done or not specified (56)	0.92 (0.89–0.94)	78.31	0.96 (0.94–0.97)	78.02
Reference standard	Definite (37)	0.92 (0.88–0.94)	74.42	0.96 (0.93–0.97)	81.02
Composite (30)	0.94 (0.91–0.96)	83.91	0.96 (0.94–0.98)	73.52
Diagnostic threshold	<2 IU/ml or <100 pg/ml (30)	0.94 (0.89–0.96)	85.84	0.96 (0.94–0.98)	85.53
2–5 IU/ml or 100–250 pg/ml (29)	0.94 (0.92–0.95)	58.82	0.96 (0.94–0.98)	77.73
>5 IU/ml or >275 pg/ml (8)	0.86 (0.76–0.92)	69.31	0.93 (0.88–0.96)	28.98

^a95% CI, 95% confidence intervals; ELISA, enzyme-linked immunosorbent assay; I², heterogeneity statistic; TB, tuberculosis.

FIG 3.

Hierarchical summary receiver operating characteristic (HSROC) plot to summarize diagnostic accuracy for unstimulated pleural fluid interferon gamma in diagnosing tuberculous pleural effusion (blue curve). Each open circle represents an individual study, with circle size proportionate to inverse standard error of sensitivity and specificity. Summary estimate of diagnostic accuracy is indicated by the red square, and the surrounding red ellipse outlines the zone of 95% confidence around this estimate.

We observed significant heterogeneity between the included studies (I² of 79.24% for sensitivity and 79.25% for specificity). On subgroup analysis, we did not find any major difference in values of the summary values of sensitivity and specificity in any prespecified category (Table 1). There was no definite shift in overall diagnostic performance with higher/lower interferon gamma thresholds (Fig. S3). The summary estimates for diagnostic accuracy were largely similar for studies employing different diagnostic threshold categories (Table 1), with some suggestion of relatively lower sensitivity from studies using interferon gamma thresholds of >5 IU/ml. There was a moderate level of evidence for the benefit of unstimulated pleural fluid interferon gamma in diagnosis of TPE for all threshold ranges (Table 2). However, based on the pooled data from studies using thresholds above 5 IU/ml, the predictive value of a positive test was quite poor in a scenario of low pretest probability of TPE (Table 2).

TABLE 2.

Summary of findings from studies evaluating unstimulated pleural fluid interferon gamma for diagnosing tuberculous pleural effusion

^aMost studies had no blinding and did not use prespecified diagnostic thresholds.

^bCoE, certainty of evidence.

None of the prespecified subgroup variables significantly influenced relative DOR in a multivariate metaregression model (Table S5). The asymmetry test on Deeks’ funnel plot revealed a slope coefficient of −6.89 (P = 0.29), which suggested overall symmetry in data, indicative of the absence of any publication bias for studies included in this review (Fig. S4).

DISCUSSION

Our study results suggest that unstimulated pleural fluid interferon gamma has a high sensitivity (0.93) and specificity (0.96) in TPE diagnosis. Thus, pleural fluid interferon gamma seems to be an excellent discriminatory investigation for this purpose. The last meta-analysis on the subject had reviewed 22 articles published until 2006 (9). We have improved this by analyzing 45 additional publications (with 5,052 additional patients) available during the subsequent years as well as earlier publications not incorporated in the previous review. In addition, we looked into the key medical problem of selecting a suitable interferon gamma threshold for correctly identifying a patient of TPE.

Pleural fluid interferon gamma estimation is conducted by several different techniques. However, we did not find any major differences in the diagnostic performance using ELISA or radioimmunoassay, the two most common assay methods. A high cost and the arduous procedure needed to estimate pleural fluid interferon gamma seem to be the major constraints for expanding the use of this investigation in routine clinical practice. However, a commercially available rapid and inexpensive immunoassay was recently developed and evaluated, with good results (21). The diagnostic accuracy estimates in the current study are also marginally better than those reported for pleural fluid ADA by our group in a recent meta-analysis (summary sensitivity, 0.92; summary specificity, 0.90) (4). Overall, this suggests that unstimulated pleural fluid interferon gamma is as useful as, or even better than, currently preferred diagnostic biomarkers for TPE. We believe that this investigation has the potential to change our clinical practice and algorithms for diagnostic evaluation of pleural effusions.

Despite 3 decades of experience, an ideal pleural fluid interferon gamma threshold remains elusive for the diagnosis of TPE. Part of the problem stems from the use of different units to describe interferon gamma levels in terms of either concentration or activity. We attempted to solve this by conversion of concentration-based results to standard activity-based values before our final analysis. The studies included in this review employed discriminative thresholds ranging from 0.01 to 23.41 IU/ml. We arbitrarily grouped threshold ranges into three categories to reflect the threshold distribution of various studies included in an earlier meta-analysis. We observed that test positivity using a threshold of >5 IU/ml may not be useful in scenarios with a low pretest probability of TPE (Table 2). As there were very few such studies, we are unable to provide a more firm recommendation. In other situations, positive and negative tests may be much more helpful to confirm or rule out TPE across all threshold ranges with modest certainty of test accuracy evidence (Table 2).

What are the clinical implications of our study? While a wide range of thresholds have been used by individual investigators, our data indicate that patients with high pleural fluid interferon gamma levels, based on diagnostic thresholds below 5 IU/ml, are highly likely to have TPE. One may think about starting antituberculous therapy empirically in such patients, especially in locations with high TB prevalence. Patients with low pleural fluid interferon gamma based on such thresholds would require further detailed evaluation to identify a nontuberculous etiology. A pleural biopsy specimen is likely to prove helpful for further characterization in such a setting.

There are several limitations to our meta-analysis. We cannot rule out a misclassification bias, since only 30 of the 67 studies used definite criteria (microbiologic and/or histopathologic) for diagnosing TPE. There was significant heterogeneity between the various studies in our review, and multivariate metaregression modeling indicated substantial residual variation related to heterogeneity. Further, many of the included studies had several limitations. Several investigators included subjects having transudative pleural effusion, which might have falsely increased estimates for specificity, even though TPE is not a usual diagnostic consideration in such patients. Likewise, some investigators studied only, or predominantly, exudative lymphocytic effusions, even though these patients form the principal group where one would suspect TPE. Most studies also derived their diagnostic thresholds in a post hoc fashion by optimizing the trade-off between specificity and sensitivity, rather than using a predefined threshold that could be more appropriate clinically for confirmation (or for exclusion) of a diagnosis of TPE. We have reviewed the diagnostic accuracy of unstimulated pleural fluid interferon gamma as an isolated investigation. Hence, we are unable to clarify its additive contribution to everyday clinical decisions once findings from other diagnostic investigations are also available concurrently. Previous studies have shown that the predictive power of pleural fluid interferon gamma improves further after considering additional clinical and other laboratory data, especially pleural fluid ADA estimation (22 to 24).

In conclusion, observations from our meta-analysis indicate that unstimulated pleural fluid interferon gamma levels provide a high diagnostic accuracy, and carry the potential for becoming a first-line test, in the diagnosis of patients with suspected TPE. However, the studies included in our review demonstrated a risk of bias. Thus, larger, well-planned multicenter studies, with predefined clinically useful thresholds of pleural fluid interferon gamma, are required.