A systematic review was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guidelines.12 The PubMed/MEDLINE and EMBASE electronic databases were consulted from their inception until July 2017. The search strategy used the following terms: (“pleural tuberculosis” OR “tuberculous pleuritis” OR “tuberculous pleurisy” OR “pleural”) AND (“adenosine deaminase” OR “ADA”). All the references of the selected articles were also reviewed. Two investigators (R.M.P. and S.B.) evaluated the studies independently and any discrepancies were resolved by mutual agreement.

Studies carried out in the Spanish population were included with no language restrictions, but had to meet 2 requirements: (1) the diagnosis of TPE was confirmed by pleural biopsy or microbiological culture of Mycobacterium tuberculosis in sputum, PF or pleural biopsy; and (2) sufficient data were available to construct a 2×2 contingency table that could be used to calculate diagnostic efficacy. Studies with duplicated or overlapping cases were excluded; if this was the case, the article with the greater sample size was selected.

The following characteristics were collected from the selected articles: authors, date of publication, study location, study design, number of patients and demographic data, methods for diagnosing TPE, etiology of non-tuberculous effusions, technique used to detect ADA in PF (Giusti, manual kinetic methods or automated kinetic methods), and the data needed to build a 2×2 table.

Two investigators (R.M.P. and S.B.) independently assessed the quality of the studies using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool.13 This consists of 4 domains: (1) patient selection, (2) index test, (3) reference standard, and (4) flow and timing. Each domain is evaluated in terms of the risk of bias (low, high or uncertain), and the first 3 also for applicability.

All the meta-analyses published on the diagnostic accuracy of pleural ADA in TB were identified,6–11 and studies performed in non-Spanish populations were selected. The complete text of these studies was read, and finally only those in which the TPE diagnosis had been confirmed by pleural biopsy or microbiological culture from a biological sample were taken into consideration. The same data as for the Spanish population, listed above, were extracted from each of the selected articles.

Continuous variables were expressed as mean or median, and categorical variables as frequencies and percentages. Agreement between observers for the QUADAS-2 tool was determined using the non-weighted Cohen's kappa statistic; the result was considered good if it was 0.6 and excellent if higher than 0.8. Sensitivity, specificity, positive and negative likelihood ratio (LR), and diagnostic odds ratio (DOR) of pleural ADA, with corresponding 95% confidence intervals (95% CI), were calculated from a 2×2 table, using bivariate models of random effects. For the calculation of the LR, a correction of 0.5 was applied to all cells of the table if any of them contained the value 0. Due to the negative correlation between sensitivity and specificity, these parameters were analyzed simultaneously using a meta-regression of random effects, according to the sensitivity data and the rate of false positives obtained from the studies included.14 The summary ROC curve was estimated from this meta-regression. The meta-regression was adjusted for the techniques used to diagnose ADA in PF (Giusti, manual or automated kinetic method), and, if there were significant differences, the analysis was stratified for each one of the techniques separately. The same analytical process was followed adjusting for the ADA cut-off point used for the diagnosis of TB. Publication bias was estimated using a funnel plot for the measurements of positive and negative LR. Heterogeneity between studies was quantified using Higgins’ I2 statistic, estimated from the univariate meta-analysis based on the method of DerSimonian and Laird15 for 3 measures of the effect (positive LR, negative LR, and DOR). The level of statistical significance was set a 0.05. Calculations were performed using the R program (R-project; http://cran.r-project.org/web/packages/mada/index.html).

In total, 1666 articles were identified, and the complete text of 52 potentially relevant papers was read. Only 16 of these met the inclusion criteria5,16–30 (Fig. 1 and Table A.1). The selected studies comprised a total of 1172 patients with TPE and 2975 with effusion due to other causes, the most notable being malignant effusions (1248, 42%; of which 78 were lymphomas) and parapneumonic effusions (539, 18%; of which 74 were empyemas). Clinical characteristics of the study population are shown in Table 1. Several different techniques used to quantify ADA: 6 studies used the Giusti method,5,17–21 5 used manual kinetic techniques,22–24,26,27 and 4 used automated kinetic techniques.25,28–30 The ADA cut-off point for the diagnosis of TPE ranged between 2323 and 45U/l,5,16,24 although the most widely used was 40U/l.20,26,27,29,30 Age and sex of patients were not specified in 817,18,21–23,25–27 and 9 studies,17,21–28 respectively; and race was not specified in any.5,16–30 The QUADAS-2 tool did not detect a high probability of bias in any of its domains, although the patient selection domain was considered “uncertain” in 8 studies.16,18–24 Interobserver concordance for the evaluation of the QUADAS-2 domains was 0.9 (95% CI: 0.78–1).

Fig. 1.

Flow chart for the search strategy and selection of Spanish studies.

(0.34MB).

Overall, pleural ADA had a sensitivity of 93% (95% CI: 88%–96%), specificity 92% (95% CI: 90%–94%) (Fig. 2), positive LR of 12 (95% CI: 9–16, negative LR of 0.08 (95% CI: 0.05–0.13), DOR of 156 (95% CI: 80–275) and area under the summary ROC curve of 0.968 (Fig. 3) for identifying TPE. There were no significant differences in sensitivity and/or specificity between different techniques for measuring ADA. Thus, the sensitivity for the Giusti and manual and automated kinetic methods was 95%, 91%, and 90% (P=0.25); and specificity was 91%, 94%, and 92%, respectively (P=0.31) (Table 2). Nor were differences detected in sensitivity or specificity between the different ADA cut-off points, when these were grouped in 3 ranges: (1) ADA 23–35U/l19,22,23,25,28 (sensitivity 92%, specificity 90%); (2) ADA 36–42U/l20,26,27,29,30 (sensitivity 88%, specificity 94%); and (3) ADA 43–45U/l5,16–18,21,24 (sensitivity 95%, specificity 92%; P=0.54 for sensitivity and P=0.1 for specificity).

Fig. 2.

Forest plot of adenosine deaminase sensitivity (A) and specificity (B) for the diagnosis of tuberculous pleural effusion. The point estimate of sensitivity and specificity of each study is shown as a solid square on a line that represents the confidence interval. The continuous vertical line represents the weighted average sensitivity (A) and specificity (B) and dotted vertical lines represent the confidence interval.

(0.42MB).

Fig. 3.

Summary receiver operating characteristics curve of pleural adenosine deaminase. Each circle represents a meta-analysis study and the asterisk represents the global estimate.

(0.06MB).

Six meta-analyses were identified,6–11 comprising a total of 195 studies, of which 122 were excluded for the following reasons: 23 because they were performed in a Spanish population (in our meta-analysis, 5 studies were excluded because of duplication of patients and 2 because the results of PF and ascites were combined); 66 due to overlap between the various meta-analyses; 20 due to lack of histological or microbiological confirmation of TPE; 5 because the diagnostic method of the TPE was not specified; 3 that were conducted in the pediatric population; 2 because they were doctoral theses with inaccessible data; 2 because they included an inappropriate comparative group (probable TB); and 1 due to duplication or overlap of patients (Fig. 4 and Table A.1). In this way, 73 articles were selected (Table A.3), including a total of 2789 patients with TPE and 3756 with pleural effusions due to other causes. Among the latter were 63 (1.6%) lymphomas and 614 (16%) parapneumonic effusions, of which 206 were empyemas. The Giusti method was used to quantify ADA in 49 studies, manual kinetic methods in 5, and automated kinetic methods in 14. The technique was not specified in another 5 studies.

Fig. 4.

Flow chart for the search strategy and selection of non-Spanish studies.

(0.23MB).

Overall, in the non-Spanish studies, pleural ADA showed a sensitivity of 88% (95% CI: 86%–90%), specificity of 88% (95% CI: 86%–90%), positive LR of 7 (95% CI: 6–9), negative LR of 0.14 (95% CI: 0.11–0.16), DOR of 54 (95% CI: 38–73), and area under the summary ROC curve of 0.937 for the diagnosis of pleural TB.

There was a non-significant trend toward greater sensitivity (93% vs 88%, P=0.06) and specificity (92% vs 88%, P=0.08) for pleural ADA in the Spanish studies compared to the non-Spanish studies. Sensitivity of the Giusti method (95% vs 89%, P=0.05) and specificity of the automated kinetic techniques (92% vs 86%, P<0.01) were also higher in Spanish studies than in non-Spanish studies. No significant differences were observed for the other techniques.

With regard to the aforementioned ranges of ADA cut-off points, the only factor distinguishing the Spanish and non-Spanish studies was the greater sensitivity of the Spanish studies for the ADA range of 43–45U/l (95% vs 87%, P=0.04), and the greater specificity for ADA values ≥36U/l (ADA 36–42U/l, 95% vs 86%, P=0.04; ADA 43–45U/l, 92% vs 87%; P=0.03, respectively) (Table 3).

With regard to the etiology of non-tuberculous effusions, more lymphomas were found in the Spanish studies (78/2975, 2.7% vs 63/3756, 1.7%; P=0.01), and more empyemas in the non-Spanish studies (206/3756, 5.5% vs 75/2975, 2.5%; P<0.01). The percentage of lymphomas and empyemas taken together was higher in the non-Spanish studies (7.2% vs 5.2%, P<0.01).

The asymmetric funnel test for LR showed no significant publication bias in the Spanish studies (P=0.46 for positive LR; P=0.62 for negative LR), or when the ADA measurement techniques or their different cut-off points were considered individually (all P>0.1). There was no significant heterogeneity in the set of Spanish studies (I2=20% for positive LR; I2<13% for negative LR and DOR), or for those conducted using the Giusti (I2=0%), manual kinetic (I2=0%) or automated kinetic methods (I2=0.2%), or when ADA cutoff points of <36U/l (I2<17.1%), between 36 and 42U/l (I2=0%) or > 42U/l (I2<17.3%) were used.