Janda and Swiston BMC Pulmonary Medicine 2010, 10:58

http://www.biomedcentral.com/1471-2466/10/58

RESEARCH ARTICLE Open Access

Diagnostic accuracy of pleural fluid NT-pro-BNP

for pleural effusions of cardiac origin: a systematic
review and meta-analysis
Surinder Janda, John Swiston*

Abstract
Background: Several studies have been published in the literature on the diagnostic accuracy of NT-pro-BNP for
pleural effusions from heart failure in the last decade. The purpose of our study was to perform a systematic
review and meta-analysis on the diagnostic accuracy of pleural fluid NT-pro-BNP for pleural effusions of cardiac
origin.
Methods: MEDLINE, EMBASE, PapersFirst, and the Cochrane collaboration and the Cochrane Register of controlled
trials were searched. All searches were inclusive as of March 2010. Studies were only included if the absolute
number of true-positive, false-negative, true-negative, and false-positive observations were available, and the
“reference standards” were described clearly. Two investigators independently reviewed articles and extracted data.
Quality was assessed with the Quality Assessment for Diagnostic Accuracy Studies (QUADAS). The bivariate model
for diagnostic meta-analysis was used to obtain a pooled sensitivity and a pooled specificity.
Results: Ten studies (total number of patients 1120) were included in the meta-analysis. The average pleural fluid
NT-pro-BNP level in effusions of cardiac origin was 6140 pg/mL. The pooled sensitivity and specificity of all studies
combined was 94% (95% CI: 90-97) and 94% (95% CI: 89-97) respectively. The pooled positive likelihood ratio was
15.2 (95% CI: 8.1-28.7) and the pooled negative likelihood ratio was 0.06 (95% CI: 0.03-0.11). The area under the
ROC curve was 0.98 (95% CI: 0.96-0.99) and the diagnostic odds ratio was 246 (95% CI: 81-745).
Conclusions: Pleural fluid NT-pro-BNP is a very useful biomarker with high diagnostic accuracy for distinguishing
pleural effusions of cardiac origin.

Background failure (CHF) were misclassified as exudative effusions

Pleural effusions arising from heart failure are usually using Light’s criteria[4]. A diagnostic dilemma often
discriminated from other causes based on clinical criteria arises when CHF-associated pleural effusions are misclas-
in association with biochemical analysis, particularly the sified as exudates which can then lead to the use of more
discrimination of transudates versus exudates, most com- expensive and sometimes invasive tests to diagnose the
monly using Light’s criteria [1]. The sensitivity of Light’s etiology of the effusion. A non-invasive and inexpensive
criteria for identifying exudative pleural effusions is very strategy to discriminate pleural effusions of a cardiac ori-
high (98%)[2], however the criteria’s ability to exclude gin would be beneficial in such circumstances.
transudative effusions is low[3]. As a result, heart failure Serum brain natriuretic peptide (BNP) or the amino-
associated pleural effusions can be misclassified as exu- terminal fragment N-terminal pro-brain natriuretic
dates using Light’s criteria, particularly after diuretics peptide (NT-pro-BNP) have an established role in the
have been used. One study showed that 28% (5 of 18) diagnosis, management, and prognosis of patients with
patients with pleural effusions from congestive heart CHF[5]. BNP, also known as B-type natriuretic peptide,
is a 32 amino acid polypeptide secreted by the ventricles
of the heart in response to excessive stretching of cardi-
* Correspondence: [email protected]
Division of Respiratory Medicine, The University of British Columbia,
omyocytes[6]. BNP is co-secreted along with a 76 amino
Vancouver, BC, Canada acid polypeptide, NT-pro-BNP, which is biologically
© 2010 Janda and Swiston; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 2 of 11
http://www.biomedcentral.com/1471-2466/10/58

inactive[7]. The half-life of BNP is approximately 20 NT-pro-BNP if available; and the reference standard
minutes whereas of NT-pro-BNP is 1-2 hours[8]. BNP used.
binds to atrial natriuretic factor receptors leading to a We used the bivariate model for diagnostic meta-
decrease in systemic vascular resistance and central analysis to obtain an overall sensitivity and an overall
venous pressure, and an increase in natriuresis[6]. specificity[11]. Instead of using the diagnostic odds
Several studies have been published in the literature ratio, as used in conventional diagnostic meta-analysis
on the diagnostic accuracy of NT-pro-BNP for pleural [12], the bivariate model uses pairs of sensitivity and
effusions from heart failure. The purpose of our study specificity as the starting point of the analysis. In addi-
was to perform a systematic review and meta-analysis tion to accounting for study size, the bivariate model
on the diagnostic accuracy of pleural fluid NT-pro-BNP estimates and incorporates the negative correlation that
for pleural effusions of cardiac origin. may arise between the sensitivity and specificity of the
index test within studies as a result of differences in test
Methods threshold between studies. The bivariate model uses a
The systematic review and meta-analysis was performed random effects approach for both sensitivity and specifi-
according to the recently published recommendations city, which allows for heterogeneity beyond chance as a
and checklist of the PRISMA statement[9]. result of clinical and methodological differences between
Searches were conducted on MEDLINE (inception-March studies. The pooled estimates of sensitivity and specifi-
2010); EMBASE (inception-March 2010), PapersFirst (incep- city were used to calculate the average positive and
tion-March 2010), and the Cochrane collaboration and the negative likelihood ratios. Publication bias through small
Cochrane Register of controlled trials for relevant studies. study effects was assessed with a regression test on the
The following key terms were used: ‘pleural fluid’ or ‘pleural diagnostic odds ratio[13].
effusion’ AND ‘brain natriuretic peptide’ or ‘B-type natriure- A receiver operating characteristic graph, with the
tic peptide’ or ‘BNP’, ‘pro-BNP’ or ‘NT-pro-BNP’ or ‘amino- y-axis representing the index test’s sensitivity (true posi-
terminal pro-BNP’ or ‘congestive heart failure’ or ‘heart fail- tive rate) and the x-axis representing 1-specificity (false
ure’ or ‘CHF’. All searches were limited to ‘humans’. We negative rate), was used to plot the individual and sum-
identified additional studies by searching the bibliographies mary points of sensitivity and specificity. Furthermore,
of retrieved articles. Two independent reviewers (SJ and JS) around the pooled estimates, we also plotted a 95% con-
performed the literature search. Studies relevant to the diag- fidence region and a 95% prediction region to illustrate
nostic value of NT-pro-BNP for pleural effusions of cardiac the precision with which the pooled values were esti-
origin were included if the following criteria were met: mated (confidence ellipse of a mean) and to show the
pleural fluid NT-pro-BNP was used for diagnosing pleural amount of between study variation (prediction ellipse;
effusions of cardiac origin; and a 2 × 2 contingency table the likely range of values for a new study). We used
could be formulated from the available data. Studies that Stata intercooled version 10.1 (StataCorp, College Sta-
used BNP or other biomarkers were excluded. tion, Texas), in particular the midas and metandi com-
All studies that appeared to fit the inclusion criteria mands, for all statistical analyses[14,15].
were identified for full review by two reviewers (SJ and
JS). Each reviewer independently selected studies for Results
inclusion in the review. Disagreement between the two Our search yielded 128 citations of which 117 were
extracting authors was resolved by consensus. excluded for various reasons based on the title and
The methodological quality of the selected studies was abstract (Figure 1). Eleven were then retrieved for full
graded independently by two reviewers (SJ and JS) with text review of which three were excluded because BNP
the Quality Assessment of Diagnostic Accuracy Studies instead of NT-pro-BNP was used as the biomarker (Fig-
(QUADAS) tool, a validated tool for the quality assess- ure 1). Ten studies were included in the final analysis
ment of diagnostic accuracy studies[10]. We performed [16-25]. One study, by Seyhan et al[22], was initially
component analysis using the QUADAS tool which was published online and then was retracted by the editors
depicted as a proportional bar graph for each of the 14 of the journal. The reason for retraction was due to a
individual criteria. Disagreement between the two violation of the journal’s Information for Authors. There-
extracting authors was resolved by consensus. fore, only the abstract was available for data extraction
The following variables were extracted from each and a quality assessment was not possible.
study: publication year; country of origin of study; study The studies were published from 2004 to 2010. The
design; patient demographics and co-morbidities; majority of the studies (7/9)[16-19,22-24] were done in
NT-pro-BNP assay type; numbers of true-positive, false- Europe whereas two were done in North America[20,25]
negative, true-negative, and false-positive observations; and one in Asia[21]. All studies except for two (Porcel
correlation statistic between pleural fluid and serum et el[23] and Long et al[25]) were of prospective design
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 3 of 11
http://www.biomedcentral.com/1471-2466/10/58

Figure 1 Flow chart of the studies included in the meta-analysis.

(Table 1). Furthermore, all studies except for two (Liao of non-cardiac origin). The control group (non-cardiac
et al[20] and Long et al[25]) used an electrochemical pleural effusions) consisted of mainly malignant pleural
luminescence immunoassay (ELCIA) performed on the effusions and infections (Table 1).
Elecsys 2010 analyzer (Roche) to measure levels of NT- As there is no gold standard for effusions of cardiac
pro-BNP whereas Liao et al[20] and Long et al[25] used origin, clinical criteria was used as the reference stan-
an enzyme-linked immunosorbent assay (ELISA) (Table dard. All studies used some combination of the follow-
1). A total of 1120 patients/pleural effusions were ing criteria: history, physical exam, chest x-ray,
included in this analysis (429 of cardiac origin and 691 electrocardiogram, echocardiogram, and response to

Table 1 Characteristics of the studies included in the meta-analysis.

Study Year Country Design Assay Type Reference Standard Control Group†
‡
Tomcsanyi et al [16] 2004 Hungary PC ECLIA(1) Clinical Criteria* 57% M; 14% I; 29% O
a, b, c, e
Porcel et al [17] 2004 Spain PC ECLIA(1) Clinical Criteria* NR
a, b, c, d, e, f
Kolditz et al [18] 2006 Germany PC ECLIA(1) Clinical Criteria* 59% M; 21% I; 30% O
a, b, c, e, f
Porcel et al [19] 2007 Spain PC ECLIA(1) Clinical Criteria* 48% M; 20% I; 32% O
a, b, c, d, e, f
Liao et al [20] 2008 USA PC ELISA Clinical Criteria* 25% M; 0% I; 75% O
a, c, e
Han et al [21] 2008 Korea PC ECLIA(1) Clinical Criteria* 24% M; 66% I; 10% O
a, b, e, f
Seyhan et al [22] 2009 Turkey PC ECLIA Clinical Criteria* N/A
N/A
Porcel et al [23] 2009 Spain RC ECLIA(2) Clinical Criteria* 44% M; 29% I; 27% O
a, b, c, e, f
Bayram et al [24] 2009 Turkey PC ECLIA(1) Clinical Criteria* 23% M; 31% I; 46% O
a, b, c, e, f
Long et al [25] 2010 USA RC ELISA Clinical Criteria* 25% M; 25% I; 25% O
a, b, c, e
* = heart failure diagnosis based on (a) history, or (b) physical exam, or (c) chest x-ray, or (d) electrocardiogram, or (e) echocardiogram, or (f) response to
diuretics; † = non-cardiac effusions; ‡ = generation of assay (one or two); ECLIA = electrochemical luminescence immunoassay; ELISA = enzyme-linked
immunosorbent assay; I = infectious effusion; M = malignant effusion; N/A = not available; NR = not recorded; O = other causes (i.e. connective tissue disease,
pleuritis, post-CABG, hepatic hydrothorax, pulmonary embolism, etc); PC = prospective cohort; RC = retrospective cohort;
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 4 of 11
http://www.biomedcentral.com/1471-2466/10/58

Figure 2 Proportion of all 14 Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool criteria that were fulfilled for seven of
eight studies included in the meta-analysis.

diuretics. Of the studies[18-20,23,25] that reported New fluid NT-pro-BNP levels, all five studies showed high corre-
York Heart Classification for the patients with effusions lation between these two parameters (Table 2). Table 2 also
of cardiac origin, all patients were either class III or IV. shows that the percentage of cardiac effusions misclassified
Overall, the quality of the reported studies was good as exudates by Light’s criteria was relatively modest (range
(Figure 2). Five studies[16,19,20,23,25] maybe subject to 7%-36%) within various studies. However, when these mis-
spectrum bias as their method of recruitment of patients classified effusions from Light’s criteria were analyzed by
consisted of recruiting a target group (patients with NT-pro-BNP levels, the absolute number of misclassified
pleural effusions of cardiac origin) and a control group effusions was significantly reduced (Table 2).
(patients with pleural effusions of non-cardiac origin) The pooled sensitivity and specificity of all studies
rather than applying the index and reference test to an combined was 94% (95% CI: 90-97) and 94% (95% CI:
unselected patient population with pleural effusions. 89-97) respectively (Figure 3). The pooled positive likeli-
Most studies may be subject to review bias as it was hood ratio was 15.2 (95% CI: 8.1-28.7) and the pooled
unclear in eight studies whether the investigators who negative likelihood ratio was 0.06 (95% CI: 0.03-0.11).
used the reference test (clinical criteria) were blinded to The area under the ROC curve was 0.98 (95% CI: 0.96-
the results of the NT-pro-BNP assay and six studies 0.99) and the diagnostic odds ratio was 246 (95% CI:
[16,18,20,21,24,25] did not state whether the laboratory 81-745). Figure 4 shows the summary receiver operating
personnel performing the index test (NT-pro-BNP characteristic graph with 95% confidence region and
assay) were blinded to the clinical diagnosis. 95% prediction region for NT-pro-BNP.
The average NT-pro-BNP level for nine of the ten studies The between study variability (i.e. heterogeneity)
was 6140 pg/mL in pleural effusions of cardiac origin (this beyond what could be expected by sampling error was
value was not available for one study[22]). Of the five stu- moderate with an I2 of 40% (p = 0.09) for the sensitivity
dies[16,18,19,21,24] that analyzed both serum and pleural results and high with an I 2 of 84% (p < 0.01) for the
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 5 of 11
http://www.biomedcentral.com/1471-2466/10/58

Table 2 Diagnostic accuracy variables of the studies included in the meta-analysis.

Study No. TP FN TN FP Se Sp NT-pro-BNP Average Misclassification Correlation
Pts/ Threshold* NT-pro- Statistic∏
Pes BNP†
Light’s‡ NT-pro-
BNP§
Tomcsanyi et al 28 14 0 14 0 100% 100% ≥ 599¶ 8236 1/14 0/1 R2 = 0.95
[16] (7%) p < 0.0001
Porcel et al[17] 117 40 4 68 5 91% 93% ≥ 1500 6931 10/35 2/10 NR
(29%)
Kolditz et al[18] 93 23 2 63 5 92% 93% ≥ 4000 10427 9/25 0/9 S = 0.96
(36%) p < 0.001
Porcel et al[19] 93 49 4 35 5 92% 87% ≥ 1500 6106 8/53 2/8 S = 0.96
(15%) p < 0.001
Liao et al[20] 40 10 0 29 1 100% 97% ≥ 2220 5390 NR NR NR
Han et al[21] 240 81 1 156 2 99% 99% ≥ 1714 3310 28/82 1/28 R2 = 0.93
(34%) p < 0.001
Seyhan et al[22] 115 47 4 61 3 92% 95% ≥ 1092 N/A 17/51 N/A N/A
(33%)
Porcel et al[23] 181 84 6 81 10 93% 89% ≥ 1500 6203 20/90 4/20 NR
(22%)
Bayram et al[24] 133 48 3 78 4 94% 95% ≥ 925 4468 19/51 0/19 R2 = 0.91
(37%) p < 0.001
Long et al[25] 80 16 4 44 16 80% 73% ≥ 2000 4189 NR NR NR
* = pleural fluid NT-pro-BNP threshold in pg/mL; † = average pleural fluid NT-pro-BNP in CHF effusions in pg/mL; ‡ = misclassification of CHF effusions as
exudates by Light’s criteria; § = misclassification of the CHF effusions already misclassified by Light’s criteria now misclassified by NT-pro-BNP; ∏ = correlation
between serum and pleural fluid NT-pro-BNP levels; ¶ = threshold from 599 to 1457 pg/mL resulted in 100% sensitivity and specificity; N/A = not available
because only abstract available for data extraction; No. = number; NR = not recorded; Pes = pleural effusions; Pts = patients; R2 = Pearson coefficient of
correlation; S = Spearman’s coefficient of rank correlation; Se = sensitivity; Sp = specificity.

specificity results. The bivariate model analysis reveals history for diagnosing pleural effusions of cardiac origin
that this heterogeneity is completely (100%) explained however these criteria misclassify transudates as exuda-
by threshold effect. Analysis of small study effects, tive effusions approximately 25% of the time, mainly
potentially a result of publication bias, yielded no signifi- related to the use of diuretics[2]. Misclassification of
cant evidence for such effects with a p value of 0.26 and transudates to exudates can lead to inappropriate man-
a funnel plot that was fairly symmetrical (Figure 5). agement or potentially more invasive diagnostic investi-
Figure 6 illustrates the post-test probabilities based on gations resulting in increased morbidity and health care
various pre-test probabilities using a Fagan nomogram. costs. As a result, tests with higher diagnostic accuracy
may prove beneficial in terms of reducing morbidity and
Discussion and Conclusion improving cost-efficiency for diagnosing pleural effu-
Pleural effusions are relatively common in medical prac- sions of cardiac origin.
tice. It is estimated that the annual incidence of pleural In the last decade several studies have been reported
effusions in the United States (US) is 1.5 million cases assessing the diagnostic utility of pleural fluid NT-pro-
[26]. The most common cause of pleural effusions is BNP in pleural effusions of cardiac origin. To summar-
CHF with an estimated incidence of 500,000 cases in ize these studies, we conducted a systematic review and
the US per year[2]. Identifying the underlying etiology meta-analysis. We included ten studies, of which seven
of pleural effusions requires a combination of strategies were from Europe, with a total of 1120 patients and/or
including clinical history, pleural fluid analysis, and pleural effusions. We showed that the pooled sensitivity
potentially more invasive procedures such as pleural tis- and specificity were identical (94% [95% CI: 90-97] and
sue biopsy. Clinical history is very important in the diag- 94% [95% CI: 89-97]) with a diagnostic odds ratio of
nosis of pleural effusions from a cardiac origin. 246 (95% CI: 81-745). When the study by Seyhan et al
However, alone, it does not appear to be very accurate. [22], which was retracted from the literature, was
Romero-Candeira et al[27] studied 64 patients with excluded from the analysis, the sensitivity, specificity,
transudative pleural effusions of which 44 were due to and diagnostic odds ratio remained the same within
CHF and showed that in 40% of cases initial clinical his- 95% confidence intervals.
tory failed to correctly classify the effusion. Pleural fluid The quality of the studies was generally good overall.
analysis using Light’s criteria is better than clinical The main limitations of the studies were the possibility
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 6 of 11
http://www.biomedcentral.com/1471-2466/10/58

Figure 3 Forrest plot of the sensitivity and specificity of each individual study, pooled sensitivity and specificity, and I2 statistic for
heterogeneity.

of spectrum bias, review bias, and population bias. Spec- within the CHF-associated pleural effusion group thus
trum bias refers to distortions in a diagnostic test’s per- potentially limiting generalizability. Furthermore, a thor-
formance caused by a distortion in the study population. acentesis in patients with a high pre-test probability for
Testing is not done across a population with the a cardiac origin of the pleural effusion with low prob-
expected distribution of disease severity, but rather lim- ability for other causes would not usually be indicated
ited subsets[28]. In our meta-analysis, five studies in most clinical scenarios thus further limiting the gen-
[16,19,20,23,25] acquired their study population by eralizability of these results.
selecting CHF patients and patients with pleural effu- Heterogeneity was moderate for the sensitivity results
sions of non-cardiac origin (control group). The pre- and high for the specificity results (Figure 3). In the
sence of spectrum bias can lead to an overestimation of summary receiver operating characteristic graph (Figure
the sensitivity and specificity of the test[28]. Review bias 4), all but three of the ten ‘sensitivity - (1-specificity)’
refers to a situation where persons interpreting the combinations of the individual studies lie on or near the
index test have knowledge of the reference standard or summary receiver operating characteristic curve. This
vice versa, when persons interpreting the reference stan- indicates that studies with a higher sensitivity have
dard have knowledge of the index test[29]. In our meta- lower specificity and vice versa. This pattern is com-
analysis, it was very unclear whether this did or did not monly attributed to differences in the threshold level for
occur because the majority of the studies did not report test positivity used in different studies[12]. As a result,
whether blinding during testing was done. Again, this the heterogeneity between the studies could well be
may have lead to an overestimation of the diagnostic explained by threshold level differences between the stu-
performance of the test. Finally, population bias refers dies which ranged from 599 to 4000 pg/mL. Another
to the generalizability of the diagnostic test to a wider potential cause for the heterogeneity may be due to the
population[30]. The studies included in our meta- various types of clinical criteria used for the reference
analysis had poor reporting of any co-morbid conditions standard (Table 1) as there is no gold standard for the
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 7 of 11
http://www.biomedcentral.com/1471-2466/10/58

Figure 4 Summary receiver operating characteristic graph with 95% confidence region and 95% prediction region for NT-pro-BNP.

diagnosis of pleural effusions from a cardiac origin. The ECLIA uses a non-competitive immunoassay that pro-
most objective criteria for heart failure is echocardiogra- duces luminescence via an electrochemical reaction and
phy and all studies except for Seyhan et al[22] used this compared to ELISA, is faster and more precise[31]. The
parameter and of those studies, the majority defined ELISA kit from Biomedica has been compared to the
CHF as an ejection fraction ≤ 40%. Furthermore, of the ECLIA kit from Roche Diagnostics revealing significant
studies that reported New York Heart Class, the CHF- disparity of results between the two types of assays[31].
associated pleural effusion patients were of New York The Roche Diagnostics assay (ECLIA) has dual antibo-
Heart Class (NYHC) III or IV. The control group dies targeting two different areas of the NT-pro-BNP
among the various studies differed and potentially could molecule whereas the Biomedica assay (ELISA) only has
have contributed to the heterogeneity of the results as a single antibody targeting one region. This may result
well. Malignant pleural effusions and parapneumonic in better recognition of NT-pro-BNP and hence better
effusions were the main groups that comprised the con- diagnostic accuracy for the Roche Diagnostics assay.
trol groups but some studies did have other causes as a Furthermore, the cut-off values for serum NT-pro-BNP
majority of the control group such as connective tissue are 10 fold higher using the Biomedica assay (ELISA)
disease, pleuritis, post-CABG, hepatic hydrothorax, and compared to the Roche Diagnostics assay (ECLIA)[31]
pulmonary embolism. Finally, eight of the ten studies which may have also added to the heterogeneity of the
used electrochemical luminescence immunoassay results.
(ECLIA) for measuring NT-pro-BNP (8 studies genera- The correlation between pleural fluid and serum levels
tion one, one study generation two, one study unknown of NT-pro-BNP are high as shown by five studies
generation) while two studies used ELISA which may [16,18,19,21,24] in our meta-analysis (Table 2). This
have added to the heterogeneity of results as well. suggests that thoracentesis could potentially be avoided
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 8 of 11
http://www.biomedcentral.com/1471-2466/10/58

Figure 5 Deeks’ funnel plot asymmetry test for publication bias.

in patients with a clinical suspicion of CHF, elevated NT-pro-BNP was greater than for pleural fluid BNP, 0.84
NT-pro-BNP, and no suspicion for a co-existing cause (95% CI: 0.72-0.95) vs 0.70 (95% CI: 0.57-0.83). Correla-
of the pleural effusion. tion between pleural fluid NT-pro-BNP and BNP levels
Most institutions use BNP to diagnose CHF. Three stu- was also good but less impressive, r = 0.57 (p < 0.001),
dies[23,25,32] have examined the diagnostic accuracy of and only explained 32% of the variance in NT-pro-BNP.
BNP for pleural effusions of cardiac origin. The first The conclusion from this study was that NT-pro-BNP
study[32] of 57 patients showed a sensitivity of 97% and a was a more stable molecule after sample processing and
specificity of 89% for plasma BNP. The second study by could be maintained for a greater duration in an in vitro
Porcel et al[23] (181 patients), showed a sensitivity of setting compared to BNP and furthermore because of the
74% and specificity of 92% for pleural fluid BNP and also higher diagnostic accuracy of NT-pro-BNP, NT-pro-BNP
showed that the area under the curve (AUC) for NT-pro- should be used over BNP to distinguish pleural effusions
BNP was higher than for pleural fluid BNP, 0.96 (95% CI: of cardiac origin in select patient populations.
0.94-0.99) vs 0.90 (95% CI: 0.86-0.95) respectively. Corre- It is difficult to suggest a threshold level for pleural
lation between pleural fluid NT-pro-BNP levels and BNP fluid NT-pro-BNP since the studies used varying thresh-
levels was good, r = 0.78 (p < 0.001). The conclusion old levels. The average pleural fluid NT-pro-BNP level
from this study was to use NT-pro-BNP rather than BNP in the studies was 6140 pg/mL. Six of the studies
for diagnosing pleural effusions of cardiac origin because [16,17,19,21,23,24] used a threshold level between 1457
of the diagnostic superiority of NT-pro-BNP versus BNP, to 1714 pg/mL (sensitivity and specificity for a threshold
greater in-vitro stability of NT-pro-BNP compared with level of 1457 pg/mL by ROC analysis in the study by
BNP, and longer half-life of NT-pro-BNP (1-2 hours) Bayram et al[24] was 84% and 98%, respectively). The
compared with BNP (20 minutes). The last study by pooled sensitivity and specificity of these six studies is
Long et al[25] (80 patients) showed that the AUC for 95% (95% CI: 91-98) and 95% (95% CI: 89-98),
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 9 of 11
http://www.biomedcentral.com/1471-2466/10/58

Figure 6 Fagan’s nomogram for NT-pro-BNP illustrating post-test probability with a fixed pre-test probability of 20% for a pleural
effusion of cardiac origin.

respectively, with a diagnostic odds ratio of 370 (95% CI: low or alternative diagnoses are considered in addition to
101-1351). As a result, a threshold level of approxi- heart failure, then a thoracentesis should be performed
mately ≥ 1500 pg/mL provides very good diagnostic and if the pleural fluid NT-pro-BNP level is high then this
accuracy for pleural fluid NT-pro-BNP. would be diagnostic of heart failure as one potential cause
The utility of NT-pro-BNP for the diagnosis of pleural for the pleural effusion. Furthermore, in undiagnosed
effusions of cardiac origin can be applied to several clinical pleural effusions, for example when Light’s criteria mis-
situations. First, if the pre-test probability of a pleural effu- classify transudative CHF effusions as exudates and the
sion due to heart failure is high and alternative diagnoses pre-test probability is equivocal for CHF, a high pleural
are less likely, then a thoracentesis can be avoided and an fluid NT-pro-BNP would also be diagnostic.
elevated serum NT-pro-BNP level will be diagnostic for a The results of our meta-analysis are corroborated by
cardiac origin of the pleural effusion. Second, if the pre- another meta-analysis done by Zhou et al[33]. They
test probability of a pleural effusion due to heart failure is included eight studies and found that the sensitivity and
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 10 of 11
http://www.biomedcentral.com/1471-2466/10/58

specificity for NT-pro-BNP for pleural effusions of car- 7. Porcel JM: The use of probrain natriuretic peptide in pleural fluid for the
diagnosis of pleural effusions resulting from heart failure. Curr Opin Pulm
diac origin was 95% (95% CI: 92-97) and 94% (95% CI: Med 2005, 11(4):329-333.
92-96), respectively, with a diagnostic odds ratio of 214 8. Bhalla V, Willis S, Maisel AS: B-type natriuretic peptide: the level and the
(95% CI: 123-374). Based on their meta-analysis, Zhou drug–partners in the diagnosis of congestive heart failure. Congest Heart
Fail 2004, 10(1 Suppl 1):3-27.
et al[33] concluded that pleural fluid NT-pro-BNP has 9. Moher D, Liberati A, Tetzlaff J, Altman DG: Preferred reporting items for
high diagnostic accuracy for differentiating cardiac from systematic reviews and meta-analyses: the PRISMA statement. Ann Intern
non-cardiac conditions in patients presenting with Med 2009, 151(4):264-269, W264.
10. Whiting P, Rutjes AW, Reitsma JB, Bossuyt PM, Kleijnen J: The development
pleural effusions. of QUADAS: a tool for the quality assessment of studies of diagnostic
In summary, misclassification of pleural effusions of accuracy included in systematic reviews. BMC Med Res Methodol 2003,
cardiac origin can lead to increased morbidity from 3:25.
11. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman AH:
further invasive testing and increased health care costs. Bivariate analysis of sensitivity and specificity produces informative
Clinical history and pleural fluid analysis with Light’s summary measures in diagnostic reviews. J Clin Epidemiol 2005,
criteria may not have sufficient diagnostic accuracy to 58(10):982-990.
12. Moses LE, Shapiro D, Littenberg B: Combining independent studies of a
diagnose these effusions in certain clinical circum- diagnostic test into a summary ROC curve: data-analytic approaches and
stances. Although there is some heterogeneity within some additional considerations. Stat Med 1993, 12(14):1293-1316.
this analysis, we conclude that where the diagnosis of 13. Dinnes J, Deeks J, Kirby J, Roderick P: A methodological review of how
heterogeneity has been examined in systematic reviews of diagnostic
pleural effusion of cardiac origin is under contempla- test accuracy. Health Technol Assess 2005, 9(12):1-113, iii.
tion, pleural fluid NT-pro-BNP is a useful biomarker 14. Dwamena B: Midas: computational and graphical routines for meta-
with high diagnostic accuracy. analytical integration of diagnostic accuracy studies in Stata. 2007.
15. Harbord R: Stata module for meta-analysis of diagnostic accuracy. 2008.
16. Tomcsanyi J, Nagy E, Somloi M, Moldvay J, Bezzegh A, Bozsik B, Strausz J:
NT-brain natriuretic peptide levels in pleural fluid distinguish between
Abbreviations
pleural transudates and exudates. Eur J Heart Fail 2004, 6(6):753-756.
BNP: brain natriuretic peptide; CHF: congestive heart failure; ELCIA:
17. Porcel JM, Vives M, Cao G, Esquerda A, Rubio M, Rivas MC: Measurement
electrochemical luminescence immunoassay; ELISA: enzyme-linked
of pro-brain natriuretic peptide in pleural fluid for the diagnosis of
immunosorbent assay; HSROC: hierarchical summary receiver operating
pleural effusions due to heart failure. Am J Med 2004, 116(6):417-420.
characteristic; NT-pro-BNP: N-terminal pro-brain natriuretic peptide; QUADAS:
18. Kolditz M, Halank M, Schiemanck CS, Schmeisser A, Hoffken G: High
quality assessment of diagnostic accuracy studies; US: United States.
diagnostic accuracy of NT-proBNP for cardiac origin of pleural effusions.
Eur Respir J 2006, 28(1):144-150.
Authors’ contributions
19. Porcel JM, Chorda J, Cao G, Esquerda A, Ruiz-Gonzalez A, Vives M:
Both SJ and JS conducted the literature search independently. Both SJ and
Comparing serum and pleural fluid pro-brain natriuretic peptide (NT-
JS analyzed the studies to be included independently. SJ retrieved the
proBNP) levels with pleural-to-serum albumin gradient for the
articles. Both SJ and JS extracted the information independently. SJ
identification of cardiac effusions misclassified by Light’s criteria.
conducted the statistical analysis with confirmation by JS. SJ wrote the
Respirology 2007, 12(5):654-659.
manuscript draft. Both SJ and JS read and approved the final manuscript.
20. Liao H, Na MJ, Dikensoy O, Lane KB, Randal B, Light RW: Diagnostic value
of pleural fluid N-terminal pro-brain natriuretic peptide levels in patients
Acknowledgements
with cardiovascular diseases. Respirology 2008, 13(1):53-57.
JS is a recipient of an In it for Life Scientist award from the Vancouver
21. Han CH, Choi JE, Chung JH: Clinical utility of pleural fluid NT-pro brain
Coastal Health Research Institute and the Vancouver General Hospital
natriuretic peptide (NT-proBNP) in patients with pleural effusions. Intern
Foundation.
Med 2008, 47(19):1669-1674.
22. Seyhan EC, Altin S, Cetinkaya E, Sokucu S, Gunluoglu MZ, Demir A,
Competing interests
Korkmaz P, Issever H: The importance of pleural fluid and serum NT-
Dr. Janda and Dr. Swiston have no competing interests.
proBNP levels in differentiating pleural effusion due to heart failure
from other causes of effusion. Intern Med 2009, 48(5):287-293.
Received: 29 August 2010 Accepted: 20 November 2010
23. Porcel JM, Martinez-Alonso M, Cao G, Bielsa S, Sopena A, Esquerda A:
Published: 20 November 2010
Biomarkers of heart failure in pleural fluid. Chest 2009, 136(3):671-677.
24. Bayram M, Ozkan G, Oztekin E, Bakan ND, Acikmese B, Bes S, Gur A,
References Camsari G: Role of serum and pleural fluid NT-proBNP levels in
1. Light RW, Macgregor MI, Luchsinger PC, Ball WC Jr: Pleural effusions: the identifying pleural effusion due to heart failure. Multidisciplinary
diagnostic separation of transudates and exudates. Ann Intern Med 1972, Respiratory Medicine 2009, 4(3):175-181.
77(4):507-513. 25. Long AC, O’Neal HR Jr, Peng S, Lane KB, Light RW: Comparison of pleural
2. Light RW: Clinical practice. Pleural effusion. N Engl J Med 2002, fluid N-terminal pro-brain natriuretic peptide and brain natriuretic-32
346(25):1971-1977. peptide levels. Chest 2010, 137(6):1369-1374.
3. Romero S, Candela A, Martin C, Hernandez L, Trigo C, Gil J: Evaluation of 26. Sahn SA: The value of pleural fluid analysis. Am J Med Sci 2008,
different criteria for the separation of pleural transudates from exudates. 335(1):7-15.
Chest 1993, 104(2):399-404. 27. Romero-Candeira S, Hernandez L, Romero-Brufao S, Orts D, Fernandez C,
4. Roth BJ, O’Meara TF, Cragun WH: The serum-effusion albumin gradient in Martin C: Is it meaningful to use biochemical parameters to discriminate
the evaluation of pleural effusions. Chest 1990, 98(3):546-549. between transudative and exudative pleural effusions? Chest 2002,
5. Tang WH, Francis GS, Morrow DA, Newby LK, Cannon CP, Jesse RL, 122(5):1524-1529.
Storrow AB, Christenson RH: National Academy of Clinical Biochemistry 28. Ransohoff DF, Feinstein AR: Problems of spectrum and bias in evaluating
Laboratory Medicine Practice Guidelines: clinical utilization of cardiac the efficacy of diagnostic tests. N Engl J Med 1978, 299(17):926-930.
biomarker testing in heart failure. Clin Biochem 2008, 41(4-5):210-221. 29. Begg CB: Biases in the assessment of diagnostic tests. Stat Med 1987,
6. Bhalla V, Maisel AS: B-type natriuretic peptide. A biomarker for all the 6(4):411-423.
right reasons. Ital Heart J 2004, 5(6):417-420.
Janda and Swiston BMC Pulmonary Medicine 2010, 10:58 Page 11 of 11
http://www.biomedcentral.com/1471-2466/10/58

30. Kelly S, Berry E, Roderick P, Harris KM, Cullingworth J, Gathercole L,

Hutton J, Smith MA: The identification of bias in studies of the diagnostic
performance of imaging modalities. Br J Radiol 1997, 70(838):1028-1035.
31. Mueller T, Gegenhuber A, Poelz W, Haltmayer M: Comparison of the
Biomedica NT-proBNP enzyme immunoassay and the Roche NT-proBNP
chemiluminescence immunoassay: implications for the prediction of
symptomatic and asymptomatic structural heart disease. Clin Chem 2003,
49(6 Pt 1):976-979.
32. Gegenhuber A, Mueller T, Dieplinger B, Lenz K, Poelz W, Haltmayer M:
Plasma B-type natriuretic peptide in patients with pleural effusions:
preliminary observations. Chest 2005, 128(2):1003-1009.
33. Zhou Q, Ye ZJ, Su Y, Zhang JC, Shi HZ: Diagnostic value of N-terminal
pro-brain natriuretic peptide for pleural effusion due to heart failure: a
meta-analysis. Heart 96(15):1207-1211.

Pre-publication history
The pre-publication history for this paper can be accessed here:
http://www.biomedcentral.com/1471-2466/10/58/prepub

doi:10.1186/1471-2466-10-58
Cite this article as: Janda and Swiston: Diagnostic accuracy of pleural
fluid NT-pro-BNP for pleural effusions of cardiac origin: a systematic
review and meta-analysis. BMC Pulmonary Medicine 2010 10:58.

Submit your next manuscript to BioMed Central

and take full advantage of:

• Convenient online submission

• Thorough peer review
• No space constraints or color figure charges
• Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution

Submit your manuscript at

www.biomedcentral.com/submit