Biomarkers of infection for the differential diagnosis of pleural effusions

Study population

Since 1994, we have been prospectively maintaining a database on all patients who undergo diagnostic thoracentesis at the Arnau de Vilanova University Hospital (Lleida, Spain), as well as a pleural bank of fluid specimens that are processed and stored at -80°C. The indication for diagnostic thoracentesis was the presence of a clinically significant pleural effusion of uncertain origin. In patients submitted to more than one thoracentesis during the hospitalisation period, only the results of the first tap were considered. Samples of pleural fluid were immediately analysed for routine biochemistries (e.g. pH, glucose and lactate dehydrogenase (LDH)), while supernatants were frozen within 4 h of collection until the measurement of the biomarkers at the end of the study.

For the present study, 308 pleural fluid samples collected from 2004 to 2008 were randomly selected for biomarkers determination. Patients were classified into seven groups according to the cause of the pleural effusion: transudates (40 patients), malignant effusions (40 patients), tuberculous pleurisy (50 patients), uncomplicated parapneumonics (UPPE; 60 patients), nonpurulent CPPE (68 patients), empyemas (30 patients) and miscellaneous exudates (20 patients). Primary tumours in malignant effusions were lung (14 patients), breast (eight patients), unknown primary (six patients), lymphoma (four patients), ovary (three patients), gastrointestinal (three patients) and mesothelioma (two patients). Among the miscellaneous exudates, there were five post-traumatic, three idiopathic, three post-coronary artery bypass surgery, two pericarditis, two pulmonary embolism, two abdominal abscess, one pancreatitis, one post-abdominal surgery and one systemic lupus erythematosus.

The local Ethics Committee (Arnau de Vilanova University Hospital, Lleida, Spain) approved this study, and all patients gave informed consent for the analysis of stored specimens for future research.

Diagnostic criteria

A pleural effusion was categorised as malignant if malignant cells were detected on cytological examination of the pleural fluid or biopsy specimen. Tuberculous pleuritis was diagnosed if Löwenstein cultures of pleural fluid, sputum or pleural biopsy tissue samples were positive (26 patients), a pleural biopsy specimen showed granulomas in the parietal pleura (five patients) or an exudative lymphocytic effusion with high adenosine deaminase levels (>40 U·L⁻¹) cleared in response to antituberculous therapy (19 patients). Transudates were effusions secondary to heart failure (30 patients), cirrhosis (six patients), nephrotic syndrome (two patients) or atelectasis (two patients). PPEs referred to those associated with bacterial pneumonia and were classified into three groups: UPPE (resolution with antibiotic treatment alone), nonpurulent CPPE (requirement of an invasive procedure such as tube thoracostomy for cure) and empyema (pus into the pleural space). Other causes of pleural effusions were determined by well-established clinical criteria.

Measurement of biomarkers

Assays for the four inflammatory markers, namely CRP, sTREM, PCT and LBP, were performed on the stored cell-free supernatants of pleural fluid samples. All samples were tested in a random order and by technicians who were blinded to the clinical diagnosis of patients at the end of the study.

Pleural fluid CRP was measured using a particle-enhanced immunoturbidimetric assay (CRPLX, Tina-quant; Roche Diagnostics GmbH, Mannheim, Germany) on Roche automated clinical chemistry analysers. The sTREM-1 concentrations in pleural fluid were measured by a sandwich ELISA (DuoSet ELISA; R&D Systems Europe, Abingdon, UK), PCT by a two-site immunoluminometric assay (Liaison Brahms PCT; DiaSorin, Saluggia, Italy) and LBP with a solid-phase, enzyme-labelled chemiluminescent immunometric assay (Immulite 2000 LBP; Siemens, Los Angeles, CA, USA), according to the manufacturer's instructions. The samples for sTREM-1 were assayed in duplicate.

Data analysis

Continuous data are presented as median (quartile range). Between-group comparisons were performed with Kruskal–Wallis tests and Dunn's multiple comparison post hoc tests for continuous variables, and Chi-squared tests with post hoc analysis of adjusted residuals for categorical variables. Receiver-operating characteristic (ROC) curves were constructed to illustrate the predictive value of various cut-off points of CRP, sTREM-1, PCT and LBP. The point with the largest sum of sensitivity and specificity was chosen as a threshold. We compared the performances of the four pleural fluid inflammatory biomarkers for discriminating infectious (bacterial and mycobacterial) versus noninfectious effusions, tuberculosis versus malignancy, PPE versus non-PPE, and CPPE versus UPPE. Measures of test efficacy included sensitivity, specificity and likelihood ratios. To adjust for confounders, a backward conditional stepwise logistic regression model estimated the simultaneous impact of each biomarker, along with other biochemical fluid findings, in the prediction of nonpurulent CPPE. The level of significance was set at p<0.05 using two-tailed tests. Data were analysed using a statistical software package (SPSS version 13.0; SPSS Inc, Chicago, IL, USA).

Pleural fluid levels of the biomarkers

The demographic data and the median pleural fluid levels of CRP, sTREM-1, PCT and LBP in each of the seven groups are shown in table 1⇓. The levels of CRP, sTREM-1 and LBP were significantly higher in the pleural fluid from patients with CPPE and lower in the transudate and malignant groups (fig. 1⇓). The levels of PCT showed the same trend but with a large overlap of values.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1—

Box plots of pleural fluid concentrations of a) C-reactive protien (CRP), b) soluble triggering receptor expressed on myeloid cells (sTREM-1), c) procalcitonin (PCT) and d) lipopolysaccharide binding protein (LBP) in different aetiologies of pleural effusion. Pleural sTREM-1 and PCT levels are plotted on a log scale. Horizontal dashed lines represent cut-off values with discriminatory properties. UPPE: uncomplicated parapneumonic effusion; CPPE: complicated parapneumonic effusion.

Discriminating infectious from noninfectious effusions

The thresholds of the four biomarkers that best discriminated between infectious (PPE and tuberculosis) and noninfectious effusions were as follows: CRP ≥20 mg·L⁻¹, sTREM-1 ≥80 pg·mL⁻¹, PCT ≥0.25 ng·mL⁻¹ and LPB ≥7 μg·mL⁻¹ (table 2⇓). While PCT had no discriminative properties (area under curve (AUC) 0.57), sTREM-1 (AUC 0.86), CRP and LBP (both AUCs 0.87) showed good abilities to separate infectious from noninfectious states. However, the overlapping 95% confidence interval values for these three biomarkers indicated that no single one could be identified as being superior to the others.

Discriminating tuberculous from malignant effusions

The AUCs for each pleural fluid biomarker (calculated with the cut-off points described above) in differentiating between tuberculous and malignant effusions are displayed in table 3⇓. In contrast to PCT (AUC 0.52), which was not useful, CRP, sTREM-1 and LPB had better AUCs that were similar in value (all AUCs >0.80).

Discriminating PPE from other aetiologies

In discriminating PPE from the other aetiologies, the respective AUCs generated from the use of CRP, sTREM-1, PCT and LBP were 0.83, 0.79, 0.63 and 0.80 (table 4⇓). Therefore, PCT lacked efficacy as a marker of PPE.

Discriminating UPPE from nonpurulent CPPE

Table 5⇓ shows the diagnostic accuracies of the four biomarkers in the differentiation of UPPE and nonpurulent CPPE at the optimum cut-off points. They were compared with the performances of pH, glucose and LDH at the cut-off values recommended in the literature. All parameters, with the exception of PCT (AUC 0.59), performed well in discriminating between groups (AUCs ∼0.80). Nevertheless, the overlap of the 95% confidence incidence values for the AUC and likelihood ratios did not allow for a selection of a superior test that clearly had the highest discriminative properties for nonpurulent CPPE.

When all the variables listed in table 5⇑ were entered into a multivariate stepwise logistic regression model, the parameters that best discriminated nonpurulent CPPE from UPPE were as follows: glucose (adjusted likelihood ratio (LR)+ 5.2, 95% CI 2.1–8; adjusted LR- 0.10, 95% CI 0.02–0.44), LBP (adjusted LR+ 4.6, 95% CI 2.1–7.1; adjusted LR- 0.14, 95% CI 0.04–0.44) and CRP (adjusted LR+ 3.4, 95% CI 1.4–6.1; adjusted LR- 0.22, 95% CI 0.07–0.69). This model had an AUC of 0.88 (95% CI 0.82–0.94).

CRP

CRP, an acute-phase reactant released from the liver, is a common diagnostic test within hospital laboratories for the screening or monitoring of infections and noninfectious inflammatory diseases. CRP levels have been studied in pleural fluid and have been found to be higher in PPEs than in other types of exudative or transudative effusions 5–7. We showed that a pleural fluid CRP level >80 mg·L⁻¹ argues for the presence of a PPE (LR+ 7.4), whereas CRP levels <20 mg·L⁻¹ are a strong indicator against an infectious pleural effusion, whether of bacterial or mycobacterial nature (LR-0.22). Indeed, a number of studies provide support for the use of CRP as a diagnostic aid in tuberculous pleuritis; low pleural CRP levels (<30 mg·L⁻¹) make this diagnosis unlikely while being more indicative of a malignancy in patients with exudates 8–10. In addition, our findings complement the scarce previous literature on the application of CRP for identifying CPPE 11, 12. This biomarker emerged as an independent predictor of nonpurulent CPPE in the multivariate analysis, although its adjusted LR+ was lower than those for glucose and LBP.

sTREM-1

TREM-1, a receptor of the immunoglobulin superfamily, amplifies the inflammatory response through its over-expression and subsequent activation of neutrophils and monocytes/macrophages in response to microbial products. TREM-1 is also shed by the membrane of activated phagocytes and can be found in a soluble form in body fluids.

Few studies have investigated the clinical significance of sTREM-1 in pleural effusions 13–18. All included small sample sizes (45–109 patients), measured sTREM-1 by ELISA or other techniques and found that patients with PPE or empyema exhibited the highest pleural fluid concentrations of this biomarker. However, there were discrepancies regarding its discriminative properties as well as its optimal cut-off point. For example, in one study, an sTREM-1 cut-off value of 768.1 pg·mL⁻¹ had a sensitivity of 86%, specificity of 93% and AUC of 0.93 in differentiating 23 bacterial effusions (including 17 empyemas) from 88 effusions with other aetiologies 17. In contrast, our findings yielded an AUC of 0.79 for identifying PPE at the best cut-off of 80 pg·mL⁻¹. The over-representation of empyemas in the first series (74% versus 19% of PPE) may help explain these discrepancies, while different sTREM-1 concentrations and cut-off points probably reflect the lack of standardisation of the ELISA technique. In a second study, a cut-off value of 114 pg·mL⁻¹ for pleural sTREM-1 achieved a sensitivity of 94% and a specificity of 93% (AUC 0.966) in differentiating 17 empyemas from 72 pleural effusions of other aetiologies 18. We think there is no point in trying to identify empyema by measuring sTREM-1, because this diagnosis is easily achieved by simple inspection. Finally, in another study, pleural sTREM-1 at a cut-off value of 374 pg·mL⁻¹ yielded a sensitivity of 93.8%, a specificity of 90.9% and an AUC of 0.93 in discriminating bacterial pleural infection (n = 22) from tuberculous pleuritis (n = 16) 14. Based on our results (sensitivity 80%, specificity 56% at a cut-off of 180 pg·mL⁻¹, AUC 0.64; data not shown), we do not support the use of sTREM-1 measurements for this particular indication. In addition, the ability to differentiate between PPEs and tuberculous effusions does not normally pose a problem in clinical practice.

Taken together, our data indicate that pleural fluid levels of sTREM-1 may be accurate enough to help differentiate between infectious and noninfectious effusions (AUC 0.86). Notably, two studies demonstrated that the levels of sTREM-1 in pleural fluid greatly exceeded those in serum 16, 17, suggesting that recruited inflammatory cells in the pleural space produce the sTREM-1 locally.

Procalcitonin

PCT, an acute-phase hormokine, has been reported to be more accurate for differentiating between bacterial infections and noninfectious causes of inflammation than other biomarkers 19. Serum levels of PCT increase in severe bacterial infections. A PCT serum concentration of 0.5 ng·mL⁻¹ has been recommended in clinical practice guidelines to rule out bacterial infection 1. There is little useful information on PCT levels in pleural fluid. One study found that the median pleural fluid PCT levels were significantly higher in 26 patients with bacterial infection (0.67 ng·mL⁻¹) than in 80 malignant effusions (0.14 ng·mL⁻¹), 33 cardiac effusions (0.06 ng·mL⁻¹) and 17 viral effusions (0.007 ng·mL⁻¹) 20. However, the authors did not fully describe the diagnostic criteria of the study population or calculate any discriminatory cut-off values for PCT. A recent investigation showed that pleural fluid PCT levels >0.18 ng·mL⁻¹ discriminated 45 PPEs from 37 non-PPEs with a sensitivity of 67%, a specificity of 77% and an AUC of 0.752 21. Interestingly, serum PCT had better diagnostic accuracy than pleural PCT and both correlated with the severity of pneumonia.

Our study failed to demonstrate any firm relationship between pleural fluid levels of PCT and the specific cause of the pleural effusion. These disappointing results could be explained by considering PCT as a potential biomarker of a state or a syndrome (e.g. severe sepsis) rather than an indicator of a disease. In fact, some studies have reported that, as a marker of bacterial infection, PCT was no better than CRP 22. However, in itself, it was considered quite useful as a marker regarding the severity of infection.

LBP

LBP, a glycosylated 58-kDa protein produced predominantly in hepatocytes, has recently been described as a promising novel diagnostic marker of bacterial infection 22. To the best of our knowledge, pleural fluid LBP concentrations have not been previously evaluated. We observed that pleural fluid LBP levels of <7 μg·mL⁻¹ substantially reduced the probability of an infectious aetiology (LR-  = 0.22). This may have implications, for instance, in the differential diagnosis of tuberculous and malignant effusions. Moreover, a LBP >17 μg·mL⁻¹ conferred an adjusted LR+ of 4.6 in identifying nonpurulent CPPE, a performance that was similar to the currently used tests, such as glucose or pH.

Selecting tests

The choice of one among different diagnostic tests depends not only on its accuracy, but also on the financial costs and technical considerations. The cost of one measurement for CRP, sTREM-1, PCT and LBP is about 1, 4, 14 and 6 \#8364;, respectively. Conversely, CRP and LBP measurements are less labour-intensive than those for PCT and sTREM-1. Therefore, given the lack of utility of measuring PCT and the similarity in the accuracy of the other three tests, cost-effectiveness and technical reasons would argue for the diagnostic use of CRP, if a selection needs to be done.

Study limitations

This study should be interpreted in the context of certain limitations. First, the use of physician judgment as the gold standard for indicating the need for pleural drainage in CPPE may have resulted in patients' misclassifications, thus compromising the diagnostic accuracy of the pleural fluid tests. However, satisfactory correction for the bias introduced by an imperfect gold standard does not exist. Secondly, except for the identification of nonpurulent CPPE, we have not compared the advantages of the new tests over alternative tests; that is, we have not answered the question of whether the new tests are better than the current diagnostic standard. Nevertheless, our primary goal was simply to determine whether the new biomarkers accurately and reliably identified cases of obvious infectious pathology in a well-defined population. Whether these biomarkers offer additional information beyond that which is normally available is a matter of speculation. Finally, our findings, which are based on a retrospective analysis, deserve further prospective evaluation in an independent validation cohort.

Conclusions

In summary, CRP, sTREM-1 and LBP measurements may be useful as an additional tool in the assessment of pleural effusions to support the differential diagnosis between infectious and noninfectious aetiologies. However, in the most challenging clinical setting, such as the identification of a nonpurulent PPE needing a tube thoracostomy, the accuracy of the three biomarkers is comparable with that of the classical pleural fluid biochemistries, such as pH, glucose or LDH. For this reason, we predict that even if these new tests are adopted, they will not have a major impact on clinical practice.