This was a retrospective study that included all consecutive patients with PE seen in the Hospital Universitario Arnau de Vilanova de Lleida between January 2001 and December 2016 who met the following criteria (derivation sample): (a) at least two diagnostic thoracenteses with available biochemical data on PF (if three or more thoracenteses were performed, only the first two were considered); (b) PF in the first thoracentesis was an exudate (according to Light's criteria8) with >50% lymphocytes in the leukocyte count and a negative cytology for malignancy; and (c) a definitive diagnosis for PE etiology or else PE determined to be idiopathic (see definition below). The Hospital Universitario Arnau de Vilanova de Lleida is a 440-bed reference hospital with a catchment population of 450000 inhabitants, and carries out around 400 diagnostic thoracenteses per year.

Patients were divided into two groups according to the final diagnosis: malignant PE and benign PE. The benign PE group was then subdivided into idiopathic and other etiologies. Absolute and relative changes in the PF biochemical variables between two successive thoracentesis considered significant in the derivation sample were validated in a population with the same inclusion criteria, seen in the Hospital de Mendaro between January 1995 and August 2015 (validation sample). Hospital de Mendaro is a 125-bed county hospital with a catchment population of 80000, and performs 50 diagnostic thoracenteses every year.

The following variables were recorded: demographic data (sex and age); date of both thoracenteses; protein and LDH serum concentrations; and red blood cell and white blood cell counts, neutrophil and lymphocyte percentages, glucose, LDH, proteins, adenosine deaminase (ADA), pH, and C-reactive protein (CRP) in PF. The local ethics committee approved the study protocol (CEIC-1592).

PE was defined as malignant if malignant cells were found in PF or pleural biopsy. PE due to heart failure was diagnosed from the clinical history, physical examination, chest X-ray, electrocardiogram, echocardiogram (if available), and good response to diuretics. Confirmation of a diagnosis of pleural tuberculosis required tuberculosis bacillus to be isolated from sputum samples, PF or pleural biopsy, or granulomas observed in the latter. Pleural tuberculosis was considered likely in patients with lymphocytic exudate and ADA≥35U/L in PF, whose PE resolved with antituberculosis treatment. Parapneumonic PEs were those associated with pneumonia, bronchiectasis, or pulmonary abscess. Finally, PE was classified as idiopathic when it resolved with no specific treatment and its etiology could not be determined despite an in-depth study that included PF analysis (biochemistry, cytology), chest and abdominal computed tomography (CT), and, in selected cases, pleural biopsy. Other diagnoses were established according to widely accepted clinical criteria.2 All patients with benign pleural diseases were followed up clinically until complete resolution of the PE, which, in the case of idiopathic PE, was at least a year.

PF was collected from the derivation population in 5mL heparinized tubes, and biochemical analysis was performed immediately after thoracentesis using an automatic analyzer, according to standard spectrophotometric methods (Hitachi 717 and 911 or Hitachi Modular PE, Roche Diagnostics, Mannheim, Germany) until August 2013, and subsequently with an AU5800 Beckman Coulte analyzer (Brea, CA, USA). The pH was analyzed using an arterial blood gases machine and ADA was determined with automated kinetic methods. Until February 2012, the PF cell count was performed manually in a Thoma counting chamber, and subsequently with automatic counters (Sysmex XN-2000 and Sysmex XN-350; Roche Diagnostics, Mannheim, Germany).

In the validation population, PF was collected in 15mL Falcon tubes containing 1mL of heparin. Biochemical variables were analyzed with a Cobas 6000 autoanalyzer (Roche Diagnostics, Mannheim, Germany). The pH was measured using an arterial blood gas analyzer, and ADA was determined with automated kinetic methods. Until December 2016, PF cell counts were determined in a Neubauer Improved counting chamber, and subsequently in an automated system (Sysmex SN-2000 and Sysmex XN-350, Roche Diagnostics, Mannheim, Germany).

Categorical variables were expressed as absolute numbers (percentages) and continuous variables as medians (25th and 75th percentiles). Benign and malignant PEs were compared with Fisher's exact test and the Mann–Whitney test, respectively. Variables associated with malignant, idiopathic and other benign PEs were compared with the Fisher's exact and Kruskal–Wallis tests, as appropriate.

The Wilcoxon test was used to compare biochemical parameters between the first and the second thoracentesis. For variables that showed significant differences between the two thoracenteses in any of the etiological groups (malignant, idiopathic, or other benign), we calculated both absolute differences (Δa=value of the variable in the second thoracentesis minus the value of the variable in the first thoracentesis) and percentage differences (Δp=value of the variable in the second thoracentesis minus the value of the variable in the first thoracentesis, divided by the value of the variable in the first thoracentesis).

An area under the curve (AUC) analysis of diagnostic efficacy was used to select the best cut-off point for continuous variables (Δa and Δp) that could predict malignant or benign PE. A multivariate logistic regression analysis, which included the dichotomous Δa or Δp of the biochemical variables that were significant in the bivariate analysis, was used to calculate the odds ratio (OR) of the Δa and Δp to identify the benign or malignant nature of the PE. Operational characteristics of the Δa and Δp were calculated using a 2×2 contingency table (sensitivity, specificity, positive and negative likelihood ratio [LR]) for both the derivation population and the validation population. Statistical significance was set at P≤.05. All statistical tests were performed using version 24.0 of the SPSS statistical package (Chicago, IL, USA).

The derivation population included 214 patients, representing 12% of all subjects with lymphocytic exudate evaluated in our hospital during the study period. Seventy (33%) had malignant PE, and the most common cancers were lung (19, 27%), breast (14, 20%) and mesothelioma (10, 14%). The final diagnosis of malignancy was obtained by subsequent cytological studies of PF (41 patients), cellular phenotyping of PF by flow cytometry (three patients), CT-guided pleural biopsy (11 patients), ultrasound-guided pleural biopsy (six patients), and thoracoscopic pleural biopsy (nine patients). Of the 144 benign PEs, 44 (31%) were classified as idiopathic. Of the latter, pleural biopsy was performed in only 11 cases (25%); in 28 (64%) the biopsy was considered a risk procedure (and was not performed), due to the presence of a PE that occupied one third or less of the hemithorax.

The validation sample included 99 patients (i.e., 15% of the lymphocytic exudates diagnosed in the corresponding hospital during the study period). Of these, 23 were diagnosed as malignant PE in subsequent cytology studies (six patients), cellular immunophenotyping of the PF by flow cytometry (three patients), and thoracoscopic biopsy (14 patients). Table 1 shows the characteristics of the derivation and validation populations. No differences were observed in the interval (days) between thoracenteses in the different etiological groups. Nor were differences observed between the overall derivation and the validation populations (medians of 8 [3–31] vs 6 [3–16] days, P=.132), or etiologic groups: malignant PE (medians of 13 [4–29] vs 6 [3–16] days, P=.147), and benign PEs (medians of 7 [3–33] vs 6 [3–18] days; P=.389).

As Table 2 shows, the percentage of neutrophils and LDH levels in PF fell in the second thoracentesis compared to the first in the subgroup of idiopathic PEs (12% vs 24%, P=.02; and 398U/L vs 476U/L, P<.01) and remaining benign PEs (16% vs 17%, P=.03; and 411U/L vs 494U/L, P<.01), but increased in the malignant PE group (12% vs 9%, P=.02; and 611U/L vs 478U/L, P<.01). Increased red blood cells and lower glucose and protein levels in the second PF sample compared to the first were observed only in the malignant PE group (11325/μL vs 4695/μL, P=.03; 99mg/dL vs 108mg/dL, P=.03; and 4.2g/dL vs 4.5g/dL, P<.01, respectively). In the group of other benign PEs, reduced leukocytes were also observed in the second thoracentesis (1084/μL vs 1518/μL, P<.01). Other biochemical variables (ADA, pH, and PCR) showed no differences between the two thoracenteses, so these parameters were ruled out as possible predictors of malignant PE.

Δa and Δp of the biochemical variables of the PFs that showed significant differences between the two thoracenteses were analyzed (Table 3). Those which best discriminated between a malignant PE and a benign PE (including idiopathic PEs) were: Δp for LDH (AUC 0.73, 95% confidence interval [95% CI]: 0.65–0.8), Δp for neutrophils (AUC 0.65, 95% CI: 0.57–0.73) and Δa for proteins (AUC 0.61, 95% CI: 0.54–0.7).

The best cut-off points for these Δa or Δp to discriminate between malignant and benign PE were: Δp LDH>0% (i.e., any increase in LDH in a second thoracentesis), Δp neutrophils>−10% (i.e., any increase or decrease of less than 10%) and Δa proteins<0.1g/dL (i.e., any decrease or an increase of less than 0.1g/dL). In this respect, according to a multivariate model, Δp LDH>0% showed an OR of 6.4 (95% CI: 3.1–12.9); Δp neutrophils>−10% showed an OR of 2.1 (95% CI: 1–4.2) and the Δa for proteins<0.1g/dL showed an OR of 3.9 (95% CI: 1.7–8.9) for identifying malignancy. The same analysis comparing malignant PEs with idiopathic PEs produced the following results: Δp LDH>0%, OR 9 (95% CI: 3.1–26.1); Δp neutrophils>−10% showed an OR of 2.7 (95% CI: 1–7.2) and the Δa for proteins<0.1g/dL showed an OR of 3.8 (95% CI: 1.2–11.8).

Table 4 shows the operational characteristics of the variables selected as predictors of malignant PE vs benign PE. The Δp LDH, Δp neutrophils, and Δa proteins were not useful on an individual basis. However, combinations of the three variables were useful. Thus, a combination of these variables (that is, the combination of Δp LDH>0%, Δp neutrophil −10% and Δa proteins<0.1g/dL) showed a positive LR of 5.6 (95% CI: 2.8–11.4). However, when at least one of the three conditions had to be met, negative LRs were 0.04 (95% CI: 0.003–0.7). Similar results were observed for discriminating between malignant PE and idiopathic PE (Table 5).

The idiopathic PE group was not considered as independent in the validation population, since it only included six patients. As shown in Table 6, the combination of the three variables that might predict malignant PE, if all three are present, showed a positive LR of 5.2 (95% CI: 2.4–11.1), while if presence of one of the three variables was considered, the negative LR was 0.39 (95% CI: 0.13–1.18); confirming, at least with regard to the combination of all three variables, the results obtained in the derivation population.