This study firstly applied a quantified nomogram predictive model in TPE and the multivariate nomogram model was demonstrated to be efficient in resource-limited conditions. The model is also easy to use as it enables practitioners to input examination results and predict TPE risk in R software using “DynNom” package. It is meaningful to establish such an efficient and less invasive but also practicable model for clinical TPE diagnosis as thoracoscopy and pathological diagnosis are source-demanding and the direct diagnostic examination is not so sensitive, especially in areas where the tuberculosis load is high.
The direct diagnostic examination, which includes mycobacterium tuberculosis culture, acid fasting staining and nucleic acid examination, has a low sensitivity as mycobacteria is scarce in TPE [9, 10]. Correspondingly, all the mycobacteria culture and acid fasting staining of tuberculous pleural effusion was negative in this study. Nucleic acid amplification test (NAAT) was reported to be more sensitive[11] but still relied on thoracoscopy to provide sufficient biopsy sample.
However, there were scarce quantified models to predict TPE. A study in 2003 reported 2 models with high efficacy. One involved 4 variables, plADA(≥ 40 U/L, 5 points), age (≤ 35 yrs, 2 points), temperature (≥ 37.8°C, 2 points) and pleural red blood cells (≤ 5×109/L, 1 point) while another compromised 6 elements, history of malignancy (none, 3 points), age (≤ 35 yrs, 2 points), temperature (≥ 37.8°C, 2 points), pleural red blood cells (≤ 5×109/L, 1 point), pleural protein (≥ 50 g/L, 1 point) and pleural fluid to serum LDH ratio (≥ 2.2, 1 point)[12]. High sensitivity and specificity though it reported, about 25% still remained unclear. Similarly, another study in 2011 proposed a scoring model, including seven risk factors, temperature (> 38°C ,1.0 point), tuberculin test (positive, 1.0 point), C-reactive protein (≥ 26 mg/L, 1.5 points), pleural fluid lymphocyte percentage (≥ 85%, 1.0 point), pleural fluid protein (≥ 49 g/L, 1.0 point), pleural fluid adenosine deaminase (≥ 43 U/L, 2.5 points) and serum and/or pleural fluid mycobacterium tuberculosis antibody (positive, 2.0 points)[13]. However, those scores for every predictor seemed to be given randomly.
Tumor markers are important tests to exclude malignant pleural effusions as malignancy and tuberculosis are both the major etiologies of exudative pleural effusion. In this study, CA153 and Cyfra21.1 were both included as negative predictors in multivariate nomogram model. CA153 is a glycan-related biomarker and its increasement in serum has been established as a biomarker for both breast and ovarian cancer diagnosis since 1980s [14]. It was also reported that serum CA153 was also a gorgeous indicator for lung cancer, lymphoma and some other noncancer diseases in a study involved 19,789 healthy individuals and patients [15]. Cyfra21-1, cytokeratin 19 fragments, has also been demonstrated to be the most sensitive tumor marker in non-small cell lung cancer, particularly the squamous cell type [16]. PlCEA was also believed to be a highly specific tumor marker associated with MPE [17, 18] but was excluded by multivariate logistic regression possibly due to collinearity with other predictors.
Thoracentesis and pleural effusion analysis are less-invasive and easier to perform than thoracoscopy or percutaneous lung biopsy.
ADA is a ubiquitous enzyme to deaminate adenosine and 2-deoxyadenosine, with highest activity in lymphoid tissues [19]. It plays a key role in adenosine homeostasis and thus immunomodulation [20]. Besides inside the cells, it also presents on the cell surface and could trigger T-cell coactivation. Furthermore, it was reported that it could potentiate CD4 + T-cell differentiation and proliferation, including effector T cells, memory T cells and regulatory T cells [21].
It has been widely reported that the elevated ADA activity in pleural effusion was an efficient biomarker of TPE, with a commonly used cut-off value at 40 IU/L. As a recent meta-analysis reported, 40 IU/L was more suitable to be an excluding threshold rather than confirmation while the diagnostic utility of plADA cut-off value depended on the geographical regions and clinical backgrounds [22]. In Brazil, where tuberculosis prevalence was reported to be about 0.077% and TPE accounted for approximately 50% in all etiologies, a meta-analysis also suggested 40 IU/L as an excluding threshold and a lower cutoff value at about 30IU/L accompanied by other examinations may be superior [23]. Correspondingly, 35 IU/L was also more recommended in Spanish [24]. It was also implied that growing threshold value couldn’t remarkably improve diagnostic performance once over 40 IU/L in an India meta-analysis [25]. This study proposed a cut-off value at 24.5 IU/L through univariate ROC analysis, which was lower than most commonly used value but closer to results mentioned above. Comparably, a recent study in China also recommended a lower threshold, as half of TPE showed plADA value below 40 IU/L [26]. Meanwhile, a Hongkong research recommended 26.5 IU/L as a TPE threshold when plADA level was less than 100 IU/L if without evidence of malignancy and non-tuberculous infection [27]. Similarly, a Taiwan study in which included lymphocyte-dominant exudative pleural effusions also indicated a threshold of 30 IU/L to differentiate TPE from MPE as the Asian population had lower ADA [28]. Taken together, a lower cut-off value of plADA may be more acceptable in Asian population to discriminate TPE from others, which also remains to be further investigation.
However, elevated plADA can also occur in many other etiologies, such as neutrophilic effusions, lymphomas and rheumatic inflammation [29–32]. Meanwhile, it can also be lower in older or critically ill groups [33]. Accordingly, combination with other tests is necessary for more accurate diagnosis.
LDH is an essential metabolic enzyme found in nearly all living cells. It was reported that plLDH was significantly higher in MPE or parapneumonic pleural effusions [34–36] than TPE, which was consistent with this study.
Interstitial cells in the pleural effusion may reveal the pleural mesothelium apoptosis or necrosis. The higher percentage in Non-TPE group may related to more invasiveness of malignant and infectious diseases than tuberculous delayed hypersensitivity.
Medical thoracoscopy is usually performed after thoracentesis implying no definite diagnosis. however, it is also resource-demanding and more invasive. Combining above 5 independent predictors, the multivariate nomogram model performed approximate accuracy to thoracoscopy (622/632) in the training cohort. Furthermore, it indeed added more benefit for patients with TPE than commonly used plADA40 model.
However, there were also some potential limitations. Firstly, the retrospective nature leaved some information inaccessible and there were about 16.3% (109/662) cases with missing data in training cohort although multiple imputation was performed and clinicopathological information comparison between patients with or without part data not available showed no significant difference in most clinicopathological characteristics (Supplement Table. S1) and LASSO selected the same 7 factors in primary completed cohort as imputed completed cohort. Secondly, participants recruited were from a university hospital rather than multiple centers, which leaved the model to be further tested or adjusted in other centers.