ABSTRACT

Objectives: Pleural tuberculosis (PlTB) diagnosis is a challenge due to its paucibacillary nature and to the need of invasive procedures. This study aimed to identify easily available variables and build a predictive model for PlTB diagnosis which may allow earlier and affordable alternative strategy to be used in basic health care units. Methods: An observational cross-sectional study compared PlTB and non-TB patients followed at a tertiary Brazilian hospital between 2010 and 2018. Unconditional logistic regression analysis was performed and a Decision Tree Classifier (DTC) model was validated and applied in additional PlTB patients with empiric diagnosis. The accuracy (Acc), sensitivity (Se), specificity (Sp), positive and negative predictive values were calculated. Results: From 1,135 TB patients, 160 were considered for analysis (111 confirmed PlTB and 49 unconfirmed PlTB). Indeed, 58 non-TB patients were enrolled as controls. Hyporexia [adjusted odds ratio (aOR) 27.39 (95% CI 6.26 - 119.89)] and cellular/biochemical characteristics on pleural fluid (PF) (polimorphonuclear in two categories: 3-14% aOR 26.22, 95% CI 7.11 - 96.68 and < 3% aOR 28.67, 95% CI 5.51 - 149.25; and protein = 5g/dL aOR 7.24, 95% CI 3.07 - 17.11) were associated with higher risk for TB. The DTC constructed using these variables showed Acc=87.6%, Se=89.2%, Sp=84.5% for PlTB diagnosis and was successfully applied in unconfirmed PlTB patients. Conclusion: The DTC model showed an excellent performance for PlTB diagnosis and can be considered as an alternative diagnostic strategy by using clinical patterns in association with PF cellular/biochemical characteristics, which were affordable and easily performed in basic health care units.

Keywords: Pleural tuberculosis; Diagnosis; Decision tree classifier.

RESUMO

Objetivo: o diagnóstico da tuberculose pleural (TBpl) é um desafio devido à sua natureza paucibacilar e à necessidade de procedimentos invasivos. Este estudo visou identificar variáveis facilmente disponíveis, bem como construir um modelo preditivo para o diagnóstico da TBpl que possa permitir a utilização de uma estratégia alternativa precoce e acessível em unidades de cuidados básicos de saúde. Métodos: um estudo transversal observacional comparou pacientes com TB e não TB seguidos em um hospital terciário brasileiro entre 2010 e 2018. Foi realizada uma análise de regressão logística incondicional que foi validado e aplicado em um modelo de classificação de árvore de decisão (CAD) em pacientes com diagnóstico empírico de TB. Foram calculados precisão (Prec), sensibilidade (Se), especificidade (Esp), assim como os valores preditivos positivos e negativos. Resultados: de 1.135 pacientes com diagnóstico de TB, 160 foram considerados para análise (111 com TBpl confirmada e 49 com TBpl não confirmada. 58 pacientes não TB foram considerados como controles. Hiporexia [odds ratio ajustado (ORa) 27,39 (95% IC 6,26-119,89)] e características celulares/bioquímicas do líquido pleural (LP) (polimorfonuclear em duas categorias: 3%-14% ORa 26,22, 95% IC 7,11-96,68; e < 3% ORa 28,67, 95% IC 5,51-149,25; e proteína ≥ 5g/dL ORa 7,24, 95% IC 3,07-17,11) foram associadas a um risco mais elevado de tuberculose. O diagnóstico de CAD construído ao empregar essas variáveis mostrou valores de Prec = 87,6%, Se = 89,2%, Esp = 84,5% para o diagnóstico de TB e foi aplicado com sucesso em doentes com TB não confirmada. Conclusão: o modelo CAD mostrou um excelente desempenho para o diagnóstico de TB e pode ser considerado como estratégia alternativa de diagnóstico, utilizando padrões clínicos em associação com características celulares/bioquímicas do LP, que são acessíveis e facilmente realizados em unidades de cuidados básicos de saúde.

Palavras-chave: Tuberculose pleural; Diagnóstico; Classificação em árvores de decisão.

INTRODUCTION
 
Tuberculosis (TB) is a major public health problem worldwide.(1) Pleural tuberculosis (PlTB) is the commonest extrapulmonary TB presentation, representing 42% of all extrapulmonary cases.(2,3) In addition to its frequency, PlTB diagnosis can be a challenge due to the paucibacillary nature of patients’ biological samples and to the need for invasive procedures.(4)
 
The Ziehl Neelsen (ZN) stain and the mycobacterial culture of pleural fluid (PF) and/or pleural tissue are the gold standard methods for tuberculous pleural effusion diagnosis, but their success rates are relatively poor.(5) The identification of granuloma in histopathological examination of the pleural tissue is also considered as diagnostic criteria.(6) However, besides being invasive, the pleural biopsy is operator-dependent, relatively expensive, may be limited by clinical contraindications and is associated with complications.(7-9) Adenosine deaminase (ADA) measurement in pleural fluid can provide a putative diagnosis of PlTB in high prevalence settings, considering that high ADA levels can also be observed in other infectious, inflammatory or malignant diseases.(6,10,11) In addition, the evaluation of biomarkers on pleural effusion configures an alternative for TB diagnosis.(12-14) Recently, our research group have shown that interferon-gamma (IFN-γ) was an excellent rule-in and rule-out test compared to other two biomarkers (IFN-γ inducible protein 10 kD and ADA) and that the combination of IFN-γ and ADA, in a reviewed cut-off point, showed to be particularly useful to PlTB confirmation.(15) However, these methods can be available only at reference centers, whereas basic health care units cannot count on them.
 
In order to surpass the limitations stated forementioned, which compels the practice of a not uncommon empiric diagnosis based on clinical and radiological criteria, as previously published,(16-19) our research group proposed the application of conventional biochemical and cellular parameters of pleural fluid, which lacks specificity when isolated considered,(4) but when in combination with each-other or with clinical features could discriminate between tuberculous pleural effusion or not. For this purpose, a predictive model for PlTB diagnosis was validated by us based on hyporexia, polymorphonuclear cells (PMN cells) and protein levels on pleural fluid, which was applied on empiric PlTB cases to perform an internal validation.
 
METHODS
 
Study design
 
A cross-sectional study was conducted based on medical records of patients with pleural effusion under investigation attended between January 2010 and January 2018 at Pedro Ernesto University Hospital from Rio de Janeiro State University (Hospital Universitário Pedro Ernesto/Universidade do Estado do Rio de Janeiro (HUPE/UERJ)), in the city of Rio de Janeiro (RJ), Brazil. Patients at least 18 years and who attended the Outpatient Clinic of Tuberculosis of HUPE with PlTB diagnosis were included. Those patients whose TB treatment outcomes were unknown (loss of follow-up) were excluded. Patients with non-TB pleural effusion were considered controls and were drawn from the Outpatient Clinic of Pleural Diseases of the same hospital.
 
The ethics committee approved the study of the institution under protocol #2612/2010.
 
Diagnostic criteria
 
PlTB was stratified as follows: i) Confirmed PlTB (C-PlTB): ZN stain positive or isolation of Mycobacterium tuberculosis on the respiratory specimen, pleural fluid or pleural tissue, or identification of granuloma on histopathological analysis, or patients with clinical manifestations of PlTB and a lymphocytic and exudative pleural effusion with ADA dosage above 40 IU/L that fully recover after at least six months of antituberculosis treatment; ii) Unconfirmed PlTB (UC-PlTB): cases with clinical manifestations of PlTB that did not fulfill the C-PlTB and fully recovered after at least six months of antituberculosis treatment.
 
Non-tuberculosis (Non-TB): patients were defined as those with pleural or pleuropulmonary diseases other than TB in which the diagnose was concluded based on clinical, laboratory, radiological, microbiological, or cytopathological/histopathological features.
 
Data collection
 
The medical records of all patients were reviewed in order to evaluate physical, clinical, and demographic information, medical history, and laboratory data. According to subjective reported presence and duration, signs and symptoms such as cough, fever, chest pain, dyspnea, night sweats, hyporexia, and weight loss were included. The Human Immunodeficiency Virus (HIV) status and other comorbidities were also recorded. Data on PF routine diagnostic tests, including a chemistry panel, total and differential cell count, ADA measurement, cytopathology, and microbiological analysis (ZN and solid media culture) were registered. In those cases where pleural biopsy with Cope’s needle was performed, histopathology analysis, ZN stain, and mycobacterial culture results were registered.
 
Statistical analysis
 
For statistical analysis, categorical and continuous data were presented in frequency (percentage) and median and interquartile range (IQR), respectively. Fisher Exact Test and T-test were used for group comparisons. Simple and multiple unconditional logistic regression analyses were performed, and odds ratios (OR) and adjusted OR (aOR) with its 95% Confidence Interval (95% CI) were calculated. The level of significance used was 0.05.
 
Decision Trees Classifier (DTC) was selected as a predictive model because it allows a straightforward and easy interpretation of the rules, and it was built with an implementation of the Quinlan’s C4.5 algorithm available in the packages ‘rpart’ version 4.1-10 for the open-source software R version 3.3.1.(20) The DTC was built with cases of C-PlTB and non-TB in a subset of nine [viral hepatitis, fever, dyspnea, hyporexia, weight loss, percentage of PMN cells, percentage of mononuclear cells (MN cells), protein and albumin on pleural fluid] of 22 variables pre-selected (age, diabetes mellitus, renal failure, cardiac failure, cancer, previous transplant, viral hepatitis, autoimmune disease, immunosuppressive treatment, fever, cough, chest pain, dyspnea, weight loss, hyporexia, night sweating, chest X-ray, percentage of PMN cells, percentage of MN cells, protein and albumin on pleural fluid) by successively eliminating at least 20% important variables (with importance as returned from random forest) using the out-of-bag error as minimization criterion from random forest (number of trees equal to 1,000) classifiers implemented in the R package ‘varSelRF’ version 0.7-5. The fitted tree was pruned to minimize the expected 10-fold cross-validation prediction accuracy. Pruning included a complexity parameter of 0.25, informing the algorithm that any split that does not improve the fit by 25% will likely be pruned off by 10-fold cross-validation. Hence, the algorithm needs not to pursue it. In addition, the performance of the built DTC was estimated by its leave-one-out cross-validation accuracy (Acc), sensitivity (Se), specificity (Sp), Positive Predictive Value (PPV) and Negative Predictive Value (NPV), and false-positive and negative ratios with 95% CI and area under the Receiver Operating Characteristics (ROC) and the Area Under the Curve (AUC).
 
RESULTS
 
From January 2010 to January 2018, 1,135 patients were diagnosed with active TB at HUPE/UERJ. Of these, 397 had an extrapulmonary presentation of TB, and 212 (53%) had PlTB, being 160 patients considered for analysis. The other 52 patients were excluded according to the exclusion criteria. The 58 non-TB cases included 34 malignancies (16 adenocarcinomas, four lymphomas, two carcinomas, one spindle epithelial, and 11 non-specified cell types), 11 renal/cardiac failures, four empyema, two systemic lupus erythematosus, one hepatic disease, one chylothorax and five cases of undefined pleural effusion (Figure 1).

 
The sociodemographic and clinical characteristics from all patients are shown in Table 1. PlTB patients were younger than non-TB and were less likely to present other comorbidities such as cancer, cardiac failure, viral hepatitis, and previous transplant. In addition, fever, chest pain, hyporexia, weight loss, and night sweats were more frequently observed in PlTB cases and higher levels of ADA measurement, MN cells frequency, total protein, and albumin levels, and reduced PMN cells frequency. Confirmed and unconfirmed PlTB were homogeneous unless for age (Table 1). Compared to non-TB, younger patients presented a higher risk for TB diagnosis (30 – 44 years: OR 4.6, 95% CI 1.63 – 12.99; 18 – 29 years: OR 6.16, 95% CI 1.79 – 21.16). This association persisted in multiple unconditional logistic analyses adjusting for gender, comorbidities, signs/symptoms, radiological appearance and cytological, and biochemical pleural fluid analyses (Table 2).

 
Fever, chest pain, weight loss, and hyporexia were the signs/symptoms considered risk factors for confirmed PlTB. Of these, hyporexia showed the highest aOR [27.39 (95% CI 6.26 – 119.89)]. Cytological characteristics on PF showed a reverse behavior once a reduced frequency of PMN cells has exhibited a greater chance of being diagnosed with PlTB (PMN cells 3 – 14%: aOR 8.78, 95% CI 3.35-22.97 and PMN cells < 3%: aOR 28.67, 95% CI 5.51-149.25) and the higher the percentage range of MN cells the greater the chance of PlTB diagnosis (MN cells 85 – 97.5%: aOR 9.04, 95% CI 3.34 – 24.42 and MN cells ≥ 97.5%: aOR 33.67, 95% CI 6.26 – 181.01). Protein levels above 5g/dL were also identified as a risk factor for the infectious disease (aOR 7.24, 95% CI 3.07 – 17.11) (Table 2).
 
From twenty-two pre-selected variables, our model chose nine to DTC building, which has included viral hepatitis, fever, dyspnea, hyporexia, weight loss, percentage of PMN cells, percentage of MN cells, protein, and albumin on PF. By the end of the analysis, the pruned DTC used only three predictive variables for PlTB discrimination: PMN cells and protein in PF and hyporexia.
 
Figure 2 depicts the DTC analysis built to discriminate confirmed PlTB from non-TB patients. Based on three variables’ results (presence of hyporexia, PMN cells levels, and protein levels, both on PF), in 21 cases (12 PlTB and nine non-TB), the diagnostic classification of the DTC was mistaken. In these cases, the results of microbiological tests, pleural tissue biopsy, and ADA dosage were of extreme importance for differential diagnosis.

 
With an area under the ROC curve of 88.7%, the DTC proved to be 90.23% accurate with a Se of 92.47% (95% CI 88.99-95.95%) and a Sp of 87.6% (95% CI 84.4-90.7) (Table 3). The performance of the DTC applied for unconfirmed PlTB classification to perform an internal validation was 88.8% accurate, 93.9% sensitive, and 84.5% specific (Table 3). Noteworthy, almost 90% of the cases would be correctly treated by applying the DTC as a diagnostic tool and using laboratory and clinical variables, which can be afforded and performed in a basic care health system.

 
DISCUSSION
 
The challenge of differential diagnosis of tuberculous pleural effusion among a range of etiologies encouraged a great number of studies that employed efforts to develop tools for differentiating PlTB from other entities using scoring systems, DTC, and artificial intelligence neural networks.(21-27) Our DTC behaved as well as them, with excellent accuracy. Besides that, we consider our model, with only three predictive variables, and applied to a broad spectrum of patients, easier to use in clinical practice. Moreover, the DTC presented here was validated using an independent sample of patients, leading to almost 90% of cases being correctly classified.
 
In this study, a DTC model was built with excellent accuracy for discriminating PlTB in an area with a high incidence of TB. The performance of the DTC was similar when either applied on C-PlTB or unconfirmed PlTB cases, with sensitivity higher than 89%, for both groups. Noteworthy, the sensitivity of the proposed DTC model is much higher than either microbiological or histopathological analyses of pleural specimens for TB diagnosis.(5,28) Moreover, the decision tree presented here can also be considered an alternative to ADA dosage, given the various performances of ADA for PlTB diagnosis(11,29,30) and the lack of availability of the test in some scenarios. In addition, physicians would be more confident to initiate anti TB therapy in cases empirically diagnosed using this DTC in scenarios without other diagnostic methods available.
 
To the best of our knowledge, there are only two studies published that have proposed DTCs that could be used in clinical practice.(26,27) In 2008, a similar model was used to discriminate between tuberculous and malignant pleural effusions based on four parameters: age > 35 years, ADA > 38 IU/L, temperature, and DHL on pleural fluid.(26) This model was 92.2% sensitive and 98.3% specific, and the validation using an independent sample showed a sensitivity of 85% and a specificity of 97%. Even though TB and cancer are the two most frequent causes of exudative pleural effusions,(31) other etiologies must be discarded during pleural diseases diagnosis. In our study, among the 58 non-TB patients, 41.4% had other etiologies than cancer. Valdés et al.(27) proposed a DTC to classify PF as tuberculous or non-tuberculous. The first proposed model included pleural fluid lymphocyte count > 31.5% and ADA > 35 IU/L with a mean accuracy of 99%. Then, to be applied in health care settings without the availability of ADA, a second model, including PF lymphocyte count > 31.5%, fever, and cough, showed to be less accurate than the first one. Only patients under 40 years old were included in that study, whereas our study population had patients from 18 to 89 years and more than 50% were older than 45.
 
Our statistical model identified PMN cells percentage instead of MN cells included in the DTC, although lymphocytic pleural effusions are most typical among tuberculous pleural fluid analyses.(9) However, in many cases, PMN rather than MN are the predominant cell type in the pleural effusion of TB cases, especially during the earlier phases of the pleural inflammatory process, as shown by Jeon et al..(9) According to Lyadova,(32) neutrophils are probably the least understood among immune cell populations, playing a dual role during TB pathophysiology. It is already known that these cells participate in the acquired immunity and granuloma formation and may kill Mtb. Notwithstanding this characteristic, at the same time, neutrophils could support mycobacterial growth and have been implicated in the transition from infection to active TB, mediate tissue destruction, disease severity, and progression.(32) However, recently published studies(33-36) showed that TB disease alters the neutrophil population, leading to the accumulation of heterogeneous subsets of immature and activated dysfunctional cells with a decline in true neutrophils.
 
Based on the observations in the present study, it would be relevant not to reduce the importance of lymphocytes but to bring out the role of neutrophils in diagnosing TB. In our study, 13 confirmed PlTB cases (13/111; 12%) had a frequency of neutrophils higher than 15%. Lin et al.(37) showed that among 354 tuberculous pleural effusion, 39 (11%) presented a PMN predominance in the pleural fluid. Interestingly, these patients presented a high mortality rate and a high risk for transmission.
 
Moreover, in our study, the addition of total protein levels of pleural fluid on the DTC contributed to correct discrimination of PlTB, considering that according to Choi et al.,(38) PlTB with predominant PMN in PF showed a more intense inflammatory response with higher levels of total protein and albumin. As already indicated, since Light’s criteria, an exudative pleural effusion is characteristic of inflammatory pleural diseases, including TB.(39) Our findings can be compared to those of Samanta et al.(30) that showed higher levels of total protein and albumin on the pleural effusion of TB patients than in lung cancer patients.
 
All of the most typical signs and symptoms of PlTB and other prevalent pleural diseases, namely fever, cough, chest pain, dyspnea, hyporexia, weight loss, and night sweat, were considered in the training of the DTC here proposed. Among them, just hyporexia was preserved in the final model. This symptom also showed the highest aOR for PlTB in the multiple unconditional logistic regression and was included in the DTC model (Figure 2). However, a literature search did not find any other publications that identified hyporexia as an essential clinical presentation for PlTB diagnosis and other classical signs/symptoms.
 
There were some limitations in our study, most of which were attributable to the use of the information from a retrospective cohort. We were dependent on documentation and interpretation of data in the medical records. However, data were reported in medical records according to standard questionnaires by the attending physicians of a university institution, where physicians, nurses, and students were trained to document data that could be used for research. In addition, many missing data regarding biochemical and cellular features of PF (40 cases) were identified. Besides that, the absence of data, which can happen in routine, was included in the DTC. Maybe, with more results available, our model could present a higher precision. Also, most non-TB cases had cancer (58.6%), with a minority of cases composed of less frequent pleural conditions such as empyema and autoimmune diseases. Our University Hospital is not a reference center for HIV patients, and this could explain the few cases of co-infected patients. Also, although some readers understand that hyporexia and protein levels in the DTC can be considered an incorporation bias, we comprehend that these isolated variables were not used to confirm or exclude TB diagnosis. Even if this bias could be found, it would not invalidate the results and only overestimate the DTC accuracy. Finally, but not a limitation of our study, but of the DTC model proposed itself, TB prevalence can affect the performance of this model in different scenarios and should be used with caution in low-prevalence settings.
 
Recently, clinical pathways have been used to optimize patient care for specific clinical conditions. Mummadi and Hahn(40) published their experience applying what they called the “pleural pathway” in an institution in the United States of America (USA). They concluded that this “pleural pathway” and a centralized pleural service are associated with reducing case charges, inpatient admissions, and length of stay for pleural conditions. We believe that the DTC proposed in this article could be included as a screening test to organize steps of diagnosis without neglecting the possibility of less frequent presentations of tuberculous pleural effusion.
 
The DTC based on pleural fluid cellular and biochemical characteristics does not replace microbiological tests for TB. Its disadvantage is associated with the failure to provide microbiological confirmation and antituberculosis sensitivity tests. Thus, we still advocate that pleural fluid and the respiratory specimen should also be obtained whenever possible. Additionally, pathological analysis of pleural fluid and fragment may be indicated for cases with high suspicion of cancer.(26) However, the variables selected by this model can be readily available in primary health care units, which cannot count on ADA’s dosage for PlTB diagnosis on a routine basis.
 
In conclusion, the DTC model proposed in this study could identify PlTB cases based on only three simple predictive variables readily available and collected in basic health care units, with a mean accuracy of almost 90%. ADA, microbiological and pathological analysis of pleural fluid should also be obtained whenever possible. Invasive diagnostic procedures, such as pleural biopsy may be reserved for specific cases considering their risks versus benefits.
 
ACKNOWLEDGMENTS
 
We would like to thank Dr. Thiago Mafort, Dr. Domenico Capone, Dr. Walter Costa and Dr. Hélio Siqueira, physicians of Pleural Diseases and Tuberculosis Outpatient Clinics of HUPE that contributed in the medical care of the patients included in the study.
 
AUTHOR CONTRIBUTIONS
 
APS: idealization, investigation, data collection, writing of the paper. MRA: methodology, statistical analysis, writing of the paper. RC, IL, MAS e JL: data collection, writing of the paper. TTM: investigation, data collection, writing of the paper. LSR and RR: idealization, methodology, statistical analysis, final revision of the paper.
 
REFERENCES
 
1.            WHO: World Health Organization. [homepage on the Internet]. Geneve: WHO; 2017. [cited 2021 Mar 24]. Available in: https://www.who.int/health-topics/tuberculosis
2.            Gomes T, Reis-Santos B, Bertolde A, Johnson JL, Riley LW, Maciel EL. Epidemiology of extrapulmonary tuberculosis in Brazil: a hierarchical model. BMC Infect Dis. 2014;14(1):9. http://dx.doi.org/10.1186/1471-2334-14-9. PMid:24400848.
3.            Light RW. Update on tuberculous pleural effusion. Respirology. 2010;15(3):451-8. http://dx.doi.org/10.1111/j.1440-1843.2010.01723.x. PMid:20345583.
4.            Greco S, Girardi E, Masciangelo R, Capoccetta GB, Saltini C. Adenosine deaminase and interferon gamma measurements for the diagnosis of tuberculous pleurisy: a meta analysis. Int J Tuberc Lung Dis. 2003;7(8):777-86. PMid:12921155.
5.            Gopi A, Madhavan SM, Sharma SK, Sahn SA. Diagnosis and treatment of tuberculous pleural effusion in 2006. Chest. 2007;131(3):880-9. http://dx.doi.org/10.1378/chest.06-2063. PMid:17356108.
6.            Trajman A, Pai M, Dheda K, van Zyl Smit R, Zwerling AA, Joshi R, et al. Novel tests for diagnosing tuberculous pleural effusion: what works and what does not? Eur Respir J. 2008;31(5):1098-106. http://dx.doi.org/10.1183/09031936.00147507. PMid:18448504.
7.            Lee LN, Chou CH, Wang JY, Hsu HL, Tsai TH, Jan IS, et al. Enzyme-linked immunospot assay for interferon-gamma in the diagnosis of tuberculosis pleurisy. Clin Microbiol Infect. 2009;15(2):173-9. http://dx.doi.org/10.1111/j.1469-0691.2008.02655.x. PMid:19154487.
8.            Liu F, Gao M, Zhang X, Du F, Jia H, Yang X, et al. Interferon-gamma release assay performance of pleural fluid and peripheral blood in pleural tuberculosis. PLoS One. 2013;8(12):e83857. http://dx.doi.org/10.1371/journal.pone.0083857. PMid:24386296.
9.            Jeon D. Tuberculous pleurisy: an update. Tuberc Respir Dis. 2014;76(4):153-9. http://dx.doi.org/10.4046/trd.2014.76.4.153. PMid:24851127.
10.          Liang Q-L, Shi H-Z, Wang K, Qin S-M, Qin X-J. Diagnostic accuracy of adenosine deaminase in tuberculous pleurisy: a meta-analysis. Respir Med. 2008;102(5):744-54. http://dx.doi.org/10.1016/j.rmed.2007.12.007. PMid:18222681.
11.          Lima DM, Colares JKB, Fonseca BAL. Combined use of the polymerase chain reaction and detection of adenosine deaminase activity on pleural fluid improves the rate of diagnosis of pleural tuberculosis. Chest. 2003;124(3):909-14. http://dx.doi.org/10.1378/chest.124.3.909. PMid:12970016.
12.          Shimokata K, Saka H, Murate T, Hasegawa Y, Hasegawa T. Cytokine content in pleural effusion. Comparison between tuberculous and carcinomatous pleurisy. Chest. 1991;99(5):1103-7. http://dx.doi.org/10.1378/chest.99.5.1103. PMid:1902158.
13.          Nemeth J, Winkler HM, Zwick RH, Rumetshofer R, Schenk P, Burghuber OC, et al. Recruitment of Mycobacterium tuberculosis specific CD4+ T cells to the site of infection for diagnosis of active tuberculosis. J Intern Med. 2009;265(1):163-8. http://dx.doi.org/10.1111/j.1365-2796.2008.02012.x. PMid:18793244.
14.          Sutherland JS, Garba D, Fombah AE, Mendy-Gomez A, Mendy FS, Antonio M, et al. Highly accurate diagnosis of pleural tuberculosis by immunological analysis of the pleural effusion. PLoS ONE. 2012;7(1): e30324.
15.          Santos AP, Corrêa RS, Ribeiro-Alves M, Silva ACOS, Mafort TT, Leung J, et al. Application of Venn’s diagram in the diagnosis of pleural tuberculosis using IFN-γ, IP-10 and adenosine deaminase. PLoS ONE. 2018;13(8):e0202481.
16.          Sané M, Ba-Fall K, Lefebvre N, Mounguengui D, Camara P, Niang A, et al. Empiric antituberculous therapy in patients with unexplained exsudative pleural effusion. Is it valid in Senegal? Rev Pneumol Clin. 2007;63(4):247-50. PMid:17978735.
17.          Seibert AF, Haynes J Jr, Middleton R, Bass JB Jr. Tuberculous pleural effusion: twenty-year experience. Chest. 1991;99(4):883-6. http://dx.doi.org/10.1378/chest.99.4.883. PMid:1901261.
18.          Richter C, Perenboom R, Swai AB, Kitinya J, Mtoni I, Chande H, et al. Diagnosis of tuberculosis in patients with pleural effusion in an area of HIV infection and limited diagnostic facilities. Trop Geogr Med. 1994;46(5):293-7. PMid:7855915.
19.          Zhai K, Lu Y, Shi HZ. Tuberculous pleural effusion. J Thorac Dis. 2016;8(7):E486-94. http://dx.doi.org/10.21037/jtd.2016.05.87. PMid:27499981.
20.          R Core Team. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing.
21.          Carrión-Valero F, Perpiñá-Tordera M. Screening of tuberculous pleural effusion by discriminant analysis. Int J Tuberc Lung Dis. 2001;5(7):673-9. PMid:11467374.
22.          Neves DD, Dias RM, Cunha AJLA. Predicitive model for the diagnosis of tuberculous pleural effusion. Braz J Infect Dis. 2007;11(1):83-8. http://dx.doi.org/10.1590/S1413-86702007000100019. PMid:17625733.
23.          Porcel JM, Vives M. Differentiating tuberculous from malignant pleural effusions: a scorring model. Med Sci Monit. 2003;9(5):CR175-80. PMid:12761453.
24.          Sales RK, Vargas FS, Capelozzi VL, Seiscento M, Genofre EH, Teixeira LR, et al. Predictive models of pleural effusions secondary to tuberculosis or cancer. Respirology. 2009;14(8):1128-33. http://dx.doi.org/10.1111/j.1440-1843.2009.01621.x. PMid:19909461.
25.          Seixas JM, Faria J, Souza JBO Fo, Vieira AF, Kritski A, Trajman A. Artificial neural network models to support the diagnosis of pleural tuberculosis in adult patients. Int J Tuberc Lung Dis. 2013;17(5):682-6. http://dx.doi.org/10.5588/ijtld.12.0829. PMid:23575336.
26.          Porcel JM, Alemán C, Bielsa S, Sarrapio J, Sevilla TF, Esquerda A. A decision tree for differentiating tuberculosis from malignant pleural effusions. Respir Med. 2008;102(8):1159-64. http://dx.doi.org/10.1016/j.rmed.2008.03.001. PMid:18573650.
27.          Valdés L, San José JM, Pose A, Gude F, González-Barcala FJ, Alvarez-Dobaño JM, et al. Diagnosing tuberculous pleural effusion using clinical data and pleural fluid analysis. A study of patients less then 40 years old in an area with a high incidence of tuberculosis. Respir Med. 2010;104(8):1211-7. http://dx.doi.org/10.1016/j.rmed.2010.02.025. PMid:20347287.
28.          Brasil. Ministério da Saúde. Departamento de Vigilância Epidemológica. Manual de recomendações para o controle da tuberculose no Brasil / Ministério da Saúde, Secretaria de Vigilância em Saúde, Departamento de Vigilância Epidemiológica. Brasília: Ministério da Saúde; 2011
29.          Valdés L, San José E, Alvarez D, Sarandeses A, Pose A, Chomón B, et al. Diagnosis of tuberculosis pleurisy using the biologic parameters adenosine deaminase, lysosyme and interferon gamma. Chest. 1993;103(2):458-65. http://dx.doi.org/10.1378/chest.103.2.458. PMid:8432137.
30.          Villegas MV, Labrada LA, Saravia NG. Evaluation of polymerase chain reaction, adenosine deaminase and interferon gamma in pleural fluid for the differential diagnosis of pleural tuberculosis. Chest. 2000;118(5):1355-64. http://dx.doi.org/10.1378/chest.118.5.1355. PMid:11083686.
31.          Villena V, Encuentra AL, Echave-Sustaeta J, Alvarez Martínez C, Escribano PM. Prospective study of 1,000 consecutive patients with pleural effusion. Etiology of the effusion and characteristics of the patients. Arch Bronconeumol. 2001;38(1):21-6. http://dx.doi.org/10.1016/S0300-2896(02)75142-9. PMid:11809133.
32.          Lyadova IV. Neutrophils in tuberculosis: heterogeneity shapes the way? Mediators of Inflammation. 2017;2017:8619307.
33.          Martineau AR, Newton SM, Wilkinson KA, Kampmann B, Hall BM, Nawroly N, et al. Neutrophil-mediated innate immune resistance to mycobacteria. J Clin Invest. 2007;117(7):1988-94. http://dx.doi.org/10.1172/JCI31097. PMid:17607367.
34.          Reyes-Ruvalcaba D, González-Cortés C, Rivero-Lezcano OM. Human phagocytes lack the ability to kill Mycobacterium gordonae, a non-pathogenic mycobacteria. Immunol Lett. 2008;116(1):72-8. http://dx.doi.org/10.1016/j.imlet.2007.11.010. PMid:18160107.
35.          Corleis B, Korbel D, Wilson R, Bylund J, Chee R, Schaible EU. Escape of Mycobacterium tuberculosis from oxidative killing by neutrophils. Cell Microbiol. 2012;14(7):1109-21. http://dx.doi.org/10.1111/j.1462-5822.2012.01783.x. PMid:22405091.
36.          Ramos-Kichik V, Mondragón-Flores R, Mondragón-Castelán M, Gonzalez-Pozos S, Muñiz-Hernandez S, Rojas-Espinosa O, et al. Neutrophil extracellular traps are induced by Mycobacterium tuberculosis. Tuberculosis. 2009;89(1):29-37. http://dx.doi.org/10.1016/j.tube.2008.09.009. PMid:19056316.
37.          Lin MT, Wang JY, Yu CJ, Lee LN, Yang PC. Mycobacterium tuberculosis and polymorphonuclear pleural effusion: incidence and clinical pointers. Respir Med. 2009;103(6):820-6. http://dx.doi.org/10.1016/j.rmed.2008.12.023. PMid:19217271.
38.          Choi H, Chon HR, Kim K, Kim S, Oh KJ, Jeong SH, et al. Clinical and laboratory differences between lymphocyte- and neutrophil-predominant pleural tuberculosis. PLoS One. 2016;11(10):e0165428. http://dx.doi.org/10.1371/journal.pone.0165428. PMid:27788218.
39.          Light RW, Macgregor I, Luchsinger C, Ball WC Jr. Pleural effusions: the diagnostic separation of transudates and exsudates. Ann Intern Med. 1972;77(4):507-13. http://dx.doi.org/10.7326/0003-4819-77-4-507. PMid:4642731.
40.          Mummadi SR, Hahn PY. Outcomes of a clinical pathway for pleural disease management: “pleural pathway”. Pulmonary Medicine. 2018;2018:2035248.