The Re-evaluation Of Differential Diagnosis Of Tuberculous Pleural Effusion

Objective: 1.To evaluate factors affecting pleural fluid adenosine deaminase(ADA)levels in tuberculous pleural effusion(TPE),and to determine the optimal ADA levels for diagnosis of TPE.2.To define the value of cancer ratio and other new parameters in the differential diagnosis of TPE.3.To explore the methods of differentiating neutrophil-predominant TPE from parapneumonic pleural effusion(PPE).4.To compare the role of different artificial intelligence machine learning algorithms in TPE differential diagnosis.Methods: 1.Data of 443 patients with pleural effusion were retrospectively collected in Shanghai Jiao Tong University Affiliated Sixth People's Hospital from January 2003 to August 2018.These 443 patients were consisted of 192 patients with TPE,54 patients with PPE,and 197 patients with malignant pleural effusion(MPE).Multivariate logistic regression analysis was used to evaluate the independent factors affecting ADA in patients with TPE.The patients were divided into two groups at the age of 50.The best cut off values and diagnosis efficiencies of ADA were assessed with ROC curve analyses in the two age groups.2.Eight markers including age/pleural fluid ADA ratio(R1),cancer ratio(serum LDH/ pleural fluid ADA ratio,R2),pleural fluid LDH/ pleural fluid ADA ratio(R3),pleural fluid neutrophil(N)/ pleural fluid lymphocyte(L)ratio(R4),pleural fluid ADA / serum CRP ratio(R5),serum LDH / pleural fluid ADA / pleural fluid L% ratio(R6),serum LDH / pleural fluid L% ratio(R7),pleural fluid ADA / serum ADA ratio(R8)were selected and compared among three types of pleural effusion in different age groups.The best cut off values were determined and diagnosis efficiencies were evaluated according to ROC curve.3.The differences in gender,age,symptoms,and laboratory parameters were compared between neutrophil-predominant TPE and PPE.Subgroup analyses by age were also performed.The ROC curves were applied to determine the optimal cut off value.For patients with repeated pleural fluid examinations,the changes of neutrophils and lymphocytes in the pleural fluid and its relationship with the course of the disease were analyzed.4.Using the Alibaba Cloud machine learning platform,the TPE diagnosis models were established with four algorithms: logistic regression,K nearest neighbor(KNN),support vector machine(SVM)and random forest(RF).The diagnosis efficiencies of four algorithms were compared.Results: 1.Age and the percentage of pleural fluid neutrophils were independent influence factors of ADA values in TPE,especially age.ADA in ?50 years old group was significantly higher than that in >50 years old group(P=0.000).The best cut off values of pleural fluid ADA for diagnosis of TPE in the two groups were 17.5U/L and 9.5U/L respectively,with sensitivity of 94.4%,94.1%,and specificity of 80.5%,67.1% respectively.2.When compared with PPE and MPE,patients with TPE had significant lower levels of R1 and R6 in ?50 years old group,R1,R3,R6 and R7 in >50 years old group.AUC of R1 and R6 were the largest in both age groups.In ?50 years old group,the best cut off values of R1 and R6 for diagnosis of TPE were 1.8 and 12.7 respectively.The sensitivity and specificity of R1 and R6 were 89.5%,80.5% and 73.8%,88.9% respectively.In >50 years old group,the best cut off values of R1 and R6 were 6.2 and 23.6.The sensitivity and specificity of R1 and R6 were 94.1%,72.4% and 84.4%,85.9% respectively.3.Compared with PPE,night sweats and chest tightness were more common in neutrophil-predominant TPE.Pleural fluid ADA still had the value in differential diagnosis among the laboratory parameters.In ?50 years old group,pleural fluid ADA combined with pleural fluid LDH had the best discriminating ability with a sensitivity of 84.6% and a specificity of 100%.In >50 years old group,the sensitivity and specificity of ADA in the diagnosis of TPE were 100%,77.4%,respectively.When combined with the percentage of pleural fluid neutrophils,the specificity was increased to 89.3%.Among the patients who received multiple pleural fluid examinations,62.5% of patients in the TPE group changed to lymphocyte-predominant pleural effusion,while in the PPE group,only 10.0% of patients changed to lymphocyte-predominant pleural effusion.The difference between the two groups was statistically significant(P=0.043).4.The sensitivity and specificity of the four TPE diagnostic models were as follows: logistic regression 80.5%,84.8%,KNN 78.6%,86.6%,SVM 83.2%,85.9%,RF 89.1%,93.6%,respectively.Conclusion: 1.Age is an important factor affecting pleural fluid ADA levels in patients with TPE.For patients ?50 years old and > 50 years old,the best cut off values of ADA are 17.5 U/L and 9.5 U/L respectively.The use of lower ADA thresholds to differentiate TPE can reduce false negative in elder patients.2.For patients ? 50 years old,pleural fluid ADA is still the best marker for identifying TPE.For patients > 50 years old,R1,R6 and pleural fluid ADA all have good diagnostic value for TPE,and R6 has the highest specificity.3.In the early stage of TPE,neutrophils can be predominant,which should be differentiated from PPE.For patients ? 50 years old,pleural fluid ADA combined with pleural fluid LDH has the best differential diagnostic value.For patients > 50 years old,pleural fluid ADA combined with the percentage of pleural fluid neutrophils can increase the specificity.Such patients should be given repeated pleural fluid examinations.It is highly suggestive of TPE when pleural effusion with neutrophils predominantly changed to lymphocyte predominantly over time.4.Using artificial intelligence machine learning algorithm to establish models for diagnosing TPE can improve diagnostic efficiency.RF is the best method among logistic regression,KNN and SVM.RF is also superior to pleural fluid ADA and other parameters.Therefore,RF can help clinicians make better diagnosis and treatment decision.