Diagnostic predictors of tuberculosis in exudative pleural effusion patients with intermediate adenosine deaminase values: a retrospective diagnostic prediction model study

Adenosine deaminase (ADA) in pleural fluid is widely used to support the diagnosis of tuberculous pleural effusion. However, ADA levels between 40 and 60 IU/L represent a gray area where both tuberculous and non-tuberculous conditions can coexist. It can potentially lead to empirical antitubercular therapy and a delay in the diagnosis of other conditions. The aim of this study was to create and validate a prediction model to improve the diagnosis of tuberculosis (TB) in patients with exudative pleural effusions and intermediate ADA levels. This retrospective diagnostic modeling study included 125 consecutive adults with exudative pleural effusions, meeting Light’s criteria, and pleural fluid ADA levels between 40 and 60 IU/L. These patients were referred to a tertiary respiratory care center in northern India. Univariable receiver operating characteristic curve analysis was used to shortlist candidate predictors with an area under the curve (AUC) of ≥0.60 and p≤0.05. A multivariable logistic regression model was developed using the forward stepwise procedure, including ADA, pleural fluid protein, and patient age. The performance of the model was assessed for discrimination, calibration, and classification and compared to the performance of ADA alone. Internal validation was done using 10-fold stratified cross-validation. Of 125 patients, 77 (62%) had TB and 48 (38%) had non‑TB aetiologies (malignancy ≈30%; acute inflammatory/parapneumonic effusion ≈12%; chronic inflammatory conditions ≈6%). The 3‑predictor model achieved an AUC of 0.89 [95% confidence interval (CI) 0.83-0.95] and an overall accuracy of 85%, with sensitivity 89%, specificity 79%, positive predictive value 84% and negative predictive value 86% at a probability cut‑off of 0.50. ADA alone yielded an AUC of 0.72 (95% CI 0.62-0.82) and accuracy of 68%, with sensitivity 70% and specificity 65%. Ten‑fold stratified cross‑validation showed stable performance (mean cross‑validated AUC 0.87±0.04; optimism 0.02). Independent predictors were ADA [odds ratio (OR) 1.24 per IU/L], protein (OR 1.65 per g/dL) and age (OR 0.93 per year; all p<0.001). In exudative pleural effusions with an ADA level of intermediate values, a simple model combining ADA, pleural protein concentration, and patient age improves the diagnostic discrimination for TB over ADA alone, with a marked improvement in negative predictive probability. External validation in a prospective multicenter study is needed before its use in clinical practice.