Abstract
BACKGROUND: Accurate differentiation between malignant and benign pleural effusions remains a diagnostic challenge, often requiring invasive procedures. This study aims to identify predictive markers and develop a model based on clinical, laboratory, and radiological parameters to distinguish non-tuberculous benign pleural effusions (NT-BPE) in patients with undiagnosed pleural effusion after thoracentesis but before pleural biopsy. METHODS: This study was designed as a retrospective predictive modeling analysis using prospectively collected patient data. Patient cohorts were derived from five studies prospectively conducted between 2008 and 2023, comprising four randomized controlled trials and one observational study, with a total of 684 individuals diagnosed with exudative pleural effusion. For model development, multivariate logistic regression was employed to identify independent predictors of NT-BPE. Model performance and discriminative ability were further evaluated using receiver operating characteristic curve analysis. RESULTS: NT-BPE was diagnosed in 193 patients (28.2%). Independent predictors included being under 65 years of age, male sex, low pleural fluid albumin, elevated glucose levels, absence of cell balls on direct fluid microscopy, low to moderate effusion volume, and the absence of rind-like pleural thickening, mediastinal involvement, or invasion of adjacent structures on imaging. The final prediction model demonstrated an area under the curve of 0.922 (95% CI: 0.894–0.945), with a sensitivity of 88.0% and specificity of 85.7%. CONCLUSION: The study findings suggest that clinically and laboratory meaningful predictors of undiagnosed benign pleural effusion after initail thoracosentesis but before biopsy can be identified, highlighting the potential role of predictive modeling in refining diagnostic decision-making. However, future multicenter studies with external validation cohorts and with integration of some other parameters such as thoracic ultrasound are required before definitive conclusions can be drawn. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12890-026-04222-1.