Bypassing pleural aspiration for early tissue diagnosis in suspected pleural malignancy: Development and external validation of a predictive model using supervised machine learning

Background and objective: The conventional diagnostic pathway for pleural malignancy involves pleural aspiration, often followed by tissue biopsy. However, pleural aspiration has limited sensitivity, which can potentially delay a definitive diagnosis. The objective of this study was to develop and externally validate a predictive model using clinical data to identify patients who could safely bypass pleural aspiration and proceed directly to tissue biopsy. Methods: This was a retrospective cohort study of patients presenting to an acute hospital in the UK with suspected pleural malignancy between 2016 and 2025 (n = 646). A Random Forest classifier was trained on this dataset, and its performance was evaluated using an independent external validation cohort from another acute hospital in the UK (n = 32). Model performance was assessed using standard metrics, including accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (ROC-AUC). Results: The model demonstrated robust performance in the internal dataset and maintained its predictive strength during external validation (AUC: 0.86 vs. 0.83; precision 87% vs. 85%; recall 79% vs. 75%). Among the variables, computed tomography findings were the most influential predictor, as quantified by Gini importance. A simplified scoring system was subsequently derived for potential bedside clinical application. Conclusion: This externally validated model provides a valuable decision-support tool for clinicians, facilitating earlier tissue diagnosis in patients with suspected pleural malignancy. By potentially avoiding non-diagnostic aspiration, the model can streamline the diagnostic pathway and expedite patient care. Successful external validation enhances confidence in the model’s generalizability and supports its potential for implementation into routine clinical practice.