Application of machine learning to develop and validate a pain risk prediction model for patients with non-small cell lung cancer after video-assisted thoracoscopic surgery: A single-center retrospective study

This study aimed to develop a machine learning (ML)-based model to identify risk factors for postoperative pain following video-assisted thoracoscopic surgery (VATS) lobectomy in non-small cell lung cancer (NSCLC) patients. This retrospective study analyzed data from 100 NSCLC patients who underwent VATS. Least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation identified predictive factors. Patients were split into training (80%) and testing (20%) sets. Seven ML algorithms were trained, with performance evaluated via receiver operating characteristic curve, sensitivity, specificity, and accuracy. The shapley additive explanations (SHAP) method interpreted the best-performing model. LASSO regression identified 11 predictors. The random forest (RF) model achieved the highest predictive performance (AUC: 0.901, 95% CI: 0.833-0.969). SHAP analysis highlighted elevated pro-gastrin releasing peptide, tumor volume, red cell distribution width, lactic dehydrogenase, and white blood cell count as risk factors, while dexmedetomidine and higher hemoglobin were protective. A simplified model retained comparable accuracy (DeLong test P = .4846). The RF-based ML model effectively predicts post-VATS pain risk in NSCLC patients, demonstrating potential to guide future research on preoperative risk assessment and personalized interventions. External validation in a larger cohort is required before clinical application.