Development of a machine learning-based nomogram for predicting chronic postsurgical pain after uniportal video-assisted thoracoscopic surgery.

Background: Chronic postsurgical pain (CPSP) remains a prevalent complication following uniportal video-assisted thoracoscopic surgery (UVATS), significantly impairing patients' quality of life. Currently, no reliable predictive tool exists. We aim to develop a machine learning-driven nomogram to stratify CPSP risk. Methods: Patients who underwent UVATS at Nanjing Brain Hospital Affiliated to Nanjing Medical University were analyzed in this retrospective cohort study. Three machine learning algorithms [least absolute shrinkage and selection operator (LASSO) regression, extreme gradient boosting (XGBoost), and random forest (RF)] were employed to identify consensus predictors. Restricted cubic splines (RSC) addressed collinearity among selected variables, followed by logistic regression to build the nomogram. The performance of the nomogram was validated internally and externally. Results: We included a total of 615 patients, who were divided into a training set (n=352), a validation set (n=149), and an external validation set (n=114). Six consensus predictors were identified from these cohorts: age, postoperative C-reactive protein (CRP), postoperative numerical rating scale (NRS) score, discharge NRS score, operative duration, and postoperative drainage tube duration. The nomogram demonstrated excellent discrimination in the training set [area under curve (AUC) =0.893, 95% confidence interval (CI): 0.856-0.929], validation set (AUC =0.861, 95% CI: 0.802-0.920), and external validation set (AUC =0.881, 95% CI: 0.804-0.958). Calibration plots showed high agreement between predicted and observed risks. Decision curve analysis (DCA) further demonstrated favorable predictive performance. Conclusions: This machine learning-driven nomogram provides a robust tool for early CPSP risk prediction after UVATS, facilitating personalized perioperative management.