Development and validation of machine learning models for preoperative prediction of perioperative outcomes in patients with thymic epithelial tumors

Development and validation of machine learning models for preoperative prediction of perioperative outcomes in patients with thymic epithelial tumors. Download PDF Abstract Thymic epithelial tumors (TETs) are rare anterior mediastinal tumors with heterogeneous perioperative recovery. We developed and externally validated machine-learning models using preoperative and preoperatively available variables to predict postoperative complications, prolonged postoperative length of stay (LOS), and moderate-to-severe postoperative pain.

This dual-center retrospective study included 726 surgically treated patients with pathologically confirmed TETs. The Qilu Hospital cohort was used for model development and internal validation, and 169 patients from the First Affiliated Hospital of Shandong First Medical University formed the external validation cohort. Least absolute shrinkage and selection operator (LASSO) regression selected 26, 14, and 13 predictors for the three outcomes, respectively.

Neural-network models were selected for complications and prolonged LOS, and a decision-tree model for pain. Areas under the receiver operating characteristic curve (AUCs) in training cross-validation, internal validation, and external validation were 0.824, 0.814, and 0.775 for complications; 0.824, 0.809, and 0.776 for prolonged LOS; and 0.695, 0.734, and 0.660 for pain. External AUC 95% confidence intervals were 0.658–0.892, 0.697–0.855, and 0.570–0.749, respectively.

Medical Disclaimer: This article is for informational purposes only and does not constitute medical advice. Always consult with qualified healthcare professionals for medical decisions and treatment options.