Postoperative success and risk analysis of patients with non-small cell lung cancer undergoing lung resection using artificial intelligence

Thoracic surgery is a comprehensive process that involves meticulous preoperative preparations filled with detailed risk analyses, significant effort and concentration during the operation, and careful postoperative follow-up requiring precise decision-making at every step. Non-small cell lung cancer (NSCLC) is the most common histopathological type among lung cancers. Our study aims to assist both patients and physicians throughout these processes, with a particular focus on the postoperative stages. The 9th edition of the TNM staging system, currently recommended and used by the IASLC, is the standard for staging. Surgical treatment is effective in early-stage and resectable locally advanced NSCLC. Lung resection and systematic mediastinal lymph node dissection form the foundation of NSCLC surgery. Postoperative complications, the need for intensive care requiring advanced treatment and care, and prolonged hospital stays are critical factors that impact mortality, morbidity, cost, hospital-acquired infections, sustainability, and social factors. Currently, there is no objective, low-cost, easy-to-use, self-improving, and highly accurate evaluation tool or application tailored to individual patients that can predict the necessity of intensive care during these stages. Artificial intelligence (AI) applications are increasingly integrated into healthcare systems and other fields, aiding decision-making through data collection, classification, storage, learning, and teaching capabilities. In our study, we aimed to predict postoperative processes in patients undergoing surgery for NSCLC using available data. We retrospectively analyzed data from 1,471 patients to create extensive datasets. Using six different AI algorithms, we predicted postoperative clinical outcomes and the need for intensive care with high accuracy and success rates based on 14 patient-specific parameters. The algorithms used were: SVM-Linear Kernel, Random Forest Classification SVM-Linear Kernel with SMOTE, Random Forest Classification with SMOTE, Linear Regression, Radial Basis SVM Correlation Method. The most successful algorithm for predicting patients requiring intensive care was Random Forest Classification with SMOTE, achieving an F1 score of 91%. The same algorithm also excelled in predicting patients not requiring intensive care, with an F1 score of 91%. For this method, the average cross-validation score was 0.9232, with a standard deviation of 0.0165. The area under the ROC curve for the most successful algorithm was 86%. Thanks to AI algorithms, postoperative clinical predictions for NSCLC surgery can now be made with high accuracy and success rates, and their widespread use in clinical practice is anticipated in the future. Keywords: Non-Small Cell Lung Cancer, NSCLC, Postoperative Complications, Intensive Care Unit, Intensive Care, Artificial Intelligence, SMOTE, Linear Regression, SVM, Support Vector Machine, Linear Kernel, Radial Basis, Random Forest, Thoracic Surgery