Machine learning-based preoperative prediction of perioperative venous thromboembolism in Chinese lung cancer patients: a retrospective cohort study

Abstract

Machine learning; Perioperative period; Venous thromboembolism

Aprendizaje automático; Período perioperatorio; Tromboembolismo venoso

Aprenentatge automàtic; Període perioperatori; Tromboembolisme venós

Background: Perioperative venous thromboembolism (VTE) is a severe complication in lung cancer surgery. Traditional prediction models have limitations in handling complex clinical data, whereas machine learning (ML) offers enhanced predictive accuracy. This study aimed to develop and validate an ML-based model for preoperative VTE risk assessment. Methods: A retrospective cohort of 1,013 lung cancer patients who underwent surgery at the First Hospital of Jilin University (April 2021–December 2023) was analyzed. Preoperative clinical and laboratory data were collected, and six key predictors—age, mean corpuscular volume, mean corpuscular hemoglobin, fibrinogen, D-dimer, and albumin—were identified using univariate analysis and Lasso regression. Eight ML models, including extreme gradient boosting (XGB), random forest, logistic regression, and support vector machines, were trained and evaluated using AUC, precision-recall curves, decision curve analysis, and calibration curves. Results: VTE occurred in 175 patients (17.3%). The XGB model demonstrated the highest predictive performance (AUC: 0.99 training, 0.66 validation; AUPRC: 0.323), with age and mean corpuscular volume identified as the most influential predictors. An online prediction tool was developed for clinical application. Conclusion: The ML-based XGB model provides a reliable preoperative risk assessment for VTE in lung cancer patients, enabling early risk stratification and personalized thromboprophylaxis.

The study was supported by Surgical standardized diagnosis and treatment research project WKZX2023WK0112, Jilin Province’s Special Project for Medical and Health Care Talents JLSWSRCZX2023-35 and JLSWSRCZX2023-96.