Ten-Year Multicenter Retrospective Study Utilizing Machine Learning Algorithms to Identify Patients at High Risk of...

Area of Science:

• Oncology
• Medical Informatics
• Surgical Outcomes

Background:

• Venous thromboembolism (VTE) is a significant complication following gastric cancer surgery.
• Accurate prediction of VTE risk is crucial for patient management and preventative strategies.

Purpose of the Study:

• To develop and validate a machine learning model for predicting postoperative VTE in gastric cancer patients.
• To identify high-risk indicators for VTE across preoperative, intraoperative, and postoperative phases.

Main Methods:

• Retrospective analysis of 1239 gastric cancer patients, with 107 developing VTE.
• Utilized Extreme Gradient Boosting (XGBoost), Random Forest, SVM, and KNN algorithms.
• Employed SHAP for model interpretation and rigorous evaluation metrics including AUC, ROC, calibration curves, and DCA.

Main Results:

• XGBoost achieved superior predictive performance with an AUC of 0.989 (training) and 0.912 (validation).
• External validation demonstrated good extrapolation with an AUC of 0.85.
• Key predictors identified include higher BMI, adjuvant therapy history, advanced tumor stage, lymph node metastasis, central venous catheter use, significant intraoperative bleeding, and prolonged operative time.

Conclusions:

• The XGBoost-based model provides a robust tool for predicting postoperative VTE in gastric cancer patients undergoing radical gastrectomy.
• This predictive model can assist clinicians in informed decision-making and personalized patient care to mitigate VTE risk.