Machine Learning Prediction of Early Recurrence in Gastric Cancer: A Nationwide Real-World Study.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Early recurrence (ER) within two years post-gastric cancer (GC) surgery is associated with poor patient prognosis.
• Accurate prediction of ER is crucial for improving patient outcomes and guiding treatment strategies.
• Machine learning (ML) offers potential for developing predictive models in oncology.

Purpose of the Study:

• To develop and validate machine learning models for predicting early recurrence (ER) in patients undergoing curative surgery for gastric cancer (GC).
• To identify key clinical, pathological, and laboratory parameters associated with ER risk.
• To create a user-friendly tool for clinical application in predicting ER.

Main Methods:

• A multicenter cohort study of 11,615 gastric cancer patients from China.
• Development and validation of ten machine learning models using training (70%) and testing (30%) cohorts.
• Model performance evaluated using AUC, calibration plots, and Brier score; SHAP used for feature interpretation.

Main Results:

• Early recurrence (ER) was observed in 15% of patients.
• A stacking ensemble model demonstrated high predictive accuracy (AUC 1.0 training, 0.8 testing).
• Significant predictors of ER included tumor staging, elevated tumor markers, lymphovascular invasion, perineural invasion, and tumor size > 5 cm; age and lymph node harvest showed U-shaped associations.

Conclusions:

• A robust machine learning model was developed to predict the risk of early recurrence (ER) after gastric cancer (GC) surgery.
• The model can assist in individualized clinical decision-making for GC patients.
• An online prediction tool is available for practical clinical use.