Predicting early gastric cancer risk using machine learning: A population-based retrospective study.

Area of Science:

• Oncology
• Medical Diagnostics
• Machine Learning in Healthcare

Background:

• Early detection and treatment of gastric cancer (GC) are vital for reducing mortality.
• Current diagnostic markers for GC have limited efficiency for early screening.
• Machine learning offers a promising approach to enhance early GC diagnosis by integrating multiple indicators.

Purpose of the Study:

• To develop and evaluate a novel machine learning model for early gastric cancer (GC) detection.
• To assess the model's predictive capability for early GC diagnosis using a prediction score.
• To investigate the model's efficacy in screening GC when conventional protein tumor markers yield negative results.

Main Methods:

• Utilized the XGBoost algorithm to construct the XHGC20 model.
• Integrated data from 20 clinical laboratory tests for patient classification.
• Evaluated model performance using metrics such as AUC, sensitivity, and specificity between 2018 and 2023.

Main Results:

• The XHGC20 model demonstrated strong performance in distinguishing GC from precancerous lesions (AUC=0.901).
• The prediction score effectively diagnosed early GC (AUC=0.888) and showed high accuracy with negative conventional markers (AUC=0.970).
• The model achieved high sensitivity and specificity in all evaluated scenarios, significantly reducing missed diagnoses.

Conclusions:

• The XHGC20 model, based on XGBoost and 20 laboratory tests, serves as a valuable tool for auxiliary laboratory diagnosis of GC.
• This machine learning model aids in the early screening of gastric cancer.
• The XHGC20 model provides a novel and effective method for improving GC diagnostic efficiency.