Prediction system for gallbladder cancer distant metastasis a retrospective cohort study based on machine learning

Area of Science:

• Oncology
• Data Science
• Medical Informatics

Background:

• Gallbladder cancer (GBC) prognosis is poor due to early distant metastasis (DM).
• Predictive models for DM risk in GBC are limited.
• This study addresses the need for improved DM risk prediction in GBC.

Purpose of the Study:

• To develop a novel machine learning (ML) model for predicting DM risk in GBC patients.
• To identify key risk and protective factors for DM development.
• To create a user-friendly online tool for personalized risk assessment.

Main Methods:

• Utilized the Surveillance, Epidemiology, and End Results (SEER) database (2000-2020) for GBC patient data.
• Applied logistic regression and six ML techniques for model development and feature selection.
• Validated models using ten-fold cross-validation and employed SHAP for interpretability.

Main Results:

• The Extreme Gradient Boosting (XGB) model achieved high predictive accuracy (AUC=0.885).
• Identified T, N, grade, and age as significant risk factors for DM.
• Rural urban continuum, marital status, and income were identified as protective factors.

Conclusions:

• The XGB model is highly effective for predicting DM in GBC.
• This predictive tool can facilitate early, personalized treatment strategies.
• Improved prediction of DM can lead to better overall prognosis for GBC patients.