Machine learning for predicting liver metastasis in colorectal cancer

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Liver metastasis is a significant complication in colorectal cancer (CRC).
• Accurate prediction of liver metastasis is crucial for effective patient management and treatment planning.
• Existing predictive models may require enhancement with advanced computational approaches.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in predicting liver metastasis in colorectal cancer (CRC) patients.
• To validate the performance of these models using both the SEER database and external clinical data.
• To identify key clinical variables contributing to liver metastasis prediction.

Main Methods:

• Utilized the Surveillance, Epidemiology, and End Results (SEER) database (2010-2023) for training and testing datasets.
• Developed and compared eight distinct machine learning models using 11 clinical variables.
• Assessed model performance via Area Under the Receiver Operating Characteristic Curve (ROC) and Area Under the Precision-Recall Curve (AUPR).
• Employed SHAP (SHapley Additive exPlanations) for model interpretation and feature importance analysis.

Main Results:

• The Gradient Boosting (GB) model achieved the highest performance in the SEER cohort with an AUC of 0.837 and AUPR of 0.294.
• External validation confirmed the GB model's predictive capability with an AUC of 0.730 and AUPR of 0.278.
• SHAP analysis identified CEA, N stage, and T stage as the most influential factors in the GB model's predictions.
• An online calculator was developed based on the GB model for practical clinical application.

Conclusions:

• The Gradient Boosting (GB) machine learning model demonstrates robust and validated performance in predicting liver metastasis in colorectal cancer patients.
• This model serves as a promising tool for clinical decision-making, aiding in the early identification of patients at risk.
• Further integration of such predictive tools can enhance personalized treatment strategies for colorectal cancer.