A machine learning-based prediction model for colorectal liver metastasis.

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Colorectal liver metastasis (CRLM) significantly worsens prognosis in colorectal cancer (CRC).
• Accurate prediction of CRLM is crucial for timely intervention and improved patient outcomes.
• Existing prediction methods often lack comprehensive validation or integration with clinical practice.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based risk prediction model for CRLM using conventional clinical data.
• To identify key clinical variables predictive of CRLM occurrence.
• To create a clinically applicable tool for forecasting CRLM risk in CRC patients.

Main Methods:

• Retrospective analysis of clinical data from 865 CRC patients (January 2018 - September 2024).
• Utilized Least Absolute Shrinkage and Selection Operator (LASSO) regression for variable selection.
• Developed and compared five ML algorithms, selecting Random Forest for the final model (CRLM-Lab6).
• Evaluated model performance using ROC curves, precision-recall curves, decision curve analysis, and calibration curves.

Main Results:

• The Random Forest-based CRLM-Lab6 model, incorporating LDH, CA199, ALT, CEA, TBIL, and AGR, demonstrated superior predictive performance.
• The model achieved an Area Under the Curve (AUC) of 0.94, with a sensitivity of 0.88 and specificity of 0.93.
• The model was integrated into a web application for enhanced clinical utility.

Conclusions:

• A novel ML-driven risk prediction model (CRLM-Lab6) for CRLM has been successfully developed and validated.
• The model leverages readily available laboratory test data for accurate and efficient risk assessment.
• This tool holds significant potential for clinical application in managing CRC patients at risk of liver metastasis.