A machine learning-based model for predicting distant metastasis in patients with rectal cancer

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Distant metastasis in rectal cancer significantly reduces survival and quality of life.
• Early identification of high-risk patients is essential for timely intervention.
• Predictive models can aid in stratifying patients for personalized treatment strategies.

Purpose of the Study:

• To develop and validate a machine learning model for predicting distant metastasis risk in rectal cancer.
• To identify key clinicopathological factors associated with distant metastasis.
• To create a user-friendly web calculator for clinical application.

Main Methods:

• Utilized eight machine learning algorithms on a SEER database cohort (23,867 patients) and validated externally on a Chinese hospital cohort (1,178 patients).
• Employed random search and tenfold cross-validation for hyperparameter tuning.
• Evaluated model performance using AUC, AUPRC, decision curve analysis, calibration curves, precision, and accuracy. Interpreted models using SHAP values.

Main Results:

• Identified age, differentiation grade, T-stage, N-stage, CEA, tumor deposits, perineural invasion, tumor size, radiation, and chemotherapy as independent risk factors.
• The extreme gradient boosting (XGB) model demonstrated superior performance, achieving an AUC of 0.855 in the internal test set and 0.814 in the external validation set.
• A web calculator was successfully developed based on the validated XGB model.

Conclusions:

• The developed XGB model accurately predicts the risk of distant metastasis in rectal cancer patients.
• This model, integrated into a web calculator, can assist physicians in clinical decision-making.
• The findings support the use of machine learning for risk stratification in rectal cancer management.