Distant metastasis prediction in renal carcinoma: a retrospective cohort study using interpretable machine learning

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Renal cell carcinoma (RCC) is a prevalent kidney cancer subtype with poor response to standard therapies.
• Distant metastasis significantly worsens prognosis for RCC patients.
• Identifying risk factors for metastasis is crucial for improving patient outcomes.

Purpose of the Study:

• To identify key clinical risk factors for distant metastasis in renal cell carcinoma (RCC).
• To develop and evaluate machine learning models for predicting distant metastasis in RCC.
• To provide a decision-support tool for personalized clinical management of RCC.

Main Methods:

• Utilized clinical data from the SEER database (2010-2021) for RCC patients.
• Employed logistic regression for factor identification and six machine learning algorithms for predictive modeling.
• Applied Shapley Additive Explanations (SHAP) for model interpretability and comprehensive performance evaluation.

Main Results:

• The Gradient Boosting Machine (GBM) model demonstrated superior predictive performance for distant metastasis in RCC.
• GBM achieved an Area Under the Curve (AUC) of 0.907, indicating high accuracy.
• SHAP analysis validated key clinical factors associated with metastasis risk.

Conclusions:

• A precise machine learning model, particularly GBM, can effectively predict distant metastasis in RCC.
• This predictive tool offers valuable decision support for clinicians in managing RCC patients.
• The findings facilitate personalized treatment strategies to improve patient prognosis.