Integration of IDPC Clustering Analysis and Interpretable Machine Learning for Survival Risk Prediction of Patients...

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Accurate survival risk prediction is crucial for esophageal squamous cell carcinoma (ESCC) patient prognosis.
• Current prediction methods for ESCC often lack sufficient fitting ability and interpretability.
• There is a need for advanced, interpretable models to enhance understanding of patient outcomes in ESCC.

Purpose of the Study:

• To develop a novel, interpretable survival risk prediction method for ESCC patients.
• To improve the accuracy and interpretability of existing black-box prediction models.
• To identify key factors influencing survival risk in ESCC.

Main Methods:

• Utilized adaptive synthetic sampling (ADASYN) to address data imbalance and generate high-risk samples.
• Employed an improved clustering by fast search and find of density peaks (IDPC) algorithm for patient stratification.
• Developed an interpretable prediction model using whale optimization algorithm-improved extreme gradient boosting (WOA-XGBoost) and visualized with Shapley additive explanations (SHAP).

Main Results:

• The proposed WOA-XGBoost and SHAP model demonstrated superior performance in survival risk prediction for ESCC.
• Achieved a high area under the receiver operating characteristic curve (AUROC) of 0.918.
• Attained an accuracy of 0.881, indicating effective prediction capabilities.

Conclusions:

• The developed interpretable survival risk prediction method offers significant improvements over existing approaches for ESCC.
• The combination of WOA-XGBoost and SHAP enhances model interpretability, aiding in the identification of prognostic factors.
• This methodology provides a valuable tool for clinicians in managing ESCC patients and predicting their prognosis.