Predicting Radiation Esophagitis in Patients Undergoing Synchronous Boost Radiotherapy Post-Breast-Conserving...

Area of Science:

• Oncology
• Radiotherapy
• Medical Informatics

Background:

• Radiation esophagitis is a common side effect of breast cancer radiotherapy.
• Accurate prediction of radiation esophagitis risk is crucial for personalized treatment planning and toxicity management.

Purpose of the Study:

• To construct and validate predictive models for radiation esophagitis occurrence in breast-cancer patients undergoing radiotherapy.
• To identify key clinical and molecular factors contributing to radiation esophagitis.
• To develop a clinically applicable risk score and nomogram for patient stratification.

Main Methods:

• Analysis of 308 breast-cancer patients' data.
• Application of Lasso regression to identify significant predictive variables.
• Development of a radiation esophagitis risk score and a nomogram.
• Internal validation using C-index, AUCs, calibration curves, and decision curve analysis.
• Model interpretation using the SHAP algorithm.

Main Results:

• Seven significant variables were identified by Lasso regression.
• Nomograms demonstrated strong predictive ability (C-index 0.795 for clinical variables, 0.784 for risk score).
• Internal validation showed good performance for risk score (AUC 0.784), nomogram (AUC 0.795), and logistic models (AUC 0.812).
• SHAP analysis highlighted radiation dose, pruritus, molecular type, and hepatic dysfunction as major contributors.

Conclusions:

• Interpretable machine learning models, including risk scores and nomograms, can effectively predict radiation esophagitis risk in breast-cancer radiotherapy.
• These models provide an intuitive tool for assessing individual patient risk, aiding in treatment personalization and toxicity reduction.
• Key factors identified can guide future research and clinical interventions to mitigate radiation esophagitis.