Machine learning models using multimodal data accurately predict chemotherapy-induced cardiotoxicity in breast cancer

Area of Science:

• Cardiology
• Oncology
• Medical Informatics

Background:

• Chemotherapy-related cardiac dysfunction (CTRCD) is a significant challenge in breast cancer treatment.
• Predicting CTRCD risk is crucial for patient management and therapeutic optimization.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting CTRCD risk in female breast cancer patients.
• To integrate multimodal data for enhanced predictive accuracy.

Main Methods:

• Retrospective analysis of 423 female breast cancer patients' data (demographics, clinical, echocardiographic, ECG, biomarkers).
• Data split into training (70%) and validation (30%) sets.
• Utilized seven feature selection methods and eight ML algorithms, including XGBoost, for model development and comparison.

Main Results:

• CTRCD occurred in 26.24% of patients.
• Key predictors identified: age, low ejection fraction, specific combination therapy, chemotherapy cycles, and abnormal ECG.
• XGBoost model achieved the highest performance with an AUC of 0.782.

Conclusions:

• The developed XGBoost model demonstrates strong predictive capability for CTRCD.
• This ML tool can aid in early risk stratification and timely clinical management of CTRCD.