Leveraging Comprehensive Health Records for Breast Cancer Risk Prediction: A Binational Assessment.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Electronic health records (EHR) offer valuable data for developing breast cancer (BC) risk prediction models.
• Accurate BC risk assessment is crucial for personalized screening and early detection.
• Existing BC risk models have limitations in capturing individual-specific risk factors.

Purpose of the Study:

• To develop and validate a personalized breast cancer risk model using machine learning and EHR data.
• To assess the model's predictive performance compared to established BC risk models.
• To identify novel clinical risk factors for breast cancer prediction.

Main Methods:

• Retrospective analysis of EHR data from 13,786 Israeli and 1,695 American women.
• Utilized 1,547 clinical features extracted from EHR, questionnaires, and radiologist reports.
• Employed machine learning algorithms to build and evaluate the BC risk prediction model.

Main Results:

• The developed personalized BC risk model demonstrated superior predictive performance.
• The model's utility was validated across different datasets and sub-cohorts.
• Identified novel risk factors that enhance BC prediction, even without imaging history.

Conclusions:

• Personalized BC risk models based on EHR and machine learning can significantly improve risk prediction accuracy.
• The model has potential for informing risk-based screening policies, especially for new patients.
• The findings highlight the importance of integrating diverse clinical data for comprehensive BC risk assessment.