Machine learning-based lifetime breast cancer risk reclassification compared with the BOADICEA model: impact on...

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• The clinical utility of machine learning (ML) for breast cancer risk prediction and its impact on screening practices remain largely unknown.
• Existing models like BOADICEA provide a benchmark for risk assessment, but their accuracy in diverse populations requires further evaluation.

Purpose of the Study:

• To compare the accuracy of ML algorithms against the BOADICEA model for lifetime breast cancer risk prediction.
• To explore the clinical implications of risk reclassification by ML on mammography surveillance recommendations.

Main Methods:

• Utilized three ML algorithms and the BOADICEA model to estimate lifetime breast cancer risk in a large cohort (112,587 individuals).
• Evaluated algorithm performance using the area under the receiver operating characteristic (AU-ROC) curve.
• Assessed risk reclassification in 36,146 women and analyzed the impact on screening based on the Swiss Surveillance Protocol.

Main Results:

• ML algorithms demonstrated superior predictive accuracy (AU-ROC 0.843–0.889) compared to BOADICEA (AU-ROC 0.639).
• ML reclassified 35.3% of women into different risk categories, with significant shifts observed in women initially classified as 'near population' risk by BOADICEA.
• Risk reclassification by ML models had the most substantial impact on screening recommendations for women under 50.

Conclusions:

• Machine learning-based risk reclassification affects approximately one-third of women, offering a more precise assessment of lifetime breast cancer risk.
• The findings highlight the importance of ML in refining screening strategies, particularly for younger women, by influencing the timing and initiation of mammography surveillance.