Predicting breast cancer risk using personal health data and machine learning models

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Breast cancer is a leading cause of mortality in women, necessitating accurate risk prediction for screening and prevention.
• Existing models like the Gail model have limitations, often requiring costly or invasive data inputs.
• Machine learning offers potential for improved risk prediction using accessible data.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting five-year breast cancer risk.
• To assess the performance of models using only Gail model inputs versus models incorporating additional personal health data.
• To compare machine learning model performance against the Breast Cancer Risk Prediction Tool (BCRAT).

Main Methods:

• Trained six machine learning models using the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial dataset.
• Evaluated models using Area Under the Receiver Operating Characteristic Curve (AUC), sensitivity, specificity, and precision.
• Compared model performance to BCRAT using Delong tests (p < 0.05).

Main Results:

• Machine learning models using only Gail inputs did not significantly outperform BCRAT.
• Logistic regression, linear discriminant analysis, and neural network models with additional personal health data significantly outperformed BCRAT.
• The enhanced models demonstrated improved predictive accuracy for breast cancer risk.

Conclusions:

• Accessible personal health data, when integrated with machine learning, can significantly enhance five-year breast cancer risk prediction.
• These models offer a non-invasive and cost-effective alternative to traditional risk assessment tools.
• Improved risk stratification can facilitate earlier detection and more effective preventative interventions for breast cancer.