Development of a Bayesian classifier for breast cancer risk stratification: a feasibility study

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• Current breast cancer risk assessment tools have limitations for younger women.
• Existing models predict population risk, not individual risk, necessitating personalized approaches.
• Bayesian classification offers a potential method for modeling clinical data for risk stratification.

Purpose of the Study:

• To develop and evaluate a proof-of-concept Bayesian classifier for breast cancer risk stratification.
• To explore the utility of Bayesian belief networks in predicting individual breast cancer risk.

Main Methods:

• A Bayesian belief network (BBN) model was trained using prospective pilot screening trial data from 591 younger women.
• Data included risk factors, demographics, electrical impedance scanning (EIS), breast imaging, and biopsy results.
• Model performance was assessed using receiver operating characteristic (ROC) curve analysis and cross-validation.

Main Results:

• Key predictors of biopsy outcome included personal breast disease history, breast size, EIS results, imaging findings, and Gail model 5-year risk.
• The BBN model demonstrated strong predictive value for biopsy outcomes (AUC 0.88, positive predictive value for malignancy 42%).
• Combined predictors significantly increased the likelihood of a malignant biopsy outcome, with probabilities ranging from 3% to 45%.

Conclusions:

• Clinical data from breast cancer screening can be effectively modeled using Bayesian classification.
• The developed BBN model shows potential for providing clinically useful, incremental risk information for individualized breast cancer risk assessment in younger women.