A Dynamic Bayesian Model for Breast Cancer Survival Prediction.

Area of Science:

• Oncology
• Biostatistics
• Data Science

Background:

• Accurate prediction of breast cancer survival is critical for patient management.
• Identifying patients at high risk of mortality is a key challenge in oncology.

Purpose of the Study:

• To develop and validate a novel Bayesian Dynamic Cox (BDCox) model for predicting 5-year overall survival in breast cancer patients.
• To compare the performance of the BDCox model against other prognostic models and feature selection methods.

Main Methods:

• A Bayesian Dynamic Cox (BDCox) model was developed using data from 12,840 women in the SEER Cancer Registry.
• Four feature selection methods (fast backward variable selection, elastic net, Bayesian Model Average (BMA), clinical expertise) were employed.
• Internal validation via bootstrapping and external validation in the Shanghai Breast Cancer Survival Study were performed.

Main Results:

• The BDCox model with 12 predictors, selected using Bayesian Model Average (BMA), demonstrated the best performance.
• BMA outperformed other feature selection methods in both internal and external validations.
• The BDCox model achieved high prediction accuracy (C-statistic: 0.802 internally) and excellent generalizability (C-statistic: 0.739 externally).

Conclusions:

• The developed BDCox model significantly outperforms other considered survival models for breast cancer prognosis.
• The model's robust internal and external validation confirms its capability for accurate survival prediction and generalizability.
• This dynamic Bayesian approach provides a powerful tool for predicting breast cancer survival and targeting high-risk individuals.