Incorporating continuous mammographic density into the BOADICEA breast cancer risk prediction model
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
BOADICEA (v7) predicts future breast cancer (BC) risk using data on cancer family history, genetic markers, questionnaire-based risk factors and mammographic density (MD) measured using the 4-category BI-RADS classification. However, BI-RADS requires manual reading, which is impractical on a large scale and may cause information loss.
Methods
We extended BOADICEA to incorporate continuous MD measurements, calculated using the automated Volpara and STRATUS software. We used data from the KARMA cohort (60,276 participants; 1,167 incident BC). Associations between MD measurements and BC risk were estimated in a randomly selected training subset (two-thirds of the dataset). Percent MD residuals were calculated after regressing on age at mammography and BMI. Hazard ratios (HRs) were estimated using a Cox proportional hazards model, adjusting for family history and BOADICEA risk factors, and were incorporated into BOADICEA. The remaining one-third of the cohort was used to assess the performance of the extended BOADICEA (v 7.2) in predicting 5-year risks.
Results
The BC HRs per SD of residual STRATUS density were estimated to be 1.48 (95%CI: 1.33-1.64) and 1.41 (95%CI: 1.27-1.56) for pre- and post-menopausal women, respectively. The corresponding estimates for Volpara density were 1.27 (95%CI: 1.15-1.40) and 1.38 (95%CI: 1.25-1.54). The extended BOADICEA showed improved discrimination in the testing dataset over using BIRADS, with a 1-4% increase in AUC across different combinations of risk factors. Based on 5-year BC risk with MD as the sole input, approximately 11% of the women were reclassified into lower risk categories and 18% into higher risk categories using the extended model.
Conclusion
Incorporating continuous MD measurements into BOADICEA enhances breast cancer risk stratification and facilitates the use of automated MD measures for risk prediction.
