To develop a breast cancer risk model to identify women at mammographic screening who are at higher risk of breast cancer within the general screening population.
This retrospective nested case-control study used data from a population-based breast screening program (2009–2015). All women aged 40–75 diagnosed with screen-detected or interval breast cancer (n = 1882) were frequency-matched 3:1 on age and screen-year with women without screen-detected breast cancer (n = 5888). Image-derived risk factors from the screening mammogram (percent mammographic density [PMD], breast volume, age) were combined with core biopsy history, first-degree family history, and other clinical risk factors in risk models. Model performance was assessed using the area under the receiver operating characteristic curve (AUC). Classifiers assigning women to low- versus high-risk deciles were derived from risk models. Agreement between classifiers was assessed using a weighted kappa.
The AUC was 0.597 for a risk model including only image-derived risk factors. The successive addition of core biopsy and family history significantly improved performance (AUC = 0.660, p < 0.001 and AUC = 0.664, p = 0.04, respectively). Adding the three remaining risk factors did not further improve performance (AUC = 0.665, p = 0.45). There was almost perfect agreement (kappa = 0.97) between risk assessments based on a classifier derived from image-derived risk factors, core biopsy, and family history compared with those derived from a model including all available risk factors.
Women in the general screening population can be risk-stratified at time of screen using a simple model based on age, PMD, breast volume, and biopsy and family history.
• A breast cancer risk model based on three image-derived risk factors as well as core biopsy and first-degree family history can provide current risk estimates at time of screen.
• Risk estimates generated from a combination of image-derived risk factors, core biopsy history, and first-degree family history may be more valid than risk estimates that rely on extensive self-reported risk factors.
• A simple breast cancer risk model can avoid extensive clinical risk factor data collection.