A 10-year breast cancer risk model based on image-data mediates modifiable risk factors of breast cancer

Image-derived AI-based risk models for breast cancer (BC) have been developed to identify high-risk women throughout 10 years of follow-up. However, a key question is the extent to which established BC risk factors influence BC risk through changes in the AI-risk score. In a Swedish case-cohort study including 2,959 women, causal mediation analysis was performed using lifestyle, breast density, benign breast disease (BBD), and family history of breast cancer (FH) risk factors as exposures. An image-derived 10-year AI-risk score was used as mediator. The percentage of exposure effects on BC risk that was mediated through the AI-risk score was reported. The mean age in the sub cohort was 54.4 (SD±9,7) and 57.5 (SD±9.4) among cases. The AI-score significantly mediated exposure effects of weight, breast density, BBD, and FH by 40.3%, 21.0%, 31.7%, and 17.3% respectively in premenopausal women. In postmenopausal women, corresponding significant mediated exposure effects were 54.6%, 50.4%, 24.2%, and 14.6%. The menopause-dependent effects varied also according to duration of follow-up. These findings suggest that the image-derived 10-year AI-risk tool manifests biological processes related to established risk factors to breast cancer. Future studies should investigate whether such AI-risk tools may support targeted primary prevention by identifying women who could benefit from risk reducing interventions.