Associations between breastfeeding and breast cancer risk through mammographic breast density in a cohort of Korean women

Abstract Background Mammographic breast density has been suggested to play a role as a mediator between the risk factors for breast cancer (BC) and BC risk. We investigated the extent to which never breastfeeding is a risk factor for BC and how this risk is further mediated by increased mammographic breast density. Methods This retrospective cohort study included 4 136 723 women aged ≥40 years who underwent mammographic screening between 2009 and 2010 and were followed up until 31 December 2020. Breastfeeding information was obtained by using a self-administered questionnaire. Mammographic breast density was extracted from national BC screening results, which were assessed by trained radiologists and categorized into dense and fatty breasts. We estimated the hazard ratios (HRs) and the proportion of the associations between breastfeeding and BC risk mediated by breast density. Results The HR of never breastfeeding on BC risk was 1.34 (95% CI, 1.32–1.37) when adjusted for only parity, body mass index and smoking status, which were selected as covariates through a directed acyclic graph and 1.21 (95% CI, 1.19–1.23) when breast density was additionally adjusted. The proportion of the association between never breastfeeding and BC risk mediated by breast density in total, pre- and post-menopausal women was 35.48%, 17.86% and 24.0%, respectively (all P < 0.001). The HR of never breastfeeding on BC risk was 1.10 (95% CI, 1.08–1.12) when adjusted for nine known breast cancer risk factors and 1.09 (95% CI, 1.07–1.12) when breast density was additionally adjusted. The proportion of the association between never breastfeeding and BC risk mediated by breast density in the total, pre- and post-menopausal women was 7.50%, 3.71% and 12.21%, respectively (all P < 0.001). Conclusions Our findings suggest that the association between never breastfeeding and BC risk may be mediated by breast density. However, the HR and proportion of medications varied according to the adjusted covariates, highlighting that variables for adjustment should be selected based on directed acyclic graphs.