Causal relationships between breast cancer mammogram risk scores based on textural features and density

Background: Mammographic risk scores based on texture (Cirrus) and density defined by different brightness thresholds Cumulus, Altocumulus, and Cirrocumulus) are associated with breast cancer risk in different ways and could reveal different information about breast cancer risk. Whether there are causal relationships between the mammographic risk measures based on texture and density is unknown. Methods We digitised mammogramsfor 371 monozygotic twin pairs, none diagnosed with breast cancer at the time. We generated normalised, age-adjusted, and standardised risk scores for Cirrus, and for three spatially independent density measures, the light areas (Cumulus minus Altocumulus), the bright areas (Altocumulus minus Cirrocumulus) and the brightest areas (Cirrocumulus). Causal inference was made using the Inference about Causation from Examination of FAmilial CONfounding (ICE FALCON) methodology. Results The risk scores were correlated within twin pairs and with each other ( r  = 0.22 to 0.81; all P < 0.005). We estimated that 8–72% of the latter associations could be attributed to familial confounding between the risk scores with remainder attributed to causal relationships. There was consistent evidence for positive causal relationships: of Cirrus, the light areas, and the bright areas on the brightest areas (accounting for 34%, 55% and 85% of the associations); and of the light areas and bright areas on Cirrus (accounting for 37% and 28% of the associations). Conclusions The lighter (less dense) areas could be having a causal effect on the brightest (highly dense) areas, including a causal pathway through the mammographic risk score based on textural features. These findings would explain the consistent observations that the associations of Cumulus with breast cancer risk (including screen-detected, younger-age-at-diagnosis, and contralateral breast cancer) attenuate to the null after adjusting for Cirrocumulus and/or Cirrus; they could be due to, at least in part, the causal relationships between the less dense areas and the highly dense areas (and/or the texture-based measure). These findings also demonstrate how ICE FALCON can decompose associations between familial biomarkers into pathways.