Association between allostatic load and breast radiomic features among diverse women undergoing screening mammography.

10514 Background: Allostatic load (AL), an indicator of chronic physiologic stress, is associated with increased breast cancer incidence and all-cause mortality in patients diagnosed with breast cancer. Mammographic density (MD) is a well-established breast cancer risk factor. We developed a deep learning, specifically convolutional neural network (CNN)-based, mammographic evaluation, which is a more accurate predictor of breast cancer risk than MD, particularly among Black and Hispanic women. We evaluated the relationship between AL and CNN risk score in a cohort of racially/ethnically diverse women undergoing breast cancer screening. Methods: We conducted a retrospective cohort study of women aged 35-74 years, without a personal history of breast cancer, with available clinical data to calculate AL score and mammograms for CNN analysis (output = risk score from 0-1), who underwent screening mammography at Columbia University Irving Medical Center (CUIMC) in New York, NY from 2014-2018. We extracted data on demographics, MD (BIRADS A-D), laboratory values, vital signs, and body mass index (BMI) from the electronic health record (EHR). We calculated AL score using biometrics from 4 physiologic systems: cardiovascular (heart rate, systolic and diastolic blood pressure), metabolic (BMI, blood glucose, albumin, alkaline phosphatase), renal (creatinine, blood urea nitrogen), and immune (white blood cell count). AL scores ranged from 0 to 10, and patients received one point for each lab value out of the reference range. High AL was defined as an AL score above the median for the overall cohort. Multivariable logistic regression analyses were conducted to evaluate the association between high AL and demographic factors, breast cancer risk factors, and CNN risk score. Results: Among 9,798 evaluable women, mean age was 57.0 (standard deviation [SD], 9.5), including 24.7% non-Hispanic White, 9.6% non-Hispanic Black, 36.1% Hispanic, and 4.0% Asian women. Median AL score was 3. In multivariable analyses, high AL was associated with older age (odds ratio [OR]=1.03, 95% confidence interval [CI]=1.02-1.03). Compared to non-Hispanic White women, Black and Hispanic women had increased odds of having high AL (OR= 2.62, 95% CI = 2.23-3.08 and OR = 2.12, 95% CI =1.90-2.37, respectively). MD was inversely associated with high AL, and there was no significant association between CNN risk score and AL. Conclusions: In this cohort of diverse women undergoing screening mammography, we observed a significant association between high AL and older age and Black and Hispanic race/ethnicity. However, MD was inversely associated with high AL, likely due to higher BMI, and CNN risk score was not associated with AL. Future studies should evaluate the association between novel breast radiomic features and genetic and hormonal biomarkers of breast cancer risk.