The Atherosclerosis Risk in Communities (ARIC) Study (RRID:SCR_021769) is a prospective cohort study consisting of 15,792 participants aged 45 to 65 years at the time of recruitment (1987–1989) from one of the four ARIC communities (Washington County, MD; Forsyth County, NC; Jackson, MS; and Minneapolis, MN) [27]. The ARIC study protocol was approved by the Institutional Review Boards (US Common Rule) of all participating centers. Written informed consent was obtained from each study participant.

Our study population was women enrolled in ARIC, at risk for first primary breast cancer, with genetic information and who consented to their data being used for studies on genetics and chronic diseases including cancer. As shown in the flowchart (Figure 1), we excluded participants with male sex; who died or were lost to follow up while premenopausal; had a premenopausal breast cancer diagnosis; had prevalent breast cancer at baseline; had a breast cancer diagnosis or were lost to follow up before the second visit (1990–1992); missing information on any of the main covariates.

Information on alcohol consumption status was obtained by interview at Visit 2. Alcohol drinking status was classified into three groups “current drinker”, “former drinker”, and “never drinker” based on the self-reported responses to the following two questions “Do you presently drink alcoholic beverages?”, and “Have you ever consumed alcoholic beverages?”. Alcohol consumption was quantified in grams of ethanol intake per week (g/wk) based on the self-reported numbers of glasses of wine per week (4 oz., 10.8 grams), bottles/cans of beer per week (12 oz., 13.2 grams), drinks of hard liquor per week (1.5 oz., 15.1 grams). For former and never drinkers, this value was set to zero. Because of the relatively large number of White participants, we further grouped White current drinkers into the following three categories: current drinker with less than 1 drink per week, current drinker with 1 drink or more but less than 7 drinks per week, and current drinker with 7 drinks or more per week, based on conventional guidelines that recommend women limit their intake to 1 drink or less per day [28]. One drink of total alcohol was defined as 13 grams of ethanol intake in our study.

Eligible participants were genotyped using DNA extracted from buffy coat (majority of samples were from Visit 2) by the Affymetrix 6.0 array [29]. For all participants, the 313-SNPs polygenic risk score (PRS) for breast cancer was calculated by summing the product of the number of risk alleles for each of the 313 SNPs and their corresponding per-allele log odds ratio for breast cancer [11]. The PRS was then categorized into lower 10^th percentile (≤ 10%), middle 80^th percentile (> 10% to ≤ 90%), and upper 10^th percentile (> 90%) separately for women who are White and Black for calculating baseline characteristics. For the main analysis for White women, PRS was categorized into seven categories: ≤ 10^th, > 10^th to ≤ 30^th, > 30^th to ≤ 50^th, > 50^th to ≤ 70^th, > 70^th to ≤ 90^th, and > 90^th percentiles, as well as two categories: lower 50^th (≤ 50^th) and upper 50^th (> 50^th) percentiles. For Black women, due to the smaller sample size, PRS was categorized into two categories, lower 50^th and upper 50^th percentiles, for the main analysis.

For participants who are Black, we calculated two additional PRSs. The first PRS, developed by Jia et al., is an 89-SNP PRS built using data from women of African ancestry (PRS_AFR) [15]. The second PRS, developed by Gao et al., is a 29,569-SNP PRS built from SNPs identified in both women of African ancestry and European ancestry (PRS_hybrid) [14]. PRS_AFR and PRS_hybrid were categorized into two categories, lower 50^th and upper 50^th percentiles.

PRS, PRS_AFR and PRS_hybrid were scaled to have a mean of 0 and a standard deviation (SD) of 1.

The outcome was first primary female, post-menopausal, invasive breast cancer, ascertained starting after Visit 2 or the first visit at which they were postmenopausal through 2015. Breast cancer cases were ascertained by linking to the cancer registries, abstracting archived hospital discharge summaries, medical records, and death certificates (underlying cause) [30]. Breast cancer was identified using the following ICD codes: ICD-O-1: 174.0–174.9; ICD-O-2: 50.0–50.9; ICD-O-3: 50.0–50.9; ICD-8 & 9: 174.0–174.9; ICD-10: 50.0–50.9.

Covariates were obtained at Visit 2, unless otherwise specified. Main covariates included race (White/Black; Visit 1), age at recruitment (years; Visit 1), height at recruitment (inches; Visit 1), body mass index (kg/m²), waist-hip ratio, education level (basic education, intermediate education, advanced education), physical activity (minutes per week; Visit 1), cigarette smoking status (current smoker, former smoker, never smoker), total energy intake (kcal; Visit 1), folic acid intake (micrograms; Visit 1), diabetes status (diabetes, at risk for diabetes, normal glycemia), age at menarche (years; Visit 1), ever taken hormone replacement therapy (yes/no), ever used oral contraceptives (yes/no; Visit 1), breast-feeding status (never, less than one year, one year or more; Visit 3).

Other covariates included parity (nulliparous/parous; Visit 3), age at first live birth (less than 35 years old/ 35 years or older; Visit 3), ever had a mammogram (yes/no; Visit 3), mother had breast cancer (yes/no; Visit 3), frequency of routine examinations (less than one ever five years/at least once every five year), health insurance status (yes/no).

For women who were postmenopausal at Visit 2 (the first available visit with genetic information for the majority of the participants), follow-up began at Visit 2 (n=5616, 99.4%). For women who were premenopausal at Visit 2 (n=35, 0.6%), follow-up began at the first visit at which they were postmenopausal.

We initially used the 313-SNPs PRS [11], and performed the entire analysis stratified by race given the difference in predictive performance of the PRS by race [14–15, 31]. For Black women, we repeated the main analyses using the PRS_AFR and PRS_hybrid to confirm the findings when using the 313-SNP PRS. We estimated the age-adjusted and fully-adjusted hazard ratio (HR) and 95% confidence intervals (CI) for the association between continuous PRS (scaled) and breast cancer incidence using multivariable Cox proportional hazards models; we reported HRs per 1 SD increase of the scaled PRS. Age-adjusted model included PRS and age (continuous) as independent variables, while the fully-adjusted models additionally included height, body mass index (BMI), waist-hip ratio, education level, physical activity, cigarette smoking status, total calories intake, folic acid intake, diabetes status, age at menarche, ever taken hormone replacement therapy, ever used oral contraceptives, and breast-feeding status as independent variables. We used the last observation carried forward method to handle missing data on alcohol drinking status, ethanol intake, cigarette smoking status, BMI, waist-hip ratio, parity, and we created missing indicator for breast-feeding status and physical activity. We then examined whether the association between continuous 313-SNP PRS and breast cancer incidence was linear on the natural logarithm scale using the likelihood ratio test comparing the model with continuous PRS and the model with penalized spline (P-spline) terms with 4 degrees of freedom for PRS [32–34]. The proportional hazards assumption was checked by visualizing the time-dependent coefficient of the PRS to determine if the line was horizontal. No violation of proportional hazards assumption was found (Supplementary Figure S1).

We estimated the age-adjusted and fully-adjusted HR and 95% CI for the association between continuous current ethanol intake and breast cancer incidence using multivariable Cox proportional hazards models, and examined whether the association was linear on the natural logarithm scale using a similar method. Time-dependent coefficient for the association between current ethanol intake and breast cancer risk was plotted. There was no violation of the proportional hazards assumption among White women, while the sample size was too small for Black women to determine if the proportional hazards assumption holds (Supplementary Figure S2)

We evaluated whether alcohol modified the association between PRS and breast cancer by estimating fully-adjusted HRs of breast cancer incidence for continuous PRS within each stratum of alcohol consumption using Cox proportional hazards models. To test for gene-alcohol interaction, likelihood ratio tests were conducted comparing the fully-adjusted model with terms for alcohol consumption categories and continuous PRS to the model additionally including interaction terms for the product of alcohol consumption categories and continuous PRS. For each alcohol consumption categories, we also ran Cox models for PRS using P-spline with 4 degrees of freedom and plotted the log HR for the P-spline PRS terms.

Additionally, we evaluated whether PRS modified the alcohol and breast cancer association by estimating fully-adjusted HRs of continuous current ethanol intake on breast cancer incidence within each PRS stratum. Likelihood ratio tests were conducted comparing model with and without interaction terms.

We conducted three sensitivity analyses. 1. We stratified the age-adjusted model for PRS by maternal history of breast cancer (collected at Visit 3), given that PRS and family history of breast cancer may be capturing the same underlying genetic susceptibility. 2. In the analysis of PRS-breast cancer association stratified by alcohol consumption categories, we excluded participants with extreme current ethanol intake using various cut points (ranging from >180 g/wk to >500 g/wk) to define an extreme intake, and modeled the alcohol as a continuous and as well as a binary variable (less than 1 drink per week vs. at least 1 drink per week; less than 7 drinks per week vs. at least 7 drinks per week). 3. To further study ethanol-breast cancer association, we restricted participant to those with health insurance, and stratified by routine examinations frequency and by mammogram status.

Statistical significance was set at P-value < 0.05 for two-sided test. All analyses were performed using R version 4.2.2 (RRID:SCR_001905).

The data that support the findings of this study are available from the corresponding author upon reasonable request (https://aric.cscc.unc.edu/aric9/publications/policies_forms_and_guidelines)

Our final analytic dataset consisted of 5651 participants, of which 4055 were of European ancestry and 1596 were of African ancestry (Figure 1). Among White women, 61.1% were current drinkers, and of them 50.7% of them drank <1 drink/week and 13.8% drank >7 drinks/week, whereas among Black women, 23.7% were current drinkers, and of them 51.6% drank <1 drink/week and 10.1% drank >7 drinks/week. Baseline characteristics for participants by alcohol intake category are shown in Supplementary Tables S1a (White) and S1b (Black). In general, current drinkers were younger, more educated, leaner, and more active, less likely to have diabetes, and more likely to take birth control pills, but had lower folic acid intake compared with never and former drinkers, in both White and Black participants. Across the three 313-SNP PRS groups, the distributions of covariates were relatively even (Supplementary Table S1c [White], S1d [Black]).

Among White women, the fully-adjusted HR of breast cancer for a 1 SD increase in PRS was 1.48 (95% CI: 1.34–1.64, P-value < 0.001), whereas among Black women, the HR was 1.15 (95% CI: 0.96–1.38, P-value = 0.13) (Table 1). These associations were linear on the natural logarithm scale (p-linearity: 0.25 [Whites] and 0.20 [Blacks]; Supplementary Figure S3).

Positive associations were not found between current ethanol intake and breast cancer risk among White (per 13 g/week fully adjusted HR: 1.00, 95% CI: 0.98–1.03), and Black (HR: 0.83, 95% CI: 0.69–1.00) women (Table 2; Supplementary Figure S4A and S4B).

Higher PRS was associated with higher breast cancer risk in non-drinkers and in each of the alcohol consumption categories among White women (Table 3). The HR of breast cancer for a 1 SD increase in PRS (HR: 1.63, 95% CI: 1.10–2.41) was highest among the “≥ 7 drinks/wk” category, while the HR of breast cancer for a 1 SD increase in PRS (HR: 1.40, 95% CI: 1.12–1.73) was lowest among the “≥ 1 to < 7 drinks/wk” category. However, these differences in the HRs for the association between PRS and breast cancer across the alcohol consumption categories were not statistically significant (P-interaction=0.81). Within each of the five alcohol consumption categories, the natural logarithm of the HR of breast cancer generally increased as the PRS increased (Supplementary Figure S5).

Among Black women, the HRs of breast cancer for a 1 SD increase PRS were 1.30 (95% CI: 0.97–1.74), 1.05 (95% CI: 0.76–1.46), 1.25 (95% CI: 0.80–1.97) for never, former, and current drinkers, respectively (Table 3; Supplementary Figure S6). These estimates did not statistically differ across alcohol consumption categories (P-interaction=0.61).

Among White women, current ethanol intake was not associated with breast cancer risk across PRS categories, and there was no interaction between continuous ethanol intake and PRS categories for breast cancer risk (Table 4). However, for Black women, the HRs of breast cancer for a 13-gram increase in ethanol per week were 0.72 (95% CI: 0.51–1.04) and 0.88 (95% CI: 0.71–1.08) for the lower and upper 50% PRS group, respectively (Table 4).

Among Black women, the fully-adjusted HR of breast cancer for a 1 SD increase in PRS_AFR was 1.12 (95% CI: 0.93–1.35, P-value = 0.22), and for PRS_hybrid, was 1.22 (95% CI: 1.02–1.46, P-value = 0.03) (Table 5). The HRs of breast cancer for a 1 SD increase in PRS_AFR were 1.08 (95% CI: 0.82–1.40), 1.38 (95% CI: 0.94–2.00), 1.02 (95% CI: 0.68–1.54) for never, former, and current drinkers, respectively. For PRS_hybrid, the corresponding HRs were 1.40 (95% CI: 1.05–1.86), 1.11 (95% CI: 0.79–1.58), and 1.16 (95% CI: 0.77–1.75). The HRs of breast cancer for a 13-gram increase in current ethanol intake per week were similar across the lower and upper 50% of PRS_AFR, whereas the HRs were 0.91 (95% CI: 0.75–1.12) and 0.70 (95% CI: 0.46–1.06) in the lower and upper 50% group of PRS_hybrid (Table 5). No statistically significant interactions between PRS and ethanol intake were observed, with all P-interaction values >0.05 (Table 5).

Among White women, the association between PRS and breast cancer risk appeared similar between those whose mother had breast cancer (HR: 1.53, 95% CI: 1.09–2.13) and those whose mother did not have breast cancer (HR: 1.47, 95% CI: 1.32–1.64; Supplementary Table S2). Among Black women, the HR for those whose mothers had breast cancer was 1.36 (95% CI: 0.59–3.15) albeit based on small numbers, while the HR was 1.17 (95% CI: 0.96–1.41) among those whose mothers did not have breast cancer.

After excluding extreme values of current ethanol intake, association between continuous ethanol intake and breast cancer risk was consistently null for Whites, while the association was inverse and borderline significant for Blacks (Supplementary Table S3). Drinking at least 1 drink per week (vs. <1 per week), or at least 7 drinks per week (vs. <7 per week) was not associated with higher breast cancer risk among Whites, while drinking at least 1 drink per week (vs. <1 per week) was associated with a borderline significant lower breast cancer risk among Blacks, although there were few cases among those who drank at least 1 drink per week.

The association between current ethanol intake and cancer risk remained unchanged after restricting to those with health insurance, and after stratifying by frequency of routine physical examinations and by ever had a mammogram status (Supplementary Table S4–S6).