Hormone-associated dietary patterns and premenopausal breast cancer risk.

The foods/food groups associated with each hormone are presented in Table 1 . Dietary patterns (five or more foods/food groups associated with the hormone of interest) were identified for luteal estradiol, luteal free estradiol, follicular estrone, luteal estrone, and free testosterone. Follicular estradiol, follicular free estradiol, follicular estrone sulfate, and luteal estrone sulfate had four or fewer food items associated with hormone levels so were not considered hormones with related dietary patterns. The identified dietary patterns were weakly correlated with each hormone of interest and explained a low percent variation in individual hormone levels. The strongest percent variations explained were for free testosterone (4.1%) and luteal estradiol (3.9%), and the lowest percent explained was 2.5% for follicular estrone ( Table 2 ). While some of the food groups included in the patterns differed when excluding BMI from RRR, Pearson correlations and percent variations were not meaningfully altered (data not shown). In secondary analyses grouping related hormones (i.e., combined luteal estradiol and free estradiol and combined testosterone and free testosterone patterns), few changes in food groups identified for pattern inclusion were observed (data not shown). During 24 years of follow-up, a total of 1,956 premenopausal breast cancer cases were ascertained among 90,341 participants. At baseline, those in the lowest quintile of the luteal free estradiol-related dietary pattern reported slightly higher weight gain since age 18 and less physical activity than participants in the highest quintile. Conversely, participants in the lowest quintile of the follicular estrone pattern reported lower weight gain since age 18 and more physical activity than those in the highest quintile. Participant characteristics, including age at menarche, BMI at age 18, and family history of breast cancer, were similar across quintiles ( Table 3 ). Luteal free estradiol and follicular estrone dietary pattern scores were positively associated with premenopausal breast cancer risk in cumulative adult diet analyses ( Table 4 ). Participants in the fifth quintile for the luteal free estradiol-associated pattern had a multivariable HR (95% CI) of 1.29 (1.11–1.49; P trend < 0.01), and the association was similar for the follicular estrone-related pattern (HR for fifth quintile = 1.28; 95% CI = 1.10–1.49; P trend < 0.01). Adjustment for BMI at age 18 and weight gain since age 18 slightly attenuated estimates for the luteal-free estradiol-related pattern (HR for fifth quintile = 1.23; 95% CI = 1.06–1.50; P trend < 0.01) but did not materially impact associations for follicular estrone-associated dietary pattern scores (HR = 1.28; 95% CI = 1.10–1.50; P trend < 0.01) ( Table 4 ). In sensitivity analyses removing alcohol from the dietary patterns before their application to the full cohort, associations with luteal free estradiol were attenuated, and only the follicular estrone-related pattern remained statistically significantly associated with overall premenopausal breast cancer risk ( Supplementary Table 2 ). No other dietary pattern scores (i.e., luteal estradiol, luteal estrone, or free testosterone) were associated with overall premenopausal breast cancer risk. When evaluating associations between each dietary pattern and premenopausal breast cancer risk by ER-positive and ER-negative tumor subtypes, luteal free estradiol pattern scores were positively associated with both ER-positive and ER-negative cases ( Table 5 ). For ER-positive cases, participants in the fifth quintile of the luteal free estrone-related pattern had a multivariable-adjusted HR (95% CI) of 1.25 (1.05–1.50; P trend = 0.01) compared to the first quintile, with a corresponding HR (95% CI) of 1.42 (0.99–2.03; P trend = 0.04) for ER-negative tumors. Higher follicular estrone (HR for fifth vs first quintile = 1.26; 95% CI = 1.04–1.51; P trend = 0.03) and luteal estrone (HR for fifth vs first quintile = 1.26; 95% CI = 1.05–1.51; P trend = 0.03) pattern scores were associated with a higher risk of ER-positive tumor subtypes, while no associations were observed between these patterns and ER-negative tumors ( Table 5 ). However, these differences were not statistically significant (all p hetereogeneity > 0.08). Additionally, no differences in our associations were observed when stratifying results by current BMI (all p interaction > 0.44) (data not shown). When examining early adulthood dietary intake, results were in similar directions but slightly attenuated for most of the examined dietary patterns (data not shown).

The NHSII is an ongoing prospective cohort study that began in 1989 and includes 116,429 female nurses aged 25–42 years living in 14 U.S. states. Using questionnaires submitted biennially, participants provide information on health and disease diagnoses, lifestyle, and sociodemographic factors[ 40 ]. The study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health and those of participating registries as required. Between 1996 and 1999, a total of 29,611 NHSII participants (ages 32–54) provided blood samples, the methods of which have previously been described[ 41 ]. In brief, premenopausal participants who had neither taken oral contraceptives nor been pregnant or breastfed within the past six months provided timed blood samples within the early follicular phase (when estrogen levels are low and derived from ovarian and non-ovarian sources [e.g., adipose tissue]) and the mid-luteal phase (when there is a secondary rise in estrogen levels derived primarily from the ovary) of their menstrual cycles[ 4 , 41 ]. The remaining participants ( n = 11,090) provided single untimed samples. Beginning in 1991, NHSII participants have completed a semi-quantitative Food Frequency Questionnaire (FFQ) containing over 130 food items every four years. The FFQs contain information on major types of food consumed and portion sizes and frequency of consumption. For most items, there were nine possible responses indicating the frequency of consumption, ranging from almost never to 6 or more times per day. The reproducibility and validity of the FFQ have been previously described[ 42 – 46 ]. For the current study, we used dietary information collected in 1991, 1995, 1999, 2003, 2007, and 2011. To derive the dietary patterns, we identified premenopausal participants who had previously assayed blood samples as part of prior nested case–control studies of breast cancer (n = 1,673), endometriosis (n = 668), ovarian cancer (n = 109), rheumatoid arthritis (n = 166), and benign breast disease (n = 492)[ 41 , 47 – 50 ]. We included only timed blood samples for estradiol, estrone, and estrone sulfate and included timed and untimed samples for testosterone[ 4 ]. For luteal estradiol, we restricted to ovulatory cycles (defined as mid-luteal progesterone ≥ 400 ng/dl)[ 51 ]. Hormone assay methods are described in Supplemental Materials [ 6 , 30 , 52 – 56 ]. To identify dietary patterns associated with our hormones of interest (estradiol, free estradiol, estrone, estrone sulfate, and free testosterone), food items from the 1995 and 1999 FFQs (the FFQs closest to the blood draw) were grouped into 34 food groups based on nutrient profiles or culinary usage to use as predictors ( Supplementary Table 1 )[ 57 ]. Pattern derivation methods are described in Supplemental Materials [ 58 – 62 ]. The food groups identified for each hormone were used to create hormone-specific simplified dietary pattern scores to apply to the full NHSII cohort using FFQ data. Since alcohol consumption has been hypothesized to increase breast cancer risk through changes in sex steroid hormone levels, we conducted sensitivity analyses removing alcohol from the derived pattern scoring before their application to the full cohort. On each biennial questionnaire, participants were asked if they had been diagnosed with breast cancer in the previous two years. The National Death Index was routinely searched for participants who did not respond to the questionnaires. All participants (or next of kin for those who had died) who reported breast cancer were asked for permission to review the relevant medical records and pathology reports to confirm the diagnosis and abstract information on estrogen receptor status. Medical record review confirmed a diagnosis of breast cancer in 99% of the participants reporting diagnosis. Only invasive breast cancer was considered in analyses; cases of carcinoma in situ were censored at the time of diagnosis. Follow-up for associations between each dietary pattern and premenopausal breast cancer risk began in 1991 when 97,813 NHSII participants completed the baseline FFQ. We excluded participants who had an implausible total energy intake ( 3,500 kcal/day), left more than 70 food items blank on the 1991 FFQ, were postmenopausal, or reported a diagnosis of cancer (except non-melanoma skin cancer) before the start of the follow-up. The final analytic study population included 90,341 participants. Participants contributed person-time from study entry (1991) until breast cancer diagnosis, diagnosis of any other cancer (except nonmelanoma skin cancer), menopause, death, loss to follow-up, or until return of the 2015 questionnaire, whichever occurred first. Two exposure windows of dietary pattern intake were examined: adult cumulative average intake and early adulthood intake (assessed using the baseline FFQ). Cox proportional hazards regression, stratified by time interval and age in months, was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for premenopausal breast cancer associated with each dietary pattern score with the lowest quintile as the reference. Linear trend tests were performed for each dietary pattern by assigning the median value of each quintile to all participants in that group. As five different dietary patterns were examined, we considered a trend statistically significant if the p-value for trend was less than 0.01. Total caloric intake was included in both age-adjusted and multivariable models[ 63 ]. We considered two multivariable-adjusted models. Model 1 adjusted for total energy intake (continuous), age at menarche (< 12, 12, 13, ≥ 14 years), age at first birth (< 25, ≥ 25 years), parity (nulliparous, 1, 2, 3 + children), history of breastfeeding (≤ 6, ≥ 7 months), family history of breast cancer (yes, no), history of benign breast disease (yes, no), height (continuous), physical activity ( 27 metabolic equivalent task-hours/week), and oral contraceptive use (ever, never). Model 2 additionally adjusted for BMI at age 18 (< 20, 20–21.9, 22–23.9, 24–26.9, ≥ 27 kg/m 2 ) and weight gain since age 18 (lost 25 kg) as BMI has been hypothesized to impact breast cancer risk through effects on endogenous hormone levels[ 64 , 65 ]. In sensitivity analyses removing alcohol from the calculation of the dietary pattern scores, we included a third model adjusting for all variables in model 2 and alcohol consumption. We used competing risk analyses to examine if associations between each dietary pattern score and premenopausal breast cancer differed by estrogen receptor (ER) status of the tumor[ 66 ]. This method allows for the estimation of separate associations of each dietary pattern score by ER status and tests whether each dietary pattern has statistically different regression coefficients for the tumor subtypes. We also assessed the association for each dietary pattern by current BMI (< 25 vs ≥ 25 kg/m 2 ) due to the potential impact of body size on sex steroid hormone levels[ 58 – 62 ]. Effect modification was assessed with a likelihood ratio test comparing the model with the cross-product term between each dietary pattern and BMI with the main effects model only. All tests of statistical significance were two-sided, and all statistical analyses were performed using SAS version 9.4(SAS Institute Inc., Cary, NC).

To our knowledge, this was the first study to derive dietary patterns correlated with premenopausal sex steroid hormone levels. Dietary patterns were identified for luteal phase estradiol and estrone, luteal free estradiol, follicular estrone, and free testosterone, but not for follicular phase estradiol, free estradiol, or estrone sulfate or for luteal estrone sulfate. However, the patterns identified explained a low percent variation in each hormone’s levels. Alcohol was observed to have the most consistent influence on estrogen levels as it was identified in four of the estrogen-associated patterns, followed by fruit juice identified in three of our four estrogen-related patterns. We observed luteal free estradiol and follicular estrone-related dietary pattern scores to be positively associated with premenopausal breast cancer risk across all analyses and a higher luteal estrone-related pattern score associated with a higher risk of ER-positive tumor subtypes. However, when alcohol was excluded from the pattern definitions, only the association with the follicular estrone-related pattern remained. Our findings suggest that dietary patterns only weakly influence endogenous hormone levels in premenopausal individuals, and the relation between sex steroid hormone levels and premenopausal breast cancer risk is not likely driven by dietary intake. Our hormone-associated dietary patterns were identified using RRR, a hypothesis-driven method that uses disease-specific biomarkers (i.e., sex steroid hormones) to determine the combination of food groups that explains the highest variation in the biomarker (i.e., hormone) of interest[ 67 ]. RRR has previously been used to derive a dietary pattern associated with estrogen levels in a cohort of primarily post-menopausal participants (the Nurses’ Health Study)[ 31 ]. Fung et al. observed the correlations between their identified dietary pattern and estradiol and estrone sulfate to be 0.22 and 0.24, with 6.5 and 5.7% of the variation of these hormones explained by this pattern, respectively. In our premenopausal population, we observed correlations of 0.20 or less for all of our patterns, with the percent variation explained ranging from 2.6–4.1%. In comparison, Schulz et al. used RRR to identify a dietary pattern associated with fatty acid intake (i.e., SFA, MUFA, n -3 PUFA, and n -6 PUFA) in the European Prospective Investigation into Cancer and Nutrition Potsdam Study[ 68 ]. Fatty acids are more strongly influenced by dietary intake, and this pattern explained over 41% of the variation in fatty acid intake and significantly predicted breast cancer in a cohort of middle-aged participants. Not unsurprisingly, our results indicate that endogenous sex steroid hormones in premenopausal people are only marginally influenced by diet, which is supported by prior work in the NHSII. Hirko et al. observed an inverse association between AHEI scores (characterized by higher intakes of vegetables, fruits, whole grains, and nuts and legumes and lower intakes of sugar-sweetened beverages and fruit juice, red/processed meats, and trans fats) and circulating levels of follicular phase estradiol and estrone and luteal phase estradiol, free estradiol, and estrone when excluding alcohol from scoring and adjusting for BMI at blood collection[ 30 ]. Consistent with our results, Hirko et al. reported that the observed associations for AHEI were primarily driven by the fruit juice/sugar-sweetened beverage component: when omitting the sugary beverage component from the AHEI score, the associations with estrogen concentrations were no longer statistically significant[ 30 ]. Our luteal free estradiol pattern—characterized by a higher intake of alcohol, fruit juice, and cream soup—was significantly associated with premenopausal breast cancer risk. However, this association was attenuated after removing alcohol from the pattern score. Alcohol intake is a probable premenopausal breast cancer risk factor according to the World Cancer Research Fund Expert Report[ 20 ] and was a component in each of our estrogen-related dietary patterns. Previous studies have suggested that higher alcohol consumption may increase circulating sex steroid hormone levels[ 69 , 70 ], impact the metabolism and clearance of hormones[ 71 , 72 ], impact menstrual cycle characteristics[ 73 ], and promote androgen aromatization to estrogens[ 74 ], all of which may potentially prolong estrogen exposure and increase breast cancer risk. Some studies have observed alcohol intake to be positively associated with estrogen and androgen levels, particularly during the luteal phase[ 14 – 17 , 70 ]. However, studies that have observed no associations between alcohol consumption and premenopausal estrogen levels have methodologic limitations, including not accounting for menstrual cycle phases in their investigations[ 7 , 10 , 69 , 75 ]. In sensitivity analyses excluding alcohol from our dietary pattern scores, only follicular estrone-related patterns remained statistically significantly associated with premenopausal breast cancer risk. This suggests that alcohol may have been a primary driver of the associations for luteal free estradiol and luteal estrone-related dietary patterns, which is consistent with evidence that alcohol particularly impacts estrogen levels during the luteal phase. The follicular estrone-related dietary pattern—characterized by a lower intake of legumes and whole grains—was positively associated with premenopausal breast cancer risk across all analyses. In some[ 76 , 77 ] but not all[ 78 – 80 ] prior research, dietary patterns including legumes and whole grains have been inversely associated with premenopausal breast cancer risk. Higher dietary fiber intake has been hypothesized to reduce the amount of estrogens that are deconjugated and reabsorbed into the bloodstream through the colon, thereby decreasing circulating estrogen levels[ 81 ]. A recent meta-analysis observed that total fiber intake was inversely associated with premenopausal breast cancer risk[ 82 ]. Additionally, several studies have observed inverse associations between dietary fiber consumption and supplementation (i.e., wheat bran supplements) and plasma estrogen levels[ 83 ], particularly during the luteal phase[ 84 , 85 ]. The observed positive association between our follicular estrone-related pattern and premenopausal breast cancer remained after removing alcohol from the pattern definition, suggesting that alcohol was not the driving factor of this association. This indicates that a dietary pattern low in whole grains and legumes could partially influence premenopausal breast cancer risk through hormonal pathways. Alternatively, evidence suggests that diets with a higher inflammatory potential are associated with increased premenopausal breast cancer risk[ 86 , 87 ], and foods higher in dietary fiber, such as legumes and whole grains, have been observed to have anti-inflammatory effects[ 88 ]. Adult BMI and weight gain since early adulthood have been consistently inversely associated with premenopausal breast cancer risk, particularly for ER-positive tumors[ 64 , 65 ]. Previous studies have observed lower circulating levels of follicular and luteal estradiol and total estrone in individuals with a higher BMI[ 58 , 60 , 61 ], indicating that BMI may influence premenopausal breast cancer risk through effects on plasma hormone concentrations. Our luteal free estradiol and follicular estrone-related patterns were positively associated with breast cancer risk regardless of BMI status, and no effect modification by BMI was observed for any of our patterns. While some of the food groups included in each of our derived dietary patterns changed when excluding BMI as a response in RRR, the percent variation in hormone levels largely remained the same. These findings indicate that the association between hormone-related dietary patterns and premenopausal breast cancer risk is likely not mediated by BMI. Our study has several strengths. To our knowledge, this is the first study to derive premenopausal hormone-related dietary patterns. Estrogen level measurements were timed within the menstrual cycle, allowing assessment in the follicular and mid-luteal phases. We also accounted for BMI in the pattern derivations using BMI-adjusted biomarker levels. Our breast cancer risk analysis was strengthened by our large prospective cohort including 1,956 premenopausal breast cancer cases. We had dietary data collected at multiple time points, allowing us to examine cumulatively averaged intake, minimizing the measurement error of a single diet assessment. Finally, we had detailed and updated information on other important covariates, limiting residual confounding. Despite these strengths, our study has limitations. Information on diet was self-reported, and some measurement error is expected. However, validation studies have observed similar corrected correlation coefficients for different food items and nutrients when comparing semiquantitative FFQs to seven-day dietary records[ 43 – 46 ]. Additionally, despite a substantial sample size for hormone assays, the distributions observed within this sample population may not reflect the distributions of these hormones in the main cohort. However, NHSII participants who provided blood samples were found to be similar to the main cohort in age (43.2 versus 42.2 years), BMI (26 versus 26 kg/m 2 ), parity (1.9 versus 1.9 children), and ever oral contraceptive use (86% versus 88%)[ 41 ]. In conclusion, in this large prospective cohort study, higher luteal free estradiol and follicular estrone-related dietary pattern scores were associated with a higher risk of premenopausal breast cancer. However, the patterns identified explained a low percent variation in each of the hormones, suggesting that associations between our dietary patterns and premenopausal breast cancer risk are more likely to be attributed to non-hormonal pathways. Future research should consider other mechanisms through which dietary patterns may influence breast cancer risk, including inflammatory processes.

Evidence suggests that circulating sex steroid hormones may be positively associated with premenopausal breast cancer risk, particularly among individuals with estrogen receptor (ER)-positive tumors[ 1 – 6 ]. Dietary intake has been extensively studied as a modifiable risk factor for breast cancer, and certain dietary factors (e.g., fiber, fat) have been hypothesized to impact breast cancer risk through effects on endogenous hormone levels[ 7 – 12 ]. However, few clear associations have emerged. One exception is alcohol, which has consistently been associated with higher levels of estrogens and androgens[ 13 – 17 ] and is an established breast cancer risk factor[ 18 – 26 ], particularly for ER-positive cases[ 27 , 28 ]. Dietary patterns, which capture the complex and cumulative interactions between individual foods and nutrients, may be more predictive of disease risk[ 29 ]. Therefore, investigating the impact of dietary patterns on endogenous hormone levels may better reveal the potential role of diet in the association between sex steroid hormones and premenopausal breast cancer risk. Prospective studies on the association between dietary patterns and premenopausal hormone levels are limited. Adherence to the Alternative Healthy Eating Index (AHEI) has been inversely associated with premenopausal estrogens and some androgens, while similar healthy dietary patterns, including the Alternate Mediterranean Diet (aMED) and the Dietary Approaches to Stop Hypertension (DASH) diet, were not associated with hormone levels[ 30 ]. However, these dietary indexes were not developed to specifically capture the potential influence of diet on intermediate biomarkers (e.g., sex steroid hormones). Dietary patterns correlated with circulating estrogens and estrogen metabolites have been previously identified among postmenopausal individuals[ 31 , 32 ], but these patterns may not be as relevant to premenopausal breast cancer risk due to fluctuations in estrogen levels across the menstrual cycle and divergent risk factors for breast cancer based on menopausal status, including body mass index (BMI)[ 2 , 6 , 33 ]. Prior work suggests that premenopausal levels of estradiol, free estradiol, estrone, estrone sulfate, and free testosterone may be positively associated with premenopausal breast cancer risk[ 1 , 34 – 39 ]. Based on these associations, we sought to: 1) identify whether dietary patterns were associated with five sex steroid hormones among a subset of premenopausal Nurses’ Health Study II (NHSII) participants, and 2) examine the associations between any identified patterns and premenopausal breast cancer risk.

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s10549-025-07689-4 .