An Adolescent and Early Adulthood Dietary Pattern Associated with Inflammation and the Incidence of Breast Cancer

Adolescence and early adulthood are highly susceptible periods for breast cancer carcinogenesis during a woman's life course. Mathematical models of breast cancer etiology have demonstrated that the years before first birth are critical in establishing breast cancer risk (1) and animal models support this period of increased vulnerability (2–5). Thus, the focus on dietary exposures during later adulthood may not target periods of heightened vulnerability.

Adult dietary patterns and breast cancer risk have been examined in several studies and recent review articles of this literature have suggested a possible inverse association between “prudent” or healthy dietary patterns and breast cancer risk but the results are not conclusive (6–9) and only one study has examined adolescent dietary patterns and breast cancer risk (10). Most previous studies on dietary patterns and breast cancer risk have used principal components analysis (PCA), which results in patterns that reflect the correlation structure between foods and consequently the derivation of the dietary patterns is independent of the endpoint of interest. Reduced rank regression, another technique for deriving dietary patterns, allows the selection of intermediate biomarkers that are specifically associated with the endpoint of interest and may be more predictive of disease risk (11).

We sought to investigate the relation between an adolescent and early adulthood dietary pattern associated with markers of inflammation and breast cancer among women in the Nurses' Health Study II. We also examined whether the associations between the adolescent and early adulthood dietary patterns and breast cancer differed by the menopausal status of the cases and the hormone receptor status of the tumor.

The Nurses' Health Study II (NHS II) was established in 1989 when 116,430 registered nurses from 14 states completed a baseline questionnaire on lifestyle factors and medical history. Follow-up questionnaires are sent to participants every 2 years to collect updated information on lifestyle factors, diseases, and other health-related topics. Implied consent was assumed upon completion and return of the questionnaire. This study was approved by the institutional review board at Brigham and Women's Hospital.

In 1997, participants were asked if they would be willing to complete a supplemental food frequency questionnaire about diet during high school (HS-FFQ; ref. 12). The HS-FFQ was completed by 47,355 women (83% of those sent the questionnaire) in 1998 when they were 33–52 years old. Women were excluded from the current analyses if they had an implausible daily caloric intake (<500 or ≥5,000 cal; n = 1,407), left more than 20 items on the HS-FFQ blank (n = 219), were missing height (n = 126), or were diagnosed with any cancer, except nonmelanoma skin cancer (n = 399) before 1989.

Adolescent diet was measured using the 124-item HS-FFQ. This questionnaire was specifically designed to include foods that were commonly consumed during the period from 1960 to 1980 when the women would have been in high school. Adult diet was assessed first in 1991 (dietary baseline), when participants were ages 27–44 years, and every 4 years after using a FFQ listing over 130 food items. Participants were asked how often, on average, they had consumed each type of food or beverage between the ages of 13–18 years (HS-FFQ) or during the previous year (adult FFQ). Nine responses were possible, ranging from never to 6 or more times a day. The validity of the HS-FFQ has been assessed in a population of young adults who had provided information 10 years earlier about their current diet while in high school. The HS-FFQ was compared to three 24-hour diet recalls and two 131-item self-administered Youth/Adolescent Questionnaires (YAQ), which were administered when the participants were ages 13–18 years (13). For the daily intake of 25 nutrients, the mean corrected correlation between the HS-FFQ and the YAQs was 0.58 (range = 0.40–0.88). In addition, within a subset of NHS II participants, recall of adolescent diet was reasonably reproducible and valid (14). To assess reproducibility, the HS-FFQ was readministered to 333 NHS II participants 4 years after the initial HS-FFQ and to assess validity, the mothers of NHS II participants reported information on their daughters' adolescent diets using the HS-FFQ. The average Pearson correlations for nutrients were 0.65 (range = 0.50–0.77) for NHS II participants and 0.40 (range = 0.13–0.59) for mothers report. The average Spearman rank correlation for foods were 0.60 (range = 0.37–0.77) and 0.30 (range = 0.10–0.61), respectively (14).

The inflammatory dietary pattern has been previously identified in a subset of women in the NHS using reduced rank regression (RRR), the derivation of this pattern has been described in detail elsewhere (11, 15, 16). In brief, biomarkers of C-reactive protein (CRP), IL6, and TNFα receptor 2 (15, 16) were assayed in a subsample of adult women who were controls from previous nested case–control and validation studies. The mean adult dietary intake from two FFQs completed within 0–3 years of blood draw was calculated and foods were grouped into up to 39 food groups. RRR was used to produce a linear function of food groups that explained the variation in the response variables (the biomarkers of interest). Food groups were retained in the pattern if the P value of the coefficient of the stepwise regression model was <0.05. A simplified pattern was then generated using stepwise linear regression with the RRR factor score as outcome and the food groups as predictors. Using the inflammatory pattern previously derived as described above, we then calculated simplified pattern scores for adolescent diet and early adult diet (1991 FFQ) by summing the intakes of the identified food groups taking into account their positive or negative associations with the original pattern scores (17). The inflammatory dietary pattern was characterized by higher intake of sugar-sweetened and diet soft drinks, refined grains (white bread, English muffins, bagels or rolls, muffins or biscuits, white rice, pasta, tortillas, pancakes, or waffles), red (hamburger, beef, lamb, pork, and meatloaf) and processed meat (hot dog, bacon, sausage, salami, and bologna), margarine, corn, other vegetables (celery, mushrooms, green pepper, eggplant, summer squash, and mixed vegetables), and fish (tuna, mackerel, salmon, sardines, bluefish, swordfish, and other fish), and lower intake of green leafy vegetables (spinach, iceberg or head lettuce, romaine or leaf lettuce), yellow vegetables (carrots, yellow/winter squash, yams), cruciferous vegetables (broccoli, brussel sprouts, cauliflower, kale/mustard/chard greens, cabbage/coleslaw), and coffee.

On each biennial questionnaire, participants were asked whether they had been diagnosed with breast cancer in the previous 2 years. All participants who reported breast cancer were asked for permission to review the relevant medical records and pathology reports, to confirm the diagnosis. Estrogen receptor (ER) and progesterone receptor (PR) status information was available for 86% of the cases in our analytic cohort. Cases of carcinoma in situ were censored at the time of diagnosis.

Information on known and potential risk factors for breast cancer was collected on the baseline and biennial questionnaires. Women were considered premenopausal if they still had periods or had at least one ovary remaining and were <46 (for smokers) or <48 (for nonsmokers) years old. Women were considered postmenopausal in the analysis if they reported being postmenopausal during follow-up due to natural menopause or surgery with bilateral oophorectomy. Women who reported hysterectomy without bilateral oophorectomy or whose type of menopause was unknown were not classified as postmenopausal until they reached the age at which 90% of the cohort had reached natural menopause (54 years for current smokers, 56 years for nonsmokers; refs. 18, 19).

We examined three exposure windows for early life dietary patterns; adolescent intake (ages 13–18), early adulthood intake (assessed in 1991 when participants were ages 27–44 years), and the average of adolescent and early adulthood intake (representing cumulative exposure from adolescence to early/middle reproductive years). In the primary analyses of these three exposures participants contributed person-time from study entry until diagnosis of breast cancer, diagnosis of any other cancer (except nonmelanoma skin cancer), death, loss to follow-up, menopause (only for the premenopausal analysis), or end of follow-up on June 1, 2011, whichever occurred first (described hereafter as the combined analysis). This analysis includes all cases of breast cancer diagnosed before and after return of the HS-FFQ. In sensitivity analyses, participants contributed person-time from return of the HS-FFQ (1998; described hereafter as the prospective analysis). This analysis was restricted to only incident cases of breast cancer diagnosed after return of the HS-FFQ. Cox proportional hazards regression models with age and questionnaire period as the time scale were used to calculate HRs and 95% confidence intervals (95% CI) with the lowest quintile as the reference. Tests for linear trend were performed by assigning the median value of each category to all participants in that group.

We considered two main covariate-adjusted models; the first adjusted for adolescent characteristics and the second additionally adjusted for adult characteristics. We included the following a priori potential confounders in the adolescent characteristics model: age (continuous), total adolescence calories (continuous), height at age 18 (continuous), age at menarche (<11, 11, 12, 13, 14, 15+ years), body mass index (BMI) at age 18 (continuous), adolescent physical activity (<21, 21–35.9, 36–53.9, 54–80.9, 81+ metabolic equivalent tasks [METS]/week), and family history of breast cancer (yes, no). In addition, the following covariates were included in the adolescent and adult characteristics model which represents the final model: age at first birth (<25, 25–30, 31+ years), parity (nulliparous, 0, 1, 2, 3, 4+), oral contraceptive use (never use, past use <5 years, past use 5+ years, current use <5 years, current use 5–9 years, current use 10+ years), adult physical activity (<3, 3–8, 9–17, 18–26, 27–41, 42+ metabolic equivalent tasks/week), alcohol consumption (nondrinker, <7.5, 7.5–15, 15–29, 30+ g/day), weight change since age 18 years (continuous), and history of benign breast disease (yes, no). In the analyses including postmenopausal women, we additionally adjusted for menopausal status/age at menopause (premenopausal, unknown menopause, <45 years, 45–46 years, 47–48 years, 49–50 years, 51–52 years, 53+ years), and hormone use (premenopausal, postmenopausal never users, postmenopausal past users, postmenopausal current users). Categories were created for missing data. The derivation of menopausal status in this cohort has been described previously (18, 19).

Competing risks analyses were used to examine whether the associations between each dietary pattern and breast cancer differed by hormone receptor status (20). This method allows for the estimation of separate associations of each dietary pattern in tumors with both ER- and PR-positive receptors and ER- and PR-negative receptors, and tests whether each dietary pattern has statistically different regression coefficients for different subtypes. All tests of statistical significance were two sided. Statistical analyses were performed using SAS Version 9.3 (SAS Institute Inc.).

Among the 45,204 women who completed the HS-FFQ, 1,477 total cases of invasive breast cancer were diagnosed during 22 years of follow-up including 870 cases of premenopausal breast cancer and 490 cases of postmenopausal breast cancer. Women in the highest quintile of adolescent inflammatory pattern score had a higher BMI at age 18, were less physically active in adulthood, had gained more weight since age 18, and were more likely to have ever used oral contraceptives than those with in the lowest quintile (Table 1). Women in the highest quintile of adolescent inflammatory pattern score had an average intake of refined grains and red meat of 3.3 and 1.2 servings/day, respectively, whereas those in the lowest quintile had an average intake of 1.5 and 0.6 servings/day, respectively. In contrast, those in the lowest quintile of adolescent inflammatory pattern score had an average intake of leafy vegetables of 0.7 servings/day compared with 0.3 servings/day in the highest quintile (Table 1). Similar differences in intake levels were observed for foods in early adulthood (Table 1). The Spearman correlation between the adolescent inflammatory pattern score and early adulthood inflammatory pattern score was 0.31, the correlation between adolescent and later adulthood was 0.24, and between early adulthood and later adulthood was 0.34.

Among all women there was no significant association between a higher inflammatory dietary pattern score in adolescence and overall breast cancer incidence; however, a significant association was observed between a higher adolescent inflammatory dietary pattern score and incidence of premenopausal breast cancer. Women in the fifth quintile of adolescent inflammatory pattern score had a multivariable adjusted HR of 1.35 (95% CI, 1.06–1.73) for premenopausal breast cancer compared with those in the first quintile (P_trend = 0.002). The association observed in the prospective analysis (n = 536 cases) was slightly attenuated but the trend was still statistically significant (HR for fifth quintile, 1.32; 95% CI, 0.97–1.80; P_trend = 0.01). No association was observed with postmenopausal breast cancer (HR, 0.84; 95% CI, 0.60–1.17; P_trend = 0.33). A similar pattern was observed for the inflammatory dietary pattern in early adulthood, with no significant association overall or among postmenopausal women and a significant increased risk of premenopausal breast cancer among those consuming an early adulthood dietary pattern associated with inflammatory markers (HR comparing the fifth to first quintile, 1.41; 95% CI, 1.11–1.78; P_trend = 0.006; Table 2).

Among women who had completed both the adolescent and early adulthood FFQ (n = 42,770), we examined the average of the two intakes. Among all women a significant association was observed between a higher average inflammatory dietary pattern score and overall breast cancer incidence (HR, 1.25; 95% CI, 1.03–1.52; P_trend = 0.04). This appeared to be driven by the association with premenopausal breast cancer (HR, 1.49; 95% CI, 1.16–1.90; P_trend = 0.002) while there was no significant association with postmenopausal breast cancer incidence (HR, 1.01; 95% CI, 0.71–1.43; P_trend = 0.62; Table 2). The associations between the adolescent and early adulthood inflammatory dietary patterns and breast cancer remained the same when we adjusted for a more recent adult inflammatory pattern (data not shown).

We also evaluated the relation between the inflammatory dietary pattern and risk of breast cancer for ER-positive/PR-positive and ER-negative/PR-negative tumors (Table 3). No significant differences were observed in the effect estimates when ER-positive/PR-positive and ER-negative/PR-negative cases were compared (all P_{heterogeneity} ≥ 0.06); however, the strongest association was observed for the average of adolescent and early adulthood intake and premenopausal ER-negative/PR-negative cases (HR for fifth quintile, 2.21; 95% CI, 1.17–4.16; P_trend = 0.009).

We also examined the individual components of the inflammatory pattern score to explore if any specific food or food groups were driving the observed association with premenopausal breast cancer. Higher adolescent intake of processed meat was the most strongly associated with premenopausal breast cancer risk (HR for fifth quintile, 1.34; 95% CI, 1.07–1.68). None of the other adolescent food groups were statistically significantly associated with increased premenopausal breast cancer risk, although a suggestion of an increased risk was also seen with adolescent refined grain intake (HR for fifth quintile, 1.23; 95% CI, 0.96–1.56). When the average of adolescent and early adulthood intake was examined, none of the individual food groups were statistically significantly associated with risk but the highest suggested risks were observed for processed meat (HR for fifth quintile, 1.20; 95% CI, 0.94–1.54), red meat (HR for fifth quintile, 1.19; 95% CI, 0.92–1.53), and refined grains (HR for fifth quintile, 1.16; 95% CI, 0.91–1.48).

Our findings suggest an increased risk of premenopausal breast cancer among women consuming an adolescent and early adulthood diet that is associated with markers of inflammation. This increased risk did not extend to postmenopausal breast cancer and was not significantly different by hormone receptor subtype.

Few studies on adolescent diet and breast cancer have been undertaken because of the difficulty in assessing diet during this time period. Among the limited number of previous studies, higher adolescent soy (21–24) and fiber (25–28) intake have been suggested to have an inverse association with breast cancer risk, whereas red meat has been suggested to increase risk only among premenopausal women (29). Dietary pattern analysis, in which different combinations of food intake are examined, is a complementary approach to the study of individual foods and nutrients that takes into account diet as whole by examining the cumulative effects of foods (30). However, adolescent and early adulthood dietary patterns have rarely been examined in the context of breast cancer risk.

Most previous studies examining dietary patterns and breast cancer risk have focused on adult dietary patterns derived using principal components analysis (PCA)/factor analysis (6–9). More recently, a dietary inflammatory index (DII) has been developed on the basis of an extensive literature review of diet and inflammatory markers (31). A meta-analysis of adult dietary patterns derived using PCA/factor analysis has suggested an 7% reduced risk of breast cancer for women in the highest category of adult prudent/healthy dietary pattern score (95% CI, 0.88–0.98; ref. 7). Among the three study populations that have examined adult dietary inflammation potential and breast cancer outcomes, two reported no significant association with overall breast cancer risk (32–34) and one reported a borderline positive association between DII and breast cancer (35). To our knowledge, the NHS II cohort is the only study to examine adolescent dietary patterns derived by PCA, whereas no studies have examined an adolescent diet DII and breast cancer risk. In an analysis including both pre- and postmenopausal breast cancer cases, we previously observed a significant inverse trend with greater intake of a “prudent” adolescent dietary pattern, characterized by high intake of vegetables, fruits, legumes, fish, and poultry but no association with a “Western” dietary pattern, characterized by high intake of refined grains, red and processed meats, sweets, and potatoes (10). In PCA, the derived patterns reflect the correlation structure between foods thus are independent of the outcome of interest. The method for deriving dietary patterns that we utilized in the current analyses reduced rank regression, may result in dietary patterns more predictive of disease risk as this method utilizes intermediate biomarkers that are associated or potentially associated with the endpoint of interest (11).

The inflammatory pattern we examined has previously been associated with type II diabetes (16) and depression (15) in the Nurses' Health Study. However, to our knowledge, no previous studies have examined the association between this dietary pattern and incidence of breast cancer. The increased risk we observed for this dietary pattern was limited to premenopausal breast cancer. The risk factor profiles for pre- and postmenopausal breast cancer differ. Postmenopausal breast cancer has been found to be more strongly driven by hormonally related risk factors, whereas fewer modifiable risk factors have been identified for premenopausal breast cancer (36–39). Some studies have suggested that hyperinsulinemia and glucose are associated with increased premenopausal breast cancer risk (40–42). The association between the biomarkers used to define our inflammatory pattern (CRP, IL6, and TNFα receptor 2) and breast cancer risk is not entirely clear; however, this may be due to previous studies of these biomarkers examining primarily postmenopausal breast cancer. CRP has generated nonsignificant positive associations (43–45), significant associations among subgroups (46), or no association (47) with predominantly postmenopausal breast cancer risk in previous studies. IL6 has been nonsignificantly associated with postmenopausal breast cancer in a pooled analysis of two studies (43) and may have prognostic value in women with breast cancer (48), whereas TNFα has been shown to contribute to tumor progression in mammary cell lines (49). Our inflammatory dietary pattern was characterized by higher intake of sweetened soft drinks, diet soft drinks, refined grains, red meat, processed meat, margarine, corn, other vegetables, fish, and lower intake of green leafy vegetables, yellow vegetables, cruciferous vegetables, and coffee. The mechanism(s) through which this type of diet during adolescence and early adulthood may influence risk is not clear; however, these results are consistent with the established association between chronic inflammation and multiple types of cancer (50). This pattern may influence the risk of premenopausal breast cancer through pathways unrelated to the specific inflammatory biomarkers used in the pattern derivation.

There are several limitations to our study. First, we did not have biomarkers available from our participants during adolescence to identify the inflammatory pattern. Instead, we utilized a dietary pattern previously derived in primarily postmenopausal women in the NHS (15, 16) and then used a simplified score approach (17) to create these patterns in the women who had completed the HS-FFQ and early adulthood FFQ. If food intake during adolescence or early adulthood has different influences on inflammatory biomarker levels, this would not be reflected in our adolescent dietary pattern. In addition, the inflammatory biomarkers were not specifically chosen for their association with breast cancer. However, regardless of how the pattern was derived, we observed a clear association between the inflammatory pattern in adolescence and early adulthood and premenopausal breast cancer that is unlikely to be entirely explained by bias or chance.

Another limitation is that diet during high school was self-reported by participants when they were 33–52 years old and some error in its measurement is expected. However, the recall of adolescent diet in the NHS II has been demonstrated to be reasonably valid and reproducible (13, 14). In addition, our combined and prospective analyses yielded consistent results, thus any misclassification was likely nondifferential resulting in an attenuation of the true effect. Because a purely prospective cohort study linking adolescent diet to breast cancer would span many decades of data collection, the use of recalled diet in a prospective analysis provides an important means for studying dietary exposures and later life outcomes.

In conclusion, we observed an association between an adolescent and early adulthood inflammatory dietary pattern, characterized by high intake of sugar-sweetened and diet soft drinks, refined grains, red and processed meat, margarine, corn, other vegetables, and fish, and lower intake of green leafy vegetables, cruciferous vegetables, and coffee and the risk of premenopausal breast cancer. This association was strongest when the cumulative effect of adolescent and early adulthood diet were considered together but did not influence risk of postmenopausal breast cancer. This is an important finding as much less is known about modifiable risk factors for premenopausal breast cancer. Whether this association is mediated through inflammatory processes or other mechanisms deserves further study.

No potential conflicts of interest were disclosed.

Conception and design: H.R. Harris, W.C. Willett, K.B. Michels

Development of methodology: H.R. Harris

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.B. Michels

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.R. Harris, W.C. Willett, R.L. Vaidya, K.B. Michels

Writing, review, and/or revision of the manuscript: H.R. Harris, W.C. Willett, R.L. Vaidya, K.B. Michels

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W.C. Willett

Study supervision: K.B. Michels

We would like to thank the participants and staff of the Nurses' Health Study II for their valuable contributions as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY.

This work was supported by Public Health Service grant R03 CA170952 from the National Cancer Institute, NIH to K.B. Michels. The Nurses' Health Study II is supported by Public Health Service grant UM1 CA176726 from the National Cancer Institute, NIH, U.S. Department of Health and Human Services to W.C. Willett.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.