The role of body size in postmenopausal breast cancer etiology has been studied extensively. A higher body mass index (BMI), as a measure of overall adiposity, has been confirmed in numerous epidemiologic studies to pose a significant risk.1–5 Other measures of body size, such as weight, adult weight gain, height and waist-hip ratio have also been shown to have a positive association.6–16 The predominant hypothesis has been that after menopause, adipose tissue becomes the primary site of estrogen production through the aromatization of androgens, and that the higher concentrations of circulating estrogens, especially estradiol, increase the risk of breast cancer.5, 17–20 Additionally, obesity increases insulin levels, which in turn inhibits the secretion of sex hormone-binding proteins, thereby further increasing estrogen levels.17 Abdominal adiposity has been found to be more important in estrogen production than adiposity at other body sites.21 The association between body size and breast cancer has been consistently found among White females. Among the few studies examining the risk in other racial/ethnic populations with varying body size traits,22–24 investigators concluded that high body weight, weight gain and waist-to-hip ratio were significant risk factors among Asian women (705 cases, 547 controls), but that obesity was not a risk factor for Hispanics (798 cases, 924 controls) or African Americans (9,542 women). In the present study, we examined the association between anthropometric measures and postmenopausal breast cancer in women from the Multiethnic Cohort (MEC), representing five ethnic groups (White, African American, Native Hawaiian, Japanese and Latino) living in Hawaii and California.
The influence of body size on postmenopausal breast cancer risk was investigated among five racial/ethnic groups in the Multiethnic Cohort. Participants were 45–75 years old at recruitment (1993–1996), living in Hawaii and California. Of the 82,971 White, African American, Native Hawaiian, Japanese and Latina women included in this analysis, 3,030 were diagnosed with invasive breast cancer. Body mass index (BMI), height, weight and adulthood weight gain were associated with a significantly higher risk and, with the exception of height, were found to vary across ethnic groups. Native Hawaiians and Japanese with a BMI ≥30.0 compared to 20.0–24.9 kg/m2 had the highest risk (hazard ratio = 1.82, 95% confidence interval: 1.31, 2.54, p-trend = 0.001, and hazard ratio = 1.59, 95% confidence interval: 1.24, 2.05, p-trend < 0.0001, respectively). Current hormone replacement therapy use modified the impact of a high BMI, as non- and former users had a significantly higher risk compared to current users. BMI also had a more pronounced risk for advanced tumors compared to localized tumors. When both BMI and adult weight gain were analyzed simultaneously, adult weight gain, rather than BMI, was a significant risk factor overall. These findings emphasize the significance of maintaining a healthy weight throughout adulthood for the prevention of postmenopausal breast cancer.
Material and Methods
The MEC Study in Hawaii and Los Angeles was established to investigate lifestyle exposures, especially diet and cancer outcomes. The respective institutional review boards (University of Hawaii, University of Southern California) approved the study proposal. The design of the MEC Study has been detailed elsewhere.25 In brief, the cohort is comprised of more than 215,000 men and women aged 45–75 years at cohort recruitment between 1993 and 1996. Driver's license files were the primary sampling frames, and African Americans, Japanese Americans, Latinos, Native Hawaiians and Caucasians were the five targeted ethnic groups. All participants initially completed a self-administered 26-page questionnaire including a detailed dietary assessment as well as sections on anthropometry; physical activity; smoking behavior; history of medical conditions; reproductive history and family history of cancer. A brief 4-page follow-up questionnaire was administered in 1998–2000, and a second 26-page questionnaire was administered in 2003–2008.
Over 99,800 women reported being postmenopausal (by natural or surgical means and at what age) on the baseline questionnaire and were eligible for analysis. Women who did not report that their menstrual periods have stopped permanently but who were over the age of 55 years at cohort entry were also assumed to be postmenopausal; they accounted for less than 10% of the eligible population. Participants not from the five major ethnic groups were excluded from this analysis (n = 6,443). Additionally, women with breast cancer prior to cohort entry as reported on the baseline questionnaire or as identified by linkage to the tumor registries were excluded (n = 4,683). Women with implausible dietary energy and macronutrient intakes were also excluded as a quality control measure (n = 3,673). Implausible energy intakes were defined as those with more than three modified standard deviations (MSD) from the mean, where the MSD was calculated after excluding individuals in the top and bottom 10% tails of the log-energy distribution assuming a truncated normal distribution, by ethnicity. Similar exclusions were made for fat, carbohydrate and protein intakes using a range of mean ± 3.5 MSD. Women who had missing data on baseline weight and height were also excluded (n = 2,081). A total of 82,971 postmenopausal women were available for analysis.
Incident, invasive breast cancer cases were identified by linkage to the Hawaii Tumor Registry, the Cancer Surveillance Program for Los Angeles County and the California State Cancer Registry. All registries are population-based and members of the National Cancer Institute's Surveillance, Epidemiology and End Results (SEER) Program. Case ascertainment was complete through December 31, 2004. Cases were divided by stage at diagnosis, localized or advanced, the latter defined as spread to tissue beyond the breast. A total of 3,080 breast cancer cases were identified, of which 31% were advanced. Linkages to the National Death Index and death certificate files in Hawaii and California provided information on vital status and causes of death.
On the baseline questionnaire (1993–1996), participants were asked to self-report their current weight, weight at age 21 years and current height. These self-reported measures were compared against values from the drivers' license files and correlated well. The Spearman correlations for male and female heights were 0.96 and 0.94, respectively, and were very high (generally above 0.90) for each sex-ethnic group. For weight, the correlations were 0.88 and 0.90 for males and females, respectively. For most sex-ethnic groups, the correlations were above 0.90, except for African Americans and Latinos who were predominantly from California. Because the driver license renewal period in California was ten years compared to four years in Hawaii, the weights reported on the California driver's licenses were likely farther in time to the date of completing the baseline questionnaire and likely less accurate. BMI was calculated as weight in kilograms divided by height in meters squared (kg/m2) and categorized into four groups (<20.0, 20.0–24.9, 25.0–29.9, ≥30), using 20.0–24.9 as the reference group. Adult weight gain from age 21 was calculated as the difference between current baseline weight and weight at age 21. Weight gain and baseline weight and height were divided into quintiles based on the distribution for all females in the MEC.
Cox proportional hazards regression was used to assess the association of the anthropometric measures and breast cancer risk while adjusting for the effects of other risk factors or potential confounders. The interval of age at cohort entry to the earliest of age at diagnosis of breast cancer, death or December 31, 2004 (closure date) was used as the time scale. The proportional hazards assumption was assessed by examination of the log(−log(survival function)) versus log(age) plots omitting the time-dependent variable, and also by assessing the Schoenfeld residuals.26 To account for missing data in the adjustment variables, an indicator variable for each variable was included in the models. Adjustment variables included follow-up time from baseline (<2, 2–4.9, ≥5 years, as a stratum variable), ethnicity (White, African American, Native Hawaiian, Japanese, Latina, as a stratum variable), age at cohort entry (continuous), family history of breast cancer (yes, no), age at menarche (≤12, 13–14, >14 years, missing), age at first live birth (<20, 21–30, >30 years, missing), age and type of menopause (natural <45, 45–49, 50–54, ≥55 years; oophorectomy <45, ≥45 years; hysterectomy <45, ≥45 years; missing), number of children (0, 1, 2–3, 4, missing), smoking status (never, former, current, missing), hormone replacement therapy (HRT) (no estrogen use, past estrogen use, current estrogen use only, current estrogen with past or current progestin use, missing), alcohol use at least once a month during the past year (yes, no), energy intake (continuous as kilocalories per day) and metabolic equivalents (METs) of physical activity per day (tertiles: ≤1.48, 1.49–1.71, >1.71, missing). The known risk factors of age at menarche, age at first live birth, number of children, age and type of menopause, HRT use, alcohol use and weight were confirmed in past analyses in the MEC.27 Age at cohort entry, family history of breast cancer, smoking, caloric intake and physical activity were included as additional adjustment variables. Height in meters (quintiles: ≤1.55, 1.56–1.57, 1.58–1.63, 1.64–1.65, >1.65) was added as an adjustment factor for models with weight, weight gain and BMI to adjust for skeletal size. Additionally, a second set of models, adjusted only for age at cohort entry and ethnicity, was fit to compare how risk estimates, for women overall, may change using a minimal number of adjustment variables. Categorical variables were modeled as indicator variables representing group membership. Hazard ratios (HR) and 95% confidence intervals (CI) were reported for the anthropometric variables of interest. To test for a linear trend, continuous variables, created by assigning the ethnic-specific median value of each BMI group, weight quintile, weight gain group or height quintile were entered into the model. Heterogeneity of risks between ethnic groups was assessed using the Wald test for the cross-product terms between the anthropometric trend variable and ethnicity. Three-way interaction terms between BMI and ethnicity, with HRT (ever, never and current, former, never), alcohol use in the past year (yes, no) and smoking status (current, former, never) were similarly examined using the Wald test, with all 2-way interaction terms included. The difference in risk factor associations for the anthropometric variables between advanced and localized breast cancers was compared in an overall model using competing risk techniques, where each stage status was a different event and the Wald test was used.26 All analyses were performed in SAS, version 9.1 (SAS Institute, Cary, NC). p values were two-sided.
Baseline characteristics of the study population are given in Table 1. Native Hawaiians, representing the youngest group, reported a higher prevalence for family history of breast cancer and an earlier age at menarche, but were less likely to be nulliparous, and, similar to African Americans, were less likely to delay childbirth. Japanese represented the oldest group and were older at first birth and at menopause, and like Latinas, were less likely to have an early menarche. Whites and Japanese were the most common users of estrogen replacement therapy or any HRT. The various body size measures differed across ethnic groups: African Americans had the highest mean height, weight and BMI at baseline, while Japanese had the lowest height, weight (baseline and age 21) and BMI (baseline and age 21). More than two-thirds of African Americans (75%), Latinas (68%) and Native Hawaiians (68%) were overweight (BMI ≥25) compared to a minority of Whites (48%) and Japanese (30%), data not shown.
Table 2 shows the associations for the anthropometric factors. Since the results from the age and ethnicity adjusted models were similar to the fully adjusted models for all body size parameters (i.e., the risks for BMI per 5 kg/m2 increase from the age- and ethnicity-adjusted model and the fully adjusted model were comparable (HR = 1.06, 95% CI: 1.04, 1.10 and HR = 1.11, 95% CI: 1.07, 1.15, respectively)), the results from the fully adjusted models are presented in all tables. Height was a significant risk factor for women overall, but the interaction with ethnicity was not significant. Increased baseline weight, adjusted for height, showed a significant positive trend overall and in each ethnic group except Latinas (p-interaction = 0.003). Japanese had the largest relative risk, and the trend was significantly steeper than for the other groups. The trends for adult weight gain were similar to those for baseline weight (overall and by ethnicity); the HRs per 5 kg in weight gain for Japanese compared to Native Hawaiians, Whites, African American and Latinas were 1.16 vs. 1.07, 1.06, 1.05 and 1.03, respectively. Similarly, the risks for increasing levels of baseline BMI showed significant positive trends overall and in all ethnic groups except Latinas (p-interaction = 0.030). Japanese, Native Hawaiians, African Americans, Whites and Latinas had an increased risk of 25%, 15%, 8%, 6% and 4% per 5 kg/m2 increase in BMI, respectively. In contrast, the findings for BMI at age 21 showed a significant inverse trend for women overall, though no evidence for an interaction with ethnicity was found. When the Latina group was subdivided into US born and non-US born (primarily Mexico, Central and South America), results from the multivariate models were similar to those for the combined Latina group. For instance, risks for BMI ≥30 vs. 20.0–24.9 kg/m2 for US born and non-US born Latinas were similar (HR = 1.22, 95% CI: 0.90, 1.65 and HR = 1.31, 95% CI: 0.88, 1.93, respectively).
Since HRT provides an exogenous source of estrogen and increases breast cancer risk,18 the joint effects of BMI and HRT use (current, former, never) were assessed in Table 3. Among all women, HRT use modified the impact of BMI (p-interaction = 0.0008), specifically current HRT use. A significant difference in trends was observed for current and never users (p = 0.0004), but not for former and never users (p = 0.83). A higher relative risk was found for BMI ≥30 kg/m2 (compared to 20.0–24.9) among never users (HR = 1.60, 95% CI: 1.36, 1.87) and former user (HR = 1.60, 95% CI: 1.27, 2.01) than among current users (HR = 1.14, 95% CI: 0.97, 1.35). These results were consistent among Japanese and Whites (p-interaction = 0.02 for the three-way interaction between all five ethnic groups, three HRT use strata, and BMI trend). Ethnic-specific differences were observed among never users (p-interaction = 0.0003) and suggested among former HRT users (p-interaction = 0.052) only.
Table 4 shows the relationship between BMI and cancer risk by stage. Among all women, BMI posed a much stronger risk for advanced than localized cancer (p-interaction = 0.0002). Women with a BMI ≥30 vs. 20.0–24.9 kg/m2 had an 82% increased risk for advanced cancer compared to a 22% increased risk for localized cancer. Risk differences between BMI and ethnicity for these cancers were not found (p-interaction = 0.53 for the three-way interaction term). Differences in risks across ethnic groups were observed for localized cancer only (p-interaction = 0.009), with significant positive trends reported only for Japanese and Native Hawaiians.
Since adult weight gain and BMI were significant risk factors in individual models (Table 2), both of these measures were included in the same model in an effort to separate the effects (Table 5). In examining them together, weight gain appeared to be a more important risk factor than BMI among all women. Ethnic-specific differences were found for adult weight gain (p-interaction = 0.01), but not for BMI (p-interaction = 0.31). The relationship between weight gain and relative risk still remained for Japanese, African Americans and Whites. Japanese women who were in the highest weight gain quintile (compared to the second lowest) had the highest risk among all ethnic groups (HR = 2.05, 95% CI: 1.40, 3.00, p-trend <0.0001). However, weight gain was no longer a significant risk factor for Native Hawaiians, but rather BMI remained a risk. When adult weight gain and baseline weight were modeled together, the results were generally similar (data not shown). Additionally, the results did not change when the first two years of cases were excluded from all analyses to avoid including women with pre-clinical disease.
In this analysis of postmenopausal women in the MEC, adulthood weight gain and adiposity, as measured by weight and BMI, were significantly associated with an increased risk of breast cancer. The magnitude of the effects was significantly different across ethnic groups, with Japanese and Hawaiians having much higher risks and Latinas showing no association. The overall adiposity associations were significantly more pronouncedly among past and never HRT users and for advanced tumors than among current HRT users and for localized tumors, respectively. Adult weight gain, rather than adult BMI itself, was a significant risk factor among Japanese, African American and White women; however, the reverse was suggested among Native Hawaiians. Neither factor was a significant risk factor among Latino women.
Obesity is a well-established risk for postmenopausal breast cancer. A pooled analysis using data from seven prospective cohort studies found that the relative risk (RR) increased up to a BMI of 28 kg/m2 and plateaued (RR = 1.26, 95% CI: 1.09, 1.46), using <21 kg/m2 as reference.4 However, our data showed a monotonic increase up to a BMI ≥30 kg/m2 (HR = 1.38, 95% CI: 1.24, 1.53), using 20.0–24.9 kg/m2 as reference. Our finding is supported by a recent meta-analysis,5 based on 31 prospective observational studies, which showed a positive linear relation (HR = 1.12, 95% CI: 1.08, 1.16 for each 5 kg/m2 increase). Authors of another recent meta-analysis, based on 10 prospective cohort studies, reported an increased risk for BMI 25–29.9 kg/m2 (RR = 1.08, 95% CI: 1.03, 1.14) and BMI ≥30 kg/m2 (RR = 1.13, 95% CI: 1.05, 1.22).28 These findings can be compared with our study, which found an increased risk per 5 kg/m2 (HR = 1.11, 95% CI: 1.07, 1.15), and for BMI 25.0–29.9 (HR = 1.22, 95% CI: 1.12, 1.33) and BMI ≥30 kg/m2 (HR = 1.38, 95% CI: 1.24, 1.53). Many prospective cohort studies found an association with weight gain from early adulthood to middle or old age.6–14, 16 In this study, BMI at baseline was highly correlated with weight gain since age 21 (Spearman's correlation of 0.81). However, weight gain was a stronger predictor than BMI when both variables were included in the same model. A possible explanation is that adult weight gain mainly reflects the accumulation of fat mass rather than lean body mass, reflecting age-related metabolic changes, whereas adult BMI reflects both fat and lean body mass without indicating changes over time.11, 13, 14
In this study, weight gain and BMI were risk factors for all ethnic groups except for Latinas. Other investigations among non-Whites found a positive association between body size and breast cancer risk. In a case–control study of Asian American women, Wu et al.22 found an 8% increased risk for every 5 kg BMI increase; an earlier case–control study of Asian American women similarly found an association with increasing height, weight and adiposity.29 In a Hawaii cohort assembled from public data sources, a nested case–control study found Hawaiians whose adult BMI increased had a significant increased risk (119%) for postmenopausal breast cancer.14 In contrast with the positive association in the MEC among African Americans between breast cancer and weight gain and BMI, the Black Women's Health Study found no relationship between anthropometry in this group.24 However, they did find that BMI at age 18 years was inversely associated with postmenopausal breast cancer, similar to our overall finding of BMI at age 21 years. Other studies have found similar results,10, 11, 30 which may be explained by an increased frequency of anovulation in young (premenopausal) obese women, resulting in lower estradiol and progesterone levels and lower rates of breast cell division.31
Hispanic women are at lower risk compared to non-Hispanic White women despite their higher degrees of adiposity.23 In this cohort, with adjustment for known risk factors including body weight, breast cancer risk of migrant Latinas was 16% lower than that of Whites.27 Latinas in the MEC were the only group that did not show a significant association, and the results were consistent among US and foreign-born Latinas. Two case–control studies compared the relationship of body size with breast cancer between Hispanic and non-Hispanic White women. One study observed that weight change and obesity were risk factors in both groups30 while the other study found this direct association in non-White Hispanics but not in Hispanics,23 similar to our findings. Several studies have compared US and non-US born Hispanics, reporting associations between a higher BMI and acculturation, neighborhood socioeconomic status and ancestry32–34; however, whether BMI, or other body size measures, is an independent risk factor for these subpopulations has not been directly addressed. A bi-national study involving the US and Mexico currently assessing how obesity is related to disease phenotypes for women of Mexican descent may provide some clarity.35
Differences in these findings across ethnic groups may be due to several factors. Endogenous sex hormonal profiles varied across ethnic groups among postmenopausal women in the MEC.36 Native Hawaiians had the highest estrogen levels and lowest sex hormone-binding globulin levels, whereas Japanese and African Americans had higher levels of estrogen than Whites, and Latinas and Whites had similar hormone profiles. Therefore, the groups in which body size has the biggest impact are also those with the highest estrogen levels, suggesting that adipose tissue may contribute to risk through estrogen production. Also, the prevalence of breast cancer by estrogen and progesterone receptor status varied significantly across ethnic groups. African American women had the highest fraction (31%) of estrogen receptor-/progesterone receptor-tumors which have weaker associations with obesity, whereas Japanese and Native Hawaiians had the lowest fraction (14%).37
To explain the ethnic variation, other factors should be examined, including the roles of physical activity and energy balance, and other adiposity measures. For example, central obesity rather than general obesity may specifically predispose one to developing breast cancer.38 Central adiposity was found to be more important in estrogen production than adiposity at other body sites.21 It may also result in increased circulating insulin and insulin-like growth factors, which act as mitogens in the body.39 The prevalence of central obesity appears to differ across ethnic groups. Preliminary results based on self-reported waist and hip circumference data from a recent MEC follow-up questionnaire suggest that the mean waist-to-hip ratio for White and Japanese women were lower than African American, Native Hawaiian and Latina women.
The association between obesity and breast cancer in this study was clearly more evident in past or never HRT users than in current users, who are at higher risk of developing breast cancer, as found in an earlier analysis in the MEC.40 Because HRT users have high levels of circulating estrogens regardless of their adiposity, the estrogen effect of obesity may not be substantial enough to further increase the risk of breast cancer.13 Therefore, similar to previous studies,9–13 these findings support the hypothesis that obesity increases risk through its estrogenic effects.
The relationship between BMI and breast cancer risk appears to vary by tumor stage. Two other large cohort studies11, 13, 41) reported that the increased risk associated with obesity was more pronounced for advanced tumors than for localized tumors. We observed the same findings as the risk for women with BMI ≥30 vs. 20.0–24.9 kg/m2 was higher for advanced cancer (HR = 1.82, 95% CI: 1.53, 2.17) than for localized cancer (HR = 1.22, 95% CI: 1.08, 1.38). The differences in the effects of BMI by stage could be attributed to socioeconomic status (SES). Obesity has been associated with lower SES,42 and heavier women, and women living in relatively poor conditions, are believed to undergo mammography and seek medical advice less frequently, which delays detection.43 In the MEC, education level is the only measure of SES available; however, we were unable to find any evidence that the varying effects of BMI by stage were attributable to education. Since educational level does not fully account for the effects of SES, we were limited in our ability to investigate this issue in depth. Heavier women were also thought to have a higher proportion of fatty breast tissue, which makes detection more difficult; however, studies have shown no difference in mammographic accuracy by BMI, which suggests more of a biologic effect on tumor development.41, 44 Overall, our findings are consistent with previous studies.
Strengths of the current study include its prospective design, large sample size and diverse ethnic composition. However, several limitations should be considered. We used self-reported weight and height, and relied on recall of weight at age 21, which are subject to measurement error. However, our self-reported values were found to agree well with driver's license height and weight, which in turn has been found to correspond acceptably to measured values.45 Our adiposity measures appear to be valid, since adult adiposity was associated with endometrial cancer across ethnic groups in the MEC.46 Still, we are unable to assess weight change patterns or fluctuation over time as weight from only two time points (age 21 and baseline) were available for this analysis. No information was available on perimenopausal weight gain, which tends to be related to a shift in body fat distribution (from a loss of subcutaneous fat to a gain of visceral fat). Also, for the present study, we had no measurement for central obesity that might contribute to postmenopausal breast cancer beyond that attributable to general obesity alone.10
In conclusion, our findings showed that excess adiposity was related to an increased risk of postmenopausal breast cancer among women in the MEC and that the magnitude of the effect varied across ethnic groups. While all groups but Latinas showed a positive relation, the impact was stronger in Japanese and Native Hawaiians. Specifically, a positive adult weight change appeared to be a stronger predictor compared to adult BMI overall, arguing that maintaining a healthy weight throughout adulthood is important for preventing postmenopausal breast cancer. This conclusion is consistent with the recommendation from the 2007 report of the World Cancer Research Fund/American Institute of Cancer Research.47 Further investigations among these diverse groups of women using additional measures of adiposity are warranted to more fully understand ethnic differences in the relationship of adiposity to breast cancer.
The authors thank Dr. Malcolm C. Pike for his contributions to this article.