Introduction

Predictors of unfavorable outcomes after breast cancer diagnosis include advanced stage at diagnosis and adverse tumor characteristics [1]. The estimated 5-year survival relative to the general population is 98% for localized disease, 84% for regional disease, and 23% for metastasized breast cancer [2]. Breast cancer is a biologically diverse disease characterized by pathologically heterogeneous tumors with different prognoses. Tumors lacking estrogen receptor (ER) and progesterone receptor (PR) expression lead to an estimated 2-fold lower survival than tumors expressing both receptors [3]. Research targeting potentially modifiable risk factors for breast cancer mortality is of public health importance given the number of deaths per year: an estimated 40,000 U.S. women were expected to die from breast cancer in 2010 [2].

Obesity represents one of the few potentially modifiable factors that may influence breast cancer outcomes, whereas epidemiologic data for other modifiable factors (e.g., physical activity and healthful diets) are less convincing [4]. Furthermore, obesity is a significant public health problem with approximately one-third of U.S. women estimated to meet the criteria for obesity, as defined by a body mass index (BMI) ≥ 30 kg/m2 [5]. Obesity has been linked to increased risk of all-cause and breast cancer-specific mortality [6], with limited data available for the effect of obesity on breast cancer-specific mortality in ethnically diverse populations [7–9]. Proposed biologic mechanisms for lower survival in obese women include elevated levels of bioavailable estradiol that contributes to the upregulation of cellular proliferation pathways and a microenvironment that promotes angiogenesis, invasion, and metastasis [10, 11]. Suboptimal treatment (e.g., under-dosage of chemotherapy) may also contribute to the lower breast cancer survival of obese women [12].

We examined the relationship between self-reported, pre-diagnostic BMI and breast cancer survival in the Multiethnic Cohort (MEC) study of African American, Native Hawaiian, Japanese American, Latino, and Caucasian postmenopausal women in Hawaii and Los Angeles. Specifically, we hypothesized that overweight or obesity would predict lower survival.

Methods

Study population

The MEC study, an ongoing, prospective study of dietary, environmental and genetic factors in relation to cancer and other chronic diseases, enrolled 96,382 men and 118,869 women in 1993 through 1996, aged 45–75 years at cohort creation from Hawaii and California (primarily Los Angeles County). Details on the study design, response rates, and baseline characteristics have been given elsewhere [13]. The population-based sampling frames included drivers’ license records in both states, supplemented with voter registration lists in Hawaii and Health Care Financing Administration (MEDICARE) files in California. Subjects entered the cohort by completing a 26-page, self-administered mailed questionnaire. The institutional review boards at the University of Hawaii and the University of Southern California approved the study protocol.

Incident cases of invasive breast cancer were identified for the women among the five main ethnic groups, African American, Caucasian, Japanese American, Native Hawaiian, and Latino, who were free of a previous self-reported or registry-detected breast cancer diagnosis at the time of the baseline questionnaire. Following entry into the MEC, incident breast cancer cases were identified by linkage with the Los Angeles County Cancer Surveillance Program, the State of California Cancer Registry, and the statewide Hawaii Tumor Registry. These tumor registries participate in the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program [14]. Dates and causes of death were identified by computerized linkages with the databases of the California and Hawaii vital records offices and the National Death Index. Complete case and/or death ascertainment was available up to December 31, 2007. Excluded from the present analysis were women younger than 50 years old at cohort entry to remove premenopausal and perimenopausal women (n = 612), women with missing (n = 60) or extreme values for height or weight that resulted in BMI outside of the 15–50 kg/m2 range (n = 11). A total of 3,842 female breast cancer cases were included in the present analyses.

Data collection

The baseline questionnaire included sections on height and body weight that were used to calculate BMI (kg/m2), medical history, reproductive history, family history of breast cancer, diet, physical activity, and demographic information. The average time from the date of the baseline questionnaire to the date of breast cancer diagnosis was 6.5 ± 3.9 years (range: 0.0–14.6 years). The presence of co-morbidities, including diabetes, hypertension, heart attack or angina, and stroke, was assessed as follows: “Has your doctor ever told you that you had any of the following conditions?” Hormone replacement therapy (HRT) use was assessed as current or past use of estrogen or progesterone for menopause or other reasons. Usual physical activity was assessed as number of hours per day spent in moderate or vigorous activities, on average, within the past year. Tumor characteristics available from the cancer registries included hormone receptor status, stage, tumor size, and first course of treatment (surgery, chemotherapy, or radiotherapy) within the 6 months after diagnosis. The tumor registries did not collect data on recurrence or comorbidities.

Statistical analysis

The primary goal of these analyses was to describe the effect of BMI (<22.5, 22.5–24.9, 25.0–29.9, >30 kg/m2) at entry into the MEC, overall, and by ethnicity (African American, Caucasian, Japanese American, Native Hawaiian, and Latino), on all-cause and breast cancer-specific survival. Survival time was modeled as years starting at the date of the invasive breast cancer diagnosis and ending at the first of the following endpoints: (1) date of breast cancer-specific death, (2) date of non-breast cancer-specific death, or (3) the date of complete case and/or death reporting (12/31/2007). Women alive at the end of 2007 were censored; in the breast cancer-specific models, women who died of other causes were also censored. Statistical computing was conducted using SAS version 9.2 (SAS Insitute Inc., Cary, NC, USA), with a P value of <0.05 considered to be statistically significant.

Ethnic-specific Kaplan–Meier survival curves were compared using a log-rank test. Hazard ratios (HR) and corresponding 95% confidence intervals (CI) were estimated using log-linear (Cox) proportional hazards models. Models were adjusted for age at breast cancer diagnosis (50–59, 60–69, ≥70 years), years between cohort entry and breast cancer diagnosis (continuous), hormone receptor status [ER+PR+, ER−PR−, ER+PR−/ER−PR+ (mixed), other/unknown], SEER summary stage (local, regional, distant, and unstaged/unknown), surgery (conserving, mastectomy, and none/unknown), radiotherapy (yes, no/unknown), chemotherapy (yes, no/unknown), and smoking status at baseline (never, former, current, and missing). The models were also adjusted for cardiovascular comorbidities (none, hypertension, heart attack/angina/stroke) because a previous study of the MEC reported ethnic differences in cardiovascular disease mortality [15]. The models were not adjusted for diabetes for two reasons: (1) diabetes may be on the pathway between obesity and poor breast cancer survival [16] and (2) obesity and diabetes were highly correlated in this study population (46.5% of women who had diabetes were obese). The addition of the following variables did not change the estimated HRs for BMI by more than 10% and were subsequently not included in the final model: family history of breast cancer (yes, no/unknown), age at menarche (<13, 13–14, >14 years), age at first birth (no children, <20, 21–30, >30 years), number of children (0, 1–2, >3), HRT (no estrogen use, current or past use of estrogen alone, current or past use of estrogen and progesterone, and missing), and physical activity (quartiles). The proportional hazards assumptions were assessed by examining log(-log(survival function)) plots for diverging or crossing survival curves over time, testing the statistical significance of time-by-covariate interaction terms, and assessing the Schoenfeld residuals. Stage and hormone receptor status were found to violate the proportional hazard assumption and, consequently, both were modeled as time-dependent variables by including cross-product terms with log-transformed survival time. Trend tests were performed by entering the categorical variable as a continuous parameter in the corresponding models.

We performed a number of additional analyses. First, a non-linear relation between BMI and survival was examined non-parametrically with restricted cubic splines and tested using the likelihood ratio test comparing the model with the linear term to the model with the linear and the cubic spline terms [17]. Second, we examined interactions with BMI, e.g., whether the association between BMI and risk for mortality varied by years between cohort entry and diagnosis (continuous log-transformed), ethnicity, hormone receptor status (ER+PR+, ER−PR−, and mixed), stage (localized, regional, and distant), age at diagnosis, or cardiovascular comorbidity. Interaction effects were assessed using subgroup specific models and global models including cross-product terms that were compared to models with main effects only. BMI was entered in the interaction models as a continuous log-transformed measure and only women with BMI > 22.5 kg/m2 were included due to the non-linear relationship present at lower BMI values.

Results

We identified 3,842 women diagnosed with incident, invasive breast cancer, with Japanese American, Caucasian, African American, Latino, and Native Hawaiian women comprising 30, 26, 19, 16, and 9% of the cases, respectively (Table 1). The mean BMI at cohort entry was 26.7 ± 5.3 kg/m2 and mean age at diagnosis was 68.8 ± 7.7 years. After a mean follow-up of 6.2 ± 3.8 years, a total of 804 deaths were reported, including 376 breast cancer-specific deaths. Japanese American women had a higher probability of 5-year survival from all-cause and breast cancer-specific death, tended to have lower BMI (<22.5 kg/m2), and were more likely to be diagnosed with localized disease or tumors ≤2 cm compared to the other ethnic groups. In contrast, African American women had a lower probability of survival and were less likely to be diagnosed with localized disease or with ER+PR+ tumors compared to the other ethnic groups. Both African American and Native Hawaiian women were more likely to be obese (BMI ≥ 30.0 kg/m2) and to have reported a cardiovascular comorbidity compared to Caucasian, Japanese American, and Latino women.

Table 1 Characteristics of breast cancer cases in the Multiethnic Cohort Study diagnosed 1993–2007
Full size table

In multivariable analyses, ethnicity, BMI at cohort entry, and surgery were statistically significant predictors of all-cause and breast cancer-specific mortality (Table 2). Japanese American women had lower risk of all-cause and breast cancer-specific mortality relative to Caucasian women; only the estimate for all-cause mortality reached statistical significance with an estimated 35% lower risk. African American and Latino women had higher breast cancer-specific mortality relative to Caucasian women; only the estimate for African American women reached statistical significance with an estimated 44% higher risk. Obese women had a statistically significant 54% higher risk of all-cause mortality and 45% higher risk for breast cancer-specific mortality relative to women with high-normal BMI (22.5–24.9 kg/m2). A higher, though non-significant, risk of mortality was observed in women with low-normal BMI (<22.5 kg/m2) relative to high-normal BMI. The suspected non-linear association of BMI with mortality, with higher risk of mortality associated with both obesity and low BMI, was confirmed with non-parametrical, restricted cubic splines (P < 0.01 for all-cause and P = 0.11 for breast cancer-specific mortality) (Fig. 1). Diabetes was significantly associated with mortality (all-cause HR = 2.05; 95% CI: 1.70, 2.48 and breast cancer-specific HR = 1.46; 95% CI: 1.06, 2.01) and adjustment for diabetes slightly attenuated the estimated HRs for the effect of obesity (all-cause HR = 1.35; 95% CI: 1.08, 1.69 and breast cancer-specific HR = 1.36; 95% CI: 0.98, 1.89).

Table 2 Risk of all-cause and breast cancer-specific mortality for breast cancer cases in the Multiethnic Cohort study diagnosed 1993–2007
Full size table
Fig. 1
figure 1

HR and 95% CI for all-cause (a) and breast cancer-specific (b) mortality predicted by BMI using restricted cubic splines with knots at 22.5, 25.0, and 30.0 kg/m2 and 25.0 kg/m2 as the reference level. Models adjusted for age at diagnosis, ethnicity, cardiovascular comorbidity, treatment, stage, hormone receptor status, smoking status, and years between cohort entry and diagnosis. Stage and hormone receptor status were included as time-dependent variables. BMI body mass index, HR hazard ratio, and CI confidence interval

Full size image

Years between cohort entry and diagnosis was not statistically significantly related to mortality (all-cause HR: 0.99; 95% CI: 0.96–1.01; breast cancer-specific HR: 0.98; 95% CI: 0.94–1.01), indicating that BMI at cohort entry likely represented the adult obesity levels that are relevant to mortality risk. Furthermore, there was limited evidence that years between cohort entry and diagnosis modified the effects of ethnicity or BMI level on all-cause and breast cancer-specific mortality (P interaction > 0.20 for all) (Table 3). Of interest was a statistically significant effect of age at diagnosis on breast cancer-specific mortality that was limited to women whose diagnosis was more than 4.25 years from entry into the cohort (P interaction = 0.03).

Table 3 Risk of all-cause and breast cancer-specific mortality by years between cohort entry and diagnosis for breast cancer cases in the Multiethnic Cohort study diagnosed 1993–2007
Full size table

Obese women, relative to women with high-normal BMI, had higher all-cause mortality across all ethnic groups, although the HR was only statistically significant for Caucasian women, and higher breast cancer-specific mortality across all ethnic groups except Native Hawaiian (Table 4). There was limited evidence of ethnic differences in the BMI effect on all-cause (P interaction = 0.87) or breast cancer-specific mortality (P interaction = 0.63). However, obese Caucasian and Native Hawaiian women had somewhat higher risk for all-cause mortality, and obese Caucasian and Japanese American women had somewhat higher risk for breast cancer-specific mortality compared to the other ethnic groups. Obese women diagnosed with ER+PR+ compared to ER−PR− tumors had a slightly higher risk of breast cancer-specific mortality (HR = 1.77; 95% CI: 0.95, 3.32 vs. HR = 1.18, 95% CI: 0.63, 2.20, respectively) (P interaction = 0.52). No significant interaction was observed between BMI and age at diagnosis, stage, radiotherapy, or cardiovascular comorbidity on all-cause or breast cancer-specific mortality (P interaction > 0.40 for all).

Table 4 Modification of all-cause and breast cancer-specific mortality by BMI level and ethnicity for breast cancer cases in the Multiethnic Cohort study diagnosed 1993–2007
Full size table

Discussion

In this ethnically diverse cohort of postmenopausal women diagnosed with breast cancer, all-cause and breast cancer-specific 5-year survival rates were 84 and 91%, respectively. After adjustment for breast cancer characteristics, including hormone receptor status, stage at diagnosis, and treatment, obese women had a modest increased risk of both all-cause (54%) and breast cancer-specific (45%) mortality compared to women with a high-normal BMI; however, being overweight did not significantly affect survival. There was limited evidence of ethnic differences in the obesity effect on all-cause or breast cancer-specific mortality, with a similar trend of higher risk with obesity across ethnic groups.

Our findings support the majority of individual studies [7, 8, 18–22] and meta-analyses [4, 6] that have provided evidence of a higher risk for breast cancer-specific mortality and/or recurrence in postmenopausal obese compared to non-obese women, with HRs ranging from 1.2 to 2.0 in obese women. Similar to our study, the higher risk of mortality associated with obesity has been shown in other studies that included ethnically diverse populations [7–9]. A higher, though non-significant, risk of mortality was observed in women with low-normal BMI relative to high-normal BMI that is consistent with other studies that have also observed a similar U-shaped association of BMI on mortality in breast cancer survivors [21, 23]. In particular, population-based studies in Korea [24] and Shanghai [21] found that being underweight (<18.5 kg/m2 relative to 18.5–24.9 kg/m2) was a risk factor for breast cancer recurrence and mortality. It is possible that high-normal BMI represents an optimal weight or that women with low-normal BMI are more susceptible to the adverse affects of breast cancer treatment or have a higher prevalence of pre-existing conditions not collected in our study. The U-shaped association is also consistent with studies of the effect of BMI on overall mortality [25, 26] including a collaborative analyses of 57 prospective studies that found a similar optimal relative weight of 22.5–25.0 kg/m2 [26].

Obese women have higher circulating levels of estrone, estradiol, and free estradiol [27] due to the aromatization of androgens in peripheral or mammary adipose tissue. The higher circulating and local free estradiol may stimulate residual neoplastic cell proliferation [11] and contribute to more biologically aggressive ER-positive tumors in postmenopausal women [28]. Estrogen-independent pathways may also mediate the obesity-induced reduced survival. Several possible pathways are possible, notably adipokine production (e.g., adiponectin and leptin) and hyperinsulinemia may contribute to an aggressive breast cancer phenotype by promoting angiogenesis, invasion, and migration [10, 29–31]. Moreover, pre-diagnostic diabetes, hyperinsulinemia, and low levels of adiponectin are associated with worse breast cancer outcomes [16, 31, 32]. Systematic under-treatment may be another factor contributing to lower survival rates among overweight and obese women. Despite compelling evidence that weight-based doses are required to achieve full benefits of adjuvant chemotherapy [33]; overweight and obese women often receive unintentionally reduced doses [12]. Diabetic women may also receive less aggressive treatment [34, 35] due to diabetic-related complications [34].

Strengths of the current study include a population-based prospective design with a large number of ethnically diverse postmenopausal women. Our study is more easily generalizable to the largest ethnic groups in the US population than previous studies of primarily white women. The MEC is also well characterized with respect to covariates of interest allowing for adjustment of several potential confounding factors, such as HRT use, smoking, and comorbidities. An additional strength is the ascertainment of incident breast cancer diagnoses through linkage with high-quality population-based tumor registries that provided detailed information on tumor characteristics as well as treatment within the 6 months of diagnosis.

Several limitations should be considered. Weight and height were self-reported and lean women tend to overestimate body weight, whereas overweight and obese women tend to underestimate BMI values [36]; however, the net effect would be attenuated HRs in that the increased mortality would be a conservative estimate of the true effect. Furthermore, health risk estimates associated with variations in BMI are similar whether based on self-report or measured BMI values [36]. The time between the completion of the baseline questionnaire and breast cancer diagnosis varied between women; however, models were adjusted for years between cohort entry and diagnosis, and no significant interaction was observed between this variable and BMI on mortality. Limited data were available for pre-existing conditions at the time of entry into the cohort and post-diagnostic BMI, weight change, or comorbidities. Given the reliability of the National Death Index linkage (e.g., sensitivity ≥87%) [37], misclassification of vital status is unlikely. As we only had information on mortality and not breast cancer recurrence, our study was limited by the small sample size for number of deaths of Native Hawaiian and Japanese American women that affected the precision of our HR estimates. Data on hormone receptor status was incomplete for 31% of the women and data on Her2/neu expression was unavailable.

Our findings are of public health importance because of the potential weight gain reported as a result of treatment regimens for breast cancer, notably adjuvant chemotherapy [38]. The available data examining the influence of pre- to post-diagnosis weight gain on survival are limited and inconclusive [20, 21, 39, 40]. Further investigations among ethnically diverse groups of women are warranted as we did not have a sufficient number of breast cancer-specific deaths to definitively evaluate ethnic-specific trends in mortality. Our findings are consistent with the literature that maintaining moderate weight throughout adult life may be beneficial for breast cancer survival in postmenopausal women and this appears to hold for all ethnic groups.