This study used data from the National Program of Cancer Registries (NPCR) Breast and Prostate Cancer Data Quality and Patterns of Care Study (POC-BP) to examine the relationships between patient BMI and mortality after breast cancer diagnosis. POC-BP data comprised breast and prostate cancer cases diagnosed in 2004 identified from population-based cancer registries in California, Georgia, Kentucky, Louisiana, Minnesota, North Carolina, and Wisconsin. Registry information was supplemented with data abstracted from medical records in hospitals and physicians’ offices. Follow-up data on women’s vital status were collected through the routine linkage of cancer registry data with death certificate data from state vital statistics departments, and through the use of the National Death Index (NDI) [45].

Women aged 20 or older diagnosed with primary breast cancer in 2004 (other than by autopsy or death certificate) were included in the POC-BP if they had no previous diagnoses of in situ or invasive breast cancer, other reportable cancers, Paget’s disease, Kaposi’s sarcoma, or lymphoma. Cases were selected from registries using random sampling stratified by race/ethnicity, as well as Appalachian/non-Appalachian region in North Carolina and Kentucky, facility type and volume in Wisconsin, and urban-rural status in Georgia.

This analysis focused on locoregional disease. A total of 6967 patients with Stage I, II, or III disease were identified. Women who did not receive any surgery for their breast cancer (n = 105), women with no vital status follow-up data (n = 6), and women with unknown BMI (n = 1462) were excluded, leaving 5394 cases.

Body mass index (kg/m²) was calculated based on weight and height, measured at the time of diagnosis, abstracted from physicians’ records.

Time to death (all-cause or overall mortality) was defined as days between diagnosis and death from any cause. Time to death from breast cancer was defined as days between diagnosis and breast cancer-specific death, as determined from ICD-10 codes in registry databases; we identified breast cancer deaths based on the presence of a primary cause-of-death code of C509. As a comparison, time to death from a cause other than breast cancer, as determined from the ICD-10 codes, was also examined. Survival analyses included data on women’s vital status through 8 years after the date of diagnosis (2004–2012). All registries contributed at least 5 years’ worth of vital status data, with some contributing up to 8 years.

Patient age was determined at time of diagnosis. Race/ethnicity information was obtained from patients’ medical records, and from information found in Indian Health Services patient registration. When Hispanic origin was unclear, a North American Association of Central Cancer Registries identification algorithm was used [45].

Insurance status was categorized as private, Medicare or other public insurance only, Medicaid, uninsured, or unknown. The “private” category included patients with Medicare plus private supplemental insurance. Women who were dual-eligible for Medicare and Medicaid were grouped in the Medicaid category.

Education and socioeconomic status were based on census-tract information. Patients were categorized as living in a higher poverty (≥ 20% of residents below the federal poverty level) or lower-poverty area, and as living in a lower-education (≥ 25% of adults age 25 or older with less than high-school education) or higher-education area. Residential areas were described as 100% urban, mixed, or 100% rural, using the 2000 U.S. Census Bureau’s urban and rural criteria [46].

Tumor pathologic stage was categorized as I, II, or III, as defined by American Joint Committee on Cancer (AJCC) staging criteria. Tumor grade was characterized by a I–IV score based on degree of cell differentiation, or as unknown. Estrogen receptor (ER) and progesterone receptor (PR) status were classified as positive (borderline was counted as positive), negative, or unknown.

A modification of the Piccirillo, or Adult Comorbidity Evaluation (ACE-27), comorbidity index [47] considered each of the 25 conditions (26 from the Piccirillo index, minus obesity, and not including the index breast cancer) and assigned a 1–3 score for “level of decompensation,” with 1 being “mild,” 2 “moderate,” and 3 “severe.” The overall comorbidity score was either based on the single highest ranking condition or assigned a value of 3 if the highest ranking ailments were grade 2 but occurred in different organ systems.

Age-adjusted, survey-weighted linear regression analyses were conducted to examine associations between BMI and other demographic and clinical variables. Next, overall mortality and breast cancer-specific mortality were compared across BMI categories, based on cutoffs defined by the World Health Organization (“underweight” if BMI < 18.5 kg/m², “normal” if BMI was 18.5–24.9 kg/m², “overweight” if BMI was 25.0–29.9 kg/m², “obese” if BMI was 30.0–34.9 kg/m², and “very obese” if BMI ≥ 35 kg/m²). Hazard ratios (HR) and 95% confidence intervals (CI) for each outcome were estimated using Cox proportional hazards models with survey weights and stratification to account for POC-BP sampling. Covariates were chosen a priori because of a known or potential relationship with BMI, and included race/ethnicity, health insurance, education, socioeconomic status, tumor registry, urban/rural residence, tumor size, tumor grade, hormone receptor status, and comorbidity. As a comparison, non-breast cancer mortality was examined in the same fashion. We calculated p values for linear trend by running a second version of each model, in which BMI was represented by a continuous variable.

After adjusting for clinical and demographic characteristics, obese women with a BMI of 30–34.9 kg/m² appeared to have a lower hazard for all-cause mortality (HR for a 5 kg/ m² difference in BMI = 0.93, 95% CI, 0.87–0.95; p value for trend = 0.02). However, the association between BMI and all-cause mortality was found to vary according to age at diagnosis (Table 3); p value for interaction = 0.01). Among women in the under-50 and 50–69 age groups, there was no significant association between BMI and overall mortality, but among women 70 and older, BMI was inversely related to overall mortality (p value for trend = 0.01). In particular, among those 70 or older, women with BMI < 18.5 kg/m² had a significantly elevated mortality hazard compared to women with a BMI of 18.5–24.9 kg/m² (HR = 1.92, 95% CI 1.11–3.34) and there was a suggestion that women in higher BMI categories had a lower mortality hazard (HR for BMI of 25–29.9 kg/m², 30–34.9 kg/m², and ≥ 35 kg/m² compared to women with a BMI of 18.5–24.9 kg/m² = 0.84, 0.73, and 0.78, respectively; 95% CIs 0.63–1.14, 0.51–1.05, and 0.52–1.17, respectively). In an analysis (not shown) in which women over 70 were examined after excluding underweight women, the association between higher BMI and lower overall mortality hazard remained significant as gaged by the p value for linear trend (p = 0.02), with hazard ratios for the overweight, obese, and very obese categories similar to those found in all women over 70 (0.84, 0.77, and 0.84, respectively).

In models stratified according to AJCC stage (Table 4), no significant differences were observed in the HR estimates for each stage, and indeed within each stage-specific model, the BMI-overall mortality relationship was not significant. There was no significant interaction between BMI and AJCC stage with regard to all-cause mortality (p = 0.83). In models stratified according to estrogen receptor status (Supplement Table 6), there was a suggestion that higher BMI might be inversely associated with lower all-cause mortality among women with ER-positive tumors (p value for trend = 0.03). Among women with ER-negative tumors, no significant overall association was found (p value for trend = 0.22). There was no significant interaction between ER status and BMI with regard to all-cause mortality (p = 0.29).

We explored whether the association between BMI and all-cause mortality might vary according to race/ethnicity or tumor registry. No significant interactions were observed.

BMI was not found to be associated with breast cancer-specific mortality in the overall model (Table 2). However, in models stratified according to AJCC stage (Table 5), specifically among patients with Stage I tumors, higher BMI was associated with greater breast cancer mortality hazard—in particular, women with BMI ≥ 35 kg/m² were found to have over four times the breast cancer mortality hazard (HR = 4.74, 95% CI 1.78–12.59) of women with a BMI of 18.5–24.9 kg/m². Among women with Stage II or III tumors, no relationship between BMI and breast cancer mortality was observed.

The association between BMI and breast cancer mortality was not found to differ according to age, estrogen receptor status (Supplement Table 7), race/ethnicity, or tumor registry.

Results for models of non-breast cancer mortality were generally similar to those for all-cause mortality (Table 2, Supplement Table 3, Supplement Table 4), including a significant inverse association between BMI and non-breast cancer mortality among women 70 or older (Supplement Table 3). The association between BMI and non-breast cancer mortality was not found to differ according to estrogen receptor status (Supplement Table 8; p = 0.35 for interaction).

In our analyses, we made use of a large sample of patients all diagnosed in the same year with locoregional disease, representing diverse geographic areas, racial/ethnic categories, and socioeconomic characteristics. The POC-BP data set was more complete than some registry-based studies due to additional resources made available to the investigators to verify data and collect more detailed information; thus, the study benefited from meticulously collected comorbidity and staging data. The study benefited from a relatively long follow-up period, with up to eight years of data from most of the tumor registries represented.

This study’s findings need to be considered in light of some limitations. Not all registries collected and provided data for the same length of time. Follow-up data for the full 8 years for all women would have improved statistical power, but we do not consider it likely that differences in follow-up duration by registry biased our results, because our statistical models accounted for tumor registry. Additionally, despite POC-BP study oversampling, there were low numbers of women in some racial/ethnic subgroups. We also could not investigate potential interactions according to menopausal status because data on menopausal status were not available. Additional data that were not available as part of the data set include information on women’s smoking status and concomitant medication use.

BMI is not a precise or consistent means of characterizing adiposity [53, 54]; it is problematic at the extremes of height, and may exhibit systematically larger inaccuracies among older women [53]. Some researchers have therefore leaned away from BMI in favor of measures such as waist-to-hip ratio. However, BMI is still frequently used, as in this retrospective analysis, in the absence of other available measures.

Missing data on height and/or weight pose a potential threat to the validity of our findings because they are unlikely to be missing at random. 21.3% of the women in the parent data set who were eligible to be included in the analytic data set were excluded due to missing information on weight, height, or both. We conducted a sensitivity analysis in which missing weight or height was imputed among women whose records showed one but not the other. Findings (not shown) were very similar to those from our main analyses reported here.

Although our investigation did not replicate some previous findings linking greater BMI to higher breast cancer mortality in the general female breast cancer population, we did identify women with Stage I disease as a group in which obesity, particularly greater than moderate obesity, was strongly linked to poorer outcomes with regard to breast cancer death. With continued improvement in diagnostic techniques and screening practices, more women will likely be diagnosed with early-stage disease, making it even more important to understand and address the disparities in outcomes within this subgroup. Another notable finding was the inverse relationship between BMI and all-cause mortality among women above age 70. The average life span of women is now well above 70, and many women are diagnosed at a later age. It is therefore of critical importance to determine whether, perhaps due to differences in tumor characteristics, comorbidities, or other factors, body weight and adiposity may have differing effects on outcomes in this subgroup.