Body Mass Index at Diagnosis and Breast Cancer Survival Prognosis in Clinical Trial Populations from NRG Oncology/NSABP B-30, B-31, B-34, and B-38

This article is featured in Highlights of This Issue, p. 1

In the United States, a woman has a one in eight chance of being diagnosed with breast cancer. Early detection combined with treatment advancements has greatly improved prognosis and survival. As of January 2014, there were more than 2.8 million American women living with a history of breast cancer including both current patients and survivors (1). Potentially modifiable lifestyle factors, such as overweight and obesity, have gained popularity as indicators for breast cancer prognosis and tools for risk reduction among these women.

Excess weight remains a large public health problem with two thirds of Americans being overweight and obese. Body mass index (BMI) is often the standardized method used for measuring and classifying excess weight. Existing literature suggests a relationship between increased BMI and an increased risk for developing breast cancer (2–6). However, associations by menopausal status have not been consistent, especially among premenopausal women (3, 5–12). Despite the inconsistencies by menopausal status, results have shown that the relationship between BMI and breast cancer risk appears to be limited to estrogen receptor (ER)-positive cancers.

In addition to breast cancer incidence, high BMI has also been associated with more aggressive or advanced tumors (13, 14) and with poorer outcomes (15–19). Earlier investigations using data from the National Surgical Adjuvant Breast and Bowel Project (NSABP) treatment trials of node-negative patients who were ER-positive (Protocol B-14) and ER-negative (Protocols B-13, B-19, and B-23) found that obesity was related to an increased risk of overall mortality, other primary cancers, and contralateral breast cancers but not recurrence (20, 21). More recently, a meta-analysis of 43 studies showed that survival among obese women with breast cancer was worse than that among nonobese women (16). However, the majority of the trials included were observational studies and included very few results from clinical trials with consistently evaluated and well documented treatment regimens and outcome measurements. In addition, a 2014 review summarized the findings of 11 studies (22), which included four retrospective investigations (23–26), four population-based epidemiologic studies (19, 27–29), and three secondary analyses from treatment efficacy clinical trials (17, 30, 31). In the review, the authors concluded that obesity is associated with negative clinical outcomes in breast cancer patients and is evident in several ethnic populations. They also observed a stronger relationship among ER-positive cancers regardless of menopausal status and no evidence of association in women with triple-negative disease. Furthermore, researchers from the Early Breast Cancer Trialists' Collaborative Group (EBCTCG) recently found obesity to be strongly related to breast cancer recurrence and mortality only in pre/peri-menopausal women with ER-positive disease (32).

Despite the abundance of literature regarding obesity, BMI, and breast cancer, there are limited data evaluating how patient characteristics, disease characteristics, and treatment may alter the association of BMI with breast cancer risk and outcomes. These are important aspects to consider, because we know there is substantial heterogeneity in breast cancer outcome by stage, type (i.e., node status, ER-status, and HER2-status), and the treatments used for these types of disease. Furthermore, it remains unclear whether the differences in findings throughout the literature have biologic basis, or are due to varied methods of the investigating studies. The purpose of this study was to investigate the consistency of the relationship between BMI at diagnosis and overall survival (OS) as well as that of BMI and breast cancer recurrence in well-documented and controlled clinical trial populations with various types of breast cancer and different treatments for the disease. The populations consisted of participants from four more recently reported breast cancer treatment trials conducted by the NSABP: B-30, B-31, B-34, and B-38.

NSABP protocols B-30, B-31, B-34, and B-38 were all multicenter, randomized, clinical trials. For each trial, OS and breast cancer recurrence were either primary or secondary endpoints. Generally, participants included women diagnosed with invasive breast cancer (IBC) who underwent either total mastectomy or lumpectomy with either an axillary dissection or sentinel node biopsy. Patients could not have had a prior history of IBC and were required to be free from other nonbreast malignancies for at least 5 years before entry. Other inclusion and exclusion criteria, along with further details of study designs have been previously published (33–36). All clinical centers obtained approval from institutional review boards, and all participants provided written informed consent. All significant values reported are statistically significant.

Protocol B-30 was conducted among women with operable, node-positive breast cancer. Between 1999 and 2004, 5,351 women were randomly assigned to one of three adjuvant treatment regimens: doxorubicin (A) and cyclophosphamide (C) followed by docetaxel (T) (AC→T), AT, or a concurrent regimen of TAC. Results were published in 2010 indicating a significant 17% reduction in mortality in AC→T versus AT, and a nonsignificant 14% reduction versus TAC. There was no difference in OS between AT and TAC (33).

Protocol B-31 comprised women with operable, node-positive breast tumors who overexpressed the HER2 protein. B-31 opened in February 2000 and began to randomly assign participants to a treatment regimen of either A and C followed by paclitaxel (P) (AC→P) or A and C followed by P plus trastuzumab (H) (AC→PH). Prior to the first interim analysis, the trial statistical plan was modified to be reported in a joint analysis with North Central Cancer Treatment Group (NCCTG) N9831. In response to the first joint interim analyses, enrollment to B-31 was stopped on April 26, 2005, at which time 2,119 patients had been randomly assigned. The joint analysis of B-31 and N9831 showed that the addition of trastuzumab reduced the rates of recurrence by half and the mortality rate by one third (34).

Protocol B-34 was a double-blind, placebo-controlled trial in a general population of early-stage breast cancer patients. A total of 3,323 women with operable breast cancer and no evidence of metastases were enrolled between 2001 and 2004. Most of these women had node-negative disease, and small tumors. Women were randomly assigned to receive either adjuvant oral clodronate for 3 years or a placebo. Local and systemic therapies were administered at the treating physician's discretion. Results were reported in 2012 and indicated no difference between treatment groups for disease-free survival (DFS), OS, recurrence-free interval (RFI), or bone metastasis-free interval (35).

Similar to B-30, protocol B-38 was limited to patients with node-positive breast cancer. Between 2004 and 2007, 4,894 women were randomly assigned to one of three treatment regimens: TAC, AC→P, and AC followed by P plus gemcitabine (G) (AC→PG). Results for B-38 were reported in 2013 (36). Adding G to AC→P improved neither DFS nor OS. There were also no significant differences in efficacy between AC→P and TAC, although toxicity profiles differed.

Collection of follow-up data for the trials was similar. Clinical, hematologic, and biochemical assessments were required at the beginning of each treatment cycle, every 6 months for the first 5 years, and annually thereafter. Patients received a physical breast examination at each follow-up visit and annual bilateral mammograms. Staff members from each clinical center performed follow-up and were required to submit documentation for each event reported. To confirm the diagnosis of each event, all documentation was centrally reviewed by medical professionals at the NSABP.

We assessed data from 15,538 breast cancer patients who participated in the four NSABP protocols. All participants with follow-up and supporting documentation were included (Fig. 1). Follow-up information for B-30, B-31, and B-34 included data as of December 31, 2012, representing an average of 9.0, 8.3, and 8.4 years, respectively. For B-38, a data cutoff of March 31, 2013, was used, representing an average follow-up of 5.9 years.

In all of the trials, the height and weight of each participant upon entry were measured by staff at each contributing center. Individual BMIs were calculated as weight in kilograms divided by height in meters squared. BMI is usually grouped into four categories of weight classification for adults: underweight (<18.5), normal (18.5–24.9), overweight (25.0–29.9), and obese (≥30.0). Approximately 1% of participants were underweight in these populations; therefore, they were combined with the normal group for analysis.

Information about important explanatory variables was also assessed upon entry. For all four studies, these included age, ER-status, number of positive nodes, and tumor size. Tumor grade was collected in all except B-38. Menopausal status was also collected; however, the defining criteria were different for each trial. In B-31, it was only collected for those who were ER-positive and thus was unknown for almost 80% of participants.

Because we know there is heterogeneity in breast cancer outcome by disease characteristics and treatment, we decided a priori to analyze each trial individually to explore the relationship of BMI with OS and recurrence among these four diverse populations. We used Cox proportional hazards regression to calculate adjusted hazards ratios (HR) for risk of recurrence and death. BMI was explored as a continuous variable and a categorical variable. However, because BMI categories are based on established cut points for overweight and obesity, and are both recognizable and clinically meaningful markers for excess weight, we only present results for the categorical classification. Breast cancer recurrence was defined as any local, regional, or distant recurrence. OS was defined as death from any cause. Time-to-event was measured from the date of assignment to the date of diagnosis or death, and all participants with follow-up were included in the analysis using the intent-to-treat method. P-values for tests for trend across BMI categories were obtained by including BMI as a single ordinal term (with values 0, 1, and 2) in Cox regression models and evaluating the global P-value for the term. Adjustment variables were treatment, age, menopausal status, ER-status, number of positive nodes, tumor size, and tumor grade, when available. We conducted separate analyses among ER-positive and ER-negative breast cancers because we had noted differences in the BMI/breast cancer risk association in previous studies (12) and it was of particular interest to investigate. We also conducted separate analyses among the different treatment groups to determine whether the effect of BMI was altered by different regimens. Forest plots were used to display results for which an HR > 1 indicates poorer survival or greater risk for recurrence when compared with participants with BMI < 25.0. Assessments of the statistical significance of interactions were based on the multivariable model for the respective study populations. P-values used to assess statistical significance of parameters in all modeling were determined using the likelihood ratio test. All tests were evaluated using a two-sided P-value of 0.05. Analyses were performed using SAS version 9.2 software (SAS Institute, Inc.).

The majority of women participating in each of the four trials were white (range, 83%–86%). The mean ages for participants of B-30, B-31, B-34, and B-38 were 51.1 (SD, 9.7), 49.6 (SD, 10.1), 54.1 (SD, 10.4), and 51.2 (SD, 9.5) years, respectively. Patient and tumor characteristics for this analysis by protocol and BMI group are presented in Table 1. Generally, obese women were older and postmenopausal with slightly larger tumors.

Results for OS are presented in Fig. 2. Among participants in B-30 (node-positive breast cancer patients), we found a statistically significant relationship between increased BMI and poorer survival. Compared with BMI <25, adjusted HRs were 1.04 for BMI 25 to 29.9 and 1.18 for BMI ≥30 (P = 0.02). Ten-year survival estimates were 79.0% for BMI < 25, 77.1% for BMI 25 to 29.9, and 74.9% for BMI ≥ 30. For the other trials, there were no statistically significant trends in OS across BMI categories. The adjusted HRs for overweight and obese women were: B-31: 0.83 and 0.94 (P = 0.58), B-34: 0.93 and 1.03 (P = 0.76), and B-38: 1.02 and 1.11 (P = 0.26). Absolute differences in 10-year (B-31 and B-34) and 7-year (B-38) survival estimates between BMI < 25 and the other two BMI categories were small ranging from 0.3% to 3.5% (Fig. 2).

Separate analyses by ER-status (Fig. 3A) showed that the statistically significant relationship in B-30 between increased BMI and OS was present in those with ER-positive breast cancer (P = 0.002) but not ER-negative (P = 0.88). The adjusted HRs for overweight and obese women were 1.03 and 1.30 for ER-positive breast cancer and 1.04 and 0.98 for ER-negative. No statistically significant results were found for either ER-positive or ER-negative disease for the other trials.

Separate analyses by treatment group (Fig. 3B) showed that the statistically significant relationship in B-30 between increased BMI and OS was present only in the AT group. The adjusted HRs for overweight and obese women in this group were 1.29 and 1.44 (P = 0.005). There were no statistically significant trends among those in the AC→T group (HRs: 0.85 and 1.15; P = 0.25) or the TAC group (HRs: 1.00 and 1.02; P = 0.87). There were also no statistically significant results among any of the treatment groups in the remaining trials.

Similarly to the results for OS, we found a statistically significant relationship between increased BMI and an increased risk of recurrence among B-30 participants only (Fig. 4). Compared with BMI < 25, adjusted HRs were 1.11 for BMI 25 to 29.9 and 1.17 for BMI ≥ 30 (P = 0.03). Ten-year recurrence-free estimates were 76.3% for BMI < 25, 74.6% for BMI 25 to 29.9, and 73.2% for BMI ≥ 30. For the remaining trials, there was no statistically significant trend across BMI categories. The adjusted HRs for overweight and obese women were: B-31: 0.78 and 1.02 (P = 0.91), B-34: 1.08 and 1.00 (P = 0.97), and B-38: 1.00 and 1.13 (P = 0.16). Absolute differences in 10-year (B-31 and B-34) and 7-year (B-38) recurrence-free estimates between BMI < 25 and the other two BMI categories ranged from 0.5% to 5.4% (Fig. 4).

Separate analyses by ER-status (Fig. 5A) again showed that the statistically significant relationship between increased BMI and recurrence in B-30 was present in those with ER-positive (HRs: 1.16 and 1.33; P = 0.001) but not ER-negative disease (HRs: 1.01 and 0.97; P = 0.78). There was also a statistically significant relationship in B-38 between increased BMI and recurrence in those with ER-positive breast cancer (HRs: 1.10 and 1.27; P = 0.03) but not ER-negative breast cancer (HRs: 0.86 and 0.93; P = 0.71). No statistically significant results were found in either ER-positive or ER-negative disease for the remaining trials.

Separate analyses of the association between BMI and breast cancer recurrence by treatment group (Fig. 5B) showed no statistically significant results in any of the protocols. In each of the trials we tested for interactions between BMI and ER-status, between BMI and treatment group, and between BMI and menopausal status when available. There were no statistically significant interactions found in any of the populations (data not shown).

Protocols B-30 and B-38 studied similar populations of node-positive breast cancer patients; therefore, similar findings were expected for these two trials. A possible explanation for why B-30 reached statistical significance and B-38 did not may be the amount of follow-up information that was available. B-38 is a newer trial and thus did not have as much follow-up data as B-30. We analyzed B-30 data that were truncated at 7 years to determine whether the association between BMI and OS would have remained significant with less follow-up. We found that although a similar pattern in HRs remained, the P-value for trend in the assessment of all B-30 patients was no longer significant (P = 0.07). However, the significant effect found among ER-positive patients remained (P = 0.01, data not shown).

In our investigation, we did not find a consistent relationship between BMI at diagnosis and OS or breast cancer recurrence. For OS, an association was evident in only one of the four trial populations: B-30 (node-positive breast cancer patients). This significant relationship was evident only in those with either ER-positive breast cancer, or those in the AT treatment group. For breast cancer recurrence, an association was again only evident in the B-30 population and only in those with ER-positive disease. However, there was also a significant relationship between increased BMI and recurrence in those with ER-positive breast cancer in B-38. This finding was expected given that B-30 and B-38 studied similar populations of node-positive breast cancer patients, but included a longer follow-up period in B-30.

Breast cancer is known to be a heterogeneous disease. Therefore, it is not surprising that its relationship with behavioral risk factors such as BMI may not be consistent when considering populations with various disease characteristics and different treatment regimens. Comparable to our results in B-30, a study by Sparano and colleagues found an association between higher BMI at diagnosis and higher risk of recurrence and death, specifically in HR-positive and HER2-negative disease (30). Another study among patients with triple-negative disease reported no significant association between obesity and OS or DFS (25). Recently, researchers from the EBCTCG provided results from 80,000 participants of 70 trials. They again found obesity to be strongly related to breast cancer mortality only in women with ER-positive disease, and furthermore only among pre/peri-menopausal women (32). Different populations with different disease characteristics and treatments likely all contribute to the difference in findings among multiple studies. It is unclear exactly why, for example, we found evidence of a positive relationship between increased BMI and breast cancer outcomes among patients with ER-positive disease in B-30 and B-38, but not in B-31 and B-34. However, both B-31 and B-34 had different disease characteristics (B-31 patients were HER2-positive, and B-34 patients were mostly node-negative with small tumors) that likely contributed to the difference in findings. Furthermore, administration of adjuvant systemic chemotherapy was at the discretion of the investigator in B-34, which may also have affected the results.

Not only did we not find a consistent relationship between BMI and breast cancer outcomes among the four trials included in this study, but we also found HRs consistently smaller than other reports. One explanation may be that we included more recently conducted trials than other investigations. A 2010 review by Goodwin and colleagues (37) speculated that lower HRs may reflect real differences in the prognostic effects of obesity in women receiving contemporary treatments. In addition, participants in clinical trial populations are more strictly monitored than in observational studies. Therefore comorbidities associated with obesity such as diabetes and cardiovascular disease may be less likely to impact breast cancer outcomes. In all four of our trials the majority of deaths were due to breast cancer (74% in B-30, 77% in both B-31 and B-38, and 52% in B-34).

In our findings and throughout the literature, increased BMI has been repeatedly associated with ER-positive breast cancer only. This is believed to be related to excess estradiol production in adipose tissue of obese participants. However, this process is only relevant in postmenopausal women because estrogen levels are controlled by the ovaries in premenopausal women. Other potential biologic factors that have been associated with obesity and could potentially affect the obesity and breast cancer relationship include insulin-like growth factors, adipocytokines, and inflammatory cytokines (37). Future studies investigating these potential mediators should also look for any interactions with host or tumor characteristics and treatment effects that could help to explain why BMI seems to only be an important factor for certain subgroups of breast cancer patients.

In addition to disease characteristics, various patient characteristics may also alter the effect of BMI on breast cancer outcomes. As mentioned above, the EBCTCG found that menopausal status affects the relationship between obesity and breast cancer mortality. We found no evidence of a difference by menopausal status in the trials in which this information was available, nor was there a difference by age group when using age 55 as a proxy for menopausal status (data not shown). Another example that could affect the association of BMI with outcome is cigarette smoking. Kroenke and colleagues published results in 2005 using data from the Nurses' Health Study, finding that weight before breast cancer diagnosis was related to worse prognosis among never smokers but not ever smokers (15). Based on a prior study by Manson and colleagues (38), they theorized that by mixing smokers and nonsmokers in analyses, studies may underestimate the impact of weight. We did not collect smoking status in these trials and so were unable to account for such findings.

This study has other limitations. First, BMI may not be the best marker of overweight and obesity because it does not consider body size or distinguish between the amount of fat and lean tissue (39). Furthermore, a one-time measure of BMI may not capture a woman's lifetime exposure to excess weight. Perhaps a combination of multiple assessments from adolescence through adulthood would be a more reliable marker. We were also unable to assess change in BMI after breast cancer diagnosis. It may be that this is a more clinically meaningful and relevant marker than BMI at diagnosis only. Concern may also arise from reports that the relationship between body size and breast outcomes may be J- or U-shaped curves (37). This would mean that combining underweight participants with those of normal weight could underestimate obesity effects. However, in our populations the number of participants who were underweight was very small (1%) and therefore unlikely to affect the results. Another limitation was that not all explanatory variables of interest were available in each of the trials. Tumor grade was not collected in B-38 and menopausal status was unknown for the majority of B-31 participants. In addition, the criteria used to define menopausal status were not the same in all trials.

Despite these limitations, this study has much strength. It was conducted among participants in clinical trials, therefore having consistently defined and measured outcomes and uniform treatment delivery. Furthermore, defined protocol eligibility requirements of clinical trials yield participants with similar disease stages at diagnosis as well as limited comorbidities at study entry. BMI was known for all participants and the measurements of height and weight used to calculate BMI were collected by trained medical staff, which is more accurate than relying on self-reported information (40). Finally, this was one the few analyses able to explore multiple populations individually, thereby representing the heterogeneity of breast cancer.

Using existing data from four separate breast cancer treatment clinical trials, we were able to conduct an exploratory analysis of the consistency of the association between increased BMI and breast cancer outcomes among various breast cancer populations receiving different types of treatment. Based on our findings, BMI does not appear to be a consistent factor for breast cancer prognosis among all breast cancer populations. However, increased BMI does appear to be related to poorer survival and an increased risk for recurrence among the specific population of ER-positive breast cancer patients with node-positive disease. Further studies are needed to confirm this finding and to better identify and target specific populations for which weight loss may be used as a treatment option to improve breast cancer outcome.

G. Tang is a consultant/advisory board member for Incyte Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.S. Cecchini, S.M. Swain, J.P. Costantino, P. Rastogi, A.H.G. Paterson, N. Wolmark

Development of methodology: R.S. Cecchini, J.P. Costantino

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.M. Swain, J.P. Costantino, S.J. Anderson, G. Tang, C.E. Geyer Jr, E.H. Romond, A.H.G. Paterson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.S. Cecchini, S.M. Swain, J.P. Costantino, J.-H. Jeong, S.J. Anderson

Writing, review, and/or revision of the manuscript: R.S. Cecchini, S.M. Swain, J.P. Costantino, P. Rastogi, J.-H. Jeong, S.J. Anderson, G. Tang, C.E. Geyer Jr, B.C. Lembersky, E.H. Romond, A.H.G. Paterson, N. Wolmark

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Wolmark

Study supervision: S.M. Swain, J.-H. Jeong, C.E. Geyer Jr

All authors received support from the following: NCI U10-CA-180868, U10-CA-180822, and UG1-CA-189867; Aventis Pharmaceuticals; Genentech, Inc.; Bayer Oy; Amgen Inc.; and Eli Lilly & Company.

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.