Recent RCT based analyses of prognostic associations of BMI in ER+ breast cancer include reports that high BMI was associated with poor outcome in the ATAC and BIG 1-98 trials (Sestak et al. 2010; Ewertz et al. 2012) which involved only women with ER+ breast cancer receiving tamoxifen or aromatase inhibitors. Analysis of completed ECOG (E1199, E5188) (Sparano et al. 2012) and NSABP studies (B30, B38) (Cecchini et al. 2013) as well as a recent meta-analysis conducted by the Early Breast Cancer Clinical Trialists Collaborative Group (Pan et al. 2014) also identified an increased risk of recurrence or death in obese (vs. non-obese) women with hormone receptor positive breast cancer (premenopausal only in the latter). A meta-analysis involving 8,874 women enrolled onto seven German adjuvant trials identified adverse prognostic associations of BMI in hormone receptor positive cases (Pajares et al. 2013). These results are consistent with results of earlier meta-analyses (Niraula et al. 2012).

Results of similar post hoc RCT analyses in women with hormone receptor negative, triple negative or HER2 positive breast cancer have been less consistent. North American investigators, using data from both ECOG (Sparano et al. 2012) and NSABP (Cecchini et al. 2013) RCTs (E3189, B30, B31, B34, B38) failed to identify significant prognostic associations of BMI in those with hormone receptor negative breast cancer. In contrast, Fontanella et al. (2013) identified adverse prognostic associations of obesity in women with triple negative breast cancers participating in a group of German neoadjuvant RCTs (chemotherapy dose was capped at 2.0 m² in three of these trials; this may have contributed to adverse obesity associations) while Pajares et al. (2013) identified worse overall and breast cancer specific survival in triple negative breast cancer patients with BMI > 35 kg/m² enrolled in a series of GEICAM RCTs. In HER2 positive patients, a significantly worse outcome in heavier women with HER2+ breast cancer was identified in two RCTs; (Pajares et al. 2013; Crozier et al. 2013) in an observational study, Mazzarella et al. (2013) identified a similar association that was present only when cancers were also estrogen receptor negative. In contrast, Sparano et al. (2012) and Turkoz et al. (2013) identified no associations of BMI with disease-free or overall survival in women with HER2+ breast cancer enrolled onto RCTs or an observational study.

In summary, adverse associations of BMI with breast cancer outcomes have been repeatedly reported in all breast cancer subtypes. It is not clear whether the inconsistency of recent data may relate to differences in study design (discussed below) or to true biologic differences in BMI associations across breast cancer subtypes. The latter should be explored in preclinical studies and in adequately powered clinical datasets that include a full, and representative, spectrum of breast cancer patients.

The more consistent results reported in ER+ breast cancer may reflect, at least in part, higher estrogen levels in obese postmenopausal women, leading to enhanced signaling through estrogen pathways in obese women. Because BMI is associated with prognosis in women receiving tamoxifen and aromatase inhibitors (Sestak et al. 2010; Ewertz et al. 2012), these treatments do not appear to fully overcome effects of higher BMI, suggesting that other obesity associated factors, such as insulin or inflammatory mediators, contribute to the BMI-prognosis association in these patients.

The less consistent results in triple negative breast cancer in particular may reflect capping of BMI at arbitrary levels (e.g. 2.0 m², 2.2 m²) when calculating chemotherapy doses [a practice that has been less common in recent years and avoided in modern RCTs and advised against in a recent American Society of Clinical Oncology guideline (Lyman and Sparreboom 2013)], leading to BMI associations that reflect under-treatment rather than biologic effects in earlier studies. This practice may have had the greatest impact in triple negative breast cancer in which chemotherapy is the primary adjuvant approach, and targeted treatments, which may overcome effects of under-dosing to some extent, are not available. The observation that BMI is associated with prognosis in recent cohorts and RCTs that avoided dose capping suggest these factors do not fully account for BMI associations. One alternative explanation is that the underlying aggressiveness of advanced triple negative breast cancers in some studies, leading to poor outcomes, may be associated with reduced prognostic impact of obesity.

Different temporal patterns of relapse of ER+, triple negative and HER2+ breast cancers and differing durations of follow-up in reported studies are unlikely to be the primary cause of inconsistent results—for example, our group has demonstrated that obesity effects are constant in periods up to 5 years, and 5–10 years and beyond in a long-term prospective study (Goodwin et al. 2012).

Inclusion of locoregional events (which contribute a greater proportion of events in the modern era of breast conserving therapy and effective systemic adjuvant therapies) in outcome measures in recent trials may have introduced noise and led to reduced power in some studies as these events have not been associated with BMI (Ewertz et al. 2012). The analysis of BMI as a categorical (vs. continuous) variable in statistical analyses, or the modelling of associations as linear (vs. quadratic which allows a curvilinear association which has been demonstrated in several studies) may also have reduced power. Importantly, power to detect associations may have been lower in subsets of receptor negative, triple negative and HER2+ breast cancer, due to smaller numbers of these cancers in some RCTs.

It is also possible that unappreciated differences in the study populations contributed to inconsistencies. Many of the earlier observational studies were population or institution based and included all women diagnosed with breast cancer, regardless of the presence or absence of associated medical conditions. Many recent RCTs involved cardiotoxic medications (e.g. anthracyclines, HER-2 targeted agents) and women with cardiac morbidity (or cardiac risk factors such as hypertension, dyslipidemia, diabetes) were commonly excluded, either explicitly through entry criteria or as a safety precaution by physicians (and patients) wanting to avoid cardiotoxicity of unproven treatments. In trials involving taxanes, women with diabetes were often excluded because of the need for steroid pre-medications; they may also have been less likely to be enrolled because of concerns about neurotoxicity. Cardiovascular disease, dysglycemia, hyperlipidemia and hypertension are components of the insulin resistance (or metabolic) syndrome; (Alberti et al. 2009) physiologic components of this syndrome (e.g. insulin, glucose, inflammation) may mediate the association of obesity with breast cancer outcomes (see below) and it is possible these recent trials preferentially enrolled metabolically healthy women who do not have the physiologic attributes that mediate obesity-breast cancer associations. This selection process has not been investigated in a breast cancer population, however, Kramer et al. (2013) have shown that obese individuals in the general population with any one of hypertension, abnormal lipids, central obesity, abnormal glucose/diabetes or high C-reactive protein (CRP—a marker of inflammation) have significantly higher greater levels of insulin resistance [reflected by homeostasis model assessment (HOMA) scores] than those who do not have any of these attributes. This issue is of relevance not only to understanding the inconsistency of recent reports; it could also impact the design of weight loss intervention trials in breast cancer survivors. If adverse associations of BMI are present only in metabolically unhealthy women, such trials should be designed to enrich for this population.

Thus, although the associations of obesity with outcome may truly differ across breast cancer subtypes, adverse associations have been repeatedly identified in all subtypes. Design differences across studies may have contributed to the identified inconsistencies.