The association between DXA‐derived body fat measures and breast cancer risk among postmenopausal women in the Women's Health Initiative

Abstract Background Most studies demonstrating an association between excess adiposity and postmenopausal breast cancer have used anthropometric measures, particularly body mass index (BMI). However, more direct body fat measures may more accurately determine the relationship between body fat distribution and breast cancer risk. Methods Cox proportional hazards regression models were created to examine the associations of dual‐energy x‐ray absorptiometry (DXA) body fat measures (at baseline and during follow‐up) with breast cancer risk among 10 931 postmenopausal women from the Women's Health Initiative cohort. A total of 639 incident invasive breast cancer cases (including 484 estrogen receptor positive (ER+) cases) were ascertained after a median follow‐up of 15.0 years. Results Excess whole body fat mass and trunk fat mass were positively associated with risk invasive breast cancer risk. These associations persisted even after additional adjustment for standard anthropometric measures. In time‐dependent analyses, we observed that both whole body fat mass and trunk fat mass, in the highest versus lowest category, were associated with a doubling of risk of invasive breast cancer overall (HR: 2.17; 95% CI: 1.54‐3.05 and 2.20; 1.55‐3.14, respectively) and of ER+ breast cancer (2.05; 1.37‐3.05 and 2.03; 1.34‐3.07, respectively). The remaining DXA measures were also positively associated with breast cancer risk in baseline and time‐dependent analyses. Conclusion These findings suggest that DXA‐derived body fat measures are positively associated with breast cancer risk after adjustment for BMI and other conventional breast cancer risk factors.


| INTRODUCTION
Excess adiposity is believed to contribute to carcinogenesis through several mechanisms, including chronic inflammation, hyperinsulinemia, elevated leptin and reduced adiponectin levels, hyperlipidemia and abnormal sex steroid hormone metabolism. 1,2 The preponderance of epidemiologic studies examining the association of body fat with risk of breast cancer (including hormone receptor-positive breast cancer) have used body mass index (BMI) as the exposure of interest. [3][4][5][6][7][8] These studies have provided consistent evidence to support a role for excess adiposity, as defined by having a BMI >25.0 kg/m 2 (overweight/obesity), in the development of breast cancer among postmenopausal women. [3][4][5][6][7][8] As a measure of adiposity, BMI is limited, in that it does not truly reflect one's body fat distribution, 9 and, therefore, may not precisely estimate breast cancer risk among persons with excess adiposity. In an attempt to better capture the influence of body fat distribution on risk of breast cancer, several epidemiological studies have investigated the associations of waist circumference (WC) and waist-to-hip-ratio (WHR), 4,8,10,11 indicators of central adiposity, with risk of postmenopausal breast cancer. In most studies, both measures have been positively associated with risk, 4,8,10,12 and some studies have suggested that these measures may better predict the risk of postmenopausal breast cancer than BMI. 4,8 BMI also fails to adequately account for differences in metabolic phenotypes among persons within the same BMI category. [13][14][15] Consistent with these limitations, some women with normal BMI have enlarged adipocytes and elevated levels of leptin, suggestive of a hyperadipose state. 16 Unlike anthropometric measures, dual-energy X-ray absorptiometry (DXA) provides estimates of fat, bone, and bonefree lean mass comparable to those from more costly methods such as computed tomography and magnetic resonance imaging. 9,17 Using data from Women's Health Initiative (WHI), we have previously shown that various DXA-derived measures of body fat were positively associated with risk of postmenopausal breast cancer. 18 More recently, we also demonstrated that among normal BMI postmenopausal women, relatively high whole body fat and trunk fat (measured using DXA) were associated with almost twofold increases in risk of estrogen receptor-positive (ER+) breast cancer. 19 In contrast to anthropometric measures such as BMI, the use of more direct measures of body fat may more accurately determine the relationship between body fat distribution and risk of breast cancer risk among postmenopausal women. Thus, in the study reported here, which was conducted in a subset of participants in the WHI and represents an extension of an earlier report, 18 we examined the association between DXA-derived measures of body fat and risk of invasive breast cancer. We also evaluated whether the DXA-derived measures influence risk of invasive breast cancer beyond that associated with the commonly used anthropometric measures and other conventional breast cancer risk factors.

| Study population and design
A detailed description of the WHI design and study population can be found elsewhere. 20 Briefly, the WHI included 161 808 postmenopausal women aged 50 to 79, from major racial/ethnic groups, who were enrolled at 40 clinical centers throughout the United States between 1993 and 1998. Women were either included in the Clinical Trial (CT) group which has three overlapping components (hormone therapy (2 trials), low-fat diet modification, and calcium-vitamin D supplementation; n = 68 132) or the Observational Study (OS) group (n = 93 676). 20 At baseline, information on demographic characteristics, menstrual history, reproductive history, exogenous hormone use, family history of breast cancer, medical history, and diet and lifestyle factors we observed that both whole body fat mass and trunk fat mass, in the highest versus lowest category, were associated with a doubling of risk of invasive breast cancer overall (HR: 2.17; 95% CI: 1.54-3.05 and 2.20; 1.55-3.14, respectively) and of ER+ breast cancer (2.05; 1.37-3.05 and 2.03; 1.34-3.07, respectively). The remaining DXA measures were also positively associated with breast cancer risk in baseline and time-dependent analyses. Conclusion: These findings suggest that DXA-derived body fat measures are positively associated with breast cancer risk after adjustment for BMI and other conventional breast cancer risk factors. was collected, and anthropometric measurements (weight, height, waist circumference, hip circumference) were made by trained staff using a standardized protocol. After completion of the original study in 2005, WHI Extension Studies (2005)(2006)(2007)(2008)(2009)(2010)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020) were conducted to gather additional follow-up data. The study was approved by the institutional review boards of all participating institutions, and all participants provided written informed consent.

| Body fat measurements
More detailed information on the ascertainment of DXA measures has been previously published. 21 Briefly, for women at three WHI BMD clinical center (Birmingham, Tucson/Phoenix, and Pittsburgh; n = 11, 393), body fat was measured at baseline by whole body DXA scans performed in fan-beam mode and obtained from Hologic QDR scanners (QDR 2000, 2000+, or 4500) (Hologic, Inc.). This subgroup of participants with DXA measures differed from those without DXA measures in that their recruitment was aimed at maximizing the number of ethnic/racial minority participants (22.6% non-Hispanic white among those with DXA measures versus 16.9% among those without DXA measures). 22 Scanners were operated by persons who were trained and certified on the basis of an evaluation of scanning and analysis technique. Scanner performance was monitored longitudinally using spine and whole body phantom scans. 21 Quality control measures, including monitoring of phantom scans, reviewing random samples of all scans and flagging those with specific issues, controlling changes in hardware and software and scanning of calibration phantoms across instruments and clinical sites, were also implemented. 22 Body fat measures included whole body fat (kg), percentage whole body fat, trunk fat (defined by the fat contained in the torso apart from head and limbs), fat mass of the legs, and the ratio of trunk to leg fat mass. In addition, the fat mass index (FMI) and trunk fat mass index (TFMI) were calculated as whole body fat mass and trunk fat mass in kilograms divided by height in meters squared, respectively. 23,24 The body fat to lean body mass ratio was also calculated as whole body fat mass divided by whole lean body mass. In addition to the baseline measures, the DXA scans and anthropometric assessments were repeated during four follow-up visits (years 1, 3, 6, 9).

| Analytic cohort
Our analytic cohort was restricted to women with available DXA measures (N = 11 393). Among these women, we T A B L E 1 Distribution of selected characteristics by octiles of trunk fat mass excluded 108 women who were missing baseline DXA measures, 325 women with a previous history of breast cancer, and 29 women with missing information on follow-up time.

Octiles of trunk fat mass
After exclusion, 10 931 women remained for our analyses, of whom 6078 were in the OS group and 4853 were in the CT intervention group (control, placebo, intervention) (See Figure S1).

| Outcome ascertainment
Participants were followed up semi-annually in the CT group and annually in the OS using in-person, mailed, or telephone questionnaires to collect information on clinical outcomes. Breast cancer cases were confirmed via central review of medical records and pathology reports by trained physician adjudicators. The tumor hormone receptor status was coded using the National Cancer Institute's Surveillance Epidemiology and End Results (SEER) coding system. 25 After a median follow-up of 15.0 years (interquartile range: 8.8-20.1 years), a total of 639 incident invasive breast cancer cases (including 484 ER+ breast cancer cases) had been ascertained.

| Statistical analysis
Cox proportional hazards regression was used to estimate the age and multivariable adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations between the body fat measures and risk of invasive breast cancer (overall and ER+) with time to diagnosis of invasive breast cancer as the underlying timescale. Participants were censored (noncases) if they died, withdrew from the study before the end of follow-up, or did not develop invasive breast cancer by the end of follow-up (February 28, 2017). Cases contributed person-time to the study from their date of enrollment until the date of diagnosis of breast cancer, and non-cases contributed person-time from their date of enrollment until the date of death, the date of withdrawal from the study, or the end of follow-up, whichever came first. For the analysis involving ER+ breast cancer, we additionally censored the non-ER+ breast cancer cases. The body fat measures were analyzed as continuous variables (per standard deviation (SD) increase and per 5 or 2 kg increase) and as octiles. We used octiles in order to evaluate the full range of risk. Covariates were selected based on existing evidence or if they caused a 10% change in the HR estimates obtained from the multivariable models. For our main analyses, all regression models were adjusted for age at enrollment (continuous), education (high school or less, post-secondary or some college, graduate school or some graduate school, missing), race/ethnicity (white, black, Hispanic, other), family body fat measures and risk of ER+ breast cancer, restricted cubic splines with three knots at the 25th, 50th, and 75th percentiles were fitted in Cox proportional hazards regression models. 26 A P value for non-linearity was determined by testing whether the coefficient of the second spline transformation was equal to zero (P < .05). In sensitivity analyses, we excluded   women with a breast cancer diagnosis within two years of enrollment to assess for possible reverse causation.
P-values for trend (P-trend) were determined by including the ordinal body fat variable as continuous variables in the regression models and using Wald tests to assess statistical significance. The proportional hazards assumption was confirmed using Schoenfeld residuals.  We also examined the associations between the DXAderived measures and risk of invasive breast cancer by hormone therapy (HT) status (never, former, current) to assess whether HT use at baseline is an effect modifier. For this analysis, the exposures were categorized by quintiles, given the relatively small number of cases in some strata. Also, women in the HT intervention arms (ie, estrogen alone intervention, estrogen and progesterone intervention) were excluded. To estimate the P-value for heterogeneity, we included an interaction term in the Cox regression models and used the Wald test to test its coefficient.
To account for the influence of changes in the measurements over time on risk of breast cancer, Cox proportional hazards regression models with time-dependent exposures were also created. We conducted sensitivity analyses whereby physical activity, alcohol consumption, and smoking were also included as time-varying covariates. However, the estimates were virtually unchanged. Therefore, only the body fat measures were fitted as time-dependent. These models were adjusted for the aforementioned covariates (except the anthropometric measures).
All statistical analyses were performed using Stata 14.1 (StataCorp). P-values were two-sided. P-values were considered to be statistically significant if <.05. Table 1 presents a summary of the participants' characteristics by octiles of trunk fat mass. Among women with trunk fat mass in the highest octile, mean levels of BMI, WC, and WHR were highest while mean physical activity level and alcohol intake were lowest. Women with elevated trunk fat mass levels (highest octile) were also less likely to be HTusers at baseline but were more likely to have started menstruating at an early age (<12).

| RESULTS
In multivariable analyses (excluding BMI, WC or WHR), we observed that all DXA-derived measures were positively associated with risk of invasive breast cancer (Table 2; Figure  S2). In particular, whole body fat mass, trunk fat mass, and the body fat indices (ie, FMI and TFMI) in the highest octile had more than a doubling of the risk of invasive breast cancer compared to women in the lowest octile (HR: 2.13, 95% CI: 1.52-2.98; 2.58, 1.83-3.65; 2.14, 1.53-3.01, and 2.17, 1.52-3.07, respectively) ( Table 2).
The multivariable models (excluding adjustment for whole body fat mass or trunk fat mass) also demonstrated positive associations of all of the anthropometric measures (examined as categorical variables) with risk of invasive breast cancer overall and with ER+ breast cancer, but these associations did not persist after further adjustment for whole body fat mass or trunk fat mass (Tables S1-S2). When considering the continuous exposures, the HRs per SD increase in the anthropometric measures (Tables S1-S2) were of similar magnitude to those seen for the DXA-derived measures (Tables 2,3).
To assess whether the DXA-derived measures explained risk of breast cancer beyond that of the anthropometric measures and other conventional breast cancer risk factors, we additionally adjusted for the anthropometric measures (separately). In these analyses, the associations for whole body fat mass and trunk fat mass remained after further adjustment (separately) for BMI, WC, and WHR (Table 2). With respect to the body fat indices, the association for FMI disappeared after additional adjustment for BMI while, for TFMI, the association was no longer evident after additional adjustment for WC. Table 4 and Figure S2 show that all DXA-derived measures were positively associated with risk of ER+ breast cancer. Our multivariable analyses (excluding BMI, WC, and WHR) also demonstrated more than a twofold increased risk of ER+ breast cancer among women with whole body fat mass, trunk fat mass, FMI, and TFMI in the highest octile (HR: 2.06, 95% CI: 1.40-3.02; 2.45, 1.64-3.66; 2.05, 1.38-3.05, and 2.21, 1.47-3.32). Whole body fat mass, trunk fat mass, and TFMI remained independently associated with risk after additional adjustment for the anthropometric measures (separately) while the association for FMI only persisted after additional adjustment for WHR. The HRs per SD increase in the anthropometric measures were also of similar magnitude to those seen for the DXA-derived measures (Tables 4 and  Tables S2). The associations between the DXA-derived measures and risk of ER+ breast cancer appeared to be nonlinear ( Figure S3).
The associations of the DXA measures with risk of breast cancer (overall and ER+) were attenuated, but remained statistically significant after additional adjustment for lean body mass (Tables S4). Furthermore, women with a relatively high whole body fat to lean mass ratio were observed to have increased risk of breast cancer (overall and ER+; Table S5). All body fat measures were also associated with risk of breast cancer in analyses restricted to women from the OS group (Tables 4, Tables S3 and S6).
In analyses where we assessed the modifying effect of HT use, the DXA-derived body fat measures were associated with increased risk of invasive breast cancer among never-users (Table 3). Among current HT users, trunk fat mass, ratio of trunk fat mass to the average of right and left leg fat mass and the body fat indices were also positively associated with risk. There was also some evidence that the association of fat mass of the right leg was modified by HT use (P heterogeneity : .031) ( Table S7).
The results of our time-dependent covariate analyses also demonstrated positive associations between all DXA-derived body fat measures and risk of invasive breast cancer (overall and ER+; Figures 1 and 2). In particular, the highest octiles of whole body fat mass (HR: 2.17; 1.54-3.05, 2.05; 1.37-3.05 for overall and ER+, respectively) and trunk fat mass (HR: 2.20; 1.55-3.14, 2.03; 1.34-3.07 for overall and ER+, respectively) were associated with over a twofold increased risk of invasive breast cancer. In these analyses, BMI and WC were also positively associated with risk of breast cancer (overall and ER+), but, WHR was only associated overall risk of invasive breast cancer (Table S8).
There were no substantial changes in the HRs after exclusion of women with a diagnosis within two years of enrollment (Tables S9).

| DISCUSSION
The results of this study confirm and extend our previous findings of positive associations between DXA-derived and anthropometric body fat measures and risk of breast cancer (overall and ER+) in postmenopausal women. 18 Notably, the associations for whole body fat mass and trunk fat mass with risk of breast cancer (overall and ER+) persisted after additional adjustment for the anthropometric indices (BMI, WC, or WHR). The associations between most of the DXAderived measures and risk of invasive breast cancer were somewhat stronger among non-HT users than current HT users. The positive association between the baseline body fat measures and risk of invasive breast cancer were substantiated in our time-dependent analyses.
In studies examining the associations between adiposity and risk of chronic disease, less robust methods than DXA, such as bioelectrical impedance analysis (BIA), have been widely used to determine levels of adiposity. 27,28 Compared to DXA, BIA is less sensitive in estimating % body fat, whole body fat mass and lean body mass among lean or obese individuals. 27,28 Despite its limitations, in agreement with the current study, recent findings from a large prospective study demonstrated strong positive associations between BIAderived body fat measures, including whole body fat mass, trunk fat mass, and body fat percent, with overall risk of breast cancer. 29 Similar positive associations were reported in other studies that investigated the associations between BIAderived whole body fat mass and body fat percent and risk of breast cancer, including ER+ breast cancer. [30][31][32] Anthropometric measures, particularly BMI, are also widely used as measures of adiposity. 9,17,33 The utility of BMI in research settings remains questionable as it has several limitations, such as an inability to differentiate between lean and fat mass, 13,15 that may limit its ability to reliably predict the risk of diseases among persons with excess adiposity. Nevertheless, BMI has been consistently shown to have a relatively strong positive association with risk of postmenopausal invasive breast cancer. 4,8,[10][11][12] Moreover in the current study, the HRs per SD increase in BMI were of similar magnitude to those of the DXA-derived measures, suggesting that DXA-derived measures are not necessarily better predictors of breast cancer risk. Notwithstanding this, we observed that the associations of whole body fat mass and measures of central adiposity, including trunk fat mass, the ratio of trunk fat mass to the average of left and right leg fat, and TFMI, with risk of breast cancer (overall or ER+) were independent of BMI and other conventional breast cancer risk factors. These findings imply that DXA-derived measures contribute to risk of postmenopausal breast cancer beyond that associated with BMI and other risk factors. 34 Body fat composition varies with height. However, epidemiological studies have principally evaluated the associations of absolute values or percentages of the DXA-derived measures, which do not reflect height differences, with risk of breast cancer. 23,35 Thus, such measures may not sufficiently explain potential differences in disease risk that may be apparent among individuals with similar body composition, but different heights. To compensate for this limitation, we used height-normalized values (namely FMI and TFMI) which were proposed by VanItallie et al 23 In the current study, both indices were associated with more than a doubling of the risk of invasive breast cancer. We are only aware of one study which examined the association of FMI with risk of breast cancer and, in this study, which was restricted to black women, no association was seen. 36 This may be partly explained by ethnic differences in the association between body fat and risk of breast cancer. Ethnic differences in the associations of the body fat measures with risk of breast cancer were not examined in this study because the number of non-Caucasian women with breast cancer was too small.
Findings from previous studies suggest that HT use modifies the association between adiposity and risk of invasive breast cancer. [37][38][39] These studies found that the positive association between BMI and risk of breast cancer was stronger among women who had never used HT. 37,38 The weaker associations among current HT users may be partly explained by improved insulin sensitivity from chronic use of therapies with low estrogen doses; thereby, mitigating the adverse effects of insulin on breast carcinogenesis. 40 However, there was little evidence in the present study to suggest that a woman's HT status modifies the associations between the DXA measures and risk of breast cancer.
Our findings are in accord with biological evidence which suggests that excess adiposity contributes to altered metabolism. Specifically, in an obesogenic environment, adipocytes become enlarged. 41 These large adipocytes, which can be found in both breast adipose and visceral adipose tissue (VAT), are insulin-resistant and hyperlipolytic, and may therefore contribute to the elevated levels of insulin, glucose, and triglycerides, [41][42][43] as seen among women in the current study who had excess central adiposity. Chronic hyperglycemia is believed to stimulate tumor cell growth by causing epigenetic modifications. 1 Moreover hyperinsulinemia can induce tumor cell proliferation and migration, at least in part by activating the PI3kinase pathway. 1 The insulin resistance that occurs in association with excess body fat leads to downregulation of SHBG which results, in turn, in elevated circulating levels of free estrogens that contribute to the development of ER+ breast cancer. 1 Adipocyte hypertrophy also enhances the production of pro-inflammatory chemokines and cytokines, such as tumor necrosis factor-α, interleukins, and leptin in human white adipose tissue, 2,43 which may stimulate mutagenesis, tumor cell proliferation, survival, migration, angiogenesis, and stem cell survival. 44,45 Moreover, a variety of molecules that are altered in association with excess body fat including the bioactive lipid prostaglandin E 2 (PGE 2 ) and leptin induce aromatase, the rate-limiting enzyme for estrogen biosynthesis. 2,43 Thus, systemic alterations, in combination with enhanced local estrogen production, are likely to contribute to the development of hormone receptor-positive breast cancer. 2,43 Lean body mass is thought to exert anti-carcinogenic benefits by inhibiting cancer-associated processes such as insulin resistance, chronic inflammation, and dysregulated lipid metabolism. 46 In keeping with this, we observed that, after adjustment for lean body mass, the associations between the body fat measures and risk of invasive breast cancer were attenuated. Notably, in line with a previous study, 47 our study demonstrated that women with higher body fat to lean body mass ratio had greater risk of breast cancer. This finding suggests that any potential benefit of lean body mass on risk of breast cancer may be offset by the deleterious effect of excess adiposity.
A strength of this study was the availability of body fat measures, which were obtained using DXA. DXA is a quick, non-invasive, and highly reproducible method for directly measuring body composition. This method, however, lacks the ability to distinguish between subcutaneous and visceral abdominal adipose tissue. However, studies have shown that DXA is almost as accurate as MRI and CT in determining skeletal muscle mass and fat mass compartments. 48 Furthermore, this is the first known study to assess the association of TFMI with risk of breast cancer. Our study also had repeated body fat measures which allowed us to assess the association of longitudinal changes in these measures with risk of breast cancer. However, a limitation of the study was that some of the variables were self-reported. Hence, these variables may have been misclassified due to measurement error. Additionally, multiple testing was performed so that the results, while strong, need to be interpreted cautiously.

| CONCLUSION
Our findings suggest that overall and central adiposity may influence the risk of postmenopausal breast cancer (overall and ER+) beyond the contributions of anthropometric measures and standard breast cancer risk factors. As this is the first known study to explore the associations between TFMI with risk of breast cancer, further studies are needed to confirm the utility of height-normalized values for predicting the risk of breast cancer among postmenopausal women. Our findings lend support for the development of behavioral or pharmacological interventions aimed at promoting weight loss among women at high risk for breast cancer due to excess body fat.