Adiposity, adult weight gain, and mammographic breast density in US Chinese women

Abstract

The association of adiposity with dense tissue area in the breast is unclear but suggests a mechanism by which adiposity might increase breast cancer risk.

Objective

We examined associations of body mass index (BMI), usual BMI from age 20–29, waist circumference, and adult weight gain with breast density in a sample of premenopausal US Chinese immigrant women.

Methods

Analyses included 415 participants in a longitudinal breast density study in Philadelphia. In addition to detailed questionnaire information, data collection included measures of anthropometry, and assessment of mammographic breast density using a computer-assisted method. We used multivariate linear regression to quantify cross-sectional associations with dense and non-dense tissue area and percent breast density assessed at baseline.

Results

In adjusted models, BMI and waist circumference were significantly positively associated with non-dense tissue area and inversely associated with percent density. BMI was also significantly positively associated with dense tissue area. Adult weight gain was associated with dense tissue area after adjusting for weight from age 20–29. In stratified analyses, BMI and adult weight gain were significantly associated with dense tissue area among women with BMI <23 kg/m2, and BMI was associated with non-dense tissue area among women with BMI >=23 kg/m2.

Conclusion

In this sample, adiposity and weight gain were associated with dense breast tissue area, although associations differed by level of adiposity. Given the potential implications of these findings for breast cancer prevention in premenopausal women, comparable studies in other population groups and with longitudinal data are needed. Reasons for the noted differences in associations by level of adiposity also warrant further investigation.

INTRODUCTION

Incidence of breast cancer is low in Chinese women, but in immigrants it increases to converge with rates of white women in the United States (US) ¹. Adiposity is an established risk factor for breast cancer ^{2, 3} that often accompanies westernization in Asian immigrants to the US ^{4, 5}, possibly as a consequence of acculturation-related changes to a less moderate diet that is higher in fat and sugar ^{4, 6, 7}. While adiposity is generally recognized to increase risk for postmenopausal breast cancer, evidence exists to suggest that it increases risk for premenopausal women, particularly Asian women, as well ^{3, 8}.

Because of its strong association with breast cancer risk ^{9, 10}, breast density, or the portion of total breast area with a mammographically dense appearance, represents a useful marker for breast cancer risk in epidemiologic studies ^10–12. The association between body mass index (BMI) and breast density is complex: although higher BMI and higher density each are associated with breast cancer, BMI is inversely correlated with percent density, suggesting that they are mutual confounders operating through separate mechanisms ¹³. That lower-risk Asian women have higher percent breast density than white women in the US but smaller areas of dense tissue ^14–17, however, supports the hypothesis that dense tissue area (absolute area of the breast that appears mammographically dense, measured for example in cm²) rather than percent density (measured as the proportion of the total area of the breast that appears dense) is the more relevant marker of breast cancer risk ¹⁸. Studies in western populations have found either no association or an inverse association between measures of adiposity or weight change and dense tissue area ^19–21. However, in studies that examined the associations of percent and area of breast density with BMI in Chinese women, BMI was negatively associated with percent density as expected but positively associated with dense area ^{22, 23}. This suggests that dense tissue in the breast may yet serve as a useful etiologic marker linking adiposity to breast cancer risk, at least in some populations. We examined associations of BMI, waist circumference, and adult weight change with breast density within a sample of US Chinese immigrant women.

MATERIAL AND METHODS

Study sample

Between October 1, 2005, and April 30, 2008, we recruited a convenience sample of 436 women into a study of mammographic breast density through local community organizations working with the recently immigrated Chinese community and contacts in social networks within this same population. Eligibility criteria included Chinese heritage, migration from Asia ≤20 years ago, and being of mammography screening age. Exclusion criteria were: postmenopausal status (no menstruation in the past year); history of breast augmentation/reduction, prophylactic mastectomy, or any cancer except non-melanoma skin cancer; current pregnancy; current breastfeeding or breastfeeding within last 9 months; or symptoms of new breast problem, such as palpable lump, skin changes, or nipple discharge. Participants received $20 as reimbursement for their time. The study was approved by the Fox Chase Cancer Center Institutional Review Board.

Data collection

Interviewers conducted detailed, language-appropriate health interviews that elicited information on sociodemographic characteristics; reproductive history, including age at menarche, pregnancy history, and oral contraceptive or other hormone use; family history of breast cancer; weight history; and smoking. Women were classified as premenopausal if they reported menses in the last three months with no decrease in predictability, early perimenopausal if they reported menses in the last three months but with decreased predictability, or late perimenopausal if they reported 3–11 months of amenorrhea ^{24, 25}. Physical activity was assessed using the short, 9-item form of the International Physical Activity Questionnaire,²⁶ which elicits information on length of time spent sitting, walking, and participating in moderate and vigorous activities over the previous 7 days. A weighted estimate of energy expenditure from physical activity in MET-hours per week is calculated as $\mathit{MET}·\mathit{hrs}·{wk}^{-1}={\displaystyle \sum _i}{t}_i·f_i·{\mathit{MET}}_i$, where t_i amount of time spent on activity category i (walking, moderate activity, vigorous activity), f is number of days per week spent on that activity category, and MET is the MET energy expenditure estimate assigned to that activity level based on the 2000 compendium of physical activities ²⁷. Each participant also completed four days of dietary recall interviews, and responses were entered into the Nutrition Data System for Research (NDS-R, Nutrition Coordinating Center, University of Minnesota). With respect to weight history, participants were asked to estimate their usual weight for each decade of their adult life from their 20s to their current age. Weight change in adulthood was calculated as the difference between their current, measured weight and their estimated weight in their 20s.

At mammographic screenings, trained research staff assessed overall and central adiposity by an anthropometric examination consisting of weight, standing height, and waist circumference for each participant, all taken in duplicate by the same interviewer following an established protocol, with the mean used in analyses. BMI (kg/m²) at these screenings (referred to in this article as ‘current’ BMI) was calculated as a measure of overall adiposity, and waist circumference, shown to be highly correlated with abdominal visceral fat as measured by computed tomography ²⁸, was used as an indicator of central adiposity.

Breast density assessment

Mammograms were conducted at Fox Chase Cancer Center or on its mobile mammography unit or van. Because breast density varies over the course of the menstrual cycle, information on date of onset of last menstruation before the mammogram was obtained to estimate menstrual cycle phase, and participants were also contacted 1–2 weeks after the mammogram in order to determine the first day of onset of their next cycle.

For most participants, cranio-caudal mammographic views were digitized using a Kodak LS-85 laser film scanner at a resolution of 100 pixels/cm. Beginning in April, 2007, Fox Chase Cancer Center began a transition to digital mammography equipment; therefore, for 46 participants recruited after that time, digital images were directly available, eliminating the need to scan and digitize images. Breast density was assessed using a highly reproducible computer-assisted method ^29–31, and average density for both breasts was calculated ³⁰. In 10% reproducibility samples, intra- and inter-batch intraclass correlation coefficients were all >0.94, indicating excellent reproducibility.

Statistical analyses

Of 436 women enrolled in the study, three women subsequently did not complete the questionnaire, we were unable to obtain mammographic images for 16, and two women were missing data on at least one of the main anthropometric variables of interest, leaving a sample of 415 women for this analysis.

We used linear regression to quantify associations between anthropometric measures and three outcomes of interest – dense and non-dense tissue areas and percent density. Anthropometric measures examined in preliminary analyses were current BMI, current weight, height, and waist circumference, estimated BMI from their 20s, 30s, and 40s, and adult weight change (change between 20s and study enrollment). Findings for BMI in their 30s and 40s was not materially different from those for current BMI and are therefore not presented. Because of high correlations of BMI with waist circumference (r=0.78) and adult weight change (r=0.67) and current weight with waist circumference (r=0.76), we followed an approach used by Han et al. ³² in analyses including these variables, modeled after a commonly used method of adjusting for energy intake in dietary studies when energy intake and a nutrient of interest are strongly correlated ³³. We adjusted for waist circumference by including residuals of the regression of waist circumference on BMI or weight, with the idea of mutual adjustment while also reducing extraneous variation in waist circumference due to variation in BMI/weight. We used a similar strategy to adjust for current BMI in models examining adult weight change.

Because results were not materially different when we conducted analyses for digitized film and digital images separately, we present results for all images combined, with adjustment for image modality. All linear regression models were adjusted at a minimum, therefore, for age (years) and original mammographic image modality (digitized film or digital). Variables were included in multivariate models as potential confounders if they were associated with at least one of the three density outcomes (dense and non-dense tissue areas or percent density); these were perimenopausal stage (premenopausal, early perimenopausal, or late perimenopausal), a combined variable representing number of live births (0–1, 2, ≥3) and age at first live birth (<25 or ≥25), and number of months of breastfeeding (none, ≤1 year, >1–2 years, >2 years). Other variables evaluated as potential confounders but found not to be significantly associated with any of the breast density measures were age at menarche, level of education, having a first or second degree relative with breast cancer, having ever used oral contraceptives or hormones, week of menstrual cycle, and level of physical activity. We examined the possibility of a difference in association by BMI in models including all women, with a cross-product term representing the predictor of interest × BMI <23.0 or ≥23.0.

RESULTS

Most women in the sample were born in China (97%), spoke no English at home (70%), and had never attended college (83%) (Table 1). Mean length of US residence among participants was 7.2 years. With respect to dietary intake, women consumed more pork than beef and had relatively high intake of fruits and vegetables. Most women (68%) were premenopausal, while 22% were categorized as early perimenopausal, and 9% were in the late perimenopause. Thirteen percent reported having ever used oral contraceptives, and only 1% of women reported a family history of breast cancer. Mean percent density was 46.5%. Dense tissue was significantly correlated with both non-dense tissue area (Pearson r=0.38, p<0.0001) and percent density (r=0.37, p=<0.0001), while non-dense tissue was inversely associated with percent density (r=−0.64, p<0.0001).

Table 1.

Descriptive characteristics of study sample (N=415).

	Mean (SD)
Age (y)	43.9 (4.5)
Length of US residence (y)	7.2 (4.9)
Age at menarche (y)a	14.9 (1.7)
Number of livebirths	2.0 (1.0)
Age at first live birth (y)	25.3 (4.6)
MET-hours/weekb	32.8 (31.2)
Dieta
Mean (SD) amount per day
Energy (kcals)	1355 (356)
Energy from fat (%)	24.2 (6.0)
Mean (SD)/median servings per week
Beef	1.7 (3.8)/0
Pork	8.1 (8.3)/6.2
Fruit	11.1 (12.5)/9.4
Vegetables	23.2 (9.7)/21.4
Breast density
Percent density	46.5 (15.8)
Dense tissue area (cm2)	36.7 (16.7)
Non-dense tissue area (cm2)	45.3 (25.9)
	%
Educationa
<8 years	48
9–12 years/technical school	35
at least some college	17
Speak English at homea
Not at all	70
A little	21
Somewhat or higher	8
1st or 2nd degree relative with breast cancer	1.2
Perimenopausal stage
Premenopausal	68
Early perimenopausal	22
Late perimenopausal	9
Total duration of breastfeeding
None	17
≤1 year	47
>1–2 years	23
>2 years	13
Ever used oral contraceptives	13
Ever used female hormones	1.7

^aNs differ due to missing data for length of US residence (N=412); age at menarche (N=413); dietary intake (N=387); education (N=414); speaking English at home (N=413).

^bBased on responses to 9-item International Physical Activity Questionnaire, MET-hours per week was calculated as product of amount of time spent on activity a given category (walking, moderate activity, vigorous activity) × number of days per week spent on that activity category × MET energy expenditure estimate assigned to that activity level ²⁷, summed over the four activity categories.

Means and standard deviations of the anthropometric measures examined in these analyses are shown in Table 2. Mean BMI in the sample was 23.4 kg/m² (mean weight 58.3 kg/m², mean height 157.7 cm), mean BMI in the 20s was 20.5 kg/m² (mean weight 51.1 kg), mean waist circumference was 79.5 cm, and women gained an average of 7.3 kg between their 20s and the study screening. Current BMI was strongly correlated with BMI in the 20s (r=0.35, p<0.0001), waist circumference (r=0.74, p<0.0001), and adult weight change (r=0.67, p<0.0001) (correlations not shown in table). Adult weight change was itself correlated with waist circumference (r=0.58, p<0.0001), and inversely correlated with BMI at age 20 (r=−0.46, p<0.0001).

Table 2.

Associations of anthropometric measures with dense area, non-dense area, and percent breast density (N=415).

		Betaa (p-value)
	Mean (SD)	Dense area	Non-dense area	Percent density
BMI (kg/m2)	23.4 (2.8)
Minimally adjustedb		0.84 (0.004)	3.99 (<0.0001)	−1.6 (<0.0001)
Multivariatec		0.90 (0.002)	3.96 (<0.0001)	−1.6 (<0.0001)
+ other anthropometric variablesd		0.74 (0.02)	3.82 (<0.0001)	−1.5 (<0.0001)
Weight (kg)	58.3 (7.7)
Minimally adjustedb		0.35 (0.0009)	1.18 (<0.0001)	−0.4 (<0.0001)
Multivariatec		0.35 (0.001)	1.20 (<0.0001)	−0.4 (<0.0001)
+ other anthropometric variablese		0.32 (0.01)	1.47 (<0.0001)	−0.6 (<0.0001)
Height (cm)	157.7 (5.4)
Minimally adjustedb		0.23 (0.13)	−0.28 (0.23)	−0.2 (0.08)
Multivariatec		0.16 (0.29)	−0.19 (0.42)	0.15 (0.28)
+ other anthropometric variablesf		−0.21 (0.24)	−1.07 (<0.0001)	0.42 (0.006)
Waist circumference (cm)	79.5 (7.6)
Minimally adjustedb		0.16 (0.14)	1.38 (<0.0001)	−0.6 (<0.0001)
Multivariatec		0.18 (0.11)	1.37 (<0.0001)	−0.6 (<0.0001)
+ other anthropometric variablesg		−0.11 (0.50)	0.22 (0.002)	−0.4 (0.001)
BMI in 20’s (kg/m2)	20.5 (2.3)
Minimally adjustedb		0.79 (0.02)	2.26 (<0.0001)	−0.9 (0.004)
Multivariatec		0.93 (0.01)	1.87 (0.0006)	−0.6 (0.06)
+ other anthropometric variablesh		0.59 (0.12)	0.49 (0.36)	−0.1 (0.75)
Adult weight change (kg)	7.3 (7.3)
Minimally adjustedb		0.12 (0.29)	0.84 (<0.0001)	−0.3 (0.0006)
Multivariatec		0.11 (0.33)	0.94 (<0.0001)	−0.4 (<0.0001)
+ other anthropometric variablesi		0.39 (0.03)	0.34 (0.17)	0.04 (0.78)

^aBeta represents cm² change in dense or non-dense area or absolute 1% change in percent density per unit change in BMI (kg/m²), waist circumference (cm), or weight change (kg).

^bAdjusted for age and image modality (digitized film vs. digital).

^cAdjusted for age, image modality, perimenopausal stage, combined variable representing number of live births and age at first live birth, and months of breastfeeding.

^dAdditionally adjusted for BMI in 20s and waist circumference (residual).

^eAdditionally adjusted for weight in 20s, waist circumference (residual), and height.

^fAdditionally adjusted for current weight, weight in 20s, and waist circumference (residual).

^gAdditionally adjusted for current BMI and BMI in 20s.

^hAdditionally adjusted for current BMI and waist circumference (residual).

ⁱAdditionally adjusted for current BMI (residual), weight in 20s, and waist circumference.

In minimally adjusted linear models, BMI, weight, BMI in 20s, waist circumference, and weight change in adulthood were associated with non-dense area and inversely with percent density (Table 2). Current BMI, weight, and BMI in 20s also predicted dense tissue area. Adjustment for other non-anthropometric covariates in multivariate models did not materially change these findings, but it did attenuate the estimate for BMI in 20s with percent density. In models that adjusted for other anthropometric variables, current BMI, weight, and waist circumference remained significant predictors of non-dense tissue area and percent density; current BMI and weight also remained significantly associated with dense tissue area. Height was inversely associated with non-dense tissue area and positively associated with percent density only after adjustment for other anthropometric variables. Adult weight change became a significant predictor of dense tissue area only after additional adjustment for weight in 20s, while estimates for non-dense tissue area and percent density were attenuated.

Because previous studies among women with generally higher BMI distributions ^19–21 showed little or an inverse association between BMI and dense tissue area, we explored whether current BMI and weight change were associated with greater dense tissue area only among women with lower BMI. As a cutpoint for stratified analyses, we selected 23.0 kg/m², suggested by the World Health Organization to define overweight in Asian women ³⁴. In stratified analyses, current BMI and adult weight change were significantly associated with dense tissue area only among women with lower BMI, although interaction p-values were not significant (Table 3). Current BMI was significantly associated with non-dense tissue area and significantly inversely associated with percent density only among women with higher BMI, but again, p-values for interaction were not significant.

Table 3.

Associations of current BMI and adult weight change with dense area, non-dense area, and percent breast density, in analyses stratified on current BMI (17.2–<23 kg/m² vs. 23.0–33.2 kg/m²) (N=415).

	Betaa (p-value)
	Dense area	Non-dense area	Percent density
Current BMI
BMI <23.0 (N=198)	1.97 (0.02)	2.42 (0.06)	−0.7 (0.37)
BMI ≥23.0 (N=217)	0.22 (0.74)	4.79 (<0.0001)	−1.8 (0.0004)
Interaction p-valueb	0.10	0.19	0.36
Adult weight change
BMI <23.0 (N=198)	0.87 (0.01)	−0.22 (0.68)	0.4 (0.20)
BMI ≥23.0 (N=217)	0.23 (0.38)	0.62 (0.07)	−0.1 (0.67)
Interaction p-value	0.14	0.21	0.10

^aBeta represents cm² change in dense or non-dense area or absolute 1% change in percent density per unit change in BMI (kg/m²), waist circumference (cm), or weight change (kg). Beta estimates for BMI were adjusted for age, image modality, perimenopausal stage, combined variable representing number of live births and age at first live birth, months of breastfeeding, BMI in 20s, and waist circumference (residual). Beta estimates for adult weight change were adjusted for age, image modality, perimenopausal stage, combined variable representing number of live births and age at first live birth, months of breastfeeding, current BMI (residual), weight at age 20, and waist circumference.

^bInteraction p-values calculated in models including all participants with an interaction term representing BMI or weight change × BMI <23.0 or ≥23.0.

DISCUSSION

Notable findings from our sample of US Chinese women were that current BMI and adult weight change were significantly associated with dense tissue area, and that these associations appeared to differ between women with lower and higher BMI. Current BMI and weight change were significantly associated with dense tissue area only among women with lower BMI. Among higher BMI women, in contrast, BMI was associated with non-dense tissue area and inversely associated with percent density.

Previous studies consistently show associations of anthropometric measures of adiposity with non-dense tissue area and inverse associations with percent density ^{13, 35}, as we did for BMI, weight, and waist circumference. These associations likely reflect correlations between body fat and fatty tissue in the breast, which then drives an inverse association with percent density. However, with some exceptions ^{22, 23, 36}, most observed an inverse association or no association between BMI and area of dense tissue ^19–21. This has led to the conclusion that BMI and percent breast density are independent predictors of breast cancer, and negative confounders for each other ¹³. With respect to weight change, although one intervention trial found that women who gained weight had an increase in dense tissue area ³⁶, other studies show no association of dense tissue area with weight gain ^{19, 20}. Reeves et al. ¹⁹ observed an association between BMI and dense tissue area in cross-sectional comparisons but no association between annual changes in BMI and dense tissue area.

Our finding of an association between BMI and dense tissue area in this premenopausal sample suggests that in certain populations, adiposity may in fact increase risk by increasing dense tissue area. Of potential relevance is the observation that the studies that showed a positive association between BMI and dense tissue area ^{22, 23}, including the present study, were conducted in Asian women, with generally lower BMI distributions than those observed in non-Asian samples. Indeed, in stratified analyses we found that BMI was associated with dense tissue area only among women with lower BMI. Among women with higher BMI, BMI was associated with fatty tissue, not dense tissue, in the breast, and hence inversely associated with percent density.

BMI is an established risk factor for postmenopausal breast cancer in western populations, likely because adipose tissue is the major source of estrogens after the menopause. Although many previous studies have reported an inverse association between BMI and premenopausal breast cancer ², recent evidence suggests that an increase in risk with greater BMI among premenopausal women is evident when models adjust for the negative confounding of mammographic density ¹³. Mechanisms by which BMI might increase proliferation and dense tissue area and/or breast cancer risk include effects on levels of androgens ^{37, 38}, insulin ^39–41, and inflammatory factors ^42–49. Previous studies that examined sex steroid hormones ^50–52 and insulin-related measures ^53–57 in relation to breast density have not consistently confirmed these mechanisms. However, most focused only on percent density rather than also examining associations with dense tissue area, and all were conducted in women with higher BMI distributions than our sample.

A question that warrants further investigation is why any of these mechanisms might be limited to women of lower BMI, as suggested by our study. Two previous studies found that an inverse association of parity with percent breast density ^{58, 59} and positive associations of age at menarche and age at first birth ⁵⁸ with percent breast density were apparent primarily among women with lower BMI. Investigators of those studies suggested that effects of these reproductive factors are most visible in the absence of effects of excess adiposity on circulating sex hormone levels. Our results may reflect a similar phenomenon, in which effects of BMI and weight gain on sex steroid hormones or insulin and subsequently dense breast tissue are most visible only among women below some threshold for adiposity, resulting in an apparent ceiling effect.

That we observed an association between BMI and dense tissue area while others did not ^19–21 might be due to other features, besides lower mean BMI, that distinguish our sample from others ^{19, 20, 60, 61} – for example, low prevalence of oral contraceptive use and older age at menarche. However, analyses stratifying on history of hormone use and age at menarche did not reveal any meaningful differences by stratum (results not shown), suggesting that these factors are unlikely to explain the difference in findings between our sample and others. Preliminary analyses also suggest some distinguishing features of food consumption in our study population (Table 1), but dietary intake in this sample has yet to be explored more fully. It is also possible that associations of BMI and weight gain with dense breast tissue area are evident only in women at low risk for breast cancer. The age-standardized breast cancer mortality rate was 17.4 per 100,000 in the US in 2000; in China it was 7.0 per 100,000 in urban areas, 4.3 per 100,000 in rural areas ⁶². Clarifying the reasons for the apparent difference in findings has implications for our understanding of breast cancer etiology. With respect to prevention, it might also help identify the subset of women for which dense tissue area can serve as a marker of the effect of change in adiposity on breast cancer risk in intervention trials.

Our observation that height was inversely associated with non-dense tissue area and positively associated with percent density is consistent with previous work ^{63, 64}. In the study by Dite et al. ⁶⁴, height was also positively associated with dense tissue area although it was not in the current study. Potential mechanisms linking height to greater percent density have focused on factors promoting pre-adolescent growth.

That participants were recruited as a convenience sample leaves open the possibility of bias. For example, the observed results might have resulted from an overrepresentation of participants who were both thinner (lower BMI) and at lower risk for breast cancer (manifested by lower breast density). Another limitation is that analyses on adult weight gain were based on participants’ recalled weight, which, while reasonably accurate, may be influenced by other characteristics such as current weight ^65–67. The accuracy of recalled weight in a lean sample of Chinese immigrant women is not known, and its potential for bias is difficult to speculate on. Nevertheless, unlike many studies in western populations, results for current BMI and weight were based on measured rather than self-reported measures. As such, the findings offer compelling evidence for a role of adiposity in increasing breast cancer risk in a unique sample of women undergoing social, cultural, and health transitions upon migration to the US. These findings, particularly with respect to weight gain, merit confirmation in longitudinal analyses.

Our study is the first to provide evidence that associations of BMI and weight gain with dense tissue area in the breast may differ by level of adiposity. Our findings support the possibility that adiposity that is modifiable in adulthood can have visible effects on breast density and possibly breast cancer risk. If confirmed, our findings point to weight gain as a modifiable risk factor for premenopausal women. These results require confirmation in other population samples, and in longitudinal data. Determining the reasons for different effects by level of adiposity also warrants investigation.

Abbreviations

BMI

body mass index

US

United States