Body Mass Index and Breast Cancer Risk According to Postmenopausal Estrogen-Progestin Use and Hormone Receptor Status

Abstract

To assess the joint relationships among body mass index, menopausal status, and breast cancer according to breast cancer subtype and estrogen-progestin medication use, we conducted a meta-analysis of 89 epidemiologic reports published in English during 1980–2012 identified through a systematic search of bibliographic databases. Pooled analysis yielded a summary risk ratio of 0.78 (95% confidence interval (CI): 0.67, 0.92) for hormone receptor–positive premenopausal breast cancer associated with obesity (body mass index (weight (kg)/height (m)²) ≥30 compared with <25). Obesity was associated with a summary risk ratio of 1.39 (95% CI: 1.14, 1.70) for receptor-positive postmenopausal breast cancer. For receptor-negative breast cancer, the summary risk ratios of 1.06 (95% CI: 0.70, 1.60) and 0.98 (95% CI: 0.78, 1.22) associated with obesity were null for both premenopausal and postmenopausal women, respectively. Elevated postmenopausal breast cancer risk ratios associated with obesity were limited to women who never took estrogen-progestin therapy, with risk ratios of 1.42 (95% CI: 1.30, 1.55) among never users and 1.18 (95% CI: 0.98, 1.42) among users; too few studies were available to examine this relationship according to receptor subtype. Future research is needed to confirm whether obesity is unrelated to receptor-negative breast cancer in populations of postmenopausal women with low prevalence of hormone medication use.

INTRODUCTION

Obesity has dramatically increased in prevalence since the early 1980s, as documented in risk factor surveillance studies (1, 2), and the consequences of this epidemic in terms of health-care expenditures and future expected mortality are staggering (3). In particular, 65 million more obese adults are projected in the United States by the year 2030, among whom over 500,000 additional cases of cancer are expected to be diagnosed (3). The greater medical costs associated with obesity demand approximately 9% of all medical spending (4). The results of numerous studies—as summarized in both systematic reviews (5) and meta-analyses (6, 7)—have accumulated to strongly suggest that obesity represented by body mass index increases the relative risk of postmenopausal breast cancer. The World Cancer Research Fund determined that evidence is convincing that so-called body fatness increases the relative risk of postmenopausal breast cancer (5). In 1 meta-analysis of 34 cohort studies including over 2.5 million women, Renehan et al. (7) reported that each 5 kg/m² increase in body mass index was associated with a 12% (95% confidence interval (CI): 8, 16) increased relative risk of postmenopausal breast cancer, whereas each 5 kg/m² increase in body mass index was associated with an 8% (95% CI: 3, 12) decreased relative risk of premenopausal breast cancer. The World Cancer Research Fund determined that the strength of the evidence suggests that the inverse association between body fatness and premenopausal breast cancer is probable (5).

Several epidemiologic studies conclude that obesity may be more strongly associated with estrogen receptor–positive postmenopausal breast cancer (6, 8–10). However, a recent publication showed that obesity was similarly related to both estrogen receptor (ER)–positive (hazard ratio = 1.35, 95% CI: 1.20, 1.51) and triple-negative (hazard ratio = 1.37, 95% CI: 0.98, 1.93) disease in a combined analysis of data from the Women's Health Initiative observational cohort and randomized trial (11).

The relationship between obesity and postmenopausal breast cancer may be modified by use of postmenopausal hormone therapy. One of the first studies to demonstrate potential heterogeneity in the association between obesity and postmenopausal breast cancer according to hormone use was published by the Nurses' Health Study in 1997 (12). In this prospective study, body mass index and adult weight gain were associated with 60% to 2-fold increased relative risks, respectively, of postmenopausal breast cancer among women who had never used hormones, whereas the relationships were attenuated among users of hormones. Although numerous studies over the past 2 decades have examined the interaction between hormone use and obesity in relation to postmenopausal breast cancer, many reports combined estrogen-alone therapy with estrogen-progestin hormone preparations in the statistical analysis (12–18). More recently, heterogeneity in the relationship between obesity and breast cancer according to type of hormone therapy has been reported, with mixed results (19, 20). Consideration of the influence of estrogen-alone hormone therapy separately from estrogen-progestin use is motivated by results from the Women's Health Initiative that showed an increased risk of postmenopausal breast cancer for estrogen-progestin (21) but not estrogen-alone (22) therapy.

Given the high prevalence of obesity, we aimed to conduct a meta-analysis of the association between body mass index and breast cancer depending on menopausal status and hormone receptor subtype. To the extent possible, relationships were investigated according to race and ethnicity. We were particularly interested in evaluating the association between obesity and postmenopausal breast cancer according to use of estrogen-progestin hormone therapy, because this association has received inadequate systematic evaluation, published studies appear to include inconsistent results, and prevalence of use of hormone medications has changed dramatically over the past decade.

METHODS

Search strategy and selection criteria

We conducted a systematic review of the literature starting with a review of the National Cancer Institute's Physician's Data Query Breast Cancer Summary of Evidence to identify key studies of the relationship between body mass index and postmenopausal hormone therapy with breast cancer incidence. After reviewing the articles referenced in the Physician's Data Query, we performed a search by topic combining Medical Subject Headings (MeSH) and non-MeSH keyword terms using PubMed (1980–2012). “Breast neoplasms” was combined with “risk factors” and permutations, variations, and abbreviations of the relevant MeSH keywords including “BMI” OR “obesity” OR “overweight” OR “Quetelet” and also “HRT” OR “hormone replacement therapy” OR “PMH” OR “postmenopausal hormonal therapy”. Additional studies were obtained through citations of review articles and reports of meta-analyses including the report prepared by the World Cancer Research Fund (5). Article titles and abstracts were reviewed to determine relevance. Only published English-language articles that provided estimates of relative risk for breast cancer were reviewed for potential inclusion in this meta-analysis; abstracts and unpublished reports were not considered.

Papers were screened and evaluated for eligibility (by G. C.) in the meta-analysis to exclude studies that did not present results according to categories of body mass index or type of hormone use, studies with mortality as the endpoint, and studies limited to special populations (e.g., BRCA mutation carriers, men, mother-daughter pairs, and so on). The remaining papers were then involved in a qualitative synthesis (by M. F. M.) to exclude studies that presented results according to incompatible risk group definitions. For example, for the analysis of body mass index, we excluded studies that used weight or waist-to-hip ratio rather than body mass index, body mass index measured on a continuous scale, or body mass index groups incompatible with obesity defined according to the World Health Organization (i.e., <25, 25–29.9, ≥30 kg/m²) (23). For the analysis of estrogen-progestin therapy, we included only those studies of the use of estrogen combined with progestin and excluded studies with nonspecific hormone therapy and studies of the use of estrogen only. The qualitative synthesis also excluded hospital-based case-control studies, studies without results presented according to menopausal status, and studies without results for white, black, or Hispanic women. In other words, our analysis included clinical trials, cohort studies, case-cohort studies, nested case-control studies, and population-based case-control studies that used incident breast cancer as the endpoint and presented the findings by using comparable measures. Hospital-based case-control studies were excluded because their results were potentially influenced by too many sources of bias.

From the selected papers, we extracted author(s), study dates, study design (e.g., case-control, cohort), risk groups, number of cases, number of controls or person-years at risk, population, population size, measure of risk (e.g., odds ratio, relative risk, hazard ratio), relative risk estimates with confidence intervals, and covariates used in multivariable-adjusted relative risk estimates. When there were multiple reports from a single population, only the results of the report with the most cases or the report with the longest follow-up were included in the analysis.

Statistical analysis

Risk ratio estimates included odds ratios, rate ratios, and hazard ratios. We categorized body mass index into 3 categories: <25.0, 25.0–29.9, and ≥30.0 kg/m² (23). If risk ratio estimates were not presented in these categories in a paper but the paper presented data so that risk ratio estimates could be reorganized into these categories, then we did so and calculated unadjusted risk ratio estimates. Results from papers with separate underweight categories (e.g., <18.5 kg/m²) were combined with the lowest body mass index group (<25 kg/m²). We also included papers with body mass index categories similar to these 3 categories; adjusted risk ratio estimates were taken directly from publications if results were shown for these body mass index categories. Lowest values of body mass index (<25 kg/m²) served as the reference category for all analyses. We categorized estrogen-progestin use into never, ever, current, and past use of estrogen-progestin therapy. Analyses for body mass index, estrogen-progestin use, and both factors jointly were conducted according to strata of menopausal status and repeated in subgroups according to estrogen receptor and progesterone receptor (PR) status and race/ethnicity (white, black, and Hispanic).

For body mass index and for estrogen-progestin therapy, we first fit a fixed-effects model to the log of the risk ratio estimates using Stata, version 12.0, statistical software (StataCorp LP, College Station, Texas) to combine results across studies weighted by the inverse standard error of the risk ratio estimate. We used the Cochran Q test and the I² statistic to examine heterogeneity among the studies (24). In the presence of significant heterogeneity (P < 0.05), we then fit a random-effects model based on the method of DerSimonian and Laird (25). Thus, risk ratios are presented from either fixed- or random-effects models depending on the Q test P value for heterogeneity. Because 2 separate models were fit for the relationship between body mass index and breast cancer (≥30 kg/m² vs. <25 kg/m² and 25–29.9 kg/m² vs. <25 kg/m²), these 2 risk ratios were not necessarily both from random (or fixed) effect models because each depended on separate Q tests. For fixed-effects models, we tested for heterogeneity between case-control and cohort studies. We also examined the probability of publication bias or small-study effects using funnel plots and Egger's test (26–30). Confidence intervals, P values, and statements of statistical significance are nominal in the sense that they are not adjusted to account for multiple comparisons. As such, they are descriptive rather than having inferential content.

RESULTS

We identified 435 candidate papers for the analysis of body mass index and, after screening and review of these papers, we included 57 papers in the analysis of body mass index (Figure 1). We identified 1,076 candidate papers for the analysis of estrogen-progestin use and, after screening and review of these papers, we included 32 papers (Figure 2). From among these 89 papers we identified 28 papers that included some analysis of the interaction of body mass index and estrogen-progestin use. Of these 28 papers, we found only 4 papers that included analyses of estrogen plus progestin use stratified by body mass index or of body mass index stratified by estrogen plus progestin use (19, 31–33) plus 1 additional paper from a study already included in the analysis of the main effects of body mass index and estrogen-progestin use (34). These 5 papers were included in the analysis of the interaction between body mass index and estrogen-progestin use.

Figure 1.

Identification, review, and selection of studies included in the meta-analysis of breast cancer in relation to body mass index (BMI).

Figure 2.

Identification, review, and selection of studies included in the meta-analysis of breast cancer in relation to postmenopausal hormone use.

Premenopausal women: body mass index

Analysis of the relationship between body mass index and premenopausal breast cancer included 22 articles comprising 12,938 participants with breast cancer in reports from case-control studies and 4,469 breast cancers in reports from cohort studies (Table 1; Figure 3). Intermediate values of body mass index (25–29.9 kg/m²) were associated with a modestly reduced risk (risk ratio (RR) = 0.95, 95% CI: 0.91, 1.00) (Figure 3) as compared with a body mass index <25 kg/m². Body mass index ≥30 kg/m² was also associated with a reduced risk ratio of premenopausal breast cancer (RR = 0.83, 95% CI: 0.75, 0.91) (Figure 4). For body mass index ≥30 kg/m², body mass index appeared to be related to ER-positive/PR-positive (RR = 0.78, 95% CI: 0.67, 0.92) but not ER-negative/PR-negative (RR = 1.06, 95% CI: 0.70, 1.60) premenopausal breast cancer (Table 2).

Table 1.

Papers Included in Meta-analysis of Body Mass Index in Relation to Breast Cancer Risk

First Author, Year (Reference No.)	Study Type	Populationa	Body Mass Indexb	Age, Years	Years of Study	No. of Cases	Adjustment Variablesc,d
Ahn, 2007 (92)	Cohort	NIH-AARP Diet and Health Study	<25	50–71	1996–2000	204	A, AF, AM, B, E, H, M, P, S
			25–29
			30–34
			≥35
Barlow, 2006 (93)	Case-control	BCSC	<25	35–84	1996–2002	3,915	A, AF, B, D, F, H, P, T
			25–29
			30–34
			≥35
Barnes, 2011 (94)	Case-control	MARIE	<25	50–74	2002–2005	3,074	None
			25–29
			≥30
Berstad, 2010 (95)	Case-control	CARE	<25	35–64	1994–1998	3,997	A, AM, F, H
			25–29
			30–34
			≥35
Boyd, 2006 (96)	Case-control	NBSS	<25.0	Average, 56.7	1984–1999	1,114	None
			25.0–27.6
			>27.6
Brinton, 2008 (66)	Cohort	NIH-AARP Diet and Health Study	<25	50–71	1995–2002	3,559	A, AF, B, F
			25–29
			≥30
Cerne, 2011 (97)	Case-control	Slovenia	<25	50–69	2006–2008	530	None
			25–29
			≥30
Chang, 2006 (98)	Cohort	PLCO	<22.4	55–74	1993–2003	764	None
			22.5–24.9
			25.0–27.4
			27.5–29.9
			≥30
Chlebowski, 2007 (99)	Randomized trial	Women's Health Initiative	<25	50–79	1995–2000	1,072	A, AF, AL, AM, F, M, P, S
			25–29
			≥30
Cotterchio, 2003 (100)	Case-control	WHS/ECSS	≤25	25–74	1995–1998	2,586	A, AF, AL, B, E, F, M, O, P, S
			25–27
			>27
Cust, 2009 (101)	Case-control	Northern Sweden Health and Disease Cohort	<24.2	Median, 52.5	1985–2005	533	A, AF, H
			24.2–27.4
			≥27.4
Eng, 2005 (102)	Case-control	Long Island Breast Cancer Study Project	<26.4	20–98	1996–1997	1,002	None
			26.4–30.1
			>30.1
Franceschi, 1996 (14)	Case-control	Italy	<25.7	23–74	1991–1994	2,562	None
			25.8–28.8
			>28.8
Friedenreich, 2002 (17)	Case-control	Alberta, Canada	<25.7	<80	1995–1997	1,233	None
			25.7–29.2
			>29.2
Gaudet, 2010 (103)	Case-control	NCI Biological Markers Project	<25	≥47	1977–1987	229	A, AF, P
			25–29
			≥30
Hall, 2000 (104)	Case-control	Carolina Breast Cancer Study	14.6–24.6	20–74	1993–1996	780	None
			24.7–30.1
			>30.1
Harlid, 2012 (105)	Case-control	MDCS, NSHDS; Iceland	<25	>50	1985–2007	1,962	A
			25–29
			≥30
Harris, 1992 (68)	Case-control	American Health Foundation	<22	<30 to >80	1987–1989	604	A, F, P
			22–27
			>27
Hines, 2010 (106)	Cohort	4-Corners Breast Cancer Study	<25	25–79	1992–2002	2,299	A, AF, AL, AM, E, F, H, M, O, P
			25–29.9
			≥30
Hislop, 1986 (107)	Case-control	Cancer Control Agency of British Columbia	<25.6	<70	1980–1982	512	None
			25.6–29.3
			>29.3
Huang, 1997 (12)	Cohort	Nurses’ Health Study	<25	30–55	1976–1992	2,517	None
			25–31
			>31
Huang, 2000 (108)	Case-control	Carolina Breast Cancer Study	<23	20–74	1993–1996	856	A, AL, F, H, M, O, P, S
			23–31
			>31
John, 2003 (109)	Case-control	San Francisco Bay Area Breast Cancer Study	<26.1	35–79	1995–1998	1,263	A
			26.1–31.9
			≥32
John, 2011 (110)	Case-control	San Francisco Bay Area, California	<25	35–79	1995–2004	672	A, AL, B, F, M, P
			25–29.9
			≥30
Jumaan, 1999 (111)	Nested case-control	Sweden	<25	50–70	1987–1990	273	A
			25–29.9
			≥30
Krebs, 2006 (112)	Cohort	Study of Osteopathic Fractures	<25	Average, 73.5	1986–1988	350	A, B, E, F, H, M, O, P, S
			25–29.9
			≥30
Li, 2000 (113)	Case-control	King County, Washington	<25	50–64	1988–1990	479	None
			25–29.9
			≥30
Li, 2006 (19)	Case-control	Seattle-Puget Sound, Washington	<25	65–79	1997–1999	926	A, T
			25–29.9
			≥30
Ma, 2006 (114)	Case-control	Los Angeles County, California	<25	20–49	1998–2003	1,725	None
			25–29.9
			≥30
Magnusson, 1998 (15)	Case-control	All native Swedish female residents	<25.8	50–74	1993–1995	2,799	None
			25.8–28.2
			>28.2
Mannisto, 1996 (45)	Case-control	Kuopio Breast Cancer Study	<25	25–75	1990–1994	327	None
			25–28
			>28
Meier, 2002 (115)	Case-control	UK General Practice Research Database	<25	50–89	1992–1997	2,507	A, S
			25–29.9
			≥30
Michels, 2006 (116)	Cohort	Nurses’ Health Study II	<25	25–42	1989–2003	1,398	None
			25–29.9
			≥30
Modugno, 2006 (117)	Case-control	Women's Health Initiative	<23.8	50–79	1993–1998	199	None
			23.8–27.4
			>27.4
Nemesure, 2009 (118)	Case-control	Barbados	<25	≥21	2002–2006	219	A, AF, B, E, F, H, M
			25–29.9
			≥30
Obi, 2009 (119)	Case-control	MARIE	<25	50–74	2002–2005	3,462	None
			25–29.9
			≥30
Palmer, 2007 (120)	Cohort	Black Women's Health Study	<25	21–69	1995–2005	949	A, AF, AM, E, F, M, P
			25–29.9
			30–34.9
Pan, 2004 (121)	Case-control	Canada	<25	20–76	1994–1997	2,362	A, AF, F, M
			25–29.9
			≥30
Peacock, 1999 (122)	Case-control	Seattle-Puget Sound, Washington	<23.3	21–45	1983–1990	845	None
			23.3–27
			>27
Phipps, 2011 (11)	Randomized trial/cohort	Women's Health Initiative	<25	50–79	1993–1998	2,898	A, F
			25–29.9
			≥30
Rinaldi, 2006 (123)	Case-control	European Prospective Investigation into Cancer and Nutrition	<25	Postmenopausal	1998–2000	613	None
			25–29.9
			≥30
Rosenberg, 2006 (124)	Case-control	All native Swedish female residents	<25.9	50–74	1993–1995	1,241	None
			25.9–28.2
			>28.2
Sarkissyan, 2011 (125)	Cohort	South Los Angeles, California	<25	>30	1995–2007	237	A
			25–29.9
			≥30
Silvera, 2006 (126)	Cohort	NBSS	<25	40–59	1980–2000	1,671	A, AF, AL, B, F, H, M, O, P, S
			25–29.9
			≥30
Suzuki, 2006 (127)	Cohort	Swedish Mammography Cohort	<25	Postmenopausal	1988–2004	1,284	None
			25–29.9
			≥30
Swanson, 1989 (128)	Case-control	BCDDP	<23.5	26–93	1973–1980	1,768	None
			23.5–27.5
			>27.5
Swanson, 1996 (129)	Case-control	Atlanta, Georgia; Seattle, Washington; New Jersey	<24.7	20–44	1990–1992	1,588	None
			24.7–28.8
			>28.8
Sweeney, 2004 (130)	Cohort	Iowa Women's Health Study	<26	55–84	1986–2001	2,286	None
			26–29.5
			>29.5
Toniolo, 1994 (131)	Case-control	NYU Women's Health Study	<23.7	35–65	1990–1991	180	None
			23.7–26.6
			>26.6
Tornberg, 1994 (132)	Cohort	Swedish Two County Study	<24	25–74	1963–1987	1,466	None
			24–27.9
			≥28
Trentham-Dietz, 1997 (133)	Case-control	Maine, Wisconsin, Massachusetts, New Hampshire	<24.9	<74	1988–1991	6,548	None
			25–27.5
			>27.5
Trentham-Dietz, 2000 (13)	Case-control	Wisconsin, Massachusetts, New Hampshire	<24.0	50–79	1992–1995	5,031	None
			24.0–29.4
			>29.4
van den Brandt, 1997 (134)	Case-cohort	Netherlands Cohort Study	<25	55–69	1986–1990	626	None
			25–29.9
			≥30
Vatten, 1992 (135)	Cohort	Norway	<24	44–67	1974–1977	291	None
			24–27
			>27
Weiderpass, 2004 (136)	Cohort	Norway and Sweden	<25	30–50	1991–1999	716	None
			25–29.9
			≥30
White, 2012 (137)	Cohort	Multiethnic Cohort	20–24.9	45–75	1993–2008	2,840	A, AF, AL, AM, E, F, H, P, S, T
			25–29.9
			≥30
Zhu, 2005 (138)	Case-control	Tennessee	<25	20–64	1995–1998	271	A, AL, B, E, F, H, M, P, S
			25–29.9
			≥30

Abbreviations: BCDDP, Breast Cancer Detection Demonstration Project; BCSC, Breast Cancer Surveillance Consortium; CARE, Women's Contraceptive and Reproductive Experiences Study; ECSS, Enhanced Cancer Surveillance Study; EPIC, European Prospective Investigation into Cancer and Nutrition; MARIE, Mammary Carcinoma Risk Factor Investigation; MDCS, Malmö Diet and Cancer Study; NBSS, Canadian National Breast Screening Study; NCI, National Cancer Institute; NIH-AARP, National Institutes of Health-AARP; NSHDS, North Sweden Health and Disease Study; NYU, New York University; PLCO, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; UK, United Kingdom; WHS, Women's Health Study.

^a Studies listed more than once in the table provide data for different subgroup analyses according to menopausal status, estrogen and progesterone receptor status, and race/ethnicity.

^b Body mass index: weight (kg)/height (m)².

^c Adjustment variables: A, age; AF, age at first birth; AL, alcohol; AM, age at menopause; B, benign breast disease; D, breast density; E, exercise or physical activity; F, family history of breast cancer; H, postmenopausal hormone use; M, age at menarche; O, oral contraceptive use; P, parity; S, smoking; T, type of menopause.

^d “None” indicates that sample sizes provided in the source articles were used to calculate unadjusted risk estimates for body mass index in 3 categories (<25, 25–29.9, ≥30 kg/m²).

Figure 3.

Summary forest plots of the association between body mass index and premenopausal breast cancer according to study design, comparing body mass index 25.0–29.9 with <25 kg/m² (Egger's test P = 0.08). Numbers of cases are shown for women with body mass index 25.0–29.9/<25 kg/m². If only 1 sample size is shown for the number of cases, then this number represents all cases in the study. CI, confidence interval.

Figure 4.

Summary forest plots of the association between body mass index and premenopausal breast cancer according to study design, comparing body mass index ≥30 with <25 kg/m² (Egger's test P = 0.14). Numbers of cases are shown for women with body mass index ≥30/<25 kg/m². If only 1 sample size is shown for the number of cases, then this number represents all cases in the study. CI, confidence interval.

Table 2.

Summary Risk Estimates^a From Meta-analysis of the Association Between Body Mass Index and Breast Cancer for Subgroups Defined by Menopausal Status, Estrogen Receptor and Progesterone Receptor Status, and Race/Ethnicity, 1980–2012

Subgroup	No. of Studies	No. of Cases	Body Mass Indexb
			25.0–29.9				≥30.0
			RR	95% CI	Heterogeneityc		RR	95% CI	Heterogeneityc
					P Value	I2 Statistic, %			P Value	I2 Statistic, %
Premenopausal Women
ER and PR status
ER+/PR+	4	2,486	0.94	0.76, 1.17	0.14	45	0.78	0.67, 0.92	0.67	0
ER−/PR−	4	1,360	1.26	1.07, 1.49	0.41	0	1.06	0.70, 1.60	0.004	77
ER+/PR unknown	2	800	1.02	0.85, 1.21	0.52	0	1.01	0.83, 1.24	0.56	0
ER−/PR unknown	2	347	0.98	0.74, 1.29	0.13	57	1.15	0.86, 1.55	0.08	67
Race/ethnicityd
White	11	5,885	0.88	0.81, 0.96	0.91	0	0.72	0.61, 0.85	0.03	50
Black	7	1,668	0.90	0.78, 1.05	0.07	48	0.87	0.70, 1.03	0.31	16
Hispanic	3	756	0.72	0.55, 0.93	0.17	44	0.71	0.54, 0.92	0.08	61
Postmenopausal Women
ER and PR status
ER+/PR+	8	6,733	1.17	1.01, 1.36	0.001	74	1.39	1.14, 1.70	0.001	81
ER−/PR−	9	2,302	1.06	0.95, 1.18	0.99	0	0.98	0.78, 1.22	0.02	57
ER+/PR unknown	6	7,965	1.08	1.02, 1.15	0.09	48	1.22	1.03, 1.45	0.002	74
ER−/PR unknown	3	831	0.96	0.78, 1.17	0.28	22	1.27	1.05, 1.55	0.77	0
Race/ethnicity
White	21	23,905	1.09	1.03, 1.15	0.01	48	1.13	1.02, 1.25	<0.001	74
ER+/PR+	4	4,123	1.11	0.95, 1.29	0.03	68	1.26	0.91, 1.76	<0.001	88
ER−/PR−	4	1,123	1.01	0.87, 1.17	0.68	0	0.84	0.55, 1.30	0.02	69
ER+/PR unknown	3	4,405	1.06	0.92, 1.23	0.05	66	1.19	0.83, 1.69	<0.001	89
Black	7	2,151	0.95	0.76, 1.20	0.04	55	1.12	0.83, 1.52	0.002	71
ER+/PR+	2	309	0.97	0.70, 1.35	0.96	0	1.38	1.00, 1.91	0.53	0
ER−/PR−	2	210	1.12	0.79, 1.60	0.79	0	0.73	0.49, 1.10	0.61	0
Hispanicd	3	969	0.97	0.81, 1.17	0.21	36	1.02	0.64, 1.63	0.03	73

Abbreviations: CI, confidence interval; ER, estrogen receptor; PR, progesterone receptor; RR, risk ratio; +, positive; −, negative.

^a Estimates from fixed-effects models if heterogeneity P value ≥ 0.05 and random-effects models if heterogeneity P value < 0.05 with body mass index <25.0 kg/m² as the reference category.

^b Body mass index: weight (kg)/height (m)².

^c Cochran's Q test P values and the I² statistic to examine heterogeneity among the studies.

^d Sample sizes were inadequate to evaluate associations according to both race/ethnicity and ER status.

Risk ratio estimates for white women suggested similar relationships between body mass index and premenopausal breast cancer as for black and Hispanic women (Table 2). In a comparison of the third with the first category of body mass index, the summary risk ratio for white women was 0.72 (95% CI: 0.61, 0.85), whereas the comparable estimate for black women was 0.87 (95% CI: 0.70, 1.03) and for Hispanic women was 0.71 (95% CI: 0.54, 0.92). Too few studies have published results for the body mass index relationship with premenopausal breast cancer to evaluate both race/ethnicity subgroups and breast cancer subtypes by hormone receptor status.

Postmenopausal women: body mass index

Analyses of the relationship between body mass index and postmenopausal breast cancer included 39 articles comprising 43,005 participants with breast cancer in case-control studies and 16,180 breast cancers in eligible cohort studies (Table 1). Greater body mass index was positively associated with postmenopausal breast cancer, as shown in Figure 5 for 25–29.9 kg/m² (RR = 1.10, 95% CI: 1.06, 1.13) and Figure 6 for ≥30 kg/m² (RR = 1.18, 95% CI: 1.12, 1.25). As among premenopausal women, the relationship between body mass index ≥30 kg/m² and postmenopausal breast cancer appeared limited to ER-positive/PR-positive (RR = 1.39, 95% CI: 1.14, 1.70) but not ER-negative/PR-negative (RR = 0.98, 95% CI: 0.78, 1.22) breast cancer (Table 2).

Figure 5.

Summary forest plots of the association between body mass index and postmenopausal breast cancer according to study design, comparing body mass index 25.0–29.9 with <25 kg/m² (Egger's test P = 0.68). Numbers of cases are shown for women with body mass index 25.0–29.9/<25 kg/m². If only 1 sample size is shown for the number of cases, then this number represents all cases in the study. CI, confidence interval.

Figure 6.

Summary forest plots of the association between body mass index and postmenopausal breast cancer according to study design, comparing body mass index ≥30 with <25 kg/m² (Egger's test P = 0.05). Numbers of cases are shown for women with body mass index ≥30/<25 kg/m². If only 1 sample size is shown for the number of cases, then this number represents all cases in the study. CI, confidence interval.

Risk ratio estimates of postmenopausal breast cancer were very similar comparing relationships limited to white women and black women (Table 2). For comparing body mass index ≥30 kg/m² with <25 kg/m², risk ratio estimates were similarly elevated for ER-positive/PR-positive breast cancer (white women: RR = 1.26, 95% CI: 0.91, 1.76; black women: RR = 1.38, 95% CI: 1.00, 1.91) and null for ER-negative/PR-negative breast cancer (white women: RR = 0.84, 95% CI: 0.55, 1.30; black women: RR = 0.73, 95% CI 0.49, 1.10). The pooled risk ratio of 1.02 (95% CI: 0.64, 1.63) for Hispanic women was null. Data were insufficient to evaluate the risk ratio of postmenopausal breast cancer jointly according to body mass index and hormone receptor status among Hispanic women.

Postmenopausal women: estrogen-progestin use

Analyses of the relationship between estrogen-progestin therapy and postmenopausal breast cancer included 32 articles comprising 32,043 participants with breast cancer from 17 reports of case-control studies and 23,541 breast cancers in 12 reports of cohort studies and 3 reports from 2 randomized trials (Table 3). Ever-use of combined estrogen-progestin hormone use was associated with an elevated risk ratio of postmenopausal breast cancer (RR = 1.34, 95% CI: 1.24, 1.46) (Table 4). The risk ratio for current estrogen-progestin use was elevated (RR = 1.72, 95% CI: 1.55, 1.92) (Table 4), whereas the risk ratio for past use was not significantly increased (RR = 1.02, 95% CI: 0.92, 1.14) (I² = 2%, Q test P = 0.41; data not shown). An elevated risk ratio associated with use of estrogen-progestin therapy was restricted to ER-positive/PR-positive breast cancer (ever use: RR = 1.40, 95% CI: 1.08, 1.82; current use: RR = 1.92, 95% CI: 1.60, 2.30) (Table 4). Too few studies presented results for estrogen-progestin use according to race/ethnicity with or without estrogen receptor status to conduct a meta-analysis.

Table 3.

Papers Included in Meta-analysis of Estrogen-Progestin Use in Relation to Breast Cancer Risk

First Author, Year (Reference No.)	Study Type	Populationa	Estrogen + Progestin Use	Age, Years	Years of Study	No. of Cases	Adjustment Variablesb,c
Bakken, 2011 (139)	Cohort	EPIC	Current	≥55	1992–2006	4,006	AF, AL, AM, BM, P, O, T
Beral, 2011 (140)	Cohort	Million Women Study	Current	50–59	1996–2001	10,419	A, AF, AL, AM, BM, P
Brinton, 2008 (66)	Cohort	NIH-AARP Diet and Health Study	Ever	50–71	1995–2002	1,886	A, AF. F
			Current
			Past
Chen, 2002 (141)	Case-control	Group Health Cooperative	Current	50–74	1990–1995	809	A
			Past
Chen, 2004 (142)	Cohort	Nurses’ Health Study	Current	30–55	1980–2000	807	A, B, AF, AL, AM, F, M, P, T
Chlebowski, 2010 (59)	Randomized trial	Women's Health Initiative	Ever	50–79	2005–2009	569	A, DT
Ewertz, 2005 (143)	Cohort	Denmark	Ever	50–67	1989–2002	678	AF, P
Fournier, 2005 (144)	Cohort	French E3N	Ever	40–66	1990–2000	899	AL, AM, B, BM, F, P, M, O
Fournier, 2008 (145)	Cohort	French E3N	Ever	Postmenopausal	1990–2002	692	A, AF, AM, B, BM, F, M, O, P, T
Hines, 2010 (106)	Case-control	4-Corners Breast Cancer Study	Current	25–79	1999–2002	1,026	A, AF, AM, B, BM, F, M, O, P, T
Hulley, 2002 (52)	Randomized trial	HERS	Ever	Postmenopausal	1993–2000	88	A, AF, BM, F, P
Kotsopoulos, 2009 (146)	Case-control	Nurses’ Health Study	Current	≥30	1971–2006	704	None
Lee, 2006 (147)	Cohort	Multiethnic Cohort Study	Past	45–75	1993–1996	746	AF, AL, AM, BM, E, F, M, P, T
Li, 2000 (148)	Case-control	King County, Washington	Current	50–64	1988–1990	22	A, T
Li, 2006 (19)	Case-control	Seattle-Puget Sound, Washington	Current	65–79	1997–1999	835	A, T
Li, 2008 (149)	Case-control	Seattle-Puget Sound, Washington	Ever	55–74	2000–2004	355	A, F
			Current
			Past
Lyytinen, 2010 (150)	Case-control	Finland	Ever	50–62	1995–2007	7,204	A, AF, P
Newcomb, 1995 (151)	Case-control	Wisconsin, Massachusetts, Maine, New Hampshire	Ever	≤75	1989–1991	2,345	AF, AL, AM, B, BM, F, M, T
			Past
Newcomb, 2002 (31)	Case-control	Wisconsin, Massachusetts, Maine, New Hampshire	Ever	50–79	1992–1994	5,289	A, AF, AL, AM, B, BM, E, F, M
Opatrny, 2008 (152)	Case-control	UK General Practices Research Database	Ever	50–75	1988–2004	5,161	AL, BM, F, O, S, T
Persson, 1999 (153)	Cohort	Uppsala Health Care Region	Ever	Average, 65	1987–1993	76	A, AF, AM, BM
Reding, 2012 (154)	Case-control	CARE	Current	35–64	1994–1998	534	None
			Past
Rosenberg, 2006 (124)	Case-control	Sweden	Ever	50–74	1993–1995	1,521	None
Rossouw, 2002 (21)	Randomized trial	Women's Health Initiative	Ever	50–79	1993–2002	290	A, DT
Saxena, 2010 (32)	Cohort	California Teacher's Study	Ever	≤80	1995–2006	1,646	A, AF, AL, AM, B, BM, F, M, P, S
Schairer, 1994 (155)	Cohort	Breast Cancer Detection Demonstration Project	Ever	<55–>75	1980–1989	609	A, AF, AM, B, F, T
			Current
			Past
Shantakumar, 2007 (69)	Case-control	Long Island Breast Cancer Study, New York	Ever	20–98	1996–1997	803	None
Sprague, 2008 (156)	Case-control	Wisconsin, Massachusetts, New Hampshire	Current	≥54	1997–2001	2,324	None
Stahlberg, 2004 (157)	Cohort	Danish Nurse Study	Ever	≥44	1993–1999	130	AF, AL, AM, B, BM, E, M, O, P, S
Ursin, 2002 (64)	Case-control	Los Angeles, California	Ever	55–72	1987–1996	1,298	None
Weiss, 2002 (158)	Case-control	CARE	Ever	35–64	1994–1998	1,361	A, T
Yang, 1992 (159)	Case-control	British Columbia, Canada	Ever	≤75	1988–1989	452	A, T

Abbreviations: CARE,Women's Contraceptive and Reproductive Experiences Study; EPIC, European Prospective Investigation into Cancer and Nutrition; French E3N,the French component of the European Prospective Investigation into Cancer and Nutrition; HERS, Heart and Estrogen/progestin Replacement Study; NIH-AARP,National Institutes of Health-AARP; UK, United Kingdom.

^a Studies listed more than once in the table provide data for different subgroup analysis according to menopausal status, estrogen and progesterone receptor status, and race/ethnicity.

^b Adjustment variables: A, age; AF, age at first birth; AL, alcohol; AM, age at menopause; B, benign breast disease; BM, body mass index; DT, diet; E, exercise or physical activity; F, family history of breast cancer; M, age at menarche; O, oral contraceptive use; P, parity; S, smoking; T, type of menopause.

^c “None” indicates that sample sizes provided in the source articles were used to calculate unadjusted risk estimates for hormone therapy use.

Table 4.

Summary Risk Estimates^a From Meta-analysis of the Association Between Estrogen-Progestin Hormone Use and Postmenopausal Breast Cancer for Subgroups Defined by Estrogen Receptor and Progesterone Receptor Status, 1980–2012

Subgroup	No. of Studies	No. of Cases	RR	95% CI	Heterogeneityb
					P Value	I2 Statistic, %
Ever Use of Estrogen-Progestin Hormones
Overall	21	33,677	1.34	1.24, 1.46	<0.001	79
ER+/PR+	3	1,872	1.40	1.08, 1.82	0.02	74
ER−/PR−	3	548	1.09	0.87, 1.37	0.40	0
Current Use of Estrogen-Progestin Hormones
Overall	13	22,917	1.72	1.55, 1.92	<0.001	79
ER+/PR+	2	876	1.92	1.60, 2.30	0.11	60
ER−/PR−	2	258	1.11	0.78, 1.57	0.98	0
ER+/PR unknown	2	3,042	2.55	1.65, 3.92	0.006	87

Abbreviations: CI, confidence interval; ER, estrogen receptor; PR, progesterone receptor; RR, risk ratio; +, positive; −, negative.

^a Estimates from random-effects (if P < 0.05 for heterogeneity) and fixed-effects (if P ≥ 0.05 for heterogeneity) models with never users of postmenopausal hormones as the reference category.

^b Cochran's Q test P values and the I² statistic to examine heterogeneity among the studies.

Postmenopausal women: joint association of body mass index and estrogen-progestin use

Five studies presented sufficient data to examine the relationship between body mass index and postmenopausal breast cancer among subgroups defined by estrogen-progestin use. Among never users of estrogen-progestin therapy, both case-control and cohort studies consistently reported an elevated risk ratio of postmenopausal breast cancer associated with greater body mass index, as shown in Table 5 and Figure 7, where the combined risk ratio was 1.42 (95% CI: 1.30, 1.55) for ≥30 compared with <25 kg/m². However, among ever users of estrogen-progestin therapy (Figure 8), the summary risk ratio estimate from cohort studies was elevated for body mass index ≥30 kg/m² (RR = 1.26, 95% CI: 1.02, 1.56), and the summary risk ratio estimate from case-control studies was null (RR = 0.98, 95% CI: 0.68, 1.42). These risk ratio estimates were not precise so that the test for heterogeneity according to study design was not significant (P = 0.25).

Table 5.

Summary Risk Estimates^a From Meta-analysis of the Association Between Body Mass Index and Postmenopausal Breast Cancer for Subgroups Defined by Use of Estrogen-Progestin Hormones, 1980–2012

Subgroup	No. of Studies	No. of Cases	Body Mass Indexb
			25.0–29.9				≥30.0
			RR	95% CI	Heterogeneityc		RR	95% CI	Heterogeneityc
					P Value	I2 Statistic, %			P Value	I2 Statistic, %
Overall	39	59,185	1.10	1.06, 1.13	0.01	38	1.18	1.12, 1.25	<0.001	72
Hormone use
Never	5	5,711	1.16	1.06, 1.27	0.81	0	1.42	1.30, 1.55	0.36	8
Ever	5	1,337	1.11	0.96, 1.29	0.50	0	1.18	0.98, 1.42	0.32	15

Abbreviations: CI, confidence interval; RR, risk ratio.

^a Estimates from fixed-effects models if the heterogeneity P ≥ 0.05 and random-effects models if the heterogeneity P < 0.05 with body mass index <25.0 kg/m² as the reference category.

^b Body mass index: weight (kg)/height (m)².

^c Cochran's Q test P values and the I² statistic to examine heterogeneity among the studies.

Figure 7.

Summary forest plots of the association between body mass index ≥30 (reference value, <25 kg/m²) and postmenopausal breast cancer according to study design, among never users of postmenopausal hormones (Egger's test P = 0.53). Numbers of cases are shown for women with body mass index ≥30/<25 kg/m². CI, confidence interval.

Figure 8.

Summary forest plots of the association between body mass index ≥30 (reference value, <25 kg/m²) and postmenopausal breast cancer according to study design, among ever users of postmenopausal hormones (Egger's test P = 0.84). Numbers of cases are shown for women with body mass index ≥30/<25 kg/m². CI, confidence interval.

An inadequate number of studies provided results jointly for body mass index and estrogen-progestin use in relation to breast cancer according to subgroups defined by race/ethnicity or ER status to conduct a meta-analysis.

Evaluation of bias

As a subjective measure of study quality, estimates were calculated separately for case-control and cohort studies (including randomized trials). Overall, retrospective and prospective studies resulted in very similar estimates (Figures 3–7). The joint analysis of body mass index and ever use of estrogen-progestin therapy suggested that risk ratios were elevated among cohort but not all case-control studies, although the test for heterogeneity according to study design was not significant (Figure 8) (P = 0.25). We did not evaluate studies on the basis of whether body mass index was measured or self-reported or on the basis of the source of data—medical records or self-report—for hormone medication use.

Funnel plots were constructed as indicators of bias due to publication bias and effect measures from small studies (Web Figures 1–3 available at http://aje.oxfordjournals.org/). Visual inspection of funnel plots and Egger's test did not suggest that most funnel plots were significantly asymmetrical (P > 0.05) (Web Figures 1 and 3) except for the association between body mass index and postmenopausal breast cancer for comparing body mass index ≥30.0 kg/m² versus <25.0 kg/m² (P = 0.05) (Web Figure 2).

DISCUSSION

Results of this meta-analysis suggest that obese women have a reduced risk of hormone receptor–positive premenopausal breast cancer and an increased risk of hormone receptor–positive postmenopausal breast cancer as compared with women of “normal” body mass. Risk ratios appeared attenuated for hormone receptor–negative breast cancer and, for postmenopausal breast cancer, among women using estrogen-progestin therapy. Results stratified according to race/ethnicity appeared consistent for premenopausal breast cancer and for white women and black women for postmenopausal breast cancer; risk ratio estimates were null for the relationship between body mass index and postmenopausal breast cancer among Hispanic women.

Results from our analysis of the main associations of body mass index with breast cancer agree with reports from other meta-analyses and large studies. The meta-analysis of cohort studies published by Renehan et al. (7), which included the Million Women Study (35), reported a risk ratio equal to 0.92 (95% CI: 0.88, 0.97) of premenopausal breast cancer for each 5 kg/m² increase in body mass index. In our analysis, summary risk ratio estimates for receptor-negative premenopausal breast cancer were null, while significantly reduced risk ratio estimates were limited to receptor-positive breast cancer. These same results were found in a meta-analysis conducted by Suzuki et al. (6) and in a pooled analysis of studies participating in the Breast Cancer Association Consortium (36). However, another meta-analysis of 11 studies reported an elevated summary odds ratio for triple-negative premenopausal breast cancer comparing body mass index ≥30 kg/m² with <30 kg/m² (pooled odds ratio = 1.43, 95% CI: 1.23, 1.65) (37); the methods for this meta-analysis differed from our approach by combining results for case-case and case-control comparisons as well as evaluating risk ratios for only 2 categories of body mass index (obese vs. nonobese). As molecular subtyping for breast cancer continues incorporating more markers than just the estrogen and progesterone receptors, increasing etiologic heterogeneity will likely be uncovered.

Meta-analysis of studies of obesity in relation to postmenopausal breast cancer overwhelmingly supports elevated relative risk in comparison with normal body mass (7). As with our study, other meta-analyses and systematic reviews using body mass index have reported increased relative risk for receptor-positive breast cancer (9). Several studies have examined different measures of body size with respect to breast cancer including weight (6, 13, 17), adult weight gain (8, 12, 13, 16, 17, 38–40), fat mass (41), and indicators of central obesity including waist circumference and waist-to-hip ratio (17, 41–46). Regardless of the measure of adiposity, relative risk estimates range close to the ones reported in this analysis for body mass index, with relative risk increased by approximately 40% for receptor-positive postmenopausal breast cancer and attenuated relative risk estimates for breast cancer overall or all other subtypes of breast cancer combined.

Limited studies have investigated the relationship between obesity and breast cancer among nonwhite populations. We found that risk ratio estimates for the relationship between obesity and pre- and postmenopausal breast cancer were similar for white women and black women. A meta-analysis of 5 studies (7) suggested that obesity may increase the risk of premenopausal breast cancer among women from Asia and the Pacific, including Japanese women in Hawaii, where the risk ratio was 1.16 (95% CI: 1.01, 1.32) per 5 kg/m² increase in body mass index; results for postmenopausal women were similar for all racial groups (P = 0.09), although there was a suggestion that the summary risk ratio estimate for Asian women was slightly greater than the estimates for whites in Europe and North America. For Hispanic women, only 4 studies contributed to this meta-analysis, with estimates for premenopausal breast cancer similar across race/ethnic groups. The summary risk ratio for postmenopausal breast cancer associated with obesity was null for Hispanic women. In another review, Sexton et al. (47) identified only 3 studies of body size in relation to breast cancer in Hispanic women; results from these 3 studies were also null, underscoring the need for adequately powered studies of breast cancer in Hispanic women (48–50).

Epidemiologic evidence has long suggested that estrogen-progestin use is associated with an increase in breast cancer (51). In 2002, the Women's Health Initiative (WHI) published the results of their large, placebo-controlled, randomized trial of estrogen plus progestin, finding that the risks of use outweighed the benefits (21). After approximately 5 years of follow-up, women assigned to the estrogen plus progestin arm were 26% (95% CI: 0, 59) more likely to develop breast cancer than women in the placebo arm. These findings were nearly identical to the results of the Heart and Estrogen/progestin Replacement Study (HERS) in the same year (52). Following these publications, there were rapid and dramatic reductions in postmenopausal hormone use in the United States (53–56). Nationally, prescriptions for postmenopausal hormone therapy fell by 38% in the first year following publication of the Women's Health Initiative results (57). Ravdin et al. (58) observed that breast cancer incidence in the United States fell by 6.7% in 1 year; this decrease was most evident in women over 50 years of age and in breast cancers that were estrogen receptor positive. Our meta-analysis concurred with this ecological analysis; the pooled risk ratio of 1.92 (95% CI: 1.60, 2.30) for current estrogen-progestin use in relation to hormone receptor–positive breast cancer was significantly elevated, but it was null at 1.11 (95% CI: 0.78, 1.57) for ER-negative/PR-negative breast cancer. These estimates were based on only 2 studies. Furthermore, an updated analysis of data from the Women's Health Initiative suggests that use of combined estrogen-progestin hormones is related similarly to both ER-positive and ER-negative disease (P_difference = 0.81) (59). Although the relative risk of receptor-negative postmenopausal breast cancer appears likely to be smaller than for receptor-positive disease, the evidence does not conclusively rule out a positive association for estrogen-progestin therapy (9).

The prevalence of estrogen-progestin use has continued to decline to very low levels since 2002. Trends in use of combined estrogen-progestin hormone use were described by using data from the National Health and Nutrition Examination Study (NHANES, 1999–2008) (60). The age-adjusted prevalence of estrogen-progestin use in 1999–2000 was 8.3% and declined to 2.9% in 2003–2004; by 2009–2010, estrogen-progestin use decreased to 1.7%. Synthesis of the evidence led the US Preventive Services Task Force to recommend that clinicians do not prescribe estrogen-progestin therapy for the prevention of chronic conditions (D grade) (61). Notably, our meta-analysis, based on 5 studies, suggests that risk ratio estimates for the association between body mass index and postmenopausal breast cancer are significantly elevated among never users—but not ever users—of estrogen-progestin therapy. Since substantially fewer women are now using postmenopausal hormones, obesity may consequently contribute to a greater proportion of the breast cancer burden.

Several studies that were not included in this meta-analysis have also reported on the modification of the association between body mass index and postmenopausal breast cancer according to estrogen-progestin use; the primary reason these studies were not included was due to presentation of the body mass index results in categories that were not compatible with the normal, overweight, and obese categories in this analysis. Two of the largest cohort studies reported the results of statistical tests for interaction (European Prospective Investigation into Cancer and Nutrition (EPIC): P_interaction = 0.003 (18); Women's Health Initiative: P_interaction = 0.001 (62)). Although 3 additional studies reported nonsignificant values for the formal test of interaction between body mass index and estrogen-progestin use in relation to postmenopausal breast cancer (13, 19, 63), many studies including both case-control (13–15, 17, 19) and cohort (12, 16, 18, 62, 63) studies reported attenuated relative risk estimates for body mass index among ever, former, or current users of estrogen-progestin therapy as compared with never users. Multiple reports (but not all (64)) have also suggested that the association between estrogen-progestin use and postmenopausal breast cancer is likewise attenuated among obese women (31–34, 51, 65–69).

Strengths and limitations

Interpretation of the results of this meta-analysis must consider several limitations. Publications were included in the analysis only if the results could be considered in the 3 categories according to well-recognized cutpoints for obesity based on body mass index as defined by the World Health Organization (23). For example, a report from the Vermont Breast Cancer Surveillance System (70) and a report from the Million Women Study (35) were not included because results were provided for 5 categories of body mass index that did not include those corresponding to the World Health Organization categories. Pooled analyses with individual-level data could address this limitation but may exclude studies based on data-sharing restrictions (51, 63). Studies of estrogen-only postmenopausal hormone use also were not included. Although estrogen-only hormone use was found to reduce risk of breast cancer incidence and mortality in the Women's Health Initiative, results from this randomized trial did not suggest that body mass index modifies the association between estrogen-only use and breast cancer incidence (P_interaction = 0.49) (71). Other systematic reviews, meta-analysis, or pooled analysis of data from individual studies will be necessary to confirm whether the association between obesity and postmenopausal breast cancer varies according to estrogen-only hormone use.

Results in the analysis were a combination of adjusted relative risk estimates taken directly from the published papers along with crude estimates if, for example, body mass index results were combined to align with the 3 categories. In most studies, multivariable relative risk estimates are very similar to unadjusted estimates, although adjusted estimates are frequently slightly attenuated as compared with the unadjusted risk ratios (38).

Meta-analysis relies on the quality of the individual studies. We did not evaluate each study on the basis of indicators of quality, such as participation rates or loss to follow-up. However, we did exclude hospital-based studies. Tests of heterogeneity from fixed-effects models suggest that pooled results for prospective studies were not significantly different from pooled results for retrospective studies. Thus, case-control, cohort, and randomized trials were generally in close agreement regarding the relationship among body mass index, estrogen-progestin therapy, and breast cancer.

Many studies collected information on body mass index, postmenopausal hormone use, and other factors by using self-report rather than medical records or measured values. This may especially have influenced the results of this meta-analysis for self-reported body mass index because heavier women are more likely to underreport their weight. In another meta-analysis that did evaluate the extent to which self-reporting of body mass index affected the results, the authors noted that the effect size between studies in which women self-reported body mass index and those in which weight and height were measured was 1.06 (95% CI: 1.02, 1.09) (7). However, several studies have shown satisfactory reproducibility for self-reported body weight (13, 72–77) and high validity of estrogen-progestin therapy reporting in comparison with medical records (78–81).

Publication bias is a common concern for reports of meta-analyses. In particular, studies of null findings may be missing from the literature, especially once the field appears to be nearing consensus. Funnel plots and statistical tests suggest that publication bias did not strongly influence the results of this meta-analysis except potentially for the evaluation of obese versus normal body mass index in relation to postmenopausal breast cancer (Egger's test P = 0.05) (Figure 6); although results for this highest category of body mass index were more potentially susceptible to bias because of the lack of small studies with positive findings, Egger's test suggests that publication bias was less of a concern for the results comparing overweight with normal body mass index in relation to postmenopausal breast cancer (Egger's test P = 0.68) (Figure 5).

Biological mechanisms

The biological mechanisms underlying the association between body mass index and breast cancer are thought to primarily involve sex hormone pathways (82). Estrogen levels are positively associated with breast cancer among postmenopausal women (83); relationships between endogenous estrogens and breast cancer development appear to be more complex among premenopausal women (84). Among premenopausal women, obesity is associated with anovulation and lower levels of circulating estrogen levels (85, 86). In contrast, estrogen levels are elevated among obese postmenopausal women because adipose tissue is the primary source of estrogen production via aromatase enzyme conversion of androgenic precursors (87). An analysis of data pooled from 8 prospective studies found that the association between body mass index and postmenopausal breast cancer was essentially eliminated by adjusting for serum estrogen concentrations (88).

The results of this meta-analysis are consistent with these proposed biological mechanisms. Breast cancer subtypes characterized by ER/PR expression are thought to be etiologically distinct diseases, with ER/PR breast cancers having a predominantly hormone-sensitive etiology (9). Consistent with this hypothesis, body mass index was associated with increased ER/PR-positive, but not ER/PR-negative, breast cancer. Further, the attenuation of the risk ratios between body mass index and breast cancer among estrogen-progestin users lends further support to the hypothesis that obesity influences risk via sex hormone pathways. Circulating estrogen levels are elevated among women using estrogen-progestin therapy, and thus the impact of adipose tissue estrogen production may be expected to be minimized.

Additional mechanisms that could contribute toward the association of body mass index with breast cancer risk among postmenopausal women include chronic hyperinsulinemia (89) and altered adipocytokine production (90). These pathways could influence both ER/PR-positive and -negative subtypes.

Conclusions and implications

This study estimates that obesity is associated with a 20% reduction in receptor-positive premenopausal breast cancer relative risk and approximately 20%–40% increased relative risk of receptor-positive postmenopausal breast cancer. As molecular characterization of breast cancer tumors increases in complexity, greater etiologic heterogeneity in terms of the relationship between body mass index and breast cancer may be revealed beyond the simple categorization of tumors according to ER and PR, or basal versus luminal, subtype. Although studies to date suggest that the relationship between body mass and breast cancer is similar for white women and black women, inconsistencies in the literature suggest that additional research is needed regarding other groups, particularly for Asian and Hispanic women.

Observational studies agree with 2 randomized trials in confirming that estrogen-progestin therapy increases risk of postmenopausal hormone use by about 35% overall and potentially as much as 2-fold for hormone receptor–positive breast cancer among current estrogen-progestin users. This analysis suggests that the relationship between body mass index and postmenopausal breast cancer was strongest in women who never took estrogen-progestin therapy. With the dramatic decline in use of postmenopausal hormones, the role of body mass index in postmenopausal women is likely elevated. A Congressionally authorized panel including federal members from several agencies, nonfederal scientists and clinicians, and breast cancer advocates has recently recognized the urgent need to increase focus on breast cancer prevention (91), and avoidance of obesity after menopause is likely an effective measure to decrease the burden of breast cancer.