Body weight and incidence of breast cancer defined by estrogen and progesterone receptor status—A meta-analysis

Authors

  • Reiko Suzuki,

    1. Institute of Environmental Medicine, Division of Nutritional Epidemiology, Karolinska Institutet, Stockholm, Sweden
    2. Cancer Research UK Epidemiology Unit, University of Oxford, Richard Doll Building, Oxford, United Kingdom
    3. Division of Epidemiology and Prevention, Research Center for Cancer Prevention and Screening,National Cancer Center, 5-1-1 Tukiji, Chuo-ku, Tokyo 104-0045, Japan
    Search for more papers by this author
  • Nicola Orsini,

    1. Institute of Environmental Medicine, Division of Nutritional Epidemiology, Karolinska Institutet, Stockholm, Sweden
    Search for more papers by this author
  • Shigehira Saji,

    1. Division of Clinical Trials and Research, Tokyo Metropolitan Cancer and Infectious Diseases Center, Komagome Hospital, Tokyo, Japan
    Search for more papers by this author
  • Timothy J. Key,

    1. Cancer Research UK Epidemiology Unit, University of Oxford, Richard Doll Building, Oxford, United Kingdom
    Search for more papers by this author
  • Alicja Wolk

    Corresponding author
    1. Institute of Environmental Medicine, Division of Nutritional Epidemiology, Karolinska Institutet, Stockholm, Sweden
    • Institute of Environmental Medicine, Division of Nutritional Epidemiology, Karolinska Institutet, Nobelsväg 13, Stockholm S-171 77, Sweden
    Search for more papers by this author
    • Fax +46-830-4571.


Abstract

Epidemiological evidence indicates that the association between body weight and breast cancer risk may differ across menopausal status as well as the estrogen receptor (ER) and progesterone receptor (PR) tumor status. To date, no meta-analysis has been conducted to assess the association between body weight and ER/PR defined breast cancer risk, taking into account menopausal status and study design. We searched MEDLINE for relevant studies published from January 1, 1970 through December 31, 2007. Summarized risk estimates with 95% confidence intervals (CIs) were calculated using a random-effects model. The summarized results of 9 cohorts and 22 case-control studies comparing the highest versus the reference categories of relative body weight showed that the risk for ER+PR+ tumors was 20% lower (95% CI = −30% to −8%) among premenopausal (2,643 cases) and 82% higher (95% CI = 55–114%) among postmenopausal (5,469 cases) women. The dose-response meta-analysis of ER+PR+ tumors showed that each 5-unit increase in body mass index (BMI, kg/m2) was associated with a 33% increased risk among postmenopausal women (95% CI = 20–48%) and 10% decreased risk among premenopausal women (95% CI = −18% to −1%). No associations were observed for ER−PR− or ER+PR− tumors. For discordant tumors ER+PR− (pre) and ER−PR+ (pre/post) the number of cases were too small (<200) to interpret results. The relation between body weight and breast cancer risk is critically dependent on the tumor's ER/PR status and the woman's menopausal status. Body weight control is the effective strategy for preventing ER+PR+ tumors after menopause. © 2008 Wiley-Liss, Inc.

Obesity could affect breast cancer risk through affecting circulating endogenous estrogen levels. Therefore, it have been postulated that the association between body weight and breast cancer risk may be heterogeneous according to the tumor's estrogen receptor (ER) and progesterone receptor (PR) status. Cumulative epidemiological evidence1–3 also suggests that the impact of body weight on breast cancer risk differs across women's menopausal status; studies indicate a weak inverse association between body weight and breast cancer risk among premenopausal women4 and a positive association among postmenopausal women (reversal association discussion5).6

Many epidemiological studies have evaluated body weight in relation to ER and PR defined breast cancer incidence. Ten studies of 9 prospective cohorts - the Iowa Women's Health Study,7, 8 the Melbourne Collaborative Cohort Study,9 the Nurses' Health Study I,10 the Nurses' Health Study II,11 the Swedish Mammography Cohort Study,12 the American Cancer Society's Cancer Prevention Study II Nutrition Cohort,13 the Japan Public Health Center-based Prospective Study,14 the Black Women's Health Study,15 and the Women's Health Initiative study Cohort16- and 22 case-control studies17–38 addressed this issue. However, the results are inconsistent. Of these 31 epidemiological studies, 7 did not specify the menopausal status.10, 20, 21, 26, 30, 32, 37

One qualitative literature review39 and one quantitative review40 reported the relation of ER and/or PR defined breast cancer risk with postmenopausal obesity. However, no study has quantitatively summarized the results with a meta-analytical approach for the relationship between body weight and the risk of breast cancer defined by ER/PR status.

We therefore summarized all available evidence to clarify the association between body weight and the incidence of breast cancer defined by ER/PR status of the tumors. We performed meta-analysis of cohort and case-control studies with stratification by menopausal status as well as study design.

Abbreviations:

BMI, body mass index; CIs, confidence intervals; E1, estrone; E2, estradiol; ER, estrogen receptor; OC, oral contraceptives; PMH, postmenopausal hormones; PR, progesterone receptor; RE, risk estimate; SD, standard deviation.

Material and methods

Selection of studies

Eligible studies were identified by searching the MEDLINE database (National Library for Medicine, Washington, DC) from January 1st 1970 through December 31th 2007 for relevant epidemiologic studies of body weight in relation to the risk of breast cancer defined by ER/PR tumors status and without any language restriction. For computer searches, we used the following Medical Subject Headings and/or text words in any field: body mass index (BMI), “weight,” “obesity,” and “breast cancer” combined with ER and PR. Studies were included in the meta-analyses if they presented data from a cohort or case-control study on the association between body weight and incidence of ER and/or PR defined breast tumors.

Data extraction and classification

The following information was extracted from each publication: the name of the first author, the year of publication, the country in which the study was conducted, study design, the name of the cohort study (cohort studies), type of controls and response rates (case-control studies), years of follow-up (cohort studies), year of data collection (case-control studies), sample size, the assay methods for determination of ER and/or PR (if provided), age/menopausal status, type of breast tumor according to ER and/or PR subtypes, the risk estimates (REs) and 95% confidence intervals (CIs), body weight categories studied and adjusted covariates.

If a study provided REs based on both relative body weight (BMI, kg/m2) and body weight gain, only the REs based on BMI were used.15, 35, 37 If the study provided the REs for both current and previous BMI,7, 8, 11, 15, 21, 22, 37 we used the results based on current BMI in the present meta-analysis. If a study provided several REs with different control factors, we extracted the REs reflecting the greatest degree of control for potential confounders by priority or those with a narrower CI. If menopausal status was not reported in a study, we assumed it as mixed-menopausal status. In total 32 studies (10 cohorts7–16 and 22 case-control studies17–38) were eligible.

Statistical analysis

We used the reported REs as the measure of association between the level of body weight and the risk of ER/PR defined breast cancer. Study-specific REs were pooled using the DerSimonian and Laird random-effect model.41 When separate REs were provided across different populations, such as race and menopausal status, these REs were independently included in the meta-analysis.

We performed meta-analyses comparing the highest vs. the reference (or lowest) body weight categories within the specific studies. We next conducted a dose-response meta-analysis for ER+PR+ tumors. We used the method proposed by Greenland and coworkers42, 43 to estimate study specific slopes from the correlated natural log of the REs across levels of body weight. For each study, the midpoint of the upper and lower boundaries in each category was assigned to each corresponding REs. If the upper boundary of the highest category was not available, the same amplitude of the lower nearest category was applied. When the lowest category was open-ended, the same amplitude of the upper nearest category was applied as the boundary of the lowest category. In one study, the results were reported according to the distribution of BMI with 3 categories (i.e., the lowest 10% as low, the upper 10% as high),21 and the cutoff points were assumed based on a previous publication.44 In the dose–response meta-analysis for ER+PR+ tumors, we excluded studies that had less than 3 quantitative exposure categories7, 25, 28 and those that did not provide the REs based on the units of BMI or the stratification by menopausal status.10, 13, 26, 29, 30

Statistical heterogeneity among studies included in the meta-analysis was assessed using the Q and I2 statistics.45 We considered p < 0.1 as a representative of statistically significant heterogeneity among studies. However, when the test indicated a statistically significant heterogeneity but all REs were in the same direction, we considered the observed statistical heterogeneity among studies as nonsubstantial (only differences in the strength of association) and results concordant. Publication bias was assessed by the Egger's regression asymmetry test.46 Statistically significant publication bias was defined as p < 0.1. Statistical analyses were performed using Stata (release 9.2; StataCorp, College Station, TX). All statistical tests were 2-sided.

Results

Characteristics of studies

A total 10 studies from 9 prospective cohorts 7–16 (Table I) and 22 case-control studies17–38 (Table II) fulfilled the criteria for inclusion in the present meta-analysis for evaluating the association between body weight and the risk for ER and/or PR defined breast cancer. Of those 32 epidemiological studies, 19 studies7, 8, 10, 11, 13, 15, 16, 17, 19, 22, 23, 25, 26, 29, 30, 33, 34, 36, 37 were carried out in the United States, 412, 32, 35, 38 in Europe, 318, 21, 27 in Canada, 39, 20, 28 in Australia and 314, 24, 31 in Asia. The study population consisted of premenopausal women in 7 studies,11, 17, 19, 25, 28, 31, 34 of only postmenopausal women in 11 studies7–9, 12, 13, 15, 16, 29, 33, 35, 36 of both premenopausal and postmenopausal in 7 studies,14, 18, 22–24, 27, 38 and of those without information about menopausal status in 7 studies10, 20, 21, 26, 30, 32, 37

Table I. Characteristics of 10 Prospective Studies (in 9 Cohort) for Body Weight and ER and/or PR Defined Breast Cancer
Author, year, country, (cohort) [ERPR assay]Follow-up, overall cases and study sizeAge and/or menopausal statusType of tumorER/PR defined cases number (approximate % known receptor)Effect estimatesExposure measurement (body weight/ BMI)Adjusted factors
  • IWHS, the Iowa Women's Health Study cohort; NHS, the Nurses' Health Study cohort; SMC, the Swedish Mammography Cohort; CPS II, the American Cancer Society's Cancer Prevention Study II Nutrition Cohort; JPHC, the Japan Public Health Center-based Prospective Study; BWHS, the Black Women's Health Study; MCC, the Melbourne Collaborative Cohort Study; WHI, the Women's Health Initiative Study; DCC, dextran-coated charcoal assay/biochemical; IHC, immunohistochemistry method/immunoperoxidase; EIA, enzyme linked immunoassay; N.A., not available; US, the United States.

  • 1

    Based on invasive breast cancer cases within 5 yr of baseline in observational study cohort and randomized clinical trial cohort.

Potter et al., 19957 US (IWHS) [N.A.]1986−1992 939 cases (37,105)55–69 yr BMI, kg/m2Age
PostER+PR+414 cases (68%)1.38 (1.12–1.71)≥30 vs. <30
 ER+PR−99 cases (16%)0.49 (0.27–0.88)≥30 vs. <30
 ER−PR+17 cases (3%)2.88 (1.11–7.46)≥30 vs. <30
 ER−PR−80 cases (13%)0.75 (0.43–1.31)≥30 vs. <30
Sellers et al., 20028 US (IWHS) [N.A.]1986−1998 1,874 cases (37,105)55–69 yr BMI, kg/m2Education, age at menarche, age at menopause, use of OC, use of PMH, parity, age at first birth, alcohol intake, smoking status, physical activity level and family history of breast caner.
PostER+1,043 cases (82%)2.00 (1.58–2.53)≥30.70 vs. ≤22.89
 ER−232 cases (18%)1.38 (0.78–2.43)≥30.70 vs. ≤22.89
 PR+993 cases (73%)2.24 (1.72–2.91)≥30.70 vs. ≤22.89
 PR−362 cases (27%)0.96 (0.62–1.49)≥30.70 vs. ≤22.89
MacInnis et al., 20049 Australia (MCC) [IHC/DCC/unknown]1990−2002 357 cases (13,598)27–75 yr (99%, 40–69) post BMIAge, country of birth, exercise, use of PMH, education.
ER+97 cases (77%)1.25 (1.05–1.49)Per 5 kg/m2
ER−29 cases (23%)0.82 (0.55–1.24)Per 5 kg/m2
PR+84 cases (67%)1.17 (0.95–1.44)Per 5 kg/m2
PR−42 cases (33%)1.11 (0.86–1.45)Per 5 kg/m2
Colditz et al., 200410 US (NHS I) [DCC/IHC]1980−2000 2,096 cases (66,145)30–70 yr Incidence BMI (ref. average)Age, duration of premenopause, menopause, pregnancy history, benign breast disease, use of PMH, height, alcohol consumption and family history of breast cancer.
MixER+PR+1,281 cases (61%)1.27 (1.15–1.39)Weight gain
   1.17 (1.05–1.30)Consistently obese
 ER+PR−318 cases (15%)0.93 (0.76–1.14)Weight gain
   0.82 (0.65–1.03)Consistently obese
 ER−PR+80 cases (4%)1.33 (1.00–1.76)Weight gain
   1.15 (0.80–1.67)Consistently obese
 ER−PR−417 cases (20%)0.96 (0.83–1.11)Weight gain
   0.90 (0.76–1.08)Consistently obese
Michels et al., 200611 US (NHS II) [N.A.]1989−2003 1,398 cases (113,130)25–42 yr   BMI, kg/m2Age, family history of breast cancer, history of benign breast disease, height, age at menarche, age at first birth, parity, alcohol consumption, physical activity and current and past OC use.
PreER+669 cases (70%)0.76 (0.59–0.97)≥30 vs. 20.0−22.4
  669 cases0.91 (0.84–0.99)Per 5 kg/m2
 ER−285 cases (30%)1.10 (0.76–1.58)≥30 vs. 20.0−22.4
  285 cases1.03 (0.91–1.15)Per 5 kg/m2
 PR+636 cases(68%)0.81 (0.63–1.05)≥30 vs. 20.0−22.4
  636 cases0.94 (0.86–1.02)Per 5 kg/m2
 PR−300 cases (32%)1.01 (0.71–1.45)≥30 vs. 20.0−22.4
  300 cases0.98 (0.86–1.10)Per 5 kg/m2
Suzuki et al., 200612 Sweden (SMC) [EIA/IHC]1987−2004 1,188 cases (51,847)    BMI, kg/m2Age, height, family history of breast cancer, age at menarche, parity, age at first birth, education, use of OC, use of PMH, the reason of menopause, history of benign breast disease, total energy intake, energy-adjusted total fiber intake and total fat intake, alcohol intake.
PostER+PR+716 cases (60%)1.67 (1.34–2.07)≥30 vs. 18.5–24.9
ER+PR−279 cases (23%)0.76 (0.49–1.17)≥30 vs. 18.5–24.9
ER−PR+50 cases (4%)1.46 (0.58–3.69) †≥30 vs. 18.5–24.9
ER−PR−143 cases (12%)0.52 (0.26-1.04)≥ 30 vs.18.5–24.9
Feigelson et al., 200613 US (CPS II) [N.A.]1992−2001 651 cases (44,161) unknown 549 cases (45% of all cases)50–74 yr post   Weight gain from 18 yr to 1992 (pounds)Age, age at menarche, age at menopause, number of live births, age at first live birth, use of OC, family history of breast cancer, history of breast cysts, history of mammography, height, education, physical activity, alcohol intake, aspirin use and race. (excluded PMH user)
 ER+PR+445 cases (68%)2.42 (1.82–3.23)≥61 vs. <5–20
 ER−PR−98 cases (15%)1.78 (0.98–3.23)≥61 vs. <5–20
Iwasaki et al., 200714 Japan (JPHC) [EIA/IHC]1993−1995 441 cases (55,537)40–69 yr   BMIAge, area (10 public health care centers), number of births, age at first birth and height.
PreER+62 cases (60%)1.04 (0.98–1.11)Per 1kg/m2
 ER−41 cases (40%)1.02 (0.93–1.13)Per 1kg/m2
 PR+53 cases (56%)1.06 (0.99–1.13)Per 1kg/m2
 PR−42 cases (44%)0.99 (0.89–1.10)Per 1kg/m2
PostER+65 cases (61%)1.08 (1.01–1.15)Per 1kg/m2
 ER−41 cases (39%)0.95 (0.84–1.06)Per 1kg/m2
 PR+46 cases (46%)1.07 (0.98–1.16)Per 1kg/m2
 PR−55 cases (54%)1.01 (0.93–1.10)Per 1kg/m2
PostER+PR+44 cases (42%)1.10 (1.01–1.18)Per 1kg/m2
 ER−PR−36 cases (35%)0.98 (0.87–1.10)Per 1kg/m2
Palmer et al., 200715 US (BWHS) [N.A.]1995−2005 442 cases (known 172) (9,542)>56 yr   Current BMI, kg/m2Age, age at menarche, parity, age at first birth, age at menopause, vigorous activity, education, family history of breast cancer and BMI at age 18 yr (excluded PMH user).
PostER+PR+84 cases (49%)1.66 (0.86–3.21)≥30 vs. <25
 ER−PR−52 cases (30%)0.88 (0.39–1.97)≥30 vs. <25
Chlebowski et al., 200716 US (WHI) [N.A.]1993− 3,236 cases (147,916)150–79 yr post   BMI, kg/m2Age group at screening, race or ethnicity, family history of breast cancer (first-degree relatives), previous breast biopsy examination, age at menarche, age at menopause, parity, age at first birth, breast feeding time, smoking, alcohol intake, duration of PMH (estrogen-only use and duration of estrogen plus progestin use).
ER+2,391 cases (84%)1.26 (1.12–1.43)≥30 vs. <25
ER−459 cases (16%)1.21 (0.92–1.60)≥30 vs. <25
Table II. Characteristics of 22 Case-Control Studies for Body Weight and ER and/or PR Defined Breast Cancer
Authors, year, countries, [major ERPR assay]Type of control (y-data collection) No. of overall cases and controlAge or menopausal statusType of tumorThe number of cases (approximate % known receptor)Effect estimatesExposure measurement(body weight/ BMI) 
  • DCC, dextran-coated charcoal assay/biochemical; IHC, immunohistochemistry method/immunoperoxidase; EIA, enzyme linked immunoassay; US, the United States.

  • 1

    Including peri menopausal women.

  • 2

    Including peri/postmenopausal women.

  • 3

    From age 50 yr to 1 yr prior to reference date.

McTiernan et al., 198617 US [DCC]Population-based 1981−1982 329 cases 332 controls (240 known)25–54 yr   Adult weight (lbs.)Age, height
PreER+143 cases (60%)1.0 (0.58–1.9)≥146 vs. ≤125
 ER−97 cases (40%)1.2 (0.63–2.5)≥146 vs. ≤125
Hislop et al., 198618 Canada [DCC]Neighborhood-based 1980−1982 512 cases 540 controls<70 yr   Weight (lbs.)Age, family history of breast cancer, parity, age at first birth, age at menarche, history of benign breast disease.
MixER+345 cases (67%)1.05 (0.78–1.41)≥160 vs. <120
 ER−167 cases (33%)0.67 (0.49–0.91)≥160 vs. <120
PreER+131 cases (68%)1.08 (0.67–1.72)≥160 vs. <120
 ER−62 cases (32%)0.30 (0.21–0.42)≥160 vs. <120.
PostER+214 cases (67%)1.00 (0.68–1.47)≥160 vs. <120.
 ER−105 cases (33%)1.06 (0.66–1.70)≥160 vs. <120.
Stanford et al., 198719 US [DCC]Population-based 1980−1982 458 cases 568 controls20–54 yr   BMI, kg/m2Age, race, menopausal status, family history of breast cancer, history of benign breast disease, use of OC.
PreER+204 cases (45%)1.54 (0.89–2.67)≥25 vs. <20
 ER−254 cases (55%)1.43 (0.85–2.40)≥25 vs. <20
Copper et al., 198920 Australia [DCC]Population-based 1982−1984 451 cases 451 controls20–74 yr   BMI, kg/m2Age at diagnosis (matched), menopausal status, menstrual and reproductive factors, family history of breast cancer, history of benign breast disease, use of PMH, smoking, and several dietary factors.
MixER+248 cases (66%)1.24 (0.79–1.94)>26 vs. <22.8
 ER−130 cases (34%)0.99 (0.55–1.80)>26 vs. <22.8
Kreiger et al., 199121 Canada [DCC]Population-based 1982−1986 528 cases 1,214 controls20–69 yr   BMI, kg/m2Age, menopausal status, age at menarche, age at menopause, number of birth, family history of breast cancer, cyctic breast disease.
MixER+334 cases (67%)1.73 (1.17–2.56)High vs. Mid
 ER−167 cases (33%)2.03 (1.06–3.87)High vs. Mid
 PR+277 cases (55%)1.68 (1.09–2.57)High vs. Mid
 PR−224 cases (45%)2.10 (1.22–3.64)High vs. Mid
Enger et al., 200022 US [DCC/IHC]Neighborhood-based Pre: 1983−1988 424 cases 714 controls post: 1987−1989 760 cases 1,091 controls    BMI, kg/m2Age at reference year (matched), socioeconomic status (matched), number of full term pregnancy, breast feeding, family history of breast cancer, the level of physical activity (For premenopaual women, age at menarche, age at first full term pregnancy) (For postmenopausal women, age at menopause, use of PMH and alcohol intake).
PreER+PR+209 cases (49%)1.11 (0.70–1.77)≥27.1 vs. <21.7
 ER+PR−51 cases (12%)0.92 (0.34–2.47)≥27.1 vs. <21.7
 ER−PR−145 cases (34%)1.07 (0.56–1.68)≥27.1 vs. <21.7
PostER+PR+450 cases (59%)2.45 (1.73–3.47)≥27.1 vs. <21.7
 ER+PR−159 cases (21%)1.29 (0.78–2.15)≥27.1 vs. <21.7
 ER−PR−127 cases (17%)1.20 (0.70–2.05)≥27.1 vs. <21.7
Huang et al., 200023 US [IHC]Population-based 1993−1996 862 cases 790 control20–74 yr   BMI, kg/m2Age at menarche, parity, history of breast feeding, history of abortion, WHR, use of OC, use of PMH, family history of breast cancer, medical radiation exposure, smoking, alcohol intake, education. (Furthermore, matched race, age at diagnosis, the offset term.).
MixER+PR+381 cases (49%)1.1 (0.7–1.7)> 31 vs. < 23
ER+PR−78 cases (10%)1.0 (0.5–2.3)> 31 vs. < 23
ER−PR+64 cases (8%)1.0 (0.4–2.4)> 31 vs. < 23
ER−PR−262 cases (33%)0.7 (0.4–1.2)> 31 vs. < 23
Pre1ER+PR+168 cases0.6 (0.3–1.2)> 31 vs. < 23
ER−PR−151 cases0.6 (0.3–1.2)> 31 vs. < 23
PostER+PR+213 cases1.6 (0.9–3.0)> 31 vs. < 23
ER−PR−111 cases0.8 (0.4–1.7)> 31 vs. < 23
Yoo et al., 200124 Japan [DCC/EIA]Hospital-based 1988−1992 1,154 cases (known 458) 21,714 control≥25 yr   BMIAge at interview, occupation, family history of breast cancer, age at menarche, menstrual regularity at age 20–29, age at menopause, age at first birth, number of full term pregnancy, breast feeding, drinking habit, smoking habit.
MixER+291 cases (64%)1.08 (1.05–1.12)per 1 kg/m2
ER−167 cases (36%)1.05 (1.00–1.10)per 1 kg/m2
PR+200 cases (44%)1.07 (1.03–1.12)per 1 kg/m2
PR−255 cases (56%)1.07 (1.03–1.11)per 1 kg/m2
PreAll 1.01 (0.98–1.04)per 1 kg/m2
ER+ 1.05 (0.99–1.12)per 1 kg/m2
ER− 1.02 (0.94–1.11)per 1 kg/m2
PR+ 1.02 (0.95–1.10)per 1 kg/m2
PR− 1.06 (0.99–1.14)per 1 kg/m2
PostAll 1.07 (1.04–1.10)per 1 kg/m2
ER+ 1.09 (1.05–1.13)per 1 kg/m2
ER− 1.05 (0.99–1.12)per 1 kg/m2
PR+ 1.09 (1.04–1.14)per 1 kg/m2
PR− 1.07 (1.02–1.11)per 1 kg/m2
Britton et al., 200225 US [DCC/IHC]Population-based 1990−1992 1,556 cases 1,397 controls20–44 yr   BMI, kg/m2Age, race, education, WHR, gravidity, abortion, parity, age at first birth, age at menarche, family history of breast cancer, menopausal status, lactation, use of OC, smoking, recreational exercise at 12−13 and in year prior to interview, alcohol intake.
PreER+PR+600 cases (51%)0.77 (0.62–0.96)>24.6 vs. ≤24.6
 ER+PR−114 cases (10%)1.04 (0.68–1.60)>24.6 vs. ≤24.6
 ER−PR+111 cases (9%)1.01 (0.66–1.56)>24.6 vs. ≤24.6
 ER−PR−344 cases (29%)0.91 (0.70–1.19)>24.6 vs. ≤ 24.6
Wenten et al., 200226 US [N.A.]Population-based 1992−1994 1,507 cases 1,039 controls30–74 yr mix   Weight change, kgAge, ethnicity, 7 heath planning district, family history of breast cancer, total METS, parity, use of OC, breast feeding, age at first birth, use of estrogens, menopausal status, and weight at 18y (1the table information was used in this study).
Hispanic   
 ER+PR+133 cases (63%)3.04 (1.47–6.26)>14 vs. <4 kg
 ER−PR−77 cases (37%)1.73 (0.71–4.25)>14 vs. <4 kg
Nonhispanic   
 ER+PR+133 cases (63%)1.05 (0.60–1.84)1>14 vs. <4 kg
 ER−PR−77 cases (37%)0.93 (0.42–2.05)>14 vs. <4 kg
Cotterchio et al., 200327 Canada [DCC/IHC]Population-based 1995−1998 3,276 known cases (2,586 for current study) 3,691 controls25–74 yr   BMI, kg/m2Age, age at menarche, parity, age at first birth, duration of OC use, alcohol intake, smoking, breast feeding, benign breast disease, family history of breast cancer. For premenopausal women, current strenuous activity. For postmenopausal women, age at menopause, duration of PMH.
PreER+PR+489 cases (65%)0.71 (0.50–1.00)>27 vs. ≤25
 ER−PR−265 cases (35%)1.35 (0.89–2.05)>27 vs. ≤25
PostER+PR+1,373 cases (75%)1.61 (1.32–1.98)>27 vs. 20–25
 ER−PR−459 cases (25%)1.48 (1.09–1.99)>27 vs. 20–25
     
McCredie et al., 200328 Australia [IHC/DCC/unknown]Population-based 1992−1999 765 cases (694 known) 564 control< 40 yr   BMI, kg/m2Study center, study period, reference age, education level, country of birth, marital status, family history of breast cancer affected first-degree relative, height, age at menarche, number of live births, use of OC.
PreER+PR+323 cases (52%)0.9 (0.7–1.2)≥23 vs. <23
 ER+PR−34 cases (6%)1.2 (0.6–2.3)≥23 vs. <23
 ER−PR+80 cases (13%)1.0 (0.6–1.6)≥23 vs. <23
 ER−PR−181 cases (29%)1.1 (0.8–1.5)≥23 vs. <23
     
Eng et al., 200529 US [N.A]Population-based 1996−1997 990 cases 1,006 controls20–98 yr   Weight change(kg)3Age at reference date, number of pregnancy, use of PMH, family history of breast cancer, history of benign breast disease, BMI at age 50 yr.
PostER+PR+387 cases (58%)2.17 (1.38–3.42)11.34–62.14 vs. 0
 ER+PR−121 cases (18%)1.25 (0.57–2.74)11.34–62.14 vs. 0
 ER−PR+30 cases (4%)1.54 (0.38–6.35)11.34–62.14 vs. 0
 ER−PR−132 cases (20%)1.50 (0.72–3.13)11.34–62.14 vs. 0
Rusiecki et al., 200530 US [IHC]Hospital-based 1994−1997 420 cases (318 known) 406 controls40–80 yr   BMI, kg/m2Age, age at menarche, null parity/age at first full-term pregnancy, lifetime lactation, menopausal status, alcohol intake, ever use of estrogen, smoking, family history of breast cancer, race.
MixER+PR+104 cases (33%)1.0 (0.6–1.9)≥30 vs. <25.0
 ER+PR−65 cases (20%)0.8 (0.4–1.8)≥30 vs. <25.0
 ER−PR+41 cases (13%)0.9 (0.3–2.3)≥30 vs. <25.0
 ER−PR−107 cases (34%)1.3 (0.7–2.3)≥30 vs. <25.0
Nichols et al., 200531 Vietnam and China [IHC]Hospital-based 1993–1999 682 cases 649 controls24–57 yr   BMI, kg/m2Age, parity, age at first birth, alcohol intake, spouse's education, (parity, and age at first birth did not adjusted for simultaneously when evaluating the effect of either variables).
PreER+271 cases (62%)0.89 (0.59–1.35)≥21.7 vs. ≤18.5
 ER−168 cases (38%)0.88 (0.54–1.42)≥21.7 vs.≤18.5
Tsakountakis et al., 200532 Greece [IHC]Hospital-based 1996–2002 384 cases 566 controlsMix    Age, residence, menopausal age, age at menarche, use of OC, use of PMH, family history of breast cancer, age at full-term pregnancy, parity, abortion, lactation, medication to suppress lactation, radiation to chest, benign breast disease.
HER-2/neu+  BMI, kg/m2
 ER+60 cases (41%)5.59 (2.58–12.13)>29 vs. ≤29
 ER−85 cases (59%)5.33 (2.59–10.94)>29 vs. ≤29
HER-2/neu−   
 ER+120 cases (52%)2.84 (1.52–5.32)>29 vs. ≤29
 ER−112 cases (48%)2.41 (1.15–5.04)>29 vs. ≤29
Li et al., 200633 US [N.A.]Population-based 1997−1999 975 cases 1,007 controls65–79 yr 92% ERPR known cases BMI, kg/m2Age at diagnosis, reference year, type of menopause.
PostAll926 cases1.2 (0.9–1.5)≥30 vs. ≤24.9
 ER+PR+615 cases (72%)1.3 (1.0–1.7)≥30 vs. ≤24.9
 ER+PR−139 cases (16%)1.1 (0.7–1.7)≥30 vs. ≤24.9
 ER−PR−95 cases (11%)0.9 (0.5–1.6)≥30 vs. ≤24.9
Ma et al., 200634 US [N.A.]Population-based 2000− 2003 1,725 cases (1,366 known) 440 controls20–49 yr   BMI 1 yr before reference date, kg/m2Age, race, education, first degree family history of breast cancer, age at menarche, gravidity, number of full term of pregnancies, alcohol intake, age at first full-term pregnancy, duration of breast feeding, use of OC, menopausal status, (use of PMH).
Pre2All1,725 cases0.88 (0.58–1.34)≥35 vs. <25
 ER+PR+854 cases (63%)0.69 (0.43–1.11)≥35 vs. <25
 ER−PR−385 cases (28%)1.18 (0.70–2.01)≥35 vs. <25
Rosenberg et al., 200635 Sweden [EIA]Population-based 1993−1995 2,643 cases 3,065 controls50–74 yr   Recent BMI, kg/m2Age, age at first birth, use of PMH (controlled by exclusion).
PostER+PR+772 cases (62%)2.2 (1.7–2.8)≥28.3 vs. <22.2
 ER+PR−202 cases (16%)1.0 (0.7–1.6)≥28.3 vs. <22.2
 ER−PR+41 cases (3%)2.2 (0.9–5.6)≥28.3 vs. <22.2
 ER−PR−226 cases (18%)1.6 (1.0–2.5)≥28.3 vs. <22.2
Han et al., 200636 US [DCC/IHC]Population-based 1996−2001 1,166 cases 2,105 controls35–79 yr   Weight change, kgAge, education, age at menarche, age at first birth, previous benign breast disease, family history of breast cancer, age at menopause, BMI residuals, use of PMH.
PostER+510 cases (79%)2.42 (1.62–3.61)>27.3 vs. 0–9.1
 ER−136 cases (21%)1.19 (0.58–2.43)>27.3 vs. 0–9.1
 PR+389 cases (60%)3.14 (1.96–5.04)>27.3 vs. 0–9.1
 PR−257 cases (40%)1.21 (0.73–2.01)>27.3 vs. 0–9.1
Slattery et al., 200737 US [N.A.]Population-based 1999−2004 2,325 cases 2,525 controls25–79 yr Mix   Recent BMI, kg/m2Age (matched), height, physical activity, energy intake, parity, alcohol consumption, age at first pregnancy, age at menopause, menopausal status, estrogen status, and center.
Nonhispanic   
 ER+812 cases (80%)0.93 (0.74–1.17)≥30 vs. <25
 ER−199 cases (20%)1.40 (0.95–2.07)≥30 vs. <25
Hispanic   
 ER+381 cases (74%)1.06 (0.76–1.46)≥30 vs. <25
 ER−133 cases (26%)0.45 (0.26–0.76)≥30 vs. <25
Sherman et al., 200738 Poland [IHC/DCC]Population-based 2000−2003 2,386 cases (842 used) 2,502 controls20–74 yr   BMIAge, study site, level of education, age at menarche, parity, age at first full term of birth, age at menopause, menopausal status, family history of breast cancer, history of benign breast disease, history of screening mammogram, and use of PMH for postmenopausal women.
PreER+ 0.47 (0.25–0.87)per 5 kg/m2
 ER− 1.08 (0.84–1.39)per 5 kg/m2
 PR+ 0.68 (0.49–0.94)per 5 kg/m2
 PR− 1.03 (0.78–1.36)per 5 kg/m2
PostER+ 0.98 (0.83–1.15)per 5 kg/m2
 ER− 0.87 (0.71–1.06)per 5 kg/m2
 PR+ 1.23 (1.05–1.43)per 5 kg/m2
 PR− 0.83 (0.69–1.00)per 5 kg/m2

Among the cohort studies,7–16 sample sizes ranged from 9,54215 to 147,91616 and the number of breast cancer patients with known receptor status ranged from 1269 to 2,850.16 Of the 22 case-control studies, the number of case-patients with known receptor status ranged from 24017 to 2,586,27 and the number of control subjects varied from 33217 to 21,714.24 Control subjects were recruited from the general population in 16 studies17, 19–21, 23, 25–29, 33, 34, 35–38 neighborhood in 2,18, 22 and hospital in 4 studies.24, 30–32

Highest vs. referent category

ER+PR+ tumors

For ER+PR+ tumors, 16 studies, of which 5 were cohort studies7, 10, 12, 13, 15 (165,839 participants and 2,940 patients with breast cancer) and 11 case-control studies,22, 23, 25–30, 33–35 (6,823 cases and 13,995 controls) were included in the meta-analysis. Among premenopausal women (based on 6 case-control studies22, 23, 25, 27, 28, 34), we observed a statistically significant 20% lower risk of developing ER+PR+ tumors (95% CI = −30% to −8%; Fig. 1). Among postmenopausal women based on 10 studies (6 case-control22, 23, 27, 29, 33, 35 and 4 cohort studies7, 12, 13, 15) we observed a statistically significant 82% higher risk (95% CI = 55–114%). We found a statistically significant heterogeneity across menopausal status (Pheterogeneity < 0.0001). The results for postmenopausal women were similar across study design; the summary of REs case-control studies = 1.89 (95% CI = 1.52–2.36) and for RE cohort studies = 1.74 (95% CI = 1.34–2.25). A statistical heterogeneity among 10 of postmenopausal women studies7, 12, 13, 15, 22, 23, 27, 29, 33, 35; Pheterogeneity = 0.007) was considered as nonsubstantial because all REs were in the same direction. No publication bias was observed for these analyses (all Ppublication bias > 0.15).

Figure 1.

ER+PR+tumors: Risk estimates from epidemiological studies estimating association between body weight (highest vs. reference categories) and the risk of ER+PR+breast cancer. Squares indicate study-specific risk estimates (size of square reflects the study-specific statistical weight, i.e. inverse of variance); horizontal lines indicate 95% confidence intervals; diamond indicates summary risk estimate with its corresponding 95% confidence interval. Test for heterogeneity overall, Q = 126.64, p < 0.0001, I2 = 85.0%. Test for heterogeneity by menopausal status (p < 0.0001); Premenopausal (6 case-control studies) RE = 0.80 (95% CI: 0.70–0.92); Q = 4.26, p = 0.513, I2 = 0.0%: (Ppublication bias = 0.78); Postmenopausal (6 case-control studies) RE = 1.89 (1.52–2.36); Q = 11.83, p = 0.037, I2 = 57.7%: (Ppublication bias = 0.82) (4 cohort studies); RE = 1.74 (95% CI: 1.34–2.25); Q = 9.54, p = 0.023, I2 = 68.6%:(Ppublication bias = 0.70) (10 studies); RE = 1.82 (95% CI: 1.55–2.14); Q = 22.67, p = 0.007, I2 = 60.3%: (Ppublication bias = 0.44). Test for heterogeneity by study design (p = 0.14); Cohort studies (all 5 cohort studies); RE = 1.58 (95% CI: 1.21–2.05); Q = 26.90, p < 0.0001, I2 = 85.1%:(Ppublication bias = 0.16). 1cc: case-control study; c: cohort study.

An additional sensitivity analysis was performed after excluding studies that did not adjust for postmenopausal hormone (PMH) use. The summary RE based on 8 studies12, 13, 15, 22, 23, 27, 29, 35 with adjustment PMH use was 2.02 (95% CI = 1.79–2.27, Pheterogeneity among studies = 0.40).

ER−PR− tumors

The analyses for ER−PR− tumors based on 16 studies (11 case-control studies22, 23, 25–30, 33–35 including 2,882 cases and 13,995 controls and 5 cohort studies7, 10, 12, 13, 15 with 165,839 participants and 790 patients). No overall association was observed comparing the highest versus the reference categories; the summary of RE = 1.04 (95% CI = 0.92–1.17; Fig. 2). According to menopausal status, the summary of results were not heterogeneous between 6 studies of premenopausal22, 23, 25, 27, 28, 34 and 10 of postmenopausal women7, 12, 13, 15, 22, 23, 27, 29, 33, 35 (Pheterogeneity = 0.57). The summary REs were heterogeneous across study design (Pheterogeneity = 0.07). A nonsignificant 9% higher risk was observed based on 11 case-control studies. The summary RE based only on cohort studies could not be assessed due to a substantial heterogeneity among those 5 studies (Pheterogeneity = 0.098). There was no evidence for publication bias among the studies (all Ppublication bias ≥ 0.30).

Figure 2.

ER−PR− tumors: Risk estimates from epidemiological studies estimating association between body weight (highest vs. reference categories) and the risk of ER − PR − breast cancer. Squares indicate study-specific risk estimates (size of square reflects the study-specific statistical weight, i.e. inverse of variance); horizontal lines indicate 95% confidence intervals; diamond indicates summary risk estimate with its corresponding 95% confidence interval. Overall RE = 1.04 (0.92–1.17). Test for heterogeneity overall, Q = 23.20, p = 0.23, I2 = 18.1% (Ppublication bias = 0.34). Test for heterogeneity by menopausal status (p = 0.569); Premenopausal (6 case-control studies) RE = 1.03 (95% CI = 0.87–1.22); Q = 5.23, p = 0.389, I2 = 4.4%, (Ppublication bias = 0.93); Postmenopausal (6 case-control and 4 cohort studies); RE = 1.06 (95% CI = 0.84–1.33); Q = 13.68, p = 0.134, I2 = 34.2% (Ppublication bias = 0.30). Test for heterogeneity by study design (p = 0.066); Case-control studies (n = 11); RE = 1.09 (95% CI = 0.96–1.23); Q = 11.97, p = 0.609, I2 = 0.0% (Ppublication bias = 0.59); Cohort studies (n = 5); RE = 0.90* (95% CI: 0.66–1.24); Q = 7.84, p = 0.098, I2 = 49.0%: (Ppublication bias = 0.97), (*should not be assessed due to heterogeneity among studies). 1cc: case-control study; c: cohort study.

ER+PR− tumors

The summary of results for ER+PR− tumors was assessed in 3 cohort studies7, 10, 12 (155,097 participants and 696 patients with breast cancer) and 7 case-control studies22, 25, 28–30, 33, 35 (885 cases and 8,426 controls). No associations were found based on 3 studies of premenopausal women,22, 25, 28 and 6 studies of postmenopausal women,7, 12, 22, 29, 33, 35 and there was no overall association based on all 10 studies (the summary RE = 0.91 (95% CI = 0.79–1.04; Fig. 3). However, there was statistically significant heterogeneity in REs between case-control and cohort studies (Pheterogeneity across study design = 0.014). Among 3 cohort studies, we observed statistically significant 25% lower risk (95% CI = −41 to −5%), but not among case-control studies. In all the analyses, there was no evidence of publication bias (all Ppublication bias ≥ 0.32).

Figure 3.

ER+PR− tumors: Risk estimates from epidemiological studies estimating association between body weight (highest vs. reference categories) and the risk of ER+PR− breast cancer. Squares indicate study-specific risk estimates (size of square reflects the study-specific statistical weight, i.e. inverse of variance); horizontal lines indicate 95% confidence intervals; diamond indicates summary risk estimate with its corresponding 95% confidence interval. Overall RE = 0.91 (95% CI = 0.79–1.04). Test for heterogeneity overall, Q = 10.l5, p = 0.43, I2 = 1.5%. Test for heterogeneity by menopausal status (p = 0.41); Premenopausal (3 case-control studies) RE = 1.06 (95% CI: 0.76–1.49); Q = 0.22, p = 0.897, I2 = 0.0%:(Ppublication bias = 0.92); Postmenopausal (4 case-control studies) RE = 1.12 (0.87–1.43); Q = 0.67, p = 0.88, I2 = 0.0%: (Ppublication bias = 0.35) (2 cohort studies); RE = 0.64 (95% CI: 0.42–0.97); Q = 1.37, p = 0.24, I2 = 27.3% (Ppublication bias = n.a.) (all 6 studies); RE = 0.93 (95% CI: 0.72–1.21); Q = 8.16, p = 0.148, I2 = 38.7%: (Ppublication bias = 0.93). Test for heterogeneity by study design (p = 0.014); Case-control studies (8 studies); RE = 1.08 (95% CI: 0.89–1.31); Q = l.58, p = 0.98, I2 = 0.0% (Ppublication bias = 0.82); Cohort studies (3 cohort studies); RE = 0.75 (95% CI: 0.59–0.95); Q = 2.53, p = 0.28, I2 = 21.1%: (Ppublication bias = 0.32). 1cc: case-control study; c: cohort study.

ER−PR+ tumors

No association was observed between body weight and the development of ER−PR+ tumors based on 5 case-control studies25, 28–30, 35 (303 cases and 5,667 controls) and 3 cohort studies7, 10, 12 (147 cases and 155,097 participants); the overall summary of RE was 1.19 (95% CI = 0.96–1.47; Pheterogeneity among studies = 0.46; Fig. 4). According to menopausal status, we observed among premenopausal women no association (2 studies25, 28) and among postmenopausal women (4 studies7, 12, 29, 35) the summary RE was 2.01 (95% CI = 1.22–3.32). In these analyses, there was no evidence of publication bias (all Ppublication bias ≥ 0.12).

Figure 4.

ER − PR + tumors: Risk estimates from epidemiological studies estimating association between body weight (highest vs. reference categories) and the risk of ER − PR + breast cancer. Squares indicate study-specific risk estimates (size of square reflects the study-specific statistical weight, i.e. inverse of variance); horizontal lines indicate 95% confidence intervals; diamond indicates summary risk estimate with its corresponding 95% confidence interval. Overall RE = 1.19 (0.96–1.47). Test for heterogeneity overall, Q = 6.72, p = 0.46, I2 = 0.0%. (Ppublication bias = 0.12). Test for heterogeneity by menopausal status (p = 0.069); Premenopausal (2 case-control studies) RE = 1.01 (95% CI: 0.73–1.39); Q = 0.00, p = 0.976, I2 = 0.0%; Postmenopausal (2 case-control studies) RE = 1.98(0.92 to 4.26); Q = 0.17, p = 0.677, I2 = 0.0%. (2 cohort studies) RE = 2.03 (95% CI: 1.04–3.95); Q = 1.01, p = 0.316, I2 = 0.5%. (all 4 studies); RE = 2.01 (95% CI: 1.22–3.32); Q = 1.18, p = 0.758, I2 = 0.0%: (Ppublication bias = 0.69). Test for heterogeneity by study design (p = 0.41); Case-control studies (n = 5); RE = 1.10 (95% CI: 0.82–1.46); Q = 2.87, p = 0.579, I2 = 0.0%: (Ppublication bias = 0.33); Cohort studies (n = 3); RE = 1.47 (95% CI: 0.89–2.43); Q = 3.16, p = 0.206, I2 = 36.7%: (Ppublication bias = 0.36). 1cc: case-control study; c: cohort study.

ER+, ER−, PR+ and PR− tumors

We performed further sensitivity analysis to evaluate the summary REs for breast cancer defined only by ER status (ER+, ER−), or by PR status (PR+, PR−). This allowed us to include additional studies (Tables I, II) that did not have specific tumor subtype information based on both receptors (ER/PR). We followed a principle of not summarizing heterogeneous results. The summary REs were substantially different between ER+PR+ and ER+PR− tumors among postmenopausal women; therefore, in analyses of postmenopausal ER+ tumors, we included 9 studies on ER+ PR+12, 13, 15, 22, 23, 27, 29, 33, 35 and additional 4 studies8, 16, 18, 36 of ER+ tumors, but the results for ER+PR− tumors12, 22, 29, 33, 35 were not included.

Table III summarized the results from Figures 1–4 and present summary REs by ER or PR status. For all ER+ and all PR+ tumors, we observed a statistically significant lower risk among premenopausal women and increased risk among postmenopausal women. Among postmenopausal women, we also observed statistically significant increased risk for ER− tumors.

Table III. Summary risk estimates (REs) with 95% confidence intervals (CI) for body weight in relation to incidence of breast cancer defined by ER and PR status - meta-analysis (Highest vs. Reference (lowest))
 PremenopausalPostmenopausal
Receptor statusStudies (total cases)RE95% CIPheterogeneityStudies (total cases)RE95% CIPheterogeneity
  • 1

    For included studies see Figures 1–4.

  • 2

    Statistical heterogeneity was considered as non-substantial because all risk estimates were in the same direction (i.e. >1.0).

  • 3

    Results for ER+/PR− tumors were not included because of a significant heterogeneity of REs between ER+PR+ tumors and ER+PR− tumors among postmenopausal women.

  • 4

    Not summarized because of a substantial heterogeneity among RE of these studies.

ER+PR+1(2,643)0.80(0.70–0.92)0.511(5,469)1.82(1.55–2.14)<0.12
ER−PR−1(1,471)1.03(0.87–1.22)0.391(1,523)1.06(0.84–1.33)0.13
ER+PR−1(199)1.06(0.76–1.49)0.901(999)0.93(0.72–1.21)0.15
ER−PR+1(191)1.01(0.73–1.39)0.971(138)2.01(1.22–3.32)0.76
ER+11 studies11, 17–19, 22, 23, 25, 27, 28, 31, 340.86(0.77–0.95)0.4113 studies8, 12, 13, 15, 16, 18, 22, 23, 27, 29, 33, 35, 361.783(1.50–2.11)<0.00012
ER− No summary RESignificant413 studies8, 12, 13, 15, 16, 18, 22, 23, 27, 29, 33, 35, 361.19(1.03–1.36)0.47
PR+7 studies 11, 22, 23, 25, 27, 28, 340.83(0.74–0.92)0.6611 studies8, 12, 13, 15, 22, 23, 27, 29, 33, 35, 361.99(1.74–2.28)<0.062
PR−7 studies 11, 22, 23, 25, 27, 28, 341.03(0.90–1.18)0.7911 studies8, 12, 13, 15, 22, 23, 27, 29, 33, 35, 361.07(0.94–1.23)0.40

Dose–response meta-analysis of ER+PR+ tumors

Nine studies (3 cohort studies12, 14, 15 and 6 case-control studies22, 23, 27, 33–35) were included in the dose–response meta-analysis for the association between BMI and the risk of ER+PR+ tumors. The summary of RE for premenopausal women was estimated based on 4 studies22, 23, 27, 34 with 1,720 breast cancer patients and that for postmenopausal women based on 8 studies12, 14, 15, 22, 23, 27, 33, 35 with 4,267 breast cancer patients. In the dose–response meta-analysis, a 5-unit increase in BMI was associated with a significant decreased risk (−10%) in premenopausal women (95% CI = −18% to −1%) and a statistically significantly 33% increased risk in postmenopausal women (95% CI = 20–48%; Fig. 5). There was evidence of heterogeneity across premenopausal and postmenopausal status (Pheterogeneity< 0.0001). We also performed further restricted analyses among never PMH-user only based on 4 studies12, 15, 33, 35 which showed that each 5-unit increase in BMI was statistically significantly associated with a 40% increased risk among postmenopausal women (95% CI = 22–60%).

Figure 5.

Scatter plot of exposure-specific risk estimates and liner trend for ER+PR+ tumors among premenopausal and postmenopausal women based on the results of the dose-response meta-analyses. The area of each circle is proportional to the precision the risk estimate (inverse of its variance). The risk estimates (linear trend) for each 5-unit increment of body mass index (BMI, kg/m2) were RE = 0.90 (95% CI: 0.82–0.99) among premenopausal women based on 4 studies22, 23, 27, 34 and 1.33 (95% CI: 1.20–1.48) among postmenopausal women based on 8 studies12, 14, 15, 22, 23, 27, 33, 35Pheterogeneity of the summarized risk estimates across menopausal status <0.0001. All heterogeneity tests were calculated by Q statistics.

Discussion

This meta-analysis of prospective cohorts and case-control studies indicates that the association between body weight and breast cancer risk varies by the ER/PR status of the tumor and by menopausal status. Among premenopausal women, the summary estimates of the risk, when comparing the highest versus the lowest (or reference) categories, showed a statistically significant 20% lower risk for the development of ER+PR+ tumors only, but there was no association with the other tumor subtypes. Among postmenopausal women there was a statistically significant 82% increased risk for ER+PR+ tumors, but no association with ER+PR− nor ER−PR− subtypes. In dose–response meta-analysis of ER+PR+ tumor, each increment in the BMI by 5 kg/m2 was associated with 10% lower risk of premenopausal and 33% increased risk of postmenopausal breast cancer. For ER+PR+ tumors, the corresponding summary of RE among never PMH-user only, we found a statistically significant 40% increased risk per 5 kg/m2 increment in BMI. The magnitude of the positive association between BMI and overall postmenopausal breast cancer risk in previous corresponding studies was reported to be between 10%2 and 18%47 per 5 kg/m2 increment in BMI. Our results for ER+PR+ tumors, which account for ∼60% of all breast tumors,48 within ranges broadly consistent with these results.

Obesity has a complex relationship with breast cancer risk that seems to be modulated by menopausal status. Obesity has been shown to have a different effect on estrogen status among premenopausal and among postmenopausal women explained by the shift of a major source of endogenous estrogen between before (ovary) and after menopause (adipose tissue). It has been reported that more frequent anovulatory cycles among obese premenopausal women,49, 50 and faster clearance rate of free estrogen in the liver among obese than among lean women51 may lead to lower levels of both estrogen and progesterone.49 In contrast, excess adipose tissue after menopause may increase endogenous estrogen production from aromatization of androgen in peripheral fat tissue.52, 53 A higher ratio of bioactive estradiol to estrone (E2/E1) in tumor and a higher expression level of 17β-HSD type 1 mRNA, which is an enzyme that converts E1 to E2, among postmenopausal compared with premenopausal patients have been also suggested.54 This dual effect of obesity on hormonal status may at least partly explain the apparently dual association between body weight and risk of hormone-dependent (ER+PR+) breast cancer, inverse among premenopausal and positive among postmenopausal women. Among postmenopausal women the obesity-induced reduction in sex hormone-binding globulin production55 may lead to further increased levels of circulating free-estrogen inducing PR expression and stimulation of ER+PR+ tumors. A previous study supported this obesity-estrogen hypothesis by showing that a significant positive association between BMI and postmenopausal breast cancer disappeared after adjustment for serum estrogen level.47 For mix-menopausal status, the results for ER+PR+ tumors were obviously scattered between pre- and postmenopausal women and diluted. This suggests the great importance for stratification by the menopausal status in evaluating for the association between body weight and ER/PR defined breast cancer risk.

Results from this meta-analysis indicate that there is no overall association between high body weight and risk for ER+PR− tumors among postmenopausal women, suggesting different etiology from the other ER+ tumors. The contrasting REs across ER+PR+ and ER+PR− tumor among cohort studies7, 10, 12 is consistent with a recent cohort study.56 This could be partly explained through a positive biological relationship between ER-mediated estrogenic action and PR induction in breast tumor cell57 as previously discussed12 with linking excess postmenopausal body weight to induction of endogenous estrogen.52, 53

Indeed, there is increasing evidence that ER+PR− breast tumors may be a distinct subset of breast cancer with a higher frequency of HER-2 over expression and HER-1 epidermal growth factor receptor (EGFR) expression and more aggressive features than ER+PR+ tumors.58 It was recently reported that ER+PR− tumors as diagnosed by clinical assays were a mixture of 3 different subtypes according to RNA profiling. With more than 1,000 breast tumors, Creighton et al. defined the specific molecular signature for ER+PR+ tumors and ER−PR− tumors.59 Whereas, ER+ PR− tumors did not have single unique profile but had a mixture of 3 different profiles: tumors manifesting ER+PR+ signature, tumors manifesting ER−PR− signature and tumors not associating with ER+PR+ or ER−PR− tumors (considered as “true” ER+PR−). This“true” ER+PR− tumors seems to represent a distinct subset with aggressive features and poor outcome. The lack of obvious correlation between body weight and risk of clinically diagnosed ER+PR− tumors may be reflected by the heterogeneous molecular profile of this subtype.

Results from this meta-analysis for ER+PR− and ER−PR+ tumors among premenopausal women and for ER−PR+ tumors among postmenopausal women are difficult to interpret and should be treated with caution because of limited number of cases included in these summary analyses (<200). The wider CI for ER−PR+ tumors among postmenopausal women compared with premenopausal women, this could be supported by a higher percentage/rate of ER−PR+ tumors in the younger women (<40 yr) than the older.10, 60

We cannot exclude a potential involvement of ER independent stimulation in developing breast tumor caused by obesity. Among specific subgroup, such as the lymph node-positive,61 HER2/neu+,32 a possible link between obesity and ER-negative breast cancer was indicated. Not only tumor's ER/PR status, but further molecular classification of breast tumor should be considered in further epidemiological investigation.

As with any meta-analysis of observational studies, our study has limitations. First, several studies included in this meta-analysis were based on self-reported anthropometric measures. The cutoff points for BMI were also different between studies. These factors may lead to exposure misclassification that may have led to dilution of the true associations.

Second, the assay methods for the measurement of receptor status and cutoff point for receptor status varied across/within studies. This may have led to misclassification of the tumor subtypes and in consequence to attenuation of the observed REs. However, most studies in this meta-analysis had been conducted after 1990 thus improvement of the receptor assessment over time could be expected. Moreover, the majority of the studies used the dextran-coated charcoal assay or immunohistochemistry assay and a high concordance (>80%) between these 2 methods has been reported.62 Furthermore, possible alteration of ER expression in breast during menstrual cycle63 might make it difficult to interpret the results among premenopausal women.

Third, the definition of menopausal status varied between studies; it was defined at time of breast cancer diagnosis in case-control studies and at time of inclusion in the cohort in prospective studies. In cohort studies, this could lead to misclassification of postmenopausal cancer (at diagnosis) into premenopausal (at baseline) category.

Fourth, we could not exclude the possibility that our observed associations were due to unknown or residual confounding. The most likely confounding factor for the relation of body weight to postmenopausal breast cancer risk is PMH use, which is positively associated with breast cancer risk64, 65 and a higher prevalence of PMH use is more likely to be observed among lean women than overweight women. However, in the sensitivity analysis restricted to studies adjusting for PMH use among postmenopausal women, a positive association between body weight and ER+PR+ tumors persisted.

Fifth, a statistical power in the analyses of the tumors with discordant receptor status (ER+PR− in pre and ER−PR+ in pre- and postmenopausal women) was low (<200 cases), which limited interpretability of these results. However, biologically, the proportion of these discordant subgroups should be smaller than that of other tumor subtype. The wider CIs could indirectly support a relatively high validity of ER/PR assessment among studies in current meta-analysis.

Finally, in the analyses of published studies it is possible that the summary RE is the result of publication bias, because publication of null papers may be limited. However, there was no evidence of publication bias in the present results, except for ER−PR− tumors among mix-menopausal women and postmenopausal women.

In summary, results from these meta-analyses show that the relation of body weight to breast cancer risk depends on tumor's receptor status as well as menopausal status. Among premenopausal women, excessive BMI was weakly inversely and among postmenopausal women positively associated with ER+PR+ tumor. In contrast, there was no overall association with ER−PR− tumors in pre- or postmenopausal women, neither with ER+PR− tumors in postmenopausal women. Small numbers of cases in other subgroups with discordant ER/PR status make these results difficult to interpret. From a public health point of view, body weight control should be considered as one of the most effective strategies for preventing postmenopausal breast cancer because ER+PR+ subtype accounts for a major proportion of these tumors.

Acknowledgements

We thank Dr. Shoichiro Tugane, Dr. Motoki Iwasaki, The Epidemiology and Prevention Division, Research Center for Cancer Prevention and Screening, National Cancer Center, Tokyo, Japan.

Ancillary