Pooled Analysis of Prospective Cohort Studies on Height, Weight, and Breast Cancer Risk

14 Pages·2000·0.13 MB·English

American Journal of Epidemiology Vol.152, No.6
Copyright ©2000 by The Johns Hopkins University School of Hygiene and Public Health Printed in U.S.A.
All rights reserved
Anthropometry and Breast Cancer Risk van den Brandt et al.
Pooled Analysis of Prospective Cohort Studies on Height,Weight, and Breast
Cancer Risk
Piet A.van den Brandt,1Donna Spiegelman,2,3Shiaw-Shyuan Yaun,4Hans-Olov Adami,5Lawrence Beeson,6
Aaron R.Folsom,7Gary Fraser,6R.Alexandra Goldbohm,8Saxon Graham,9Larry Kushi,7James R.Marshall,10
Anthony B.Miller,11Tom Rohan,11Stephanie A.Smith-Warner,12Frank E.Speizer,4,13Walter C.Willett,2,4,12,14Alicja
Wolk,5and David J.Hunter2,4,14
D
o
w
n
The association between anthropometric indices and the risk of breast cancer was analyzed using pooled data loa
d
from seven prospective cohort studies.Together, these cohorts comprise 337,819 women and 4,385 incident e
d
invasive breast cancer cases.In multivariate analyses controlling for reproductive, dietary, and other risk factors, fro
the pooled relative risk (RR) of breast cancer per height increment of 5 cm was 1.02 (95% confidence interval (CI): m
h
0.96, 1.10) in premenopausal women and 1.07 (95% CI:1.03, 1.12) in postmenopausal women.Body mass index ttp
(BMI) showed significant inverse and positive associations with breast cancer among pre- and postmenopausal s
women, respectively;these associations were nonlinear.Compared with premenopausal women with a BMI of less ://a
c
a
than 21 kg/m2, women with a BMI exceeding 31 kg/m2had an RR of 0.54 (95% CI:0.34, 0.85).In postmenopausal d
e
women, the RRs did not increase further when BMI exceeded 28 kg/m2;the RR for these women was 1.26 (95% m
ic
CI:1.09, 1.46).The authors found little evidence for interaction with other breast cancer risk factors.Their data .o
u
indicate that height is an independent risk factor for postmenopausal breast cancer;in premenopausal women, this p
.c
relation is less clear.The association between BMI and breast cancer varies by menopausal status.Weight control o
m
may reduce the risk among postmenopausal women.Am J Epidemiol 2000;152:514–27. /a
je
/a
body height;body weight;breast neoplasms;prospective studies rtic
le
-a
b
s
The relation between body size and breast cancer risk has studies, but only some cohort studies, showed a clear posi- tra
c
been the subject of numerous investigations.Many of these tive association with postmenopausal breast cancer risk and t/1
5
studies have been focused on the association between that the relative risks found in cohort studies were much 2
/6
weight (typically corrected for height) and breast cancer. closer to the null value than were those from case-control /5
1
Attention to the relation with height has increased in recent studies. On the other hand, inverse associations with BMI 4
/7
years due to increased interest in the effects of early diet on were found for premenopausal breast cancer in most cohort 5
8
breast cancer risk in later life (1); adult height may serve as studies, whereas case-control studies showed both inverse 07
an indicator of childhood or adolescent nutrition and energy and direct associations (3, 4). For adult height, a modest by
balance (2). Studies have revealed that associations with positive association with breast cancer risk was found in gu
e
body size may vary by menopausal status. Hunter and most, but not all, case-control studies. These positive asso- s
Willett (3) concluded that for body mass index (BMI, ciations were predominantly found for postmenopausal t on
defined as weight (kg)/height2(m2)), most large case-control breast cancer. Cohort studies have shown positive associa- 10
A
p
ril 2
Received for publication June 7, 1999, and accepted for publica- Minnesota, Minneapolis, MN. 01
tion November 17, 1999. 8Department of Epidemiology, TNO Nutrition and Food Research 9
Abbreviations: BMI, body mass index; HRT, hormone replace- Institute, Zeist, the Netherlands.
ment therapy;IGF, insulin-like growth factor;RR, relative risk. 9Department of Social and Preventive Medicine, State University
1Department of Epidemiology, Maastricht University, Maastricht, of New York at Buffalo, Buffalo, NY.
the Netherlands. 10Arizona Cancer Center, University of Arizona, Tucson, AZ.
2Department of Epidemiology, Harvard School of Public Health, 11Department of Public Health Sciences, Faculty of Medicine,
Boston, MA. University of Toronto, Toronto, Ontario, Canada.
3Department of Biostatistics, Harvard School of Public Health, 12Department of Nutrition, Harvard School of Public Health,
Boston, MA. Boston, MA.
4Channing Laboratory, Department of Medicine, Brigham and 13Department of Environmental Health, Harvard School of Public
Women’s Hospital and Harvard Medical School, Boston, MA. Health, Boston, MA.
5Department of Medical Epidemiology, Karolinska Institute, 14Harvard Center for Cancer Prevention, Boston, MA.
Stockholm, Sweden. Reprint requests to Prof.Piet A.van den Brandt, Department of
6Center for Health Research, Loma Linda University School of Epidemiology, Maastricht University, P. O. Box 616, 6200 MD
Medicine, Loma Linda, CA. Maastricht, the Netherlands (e-mail: [email protected]
7Division of Epidemiology, School of Public Health, University of maas.nl).
514
Anthropometry and Breast Cancer Risk 515
tions with height in both pre- and postmenopausal women sponds to the period 1986–1991. Self-administered ques-
(3), but several used databases in which information on tionnaires were used to assess anthropometric factors, repro-
height, but not reproductive and other risk factors, was ductive factors, medical history, and family history. Each
available, leaving open the possibility of uncontrolled con- study measured dietary intake by using food frequency
founding. questionnaires. Incident breast cancers were ascertained by
Against this background, we studied the relation between using follow-up questionnaires and inspection of medical
height, weight, and BMI and breast cancer risk in the records and/or tumor registry linkage. In all cohorts, follow-
Pooling Project of Diet and Cancer (5). In this project, pri- up was estimated to be more than 90 percent complete.
mary data from seven major prospective studies have been The questionnaires yielded self-reported information on
combined to evaluate associations between dietary factors height and weight at baseline. Weight at ages other than
and breast cancer risk by using a standardized approach. baseline age was not available for most cohorts; therefore,
Since these studies also provided information on anthro- these variables and weight change could not be taken into
pometry and many covariates that can act as confounders or account in this analysis. All height and weight data were
D
effect modifiers (e.g., reproductive factors, family history of converted into meters and kilograms, respectively. BMI ow
cancer, and exogenous hormone use), we decided to use the (Quetelet index) was used as an indicator of relative weight. nlo
pooled data for the current analysis of anthropometry and The correlation coefficients between height and BMI varied ad
e
breast cancer as well. The pooling approach enables use of from –0.05 in the Nurses’ Health Study to –0.13 in the d
uniform categories of anthropometric variables and covari- Netherlands Cohort Study and the New York State Cohort. fro
m
ates and evaluation of potential effect modification by Participants were excluded from these analyses if they h
covariates. met study-specific exclusion criteria, if their reported energy ttp
s
intake was greater than three standard deviations from the ://a
c
study-specific log-transformed mean energy intake of the a
MATERIALS AND METHODS e d
baseline population, if they reported a history of cancer e
m
The Pooling Project has been described previously (5). other than nonmelanoma skin cancer at baseline, or if they ic.o
Briefly, seven prospective studies (6–12) (table 1) were had missing data on anthropometry. As a result of these up
identified in 1992 that met the following predefined criteria: additional exclusions and expanded follow-up in some stud- .co
the study 1) had at least 200 incident cases of breast cancer ies, the baseline cohort size and the number of cases m/a
available for analysis; 2) assessed long-term intake of foods reported in these analyses may differ from original study- je
acnodm pnluettreiden at sv, ailnidcalutidoinn gs tuendeyr goyf tihnet adkiee,t aast sbeasssmeliennet; manetdh o3d) sreppeocirftiecd pPuobolliicnagti oPnrso j(e7c,t 1p0u,b l1ic3a–t1io5n) s a(n5d, 1fr6o)m. T phree vpioooulselyd /article
or a closely related instrument. For Pooling Project analyses analysis of anthropometry and breast cancer is based on -ab
s
(5), the Nurses’Health Study was split into two studies since 4,385 incident cases of invasive breast cancer (table 1). tra
it had repeated assessments of dietary intake and a longer Risk factors, including anthropometric factors, may differ ct/1
follow-up period than did the other studies. To make the for premenopausal and postmenopausal breast cancer (3). 5
2
length of follow-up more comparable with the other cohorts, However, most of the cohort studies included had informa- /6
/5
Nurses’ Health Study (a) corresponds to the 1980–1986 tion on menopausal status at baseline only. As has been 1
4
follow-up period, while Nurses’ Health Study (b) corre- described before (16), to assign changing menopause status /7
5
8
0
7
b
y
g
u
TABLE 1. Characteristics of the cohort studies included in the pooled analysis of anthropometry and e
s
breast cancer,the Pooling Project of Diet and Cancer t o
n
1
Baseline Age at baseline (years) Years No.of cases∗ 0 A
Study cohort of Pre- Post- p
size Median Range follow-up Total menopausal menopausal ril 2
0
AHS† 15,172 52 28–90 1976–1982 122 20 87 1
9
CBSS† 56,837 50 40–59 1982–1987 392 122 242
IWHS† 34,406 61 55–69 1986–1991 643 0 643
NLCS† 62,412 61 55–69 1986–1989 420 0 420
NYSC† 18,475 68 50–93 1980–1987 358 0 358
NHS(a)† 89,046 47 34–59 1980–1986 1,015 383 571
NHS(b)† 68,817 53 40–65 1986–1991 787 130 613
SMC† 61,471 52 40–76 1987–1993 648 68 274
Total 337,819 4,385 723 3,208
*Cases consisted of women diagnosed with invasive breast cancer, with complete data on height and weight.
A total of 454 cases were considered to have uncertain menopausal status at diagnosis (see text for details).
† AHS, Adventist Health Study; CBSS, Canadian National Breast Screening Study; IWHS, Iowa Women’s
Health Study;NLCS, Netherlands Cohort Study;NYSC, New York State Cohort;NHS(a), Nurses’Health Study
(a);NHS(b), Nurses’Health Study (b);SMC, Sweden Mammography Cohort.
Am J Epidemiol Vol.152, No.6, 2000
516 van den Brandt et al.
during follow-up, an algorithm was developed based on an covariate data. Two-sided 95 percent confidence intervals
analysis of 42,531 Nurses’ Health Study participants who were calculated. The “random effects” model developed by
were followed up biennially, were premenopausal in 1976, DerSimonian and Laird (20) was used to combine log rela-
e
and remained premenopausal or had a natural menopause by tive risks from the multiple studies.
1992. On the basis of the algorithm, 723 (16.5 percent) inci- In these analyses, anthropometric variables were entered
dent cases with complete anthropometry data were pre- as categorical variables, and in addition, models were run
menopausal, 3,208 (73.2 percent) cases were post- with continuous variables. Analyses were conducted for the
menopausal, and 454 (10.3 percent) cases were of uncertain total group and in pre- and postmenopausal cases only.
menopausal status. Because there were only 20 pre- Potential effect modification by other covariates was also
menopausal cases in the Adventist Health Study, leading to evaluated by examining the associations between anthro-
very unstable effect estimates, this study was excluded from pometry and breast cancer within each category of the
the pooled analysis among premenopausal cases. covariate and statistical testing of the pooled linear interac-
tion term. Pooled pvalues for interaction were based on the D
squared pooled estimate of the continuous interaction term ow
Statistical analysis n
divided by its estimated variance. We used restricted cubic lo
a
Each study was analyzed by using a method consistent splines as a graphic method of presenting dose-response d
e
wHietahl tihts S sttuuddyy, Idoewsiag Wn. oFmiveen ’osf Htheea lctho hSoturtd syt,u Ndieews (YAodrkv eSnttaitset cthuer vceusr vthe.a tT mheaskee n noon paa prarmioerit raicss ruemgrpetsisoinosn acbuoruvte sth cea nsh aalpsoe boef d from
Cohort, Nurses’Health Study (a) and (b), and the Sweden compared with the linear model to evaluate linearity of the h
Mammography Cohort) were analyzed as nested case- associations by using a likelihood ratio test, as described pre- ttps
control studies with a 1:10 ratio of cases diagnosed with viously (16). For these analyses, the studies were combined ://a
c
invasive breast cancer to controls free of diagnosed cancer. into a single data set stratified by study, since there was no ad
Controls were matched to the cases on the basis of the year between-study heterogeneity in the effects of the other model em
of diagnosis and the year of birth of the cases. In addition, covariates (21) and because height and weight were mea- ic.o
controls did not have a history of breast cancer prior to the sured similarly in all studies. All pvalues are two-sided. up
year of diagnosis of the cases. Controls were sampled with- .co
m
out replacement within each year, but were eligible to be RESULTS /a
chosen again as controls or to be classified as cases in sub- je
/a
sequent years. We used a nested case-control approach Table 1 shows the characteristics of the cohort studies and rtic
because of computational limitations. Anested case-control the number of incident breast cancer cases with complete le
design also was used for the Canadian National Breast anthropometry data that are available for this pooled analy- -ab
s
Screening Study; the investigators of that study selected two sis. The results of the pooled multivariate analyses using tra
controls matched to each case on the basis of age (±2 categories of anthropometric factors are presented in table 2. ct/1
months) and then processed previously administered ques- Height was positively associated with breast cancer risk in 52
tionnaires for the cases and matched controls. The both pre- and postmenopausal women; the relative risks /6/5
Netherlands Cohort Study used a case-cohort design (17); (RRs) for women 1.75 m or taller compared with women 14
cases were identified within the total cohort and compared shorter than 1.60 m were 1.42 for premenopausal women, /75
8
with the person-time of a subcohort of 1,812 women ran- 1.28 for postmenopausal women, and 1.22 for all women. 0
7
domly sampled at baseline. The tests for linear trend were significant in postmenopausal b
y
For the nested case-control studies, relative risks associ- women and in the total group. g
u
ated with height, weight, and BMI were estimated by condi- The association of weight with breast cancer depended on es
tional logistic regression with the use of SAS PROC menopausal status. There was an inverse association in pre- t o
n
PHREG (18). These models were stratified by year of birth menopausal women, concentrated in the top weight cate- 1
0
and calendar year (in single years) except for the models gories. For postmenopausal women, a positive association A
p
using the Canadian National Breast Screening Study, which was found, with statistically significant relative risks in all ril 2
were stratified by year of birth and additional study-specific categories. The RRs for women who weighed 80 kg or more 0
1
matching factors (8). For the Netherlands Cohort Study, compared with those who weighed less than 60 kg were 0.58 9
EPICURE software (19) was used to estimate the relative for premenopausal and 1.25 for postmenopausal women.
risk; models were stratified on age at baseline. Multivariate The tests for trend were significant in both groups. BMI
relative risks were adjusted for menopausal status at diag- showed a significant inverse association with breast cancer
nosis; age at menarche; parity; age at birth of the first child; risk among premenopausal women (pvalue, test for trend (cid:2)
postmenopausal hormone use; oral contraceptive use; his- 0.007) but the inverse association was concentrated in
tory of benign breast disease; maternal history of breast can- women in the highest BMI categories. When BMI of 31
cer; history of breast cancer in a sister; smoking status; edu- kg/m2 or more was used as the highest BMI category, the
cation; BMI or height (depending on the analysis); and RR was 0.54 (95 percent CI: 0.34, 0.85) (data not shown).
intake of fat, fiber, energy, and alcohol (see table 2 for the On the other hand, the association with BMI among post-
definition of the categories used in the analysis). An indica- menopausal women was significantly positive (pvalue, test
tor variable for missing responses was created for each for trend (cid:2) 0.001), with an increased risk seen in all BMI
covariate to minimize loss of observations due to missing categories. The RRs for women with a BMI of 33 kg/m2or
Am J Epidemiol Vol.152, No.6, 2000
Anthropometry and Breast Cancer Risk 517
TABLE 2. Pooled multivariate relative risks and 95 percent confidence intervals for breast cancer according to categories of
height,weight,and body mass index,the Pooling Project of Diet and Cancer
Premenopausal† Postmenopausal Total
Anthropometric
variable No.of RR‡ 95% CI§ No.of RR‡ 95% CI No.of RR¶ 95% CI
cases cases cases
Height (m)
<1.60 149 1.0 Reference 724 1.0 Reference 970 1.0 Reference
1.60–<1.65 202 1.21 0.94, 1.55 921 1.09 0.98, 1.21 1,261 1.10 1.00, 1.20
1.65–<1.70 196 1.06 0.82, 1.36 916 1.23 1.11, 1.37 1,253 1.20 1.09, 1.32
1.70–<1.75 117 1.14 0.86, 1.52 491 1.24 1.09, 1.41 688 1.24 1.11, 1.38
≥1.75 39 1.42 0.95, 2.12 156 1.28 0.94, 1.76 213 1.22* 0.90, 1.65
pvalue, test for trend 0.41 <0.001 0.001
D
o
Weight (kg) w
n
<60 300 1.0# Reference 812 1.0# Reference lo
a
60–<65 132 0.85 0.66, 1.09 608 1.16 1.03, 1.31 d
e
65–<70 100 1.09 0.83, 1.45 549 1.22 1.01, 1.47 d
70–<75 83 1.12 0.83, 1.52 464 1.22 1.01, 1.47 fro
m
75–<80 35 0.95 0.51, 1.76 289 1.43 1.14, 1.80 h
≥80 53 0.58 0.40, 0.83 486 1.25 1.02, 1.52 ttp
s
pvalue, test for trend 0.02 0.003 ://a
c
a
BMI (kg/m2) d
e
<21 158 1.0 Reference 363 1.0 Reference m
ic
21–<23 223 1.24 0.97, 1.57 632 1.14 0.99, 1.33 .o
23–<25 131 1.03 0.78, 1.35 699 1.15 1.00, 1.34 up
25–<27 82 1.08 0.79, 1.48 564 1.26 1.09, 1.47 .co
m
27–<29 47 0.97 0.66, 1.44 401 1.43 1.21, 1.67
/a
29–<31 32 0.96 0.60, 1.52 224 1.21 1.01, 1.46 je
3≥313–<33 1200 00..5558 00..2364,, 11..1050 114805 11..2297 11..0033,, 11..6505 /artic
le
pvalue, test for trend 0.007 0.001 -a
b
s
*Test for heterogeneity between studies, p< 0.05. trac
† The Adventist Health Study was not included in analyses of premenopausal women. t/1
‡ Multivariate relative risks (RR) were adjusted for age at menarche (≤11, 12, 13, 14, ≥15 years), parity (0, 1–2, ≥3), age at birth of first 52
child (≤20, 21–25, 26–30, >30 years), postmenopausal hormone use (ever, never), oral contraceptive use (ever, never), history of benign /6
/5
breast disease (no, yes), maternal history of breast cancer (no, yes), history of breast cancer in a sister (no, yes, no sisters), smoking status 1
4
(ever, never), education (less than high school graduation, high-school graduation, more than high-school graduation), fat intake (quintiles), /7
fiber intake (quintiles), energy intake (continuous), and alcohol intake (0, >0–<1.5, 1.5–<5, 5–<15, 15–<30, ≥30 g/day). 5
8
§ CI, confidence interval. 0
7
¶ The model included all of the above terms, and menopausal status at diagnosis (premenopausal, postmenopausal, uncertain). b
y
# Further adjusted for height. g
u
e
s
t o
n
more compared with women with a BMI of less than 21 percent CI: 1.02, 1.11) per height increment of 5 cm (data 10
kg/m2 was 0.58 (95 percent CI: 0.34, 1.00) for pre- not shown). A
p
menopausal women and 1.27 (95 percent CI: 1.03, 1.55) for Both continuous weight (with height in the model) and ril 2
postmenopausal women. BMI showed significantly inverse associations with breast 0
1
Table 3 shows study-specific and pooled relative risk cancer risk in premenopausal women in each study, without 9
estimates for multivariate models with continuous anthro- statistical evidence of heterogeneity. The pooled RR per
pometric data. The pooled RR per height increment of 5 cm weight increment of 10 kg was 0.90, and for a BMI incre-
was 1.02 for premenopausal women, but the estimates from ment of 4 kg/m2, it was 0.89 (table 3). In postmenopausal
the four studies for which data on premenopausal women women, continuous BMI was positively associated with
were available varied: Only Nurses’ Health Study (a) breast cancer in all studies except the Netherlands Cohort
showed a positive association (pvalue, test for heterogene- Study and Nurses’ Health Study (a), in which there was
ity (cid:2) 0.74). For postmenopausal women, all studies essentially no association. For weight itself, a similar pattern
showed a positive association except the Canadian was seen. There was no statistical evidence of heterogeneity
National Breast Screening Study and the New York State between the studies. The pooled RR was 1.07 per 4-kg/m2
Cohort, in which no association was found. The pooled RR increment in BMI and 1.06 per 10-kg increment in weight.
was 1.07 (pvalue, test for heterogeneity (cid:2)0.17) (table 3). The nonparametric regression curve (figure 1) indicated a
For the total group of women, the pooled RR was 1.07 (95 linear association with height; the test for deviation from
Am J Epidemiol Vol.152, No.6, 2000
5
1
8
v
a
n
d
e
n
B D
ra ow
n n
d lo
TABLE 3. Study-specific and pooled multivariate relative risks*of breast cancer for continuous linear effects of height,weight,and body mass index,the Pooling t e ad
Project of Diet and Cancer t a ed
l. fro
m
Anthropometric variable 9A5RH%RS C†I 9C5BR%SR SC†I 9IW5R%HR SC†I 9N5LR%CR SC†I N95YR%SR CC†I N9H5R%SR( aC)I† N9H5R%SR( bC)I† 9S5MR%RC C†I Po9o5l%ed C RI†R† bpehvtewateleureeon,g etsentsuetd itfiyoers https://a
Height (5-cm increment) ca
d
Premenopausal —‡ 0.98 — — — 1.05 0.97 0.98 1.02 0.74 e
m
0.75, 1.28 0.97, 1.15 0.83, 1.12 0.78, 1.23 0.96, 1.10 ic
Postmenopausal 1.10 0.99 1.10 1.20 1.00 1.04 1.07 1.09 1.07 0.17 .o
u
0.91, 1.32 0.85, 1.16 1.03, 1.17 1.09, 1.32 0.92, 1.09 0.97, 1.12 1.00, 1.15 0.97, 1.23 1.03, 1.12 p
.c
o
m
Weight§ (10-kg increment) /a
Premenopausal —‡ 0.86 — — — 0.87 0.95 0.93 0.90 0.80 je
/a
Postmenopausal 1.18 0.610.,1 11.22 1.11 1.03 1.04 0.719.,0 01.97 0.812.,0 13.11 0.616.,1 15.30 0.813.,0 06.97 0.36 rticle
0.97, 1.44 0.91, 1.36 1.04, 1.18 0.91, 1.16 0.95, 1.15 0.93, 1.09 0.97, 1.11 1.02, 1.30 1.03, 1.10 -a
b
s
Body mass index (4 kg/m2 trac
increment) t/1
5
Premenopausal —‡ 0.86 — — — 0.86 0.95 0.90 0.89 0.82 2
/6
0.60, 1.25 0.77, 0.96 0.80, 1.12 0.62, 1.32 0.81, 0.97 /5
Postmenopausal 1.21 1.11 1.11 0.99 1.05 1.01 1.04 1.15 1.07 0.34 14
0.98, 1.50 0.91, 1.36 1.04, 1.19 0.86, 1.13 0.95, 1.16 0.93, 1.10 0.97, 1.12 1.01, 1.32 1.02, 1.11 /75
8
*Relative risks were adjusted for age at menarche (≤11, 12, 13, 14, ≥15 years), parity (0, 1, 2, ≥3), age at birth of the first child (≤20, 21–25, 26, 30, ≥30 years), postmenopausal 07
Am hormone use (ever, never), oral contraceptive use (ever, never), history of benign breast disease (no, yes), maternal history of breast cancer (no, yes), history of breast cancer in a sis- by
J E tinetra (kneo ,( qyueisn,t inleos s),i setenresr)g, ys minotakkineg ( csotanttuinsu (oeuvse)r,, annedve arl)c, oehdoulc ianttiaokne ( l(e0s, s> t0h–a<n1 h.5ig, h1 .s5c–h<o5o, l 5g–ra<d1u5a, t1io5n–,< h3i0g,h ≥ s3c0h ogo/dl agyra).duation, more than high school graduation), fat intake (quintiles), fiber gue
pid † AHS, Adventist Health Study;CBSS, Canadian National Breast Screening Study;IWHS, Iowa Women’s Health Study;NLCS, Netherlands Cohort Study;NYSC, New York State st o
e Cohort;NHS(a), Nurses’Health Study (a);NHS(b), Nurses’Health Study (b);SMC, Sweden Mammography Cohort;RR, relative risk;CI, confidence interval. n
miol ठTFhuert hAedrv aednjtuisstt eHde afoltrh h Seitguhdty. was not included in analyses of premenopausal women. 10 Ap
Vo ril 2
l. 01
1 9
5
2
, N
o
.
6
, 2
0
0
0
Anthropometry and Breast Cancer Risk 519
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
.c
FIGURE 1. Nonparametric regression curve for the relation between height and breast cancer, the Pooling Project of Diet and Cancer. o
m
/a
je
/a
rtic
linearity was also not significant (p(cid:2)0.65). The results for tions with height and BMI were evaluated according to cat- le-a
BMI are shown in figure 2 for pre- and postmenopausal egories of age at diagnosis, the association with height and bs
women separately. Although the test for deviation from lin- BMI appeared to be somewhat stronger in older, post- tra
c
earity in premenopausal women was technically not signifi- menopausal women. In premenopausal women, the associa- t/1
cant (p(cid:2)0.06), power was limited due to small numbers in tion with BMI was more inverse in older premenopausal 52
/6
this group. Among postmenopausal women, there was sig- women. However, the interactions with age at diagnosis /5
nificant deviation from linearity (p (cid:2) 0.004). The curve were not significant in either pre- or postmenopausal 14
/7
among postmenopausal women suggests that the relative women (table 4). 5
8
risk did not increase further above BMI of more than 28 Table 5 shows results of interaction analyses for other 07
kg/m2. When BMI more than 28 kg/m2was used as the high- covariates. No significant interactions with height or BMI by
est category in the categorical analyses, the RR was 1.26 (95 on breast cancer risk were found except for the interaction gu
e
percent CI: 1.09, 1.46) (data not shown). between a maternal history of breast cancer and height. s
The results of the analyses on interaction are given in Women with a maternal history of breast cancer had a t on
tables 4 and 5. In these analyses, height and BMI were pooled RR of 1.23 per height increment of 5 cm, whereas 10
entered as continuous variables into the models, and relative women without such a history had a pooled RR of 1.07. This Ap
risks for these variables were estimated in each category of interaction was noted in every cohort except the Sweden ril 2
the covariates. The analysis by menopausal status (table 4) Mammography Cohort. However, the association with 01
9
showed no evidence for interaction between menopausal height was opposite (although not significant) in those with
status and height (p value, test for interaction (cid:2) 0.13), but a history of breast cancer in a sister, which was consistent
there was significant interaction with BMI (pvalue, test for across the cohorts. The association of BMI and breast can-
interaction (cid:2) 0.004). Therefore, in the ensuing analyses, cer among premenopausal women also showed variation
interactions with height were evaluated in the total group, according to the history of breast cancer in a sister. Although
whereas interactions with BMI were evaluated in pre- and the same pattern occurred in every cohort, this interaction
postmenopausal women separately. Before further evalua- was not significant. No interactions were seen between
tion of interactions with other covariates, the mutual inter- overall family history of breast cancer and height or BMI
action between height and BMI was considered. As table 4 (data not shown). For height, a borderline significant inter-
shows, the effect of height was not significantly different action was seen with a history of benign breast disease,
between BMI categories, nor was the effect of BMI signifi- which was apparent in every cohort except the Sweden
cantly different between height categories in the pre- and Mammography Cohort. Although the interaction was not
postmenopausal women. When the breast cancer associa- significant, it is of interest, given the possible role of estro-
Am J Epidemiol Vol.152, No.6, 2000
520 van den Brandt et al.
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
FIGURE 2. Nonparametric regression curve for the relation between body mass index and breast cancer, the Pooling Project of Diet and .c
o
Cancer. m
/a
je
/a
rtic
gens in the etiology of breast cancer, that the association The studies we used are a subset of prospective studies on le-a
between BMI and breast cancer was stronger and was sig- anthropometric factors and breast cancer risk, but represent bs
nificant only among women who never used post- all studies with data on food and nutrient intakes that could tra
c
menopausal hormone replacement therapy (HRT). There potentially confound the anthropometric associations. The t/1
5
was also a tendency for the association between BMI and Pooling Project includes cohort studies with detailed dietary 2
/6
breast cancer in postmenopausal women to decrease with data and other relevant breast cancer risk factors. Because, in /5
1
increasing alcohol intake. Analyses with categorical BMI the current analysis of anthropometric indices and breast can- 4
(<21, 21–<25, 25–<28, and ≥28 kg/m2) in the model cer, we wanted to be able to adjust for possible confounding /75
8
revealed the same pattern of interaction with age at diagno- by both nondietary and dietary risk factors, we decided to use 07
sis, HRTuse, and alcohol as was observed with continuous the originally collected pooled data set. For example, alcohol by
BMI in postmenopausal women (data not shown). intake is a risk factor for breast cancer (16), which also may gu
e
be associated with BMI. Our pooling results regarding height s
DISCUSSION are in agreement with results of major cohort studies not t on
included that lack reproductive or dietary data (22–24). For 10
In this pooled analysis of seven major prospective cohort BMI, the results among premenopausal women are also com- A
p
studies on risk factors for breast cancer, we found a signifi- parable (22, 24), but less so in postmenopausal women (24). ril 2
cant positive association between height and the risk of The availability of information on many potential con- 0
1
postmenopausal breast cancer. In premenopausal women, founders is a major strength of our pooled analysis to rule out 9
the association was less clear and was not significant, and confounding bias, and it also provides an opportunity to
the interaction by menopausal status was not significant. examine potential effect modification. Adisadvantage is that
Weight and BMI showed significant inverse associations our analysis of premenopausal breast cancer was based on
with risk of premenopausal breast cancer and significant only four cohorts. Although this comprised over 700 pre-
positive associations with postmenopausal breast cancer. menopausal cases, the estimates were clearly less precise
This interaction by menopausal status was highly signifi- than were those for postmenopausal women.
cant. There was no statistically significant heterogeneity Another potential drawback of this analysis is the use of
among the studies for these associations. The associations of self-reported data on height and weight in our cohort stud-
these anthropometric factors were independent of other risk ies. Although some attenuation of relative risk estimates
factors for breast cancer and were not significantly modified might have occurred because of imperfect measurements,
by most of these factors, except for a possible interaction the potential for bias is small, considering that validation
between height and maternal history of breast cancer. studies of self-reported measurements show high correla-
Am J Epidemiol Vol.152, No.6, 2000
Anthropometry and Breast Cancer Risk 521
TABLE 4. Pooled multivariate relative risks*(95% confidence intervals) for a 5-cm increment in height and a 4-kg/m2increment
in body mass index by menopausal status,anthropometric factors,and age at diagnosis†,the Pooling Project of Diet and Cancer
Covariate and pvalue,
anthropometric Category of covariate test for
variable interaction
Menopausal status‡ Premenopausal Postmenopausal
Height 1.02 (0.96, 1.10) 1.07 (1.03, 1.12) 0.13
BMI§ 0.89 (0.81, 0.97) 1.07 (1.02, 1.11) 0.004
BMI (kg/m2) <21 21–<23 23–<25 25–<29 ≥29
Height 1.10 (1.02, 1.17) 1.08 (0.98, 1.19) 1.09 (1.03, 1.15) 1.08 (1.03, 1.14) 1.02 (0.93, 1.12) 0.12
Height (m) <1.60 1.60–<1.65 1.65–<1.70 1.70–<1.75 ≥1.75 D
o
BMI w
n
Premenopausal 0.85 (0.62, 1.17) 0.92 (0.79, 1.07) 0.94 (0.81, 1.09) 0.87 (0.70, 1.08) 0.93 (0.64, 1.36) 0.40 lo
a
Postmenopausal 1.10 (1.03, 1.18) 1.10 (1.02, 1.18) 1.06 (0.99, 1.14) 1.02 (0.92, 1.12) 1.02 (0.81, 1.29) 0.30 d
e
d
Age at diagnosis (years)‡ fro
in premenopausal m
women <45 45–<50 >50 http
HBMeiIght 01..9084 ((00..8704,, 11..2406)) 10..0965 ((00..9854,, 11..1078)) 10..0713 ((00..8554,, 10..2918)) 00..7008 s://a
c
a
d
Age at diagnosis (years) in e
m
postmenopausal ic
women <55 55–<60 60–<65 65–<70 ≥70 .o
u
p
Height 1.12 (1.00, 1.26) 1.06 (0.95, 1.18) 1.04 (0.94, 1.16) 1.14 (1.06, 1.23) 1.15 (1.02, 1.30) 0.24 .c
o
BMI 0.97 (0.86, 1.10) 1.04 (0.94, 1.16) 1.06 (0.95, 1.17) 1.15 (1.07, 1.24) 1.09 (0.99, 1.20) 0.13 m
*Relative risks were adjusted for age at menarche (≤11, 12, 13, 14, ≥15 years), parity (0, 1, 2, ≥3), age at birth of the first child (≤20, /aje
21–25, 26–30, ≥30 years), postmenopausal hormone use (ever, never), oral contraceptive use (ever, never), history of benign breast disease /a
(no, yes), maternal history of breast cancer (no, yes), history of breast cancer in a sister (no, yes, no sisters), smoking status (ever, never), rtic
education (less than high school graduation, high school graduation, more than high school graduation), fat intake (quintiles), fiber intake le
(quintiles), energy intake (continuous), and alcohol intake (0, >0–<1.5, 1.5–<5, 5–<15, 15–<30, ≥30 g/day). -ab
†‡ TThhee IAodwvae nWtiostm Heena’sl thH eSatulthd yS wtuadsy ,n tohte i nNceluthdeerdla innd asl l Caonhaolyrst eSst uodf yp, raenmde tnhoep Naueswa Yl owrokm Setant.e Cohort were not included in this analysis. strac
§ BMI, body mass index. t/1
5
2
/6
/5
1
4
/7
5
8
tions (r> 0.8) with measured anthropometric data (25–27), cant positive association among young premenopausal 07
and all data here were reported prior to the occurrence of women (37). by
outcomes. Furthermore, various prospective studies in Most cohort studies (13, 15, 23, 24, 28, 29, 38–40), but gu
e
which height and weight were actually measured (e.g., 22, not all (41–43), reported a positive association between s
24, 28, 29) generally reported associations between height, height and breast cancer. Several of these studies were able t on
weight, and breast cancer that were consistent with those to compare menopausal strata and reported weaker associa- 10
from prospective studies relying on self-reported data. tions among premenopausal women (13, 24, 29, 38). Our A
p
Selection bias due to incomplete follow-up is unlikely pooled analysis showed an overall positive association with ril 2
because all cohorts show high degrees of completeness of premenopausal breast cancer, but the effect was less clear 01
9
follow-up. than that in postmenopausal cases and showed more vari-
ability among studies.
In our analysis of effect modification of height and breast
Height and breast cancer
cancer, we found a significant interaction only with a mater-
Apositive association between attained height and breast nal history of breast cancer, with a much stronger associa-
cancer risk has been found in many earlier studies (3). The tion among women with a positive maternal history.
positive association with height was less clear for pre- Although this is in agreement with earlier findings regarding
menopausal than for postmenopausal breast cancer in large overall family history of breast cancer (44, 45), it is not con-
(>500 cases) case-control studies (3), but recent, large case- sistent with our opposite finding of a lower degree of risk
control studies showed comparable positive associations in among women with a sister with breast cancer. The direc-
both menopausal strata (30–34). Two other recent, large tions of the effect modification were quite consistent across
case-control studies reported no association in pre- or post- the cohorts, suggesting that these interactions may not be
menopausal women (35, 36), but others reported a signifi- due to chance. These interactions merit further attention.
Am J Epidemiol Vol.152, No.6, 2000
522 van den Brandt et al.
TABLE 5. Pooled multivariate relative risks*(95% confidence intervals) for a 5-cm increment in height and a 4-kg/m2increment
in body mass index by levels of nondietary and dietary factors†,the Pooling Project of Diet and Cancer
Covariate and pvalue,
anthropometric Category of covariate test for
variable interaction
Age at menarche (years)‡ <12 12 13 14 ≥15
Height 1.03 (0.92, 1.15) 1.12 (1.01, 1.24) 1.03 (0.97, 1.08) 1.13 (1.01, 1.26) 1.07 (0.99, 1.17) 0.70
BMI
Premenopausal 0.88 (0.72, 1.09) 1.00 (0.87, 1.16) 0.84 (0.60, 1.17) 0.89 (0.64, 1.24) 0.94 (0.66, 1.35) 0.94
Postmenopausal 1.07 (0.99, 1.16) 1.07 (1.00, 1.16) 1.05 (0.98, 1.13) 1.05 (0.94, 1.18) 1.11 (0.96, 1.29) 0.77
Parity 0 1–2 ≥3
Height 1.05 (0.97, 1.15) 1.08 (1.02, 1.15) 1.06 (1.02, 1.11) 0.89 D
o
BMI w
n
Premenopausal 0.98 (0.58, 1.66) 0.85 (0.71, 1.02) 0.93 (0.82, 1.05) 0.76 lo
a
Postmenopausal 1.08 (0.99, 1.19) 1.11 (1.02, 1.20) 1.03 (0.98, 1.08) 0.23 d
e
d
Age at first birth (years) ≤20 >20–25 >25–30 >30 fro
m
Height 1.07 (0.94, 1.22) 1.07 (1.03, 1.12) 1.07 (1.01, 1.13) 1.05 (0.96, 1.14) 0.54 h
BMI ttp
s
Premenopausal 0.89 (0.64, 1.25) 0.94 (0.75, 1.17) 0.89 (0.75, 1.05) 0.87 (0.60, 1.25) 0.99 ://a
Postmenopausal 1.06 (0.96, 1.18) 1.07 (1.01, 1.13) 1.10 (1.00, 1.21) 0.98 (0.86, 1.12) 0.86 c
a
d
e
Benign breast disease‡ No Yes m
ic
Height 1.10 (1.05, 1.15) 1.02 (0.97, 1.09) 0.06 .o
u
BMI p
.c
Premenopausal 0.87 (0.78, 0.96) 0.95 (0.69, 1.33) 0.60 o
m
Postmenopausal 1.06 (1.00, 1.12) 1.07 (0.97, 1.17) 0.82 /a
je
Maternal history of breast /a
cancer‡ No Yes rtic
le
Height 1.07 (1.02, 1.12) 1.23 (1.11, 1.37) 0.005 -a
b
BMPIremenopausal 0.87 (0.79, 0.97) 0.78 (0.46, 1.34) 0.64 strac
Postmenopausal 1.07 (1.01, 1.13) 1.08 (0.95, 1.24) 0.77 t/1
5
2
History of breast cancer in /6/5
a sister‡ No Yes 1
4
Height 1.07 (1.02, 1.12) 0.96 (0.85, 1.09) 0.08 /75
8
BMI 0
7
Premenopausal 0.82 (0.73, 0.92) 1.88 (0.63, 5.61) 0.14 b
y
Postmenopausal 1.07 (1.03, 1.12) 1.08 (0.93, 1.26) 1.00 g
u
e
s
Oral contraceptive use‡ Never Ever t o
n
Height 1.09 (1.04, 1.15) 1.03 (0.98, 1.09) 0.21 1
0
BMI A
p
PPoresmtmeennooppaauussaall 01..8058 ((01..7031,, 01..9186)) 01..9016 ((00..8925,, 11..0129)) 00..4947 ril 2
0
1
Table continues 9
Because attained height might be an indicator of child- the end of the War (15, 24, 29, 40) compared with US cohort
hood energy intake in situations in which there is sufficient studies (3). It is also consistent with the prior observations
variation in energy intake (2), this has been proposed as an that associations were found in US Black women (33) as
explanation for the observation that the positive association opposed to White women (36) and in US cohorts with over-
between height and breast cancer was found more fre- representation of women potentially undernourished during
quently in European populations that experienced food childhood or adolescence (28), as well as with many animal
shortages during World War II (13, 24). Indeed, stronger studies showing that energy restriction clearly reduces
associations have generally been found in studies of the experimentally induced and spontaneous mammary tumor
affected birth cohorts in Norway, the Channel Islands, and risk in rodents (46). However, more recent studies show that
the Netherlands, where severe food deprivation occurred at height is also rather consistently related to breast cancer risk
Am J Epidemiol Vol.152, No.6, 2000
Anthropometry and Breast Cancer Risk 523
TABLE 5. Continued
Covariate and pvalue,
anthropometric Category of covariate test for
variable interaction
Hormone replacement
therapy use‡ Never Ever 0.45
Height (postmenopausal) 1.06 (1.00, 1.13) 1.09 (1.03, 1.15) 0.24
BMI (postmenopausal) 1.09 (1.04, 1.14) 1.04 (0.92, 1.18)
Education§ Less than High More than
high school school high school
Height 1.10 (1.04, 1.17) 1.11 (1.02, 1.20) 1.00 (0.93, 1.08) 0.28
BMI D
o
Premenopausal 0.89 (0.62, 1.27) 0.66 (0.31, 1.41) 1.28 (0.85, 1.93) 0.17 w
Postmenopausal 1.12 (0.99, 1.25) 1.04 (0.93, 1.16) 1.08 (0.99, 1.17) 0.37 nlo
a
d
Dietary fiber intake Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 ed
Height 1.09 (1.02, 1.17) 1.01 (0.95, 1.07) 1.10 (1.03, 1.16) 1.11 (1.02, 1.21) 1.03 (0.97, 1.09) 0.84 fro
m
BMI h
Premenopausal 0.85 (0.60, 1.19) 0.86 (0.70, 1.04) 0.89 (0.72, 1.10) 0.95 (0.80, 1.13) 1.00 (0.78, 1.28) 0.34 ttp
Postmenopausal 1.09 (1.01, 1.18) 1.04 (0.93, 1.17) 1.01 (0.93, 1.11) 1.05 (0.95, 1.17) 1.13 (1.03, 1.23) 0.52 s://a
c
Total fat intake Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 ad
e
m
Height 1.01 (0.95, 1.08) 1.08 (1.02, 1.15) 1.11 (1.05, 1.18) 1.04 (0.98, 1.11) 1.07 (0.99, 1.15) 0.31
ic
BMI .o
u
Premenopausal 0.86 (0.70, 1.05) 0.77 (0.62, 0.96) 0.90 (0.73, 1.11) 0.95 (0.78, 1.16) 0.96 (0.82, 1.12) 0.14 p
.c
Postmenopausal 1.10 (0.98, 1.24) 1.07 (0.99, 1.16) 0.99 (0.92, 1.07) 1.10 (1.00, 1.20) 1.08 (0.97, 1.20) 0.93 o
m
/a
Alcohol intake (g/day) 0 1–<30 ≥30 je
/a
Height 1.08 (1.03, 1.13) 1.05 (1.00, 1.12) 1.06 (0.94, 1.19) 0.76 rtic
BMI le
Premenopausal 0.89 (0.77, 1.02) 0.93 (0.75, 1.14) 0.92 (0.57, 1.49) 0.92 -ab
s
Postmenopausal 1.07 (1.02, 1.13) 1.04 (0.99, 1.10) 1.00 (0.75, 1.35) 0.78 tra
c
Smoking¶ Never Ever t/15
2
Height 1.05 (1.00, 1.11) 1.07 (0.99, 1.16) 0.56 /6
BMI /51
4
Premenopausal 0.89 (0.78, 1.01) 0.88 (0.78, 1.00) 0.94 /7
Postmenopausal 1.05 (1.00, 1.10) 1.06 (1.00, 1.14) 0.39 58
0
*Relative risks were adjusted for age at menarche (≤11, 12, 13, 14, ≥15 years), parity (0, 1, 2, ≥3), age at birth of the first child (≤20, 7 b
21–25, 26–30, >30 years), postmenopausal hormone use (ever, never), oral contraceptive use (ever, never), history of benign breast disease y
g
(no, yes), maternal history of breast cancer (no, yes), history of breast cancer in a sister (no, yes, no sisters), smoking status (ever, never), u
e
education (less than high school graduation, high school graduation, more than high school graduation), fat intake (quintiles), fiber intake s
(quintiles), energy intake (continuous), and alcohol intake (0, >0–<1.5, 1.5–<5, 5–<15, 15–<30, ≥30 g/day). t o
n
† The Adventist Health Study was not included in analyses of premenopausal women. 1
‡ The New York State Cohort was not included in this analysis. 0 A
§ The Nurses’Health Studies (a) and (b) were not included in this analysis. p
¶ The Sweden Mammography Cohort was not included in this analysis. ril 2
0
1
9
in relatively affluent populations that have not experienced closure during puberty and, thus, attained height. Furthermore,
energy restriction during growth periods and in which Stoll (49) has suggested that better nutrition accelerates
height is presumably primarily genetically determined (30, growth hormone release which, in turn, increases levels of
31, 37, 47). It has also been hypothesized that height reflects insulin-like growth factor (IGF). The adolescent growth spurt
mammary gland mass (or more precisely, the number of involves stimulation by growth hormone, insulin, IGF, and
ductal stem cells that develop in the breast in utero), which sex steroids, and Stoll hypothesizes that the combination of
could be related to breast cancer risk (48). IGF and sex steroids results in mitogenic effects on develop-
Considering that nutritional inadequacy does not seem to ing mammary tissue in adolescence and a concomitant
be the only factor responsible for the association between increased risk of epithelial atypia and carcinogenesis. This is
height and breast cancer, Ballard-Barbash (45) has proposed compatible with the observation (50) that plasma levels of
that genetic and environmental factors (e.g., diet and physical IGF-1 predicted the incidence of premenopausal, but not
activity) may affect the hormones that influence epiphyseal postmenopausal, breast cancer. Dorgan et al. (51) have sug-
Am J Epidemiol Vol.152, No.6, 2000