American Journal of Epidemiology
© The Author 2012. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 177, No. 1
DOI: 10.1093/aje/kws270
Advance Access publication:
November 7, 2012
Practice of Epidemiology
Scaling of Weight for Height in Relation to Risk of Cancer at Different Sites in a
Cohort of Canadian Women
Geoffrey C. Kabat*, Moonseong Heo, Anthony B. Miller, and Thomas E. Rohan
* Correspondence to Dr. G. C. Kabat, Department of Epidemiology and Population Health, Albert Einstein College of Medicine, 1300 Morris
Park Avenue, Bronx, NY 10461 (e-mail: [email protected]).
Initially submitted February 24, 2012; accepted for publication May 15, 2012.
Many studies have examined the associations of body mass index (weight (kg)/height (m)2) with risk of
various cancers. However, optimal scaling of weight for height may depend on the population studied. The
authors used data from a large cohort study of women (Canadian National Breast Cancer Screening Study,
1980–2000; n = 89,835) to examine how the scaling of weight for height (W/Hx) influenced the association with
risk of 19 different cancers. Cox proportional hazards models were used to estimate the hazard ratio for each
cancer site with W/Hx, with x increasing from 0 to 3.0 by increments of 0.1. The correlation between weight and
W/Hx decreased monotonically with increasing x, whereas W/Hx was minimally correlated with height when
x = 1.4. W/Hx showed significant positive associations with postmenopausal breast cancer, endometrial cancer,
kidney cancer, and lung cancer in never smokers. W/Hx was inversely associated with lung cancer in ever
smokers. The value of x for which W/Hx produced the largest statistically significant hazard ratio ranged from 0.8
(endometrial cancer) to 1.7 ( postmenopausal breast cancer). For lung cancer in ever smokers, the inverse association was statistically significant for all values of x. These findings suggest that the scaling of weight for height
may vary depending on the cancer site and that optimal scaling may be considerably different from W/H2 or,
alternatively, that a range of scaling should be considered when examining the association of body weight with
risk of disease.
adiposity, body mass index, cancer, cohort studies, scaling
Abbreviations: BMI, body mass index; CNBSS, Canadian National Breast Screening Study; W/Hx, weight/heightx.
Body mass index (BMI, determined as weight in kilograms divided by height in meters squared), which was
originally referred to as the Quetelet Index after the Belgian
mathematician Adolphe Quetelet who first proposed it in
1835 (1), has become the most widely used weightfor-height index of adiposity in large population studies of
children and adults (2–4). The fact that many large epidemiologic studies only have information on weight and
height and the ease of computing BMI appear to have contributed to its popularity. BMI is positively associated with
increased risk of a number of cancers (including endometrial, postmenopausal breast, colorectal, esophageal adenocarcinoma, and renal cancers) (5–7) and with increased cancer
incidence and mortality overall (6, 8), as well as with increased risk of diabetes and cardiovascular disease (9).
Inverse associations of BMI with a number of cancers have
also been observed, specifically for premenopausal breast,
lung, and oropharyngeal cancers and squamous cell carcinoma of the esophagus (6, 7).
Although BMI has been adopted as a proxy for adiposity and has been widely used to assess the impact of
obesity on health, the limitations of BMI as a marker of
adiposity have long been recognized (10–14). BMI does
not distinguish between body fat and lean mass, and individuals who are physically fit will tend to have a higher
BMI because of their increased muscle mass (10). On the
other hand, postmenopausal women are more likely to have
BMIs that reflect reduced lean mass, and so they may have
a relatively low BMI but more fat than do younger women.
BMI was adopted because in the original populations
93
Am J Epidemiol. 2013;177(1):93–101
94 Kabat et al.
studied ( predominantly middle-aged men) (2, 13, 15), it
was more strongly correlated with direct measures of adiposity (e.g., skinfold thickness and hydrostatic weighing)
than was weight alone and because it was strongly correlated with weight and minimally correlated with height.
However, other studies suggest that the optimal weight-forheight index is dependent on the population and that
weight (kg)/height (m)x (W/Hx), where the scaling factor x
takes values other than 2, performs better than BMI
because it is highly correlated with weight and virtually independent of height (14, 16–18).
The fact that a number of cancers are associated with
height (19, 20) (and also with BMI) further underscores the
need for a weight-for-height measure that is minimally correlated with height so as to allow independent assessment
of the associations of indices of obesity and of height with
cancer. Renehan (21) recently pointed out that the association of BMI with cancer at different sites may be confounded by height, and he recommended that future studies
should test additional weight-for-height indices that may be
less sensitive to the influence of height.
Few studies have used weight-for-height indices other
than BMI or compared the performances of these indices in
estimating of the independent associations of obesity
indices with the risk of specific cancers. It is possible that
the optimum weight-for-height index differs for different
cancer sites both because weight and height may each have
different associations with specific cancers and because
body weight may increase (or decrease) the risk of cancer
at different sites via different mechanisms (22, 23).
We therefore used measured weight and height data from
89,835 women enrolled in the Canadian National Breast
Screening Study (CNBSS) prospective cohort to compare
the performance of a range of weight-for-height measures
as predictors of the risk of different cancers, both with and
without adjustment for potential confounding factors.
MATERIALS AND METHODS
Study population and questionnaire
The CNBSS is a randomized controlled trial of screening
for breast cancer, the details of which have been described
elsewhere (24, 25). The Human Experimentation Committee at the University of Toronto and Human Experimentation Committees at 15 CNBSS collaborating centers
approved the study. In brief, 89,835 women aged 40–59
years were recruited from the general Canadian population
between 1980 and 1985. On enrollment into the study, information was obtained from participants on demographic,
hormonal, reproductive, and lifestyle characteristics using a
self-administered questionnaire. Participating women had
their weight and height measured at the initial examination.
Starting in 1982, a self-administered food frequency
questionnaire developed for the CNBSS (26) was distributed to all new attendees at all screening centers and to
women returning to the centers for rescreening. The food
frequency questionnaire elicited information on usual
portion size and consumption of 86 food and beverage
items, including alcoholic beverages, as well as 2 questions
about usual level of moderate and vigorous physical activity during the past month (hours per day). The questionnaire
included photographs of portion sizes to assist respondents
in quantifying intake. Responses were used to estimate the
daily intake of calories, alcohol, and selected nutrients
using a nutrient database developed by modifying the US
Department of Agriculture’s food composition tables to
include typically Canadian foods (27). A total of 49,654
dietary questionnaires were returned and were available for
analysis. Use of the self-administered questionnaire was
validated by comparison with an interviewer-administered
version of the same questionnaire (26).
Ascertainment of index cancers and deaths
Incident cases of cancer and deaths from all causes were
ascertained by means of computerized record linkages to
the Canadian Cancer Database and the National Mortality
Database, respectively. The linkages yielded data on cancer
incidence and death to December 31, 2000, for women in
Ontario, December 31, 1998, for women in Quebec, and
December 31, 1999, for women in other provinces. For the
present analyses, study participants were considered at risk
from their date of enrollment until the date of diagnosis of
1 of 19 different types of cancer, termination of follow-up
(the date until which cancer incidence data were available
for women in the corresponding province), or death, whichever occurred first. We excluded women who were missing
height or weight measurements at baseline (n = 868) and
women who were diagnosed with cancer before their date
of enrollment (n = 711). After exclusions, during an
average of 16.4 years (1,429,734 person-years) of followup, a total of 5,679 incident invasive cancers were ascertained among 88,256 women. Where an individual had
more than 1 cancer diagnosis, we selected the first diagnosis for inclusion, and follow-up of that individual stopped
at that point. We selected types of cancer for which there
were more than 90 cases for inclusion in the analysis. The
number of cases by site and person-years of follow-up are
given in Table 1. Analyses of endometrial cancer were restricted to women with intact uteri at baseline, and cases of
ovarian cancer were restricted to women with at least 1
intact ovary at baseline.
Statistical analysis
We examined the correlation of W/Hx with weight and
height for increments of x of 0.1, ranging from x = 0 to
x = 3.0. Cox proportional hazards models were used to estimate hazard ratios and 95% confidence intervals for the association of W/Hx with each cancer site, where x ranged
from 0 to 3.0 in increments of 0.1. Specifically, for each
cancer site, we fitted 31 age-adjusted models (model 1) and
another 31 multivariable-adjusted models (model 2), resulting in a total of 1,178 model fittings and the same number
of hazard ratios.
The following covariates were included in the multivariable model: age at enrollment (continuous), oral contraceptive use (ever vs. never), hormone therapy (ever vs. never),
menopausal status ( premenopausal, perimenopausal, or
Am J Epidemiol. 2013;177(1):93–101
Scaling of Weight for Height and Cancer in Women 95
Table 1. Number of Incident Cancer Cases and Person-Years of
Follow-up at Different Sites in the Canadian National Breast
Screening Study, 1980–2000
Cancer Site or Type
No. of Cases
Breast (premenopausal)
1908
Breast (postmenopausal)
2316
751,200
Colorectuma
1096
1,450,030
Colon
758
1,451,736
Rectum
327
1,453,350
Endometrium
783
1,010,599
Ovary
476
1,443,769
Lung (ever smokers)
657
690,845
Lung (never smokers)
100
762,652
Kidney
196
1,454,059
Bladder
158
1,454,167
Thyroid
167
1,453,873
Cervix
Person-Years
627,798
91
1,454,461
Pancreas
182
1,454,872
Melanoma
327
1,452,761
Brain
138
1,454,860
Leukemia
155
1,454,551
Non-Hodgkin’s lymphoma
112
1,454,598
Multiple myeloma
113
1,454,736
cancer, we performed analyses stratified by menopausal
status ( premenopausal vs. postmenopausal), and for lung
cancer, we performed analyses after stratification by
smoking status (ever vs. never). For breast cancer, the
models stratified by menopausal status also included age at
menarche (<12, 12, 13, or ≥14 years), parity (0, 1, 2, or ≥3
children), family history of breast cancer in a first degree
relative (yes vs. no), and history of benign breast disease
(yes vs. no). In addition to the analysis of the entire cohort,
we carried out analyses for 6 sites in the dietary subcohort
to include covariates that were available only for women
who completed the food frequency questionnaire
(n = 49,654). For the 6 cancer sites, we compared the
results of the multivariable model used in the entire cohort
but restricted to the dietary subcohort with the results of the
models including additional variables, as follows: premenopausal and postmenopausal breast cancer (alcohol intake in
g/day (continuous) and physical activity level in hours/day
(continuous)); colorectal cancer, colon cancer, and rectal
cancer (alcohol intake, physical activity level, dietary
folate, and red meat intake); and endometrial cancer ( physical activity).
All statistical significance tests were 2-sided. All analyses were performed using SAS, version 9 (SAS Institute,
Inc., Cary, North Carolina).
RESULTS
a
A total of 11 women who had both colon and rectal cancer were
excluded from the subsite analyses.
postmenopausal), years of education (<12, 12, or >12), and
pack-years of smoking (never smoked, 1–9, 10–19, 20–29,
30–39, or ≥40). The model for colorectal cancer also included terms for parity (0, 1, 2, or ≥3 children). For breast
Table 2 presents the distribution of personal characteristics
of study participants by quintiles of weight and BMI. Mean
age was 49.0 years. Twenty-seven percent of women had
more than 12 years of education and 53.4% were postmenopausal. The percentages of postmenopausal women and
women who experienced menarche before 12 years of age
were strongly and positively associated with BMI, whereas
Table 2. Baseline Characteristics by Quintiles of Body Mass Index in the Canadian National Breast Screening Study, 1980–2000
BMIa
<21.6
21.6–23.2
23.3–24.9
25.0–27.8
≥27.9
17,734
17,853
17,807
17,784
17,789
Variable
Mean (SD)
%
Mean (SD)
%
Mean (SD)
%
Mean (SD)
%
Mean (SD)
P Value
%
Measured BMIa
20.3 (1.1)
22.5 (0.5)
24.1 (0.5)
26.3 (0.8)
32.2 (5.3)
<0.0001
Age, years
48.1 (5.5)
48.6 (5.6)
49.1 (5.5)
49.5 (5.6)
49.6 (5.5)
<0.0001
Height, cm
162.6 (6.4)
162.0 (6.0)
161.7 (6.0)
161.2 (6.1)
160.5 (6.6)
<0.0001
>12 years of education
33.7
30.4
27.6
24.4
20.0
<0.0001
Postmenopausal
45.2
49.5
54.4
57.9
60.0
<0.0001
Nulliparous
18.4
14.9
14.1
12.6
12.2
<0.0001
Age at menarche <12
years
13.2
14.6
16.4
18.8
24.5
<0.0001
Age at first birth ≥30
years
11.6
10.4
10.2
10.1
9.3
<0.0001
Oral contraceptive use,
ever
64.9
62.0
59.3
55.1
51.3
<0.0001
Hormone therapy, everb
42.9
42.6
43.5
42.7
38.5
<0.0001
25.8
22.3
21.2
20.3
19.5
<0.0001
Current smokers
a
b
2
Weight (kg)/height (m) .
Among postmenopausal women.
Am J Epidemiol. 2013;177(1):93–101
96 Kabat et al.
Figure 1. Correlation of weight (kg)/height (m)x with weight and height in 89,835 women in the Canadian National Breast Screening Study,
1980–2000. At x = 1.4, the correlation with weight is 0.95 and that with height is 0.
the percentages of women who had more than 12 years of
education, who were nulliparous, or who were ever contraceptive users were strongly and inversely associated with BMI.
In the total population, W/Hx was (necessarily) maximally correlated with weight when x = 0 (i.e., W/H0 = W), and
the correlation declined monotonically, yet very slowly,
with increasing x (Figure 1). On the other hand, W/Hx was
minimally correlated with height when x = 1.4. Thus, it
appears that the ideal scaling in this population is x = 1.4,
where the correlation with height is 0 and that with weight
is still very high (0.95).
In Figure 2, age- and multivariable-adjusted hazard ratios
are presented for the association of W/Hx with 19 different
cancers as x increased from 0 to 3.0 in increments of 0.1. In
multivariable-adjusted models, W/Hx showed significant
positive associations with postmenopausal breast cancer, endometrial cancer, kidney cancer, and lung cancer in never
smokers and was inversely associated with lung cancer in
ever smokers. The values of x that produced the largest statistically significant hazard ratio were 1.7 for postmenopausal
breast cancer, 1.1 for kidney cancer, 0.8 for endometrial
cancer, and 1.3 for lung cancer in never smokers. For lung
cancer in ever smokers, the hazard ratio decreased monotonically with increasing x and was highly statistically significant for all values of x. For the 5 sites that showed
statistically significant results, the estimates from the ageand multivariable-adjusted models were generally similar.
For pancreatic and cervical cancers, W/Hx showed positive
associations in age-adjusted models but not in multivariableadjusted models. The following sites/types showed no association with W/Hx: premenopausal breast, colorectum,
colon, rectum, thyroid, bladder, ovary, brain, melanoma, leukemia, non-Hodgkin’s lymphoma, and multiple myeloma.
The results of analyses carried out in the dietary subcohort for 5 cancer sites were similar to the results reported
for the entire cohort in Figure 2; however, for postmenopausal breast cancer in the expanded model, the range of
statistically significant hazard ratios was limited to x = 0 to
0.2 (results not shown).
To further examine the effect of different scaling, we
compared the association of 4 indices (W, W/H, W/H1.4,
and W/H2) with the risk of cancer at the 5 sites that showed
significant associations in the primary analysis (Table 3).
For 3 types of cancer (endometrial cancer, kidney cancer,
and lung cancer in ever smokers), quintiles of weight
showed a steeper trend compared with the other indices.
For kidney cancer and lung cancer in never smokers, there
was a trend toward higher P values with increasing scaling.
Kidney cancer showed the greatest range in P values,
which decreased with increasing values of the scaling
factor. However, all indices showed significant associations
with all 5 sites. Particularly for endometrial cancer and
lung cancer in ever smokers, all 4 indices gave comparably
strong associations. Addition of height as a covariate in
models for all 4 indices did not materially affect the results
that are presented in Table 3.
DISCUSSION
We examined the scaling of W/Hx, where x ranged from
0 to 3, in relation to the risk of cancer at 19 anatomic sites.
For the 4 sites that showed a statistically significant positive
association with W/Hx, the highest value of x (that was statistically significant) ranged from 0.8 (endometrial cancer)
to 1.7 ( postmenopausal breast cancer). For lung cancer in
ever smokers, all values of x between 0 and 3.0 yielded
statistically significant inverse associations. Two sites
(cervix uteri and pancreas) showed positive associations in
the age-adjusted model but not in the multivariable-adjusted model. The 12 remaining sites were not associated with
Am J Epidemiol. 2013;177(1):93–101
Scaling of Weight for Height and Cancer in Women 97
W/Hx for any value of x between 0 and 3. There was a high
degree of consistency when scaling was between 0 and 2 as
to whether a significant association with a cancer outcome
was observed. Furthermore, inclusion of height as a covariate did not materially affect the associations of weight-forheight indices with cancer outcomes. To the best of our
knowledge, no previous study has examined the performance of a range of different weight-for-height indices in
association with different cancer sites.
BMI is by far the most widely used index of obesity/adiposity in epidemiologic studies. However, it has long been
recognized that BMI is an imperfect indicator of adiposity
because it does not distinguish between fat mass and lean
mass (10, 13, 14). Three issues in particular are relevant to
determining the optimal weight-for-height measure to
assess the association of adiposity/obesity with disease risk
in epidemiologic studies: 1) the ideal weight-for-height
index should be maximally correlated with weight (itself a
proxy for adiposity) and minimally correlated with height;
2) the index should be maximally correlated with adiposity
in validation studies using more accurate techniques to
measure body fat, such as hydrostatic weighing, bioelectrical
impedance, and dual energy x-ray absorptiometry; and
3) the index should show consistent associations with obesityrelated outcomes. We discuss each of these issues below.
First, numerous studies have addressed the question of
what the optimal weight-for-height index to serve as an indicator of adiposity/obesity in large epidemiologic studies
is (2, 13–18, 28–31). Although there is general agreement
that the optimal index is that which is maximally correlated
with weight and least correlated with height, there is disagreement about which index performs best. Several studies
(mainly limited to men) (2, 13, 15) that compared W/H,
W/H2, and W/H3 concluded that W/H2 was the best of the
3 indices. However, other studies suggested that the best
index may differ between men and women (14, 17, 18) and
may depend on the population studied (16–18). Benn (16)
argued that the exponent should be derived from the population under study and proposed use of W/Hp, where
p could be a noninteger. In studies by Florey (14), Lee
et al. (17), and Micozzi et al. (18), x was consistently lower
in females than in males. The work of Benn (16), Florey
(14), and Lee et al. (17) suggests that the best weight-forheight index will depend on the population studied.
Second, studies have attempted to validate different
weight- and height-based indices of obesity against more
direct measurements of body fat, including skinfold thickness, hydrostatic weighing, and dual-energy x-ray absorptiometry (2, 13, 14, 18, 28, 29, 31). Benn’s index and W/H2
have generally shown the strongest correlations with direct
measures of adiposity (28, 29). In an analysis data from the
National Health and Nutrition Examination Survey,
Micozzi et al. (18) found that in men, W/H2 showed a high
correlation with weight, the highest correlation with measured body fat, and no significant correlation with height,
whereas in women W/H and W/H1.5 showed the highest
correlations with weight and subscapular skinfold thickness
and no significant correlation with height.
Larsson et al. (31) examined optimized predictions of absolute and relative amounts of body fat, measured using
Am J Epidemiol. 2013;177(1):93–101
dual-energy x-ray absorptiometry, by different weightfor-height indices in which the power for height ranged
from 0 to 3.0. Their study population consisted of over 1,100
randomly selected subjects from the general populations of
2 Swedish cities and 149 obese subjects from the same 2
cities. Correlations between W/H and W/H2 and absolute
body fat measured using dual-energy x-ray absorptiometry
in men and women were high (0.91 and 0.88, respectively,
in men and 0.97 and 0.96, respectively, in women). The
optimal value of x in W/Hx in predicting absolute body fat
was close to 1.0 in both men and women, whereas W/H2
was the optimal predictor of percent body fat, although
BMI was not proportional to either measure of body fat in
severely obese subjects. The authors concluded that W/H
might be a more optimal weight-for-height index than
BMI, particularly at high body weights.
Third, we are unaware of previous studies in which investigators have examined the performance of a range of
weight-for-height indices in relation to the risk of cancer at
different sites. Lee and Kolonel (32) compared the performance of 4 weight-for-height indices (W/H, W/H2, W/H3,
and W/Hp (Benn’s index)) in relation to the risk of lung
cancer in a small, nested case-control study. Although all 4
indices showed an inverse association with risk, only W/H
and W/Hp showed monotonic decreasing trends in the odds
ratio by increasing quartiles. The authors concluded that
different body mass indices were not interchangeable and
that W/Hp was the “body mass index of choice in epidemiologic analysis” (32, p. 377).
Our results suggest that the optimal scaling of W/Hx in
measuring adiposity may differ by cancer site and may be
below the value of 2.0, which is used for BMI. In the
present study, the value of x that was least correlated with
height and maximally correlated with weight was 1.4.
However, the optimal value of x differed by cancer site/
type. All 5 sites that showed significant associations with
W/Hx also showed significant associations when x = 2 (i.e.,
for BMI); however, for 4 of the 5 sites, the magnitude of
the hazard ratios (absolute value) was larger for values of
x < 2 than for x = 2. This suggests that the statistical power
to detect a significant association with W/Hx may be
greater for the optimal value of x compared with using
BMI. Nevertheless, the fact that for specific cancers W/Hx
showed the largest hazard ratio for specific values of x does
not necessarily imply that that value of x is maximally associated with reliably measured body fat or that similar correlations would be obtained using different sample
populations. At the very least, our results indicate that it
may be useful to consider a range of weight-for-height
indices when examining the association of body weight
with disease.
There has been inconsistency among different studies as
to the association of BMI with certain cancer sites (e.g.,
thyroid cancer (33)), and there is inconsistency within
some studies by sex (e.g., studies of colorectal cancer (34)).
Some of these inconsistencies may be due to confounding
by height, which in some studies was positively associated
with thyroid and colorectal cancers (19, 20, 33). The need
for a weight-for-height index of adiposity that is uncorrelated with height takes on added importance in light of the
98 Kabat et al.
Figure 2. Age- and multivariable-adjusted hazard ratios for the association of weight/heightx with 19 cancer sites, as x ranges from 0 to 3.0 in
increments of 0.1, Canadian National Breast Screening Study, 1980–2000. Dashed lines denote the age-adjusted hazard ratio (model 1); solid
lines denote the multivariable-adjusted hazard ratio (model 2). Arrows indicate corresponding range of statistically significant hazard ratios.
fact that a number of cancers are positively associated with
height (19, 20). Renehan (21) recently pointed out that the
association of BMI with a given outcome may be confounded by height. However, our analyses of specific
cancers indicated that confounding of the association of a
range of weight-for-height indices by height was minimal.
Further studies are needed that examine the independent
associations of indices of adiposity and height with the risk
of specific cancers. Furthermore, although obesity and leanness are associated with a number of cancers, the association
of obesity with different cancers may operate through different pathways (22, 23, 35). For example, the associations of
obesity with postmenopausal breast cancer and endometrial
cancer may involve estrogen, whereas that of obesity with
colorectal cancer may operate through insulin and related
growth factors and that of obesity with renal cancer may
operate through effects on blood pressure (35).
Strengths of the present study include the availability of
measured height and weight from a large population, large
numbers of cancers at specific sites, completeness of followup due to linkage to national cancer and death registries, and
the availability of information on potential confounding variables. However, a major limitation is that we were not able
to examine the associations of W/Hx in men, which would
have enabled us to assess consistency between the 2 sexes
for cancer sites that occurred in both men and women. Furthermore, we were not able to adjust for several covariates,
including alcohol intake and physical activity level, in the
full cohort; however, the results were unchanged in analyses
in which we controlled for these variables in the dietary subcohort. Data on other key covariates, such as family history
of colorectal cancer, were unavailable.
In conclusion, in a cohort of nearly 90,000 Canadian
women, the scaling of weight for height that was most
Am J Epidemiol. 2013;177(1):93–101
Scaling of Weight for Height and Cancer in Women 99
Figure 2. Continued.
strongly associated with weight and least associated with
height was 1.4. W/Hx was positively associated with the risk
of 4 cancers (postmenopausal breast cancer, endometrial
cancer, kidney cancer, and lung cancer in never smokers),
with x ranging from 0.8 (endometrial cancer) to 1.7 (breast
cancer). W/Hx was inversely associated with lung cancer in
ever smokers for all values of x. Our results suggest that
the optimal scaling of x may be below 2.0 and may differ
Am J Epidemiol. 2013;177(1):93–101
for different cancer sites or, alternatively, that a range of
scaling of weight for height may be equally informative.
Further studies are needed to confirm and extend these
findings in other, more diverse populations. Potential implications of our findings are that epidemiologists should consider using different weight-for-height scaling to examine
the association of body size with risk of disease in different
populations.
100 Kabat et al.
Table 3. Hazard Ratiosa and 95% Confidence Intervals for the Association of Selected Weight-for-Height Indices With Specific Cancer Sites
in the Canadian National Breast Screening Study, 1980–2000
Quintiles of Selected Weight-for-height Indices
Cancer Site and
Variable
No. of Cases
Postmenopausal
breast
1
2
3
4
P
for trend
5
HR
95% CI
HR
95% CI
HR
95% CI
HR
95% CI
2,356
Weight
1.00
1.07
0.92–1.24
1.26
1.10–1.45
1.33
1.16–1.53
1.43
1.24–1.64
<0.0001
W/H
1.00
1.13
0.98–1.31
1.18
1.03–1.37
1.43
1.25–1.64
1.45
1.26–1.66
<0.0001
W/H1,4
1.00
1.15
1.00–1.34
1.16
1.27–1.67
1.46
1.27–1.67
1.43
1.25–1.65
<0.0001
1.00
1.19
1.04–1.37
1.15
1.00–1.33
1.42
1.24–1.62
1.44
1.26–1.66
<0.0001
Weight
1.00
0.99
0.75–1.30
1.12
0.86–1.46
1.39
1.08–1.78
2.92
2.33–3.67
<0.0001
W/H
1.00
0.88
0.67–1.15
1.17
0.91–1.51
1.22
0.95–1.56
2.79
2.24–3.48
<0.0001
W/H
1.00
0.87
0.67–1.14
1.11
0.86–1.43
1.28
1.00–1.63
2.68
2.15–3.33
<0.0001
W/H2
1.00
1.03
0.79–1.33
1.21
0.94–1.56
1.30
1.01–1.67
2.78
2.22–3.48
<0.0001
Weight
1.00
1.34
0.78–2.29
1.19
0.69–2.06
1.64
0.99–2.72
2.05
1.24–3.37
0.001
W/H
1.00
1.21
0.72–2.04
1.10
0.65–1.87
1.71
1.06–2.78
1.70
1.05–2.77
0.01
W/H1,4
1.00
1.00
0.59–1.69
1.12
0.67–1.86
1.49
0.93–2.41
1.64
1.02–2.63
0.008
W/H2
1.00
0.93
0.56–1.56
0.92
0.55–1.53
1.54
0.97–2.43
1.43
0.89–2.28
0.02
Weight
1.00
0.84
0.67–1.05
0.57
0.45–0.73
0.63
0.50–0.79
0.50
0.39–0.64
<0.0001
W/H
1.00
0.80
0.64–1.00
0.56
0.44–0.72
0.58
0.46–0.74
0.53
0.41–0.67
<0.0001
W/H1,4
1.00
0.73
0.59–0.92
0.61
0.48–0.78
0.50
0.39–0.64
0.54
0.42–0.68
<0.0001
W/H2
1.00
0.84
0.67–1.05
0.65
0.52–0.83
0.53
0.41–0.68
0.57
0.45–0.73
<0.0001
Weight
1.00
0.98
0.46–2.09
1.08
0.52–2.25
1.37
0.69–2.71
1.98
1.03–3.80
0.007
W/H
1.00
1.16
0.54–2.47
1.49
0.73–3.05
1.36
0.66–2.82
2.13
1.09–4.17
0.01
W/H1,4
1.00
0.71
0.31–1.62
1.57
0.80–3.12
1.43
0.72–2.85
1.81
0.93–3.51
0.02
W/H2
1.00
1.26
0.56–2.85
2.07
0.99–4.34
1.88
0.89–3.98
2.02
0.96–4.23
0.07
2
W/H
Endometriumb
795
1,4
Kidney
Lung (ever smokers)
Lung (never smokers)
206
685
104
Abbreviations: CI, confidence interval; HR, hazard ratio; W/H, weight for height.
Covariates in the main model included age at entry (continuous), oral contraceptive use (ever vs. never), hormone use (ever vs. never),
menopausal status (premenopausal, perimenopausal, postmenopausal, or unknown), years of education (<12, 12, or >12), and pack-years of
smoking (never smoked, 1–9, 10–19, 20–29, 30–39, or ≥40). For postmenopausal breast cancer, the model additionally included age at
menarche (<12, 12, 13, or ≥14 years), parity, family history of breast cancer in a first degree relative (yes vs. no), and history of benign breast
disease (yes vs. no). The model for lung cancer in never smokers did not include pack-years.
b
Analyses of endometrial cancer were restricted to women with an intact uterus.
a
ACKNOWLEDGMENTS
Author affiliations: Department of Epidemiology and
Population Health, Albert Einstein College of Medicine,
Bronx, NY (Geoffrey C. Kabat, Moonseong Heo,
Thomas E. Rohan); and Dalla Lana School of Public
Health, University of Toronto, Toronto, Canada (Anthony
B. Miller).
This work was supported by institutional funds to the
Albert Einstein College of Medicine. Dr. Rohan is supported by a grant from the Breast Cancer Research Foundation.
Conflict of interest: none declared.
REFERENCES
1. Quetelet LAJ. A treatise on man and the development of his
faculties. Obes Res. 1994;2(1):72–85.
2. Keys A, Fidanza F, Karvonen MJ, et al. Indices of relative
weight and obesity. J Chron Dis. 1972;25(6):329–343.
3. Bray GA. Evaluation of obesity. Who are the obese?
Postgrad Med. 2003;114(6):19–27.
4. Gallagher D, Visser M, Sepulved D, et al. How useful is body
mass index for comparison of body fatness across age, sex,
and ethnic groups? Am J Epidemiol. 1996;143(3):228–239.
5. World Cancer Research Fund. Food, Nutrition, Physical
Activity, and the Prevention of Cancer: A Global Perspective.
Am J Epidemiol. 2013;177(1):93–101
Scaling of Weight for Height and Cancer in Women 101
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
2nd ed. Washington, DC: American Institute for Cancer
Research; 2007.
Reeves GK, Pirie K, Beral V, et al. Cancer incidence and
mortality in relation to body mass index in the Million Women
Study: cohort study. BMJ. 2007;335(7630):1134–1200.
Renehan A, Tyson M, Egger M, et al. Body mass index and
incidence of cancer: a systematic review and meta-analysis of
prospective observational studies. Lancet. 2008;371(9612):
569–578.
Manson JE, Willett WC, Stampfer MJ, et al. Body weight
and mortality among women. New Engl J Med. 1995;
333(11):677–685.
Prospective Studies Collaboration, Whitlock G, Lewington S,
Sherliker P, et al. Body-mass index and cause-specific
mortality in 900 000 adults: collaborative analyses of 57
prospective studies. Lancet. 2009;373(9669):1083–1096.
Prentice AM, Jebb SA. Beyond body mass index. Obesity
Rev. 2001;2(3):141–147.
Franzosi MG. Should we continue to use BMI as a
cardiovascular risk factor? Lancet. 2006;368(9536):624–625.
Renehan AG, Soerjomataram I, Leitzmann MF. Interpreting
the epidemiological evidence linking obesity and cancer: a
framework for population-attributable risk estimations in
Europe. Europ J Cancer. 2010;46(14):2581–2592.
Billewitz WZ, Kemsley WF, Thomson AM. Indices of
adiposity. Br J Prev Soc Med. 1962;16:183–188.
Florey C du V. The use and interpretation of ponderal index
and other weight-height ratios in epidemiological studies.
J Chron Dis. 1970;23(2):93–103.
Khosla T, Lowe CR. Indices of obesity derived from body
weight and height. Brit J Prev Soc Med. 1967;21(3):O122–128.
Benn RT. Some mathematical properties of weight-for-height
indices used as measures of adiposity. Brit J Prev Soc Med.
1971;25(1):42–50.
Lee J, Kolonel LN, Hinds MW. Relative merits of the weightcorrected for height indices. Am J Clin Nutr. 1981;34(11):
2521–2529.
Micozzi MS, Albanes D, Jones DY, et al. Correlations of
body mass indices with weight, stature, and body
composition in men and women in NHANES I and II. Am
J Clin Nutr. 1986;44(6):725–731.
Green J, Cairns BJ, Casabonne D, et al. Height and cancer
incidence in the Million Women Study: prospective cohort,
and meta-analysis of prospective studies of height and total
cancer. Lancet Oncol. 2011;12(8):785–794.
Sung J, Song Y-M, Lawlor DA, et al. Height and site-specific
cancer risk: a cohort study of a Korean adult population. Am
J Epidemiol. 2009;170(1):53–64.
Am J Epidemiol. 2013;177(1):93–101
21. Renehan A. Height and cancer: consistent links,
but mechanisms unclear. Lancet Oncol. 2011;12(8):
716–717.
22. Roberts DL, Dive C, Renehan AG. Biological mechanisms
linking obesity and cancer risk: new perspectives. Annu Rev
Med. 2010;61:301–316.
23. Khandekar MJ, Cohen P, Spiegelman BM. Molecular
mechanisms of cancer devepment in obesity. Nat Rev.
2011;11(12):886–895.
24. Miller AB, Baines CJ, To T, et al. Canadian national breast
screening study 1. Breast cancer detection and death rates
among women aged 40–49 years. Can Med Assoc J.
1992;147(10):1459–1476.
25. Miller AB, Baines CJ, To T, et al. Canadian national breast
screening study 2. Breast cancer detection and death rates
among women aged 50–59 years. Can Med Assoc J.
1992;147(10):1477–1488.
26. Jain M, Harrison L, Howe GR, et al. Evaluation of a selfadministered dietary questionnaire for use in cohort study.
Am J Clin Nutr. 1982;36(5):931–935.
27. Jain M, Howe GR, Rohan T. Dietary assessment in
epidemiology: comparison of a food-frequency and a diet
history questionnaire with a 7-day food record. Am
J Epidemiol. 1996;143(9):953–960.
28. Garn SM, Pesick SD. Comparison of the Benn index and
other body mass indices in nutritional assessment. Am J Clin
Nutr. 1982;36(4):573–575.
29. Revicki DA, Israel RG. Relationship between body mass
indices and measures of body adiposity. Am J Publ Health.
1986;76(8):992–994.
30. Lee J, Kolonel LN, Hinds MW. Relative merits of old and
new indices of body mass: a commentary. Am J Clin Nutr.
1982;36(4):727–728.
31. Larsson I, Henning B, Lindroos AK, et al. Optimized
predictions of absolute and relative amounts of body fat from
weight, height, and other anthropometric predictors, and age.
Am J Clin Nutr. 2006;83(2):252–259.
32. Lee J, Kolonel LN. Are body mass indices interchangeable in
measuring obesity-disease associations? Am J Publ Health.
1984;74(4):376–377.
33. Kabat GC, Kim MY, Thomson CA, et al. Anthropometric
factors and physical activity and risk of thyroid cancer in
postmenopausal women. Cancer Causes Control. 2012;
23(3):421–430.
34. Bianchini F, Kaaks R, Vaino H. Overweight, obesity, and
cancer risk. Lancet Oncol. 2002;3(9):565–574.
35. Wolin KY, Carson K, Colditz GA. Obesity and cancer.
Oncologist. 2010;15(6):556–565.