American Journal of Epidemiology
© The Author 2015. 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. 182, No. 5
DOI: 10.1093/aje/kwv073
Advance Access publication:
July 23, 2015
Original Contribution
Weight Cycling and Cancer Incidence in a Large Prospective US Cohort
Victoria L. Stevens*, Eric J. Jacobs, Alpa V. Patel, Juzhong Sun, Marjorie L. McCullough,
Peter T. Campbell, and Susan M. Gapstur
* Correspondence to Dr. Victoria L. Stevens, Epidemiology Research Program, American Cancer Society, 250 Williams Street, N.W., Atlanta,
GA 30303 (e-mail: [email protected]).
Initially submitted November 21, 2014; accepted for publication March 19, 2015.
Weight cycling, which consists of repeated cycles of intentional weight loss and regain, is common among individuals who try to lose weight. Some evidence suggests that weight cycling may affect biological processes that
could contribute to carcinogenesis, but whether it is associated with cancer risk is unclear. Using 62,792 men
and 69,520 women enrolled in the Cancer Prevention Study II Nutrition Cohort in 1992, we examined the association between weight cycling and cancer incidence. Weight cycles were defined by using baseline questions that
asked the number of times ≥10 pounds (4.54 kg) was purposely lost and later regained. Multivariable-adjusted hazard ratios and 95% confidence intervals for all cancer and 15 individual cancers were estimated by using Cox proportional hazards regression. During up to 17 years of follow-up, 15,333 men and 9,984 women developed cancer.
Weight cycling was not associated with overall risk of cancer in men (hazard ratio = 0.96, 95% confidence interval:
0.83, 1.11 for ≥20 cycles vs. no weight cycles) or women (hazard ratio = 0.96, 95% confidence interval: 0.86, 1.08)
in models that adjusted for body mass index and other covariates. Weight cycling was also not associated with any
individual cancer investigated. These results suggest that weight cycling, independent of body weight, is unlikely to
influence subsequent cancer risk.
body weight changes; cancer risk; obesity
Abbreviations: BMI, body mass index; CI, confidence interval; CPS-II, Cancer Prevention Study II; HR, hazard ratio.
Weight cycling, which is also referred to as “yo-yo dieting,” is the repeated intentional loss of weight followed by
regain. With almost half of American adults attempting to
lose weight (1) and most weight loss not maintained (2),
weight cycling is very common. Weight cyclers tend to be
heavier (3, 4) and to gain more weight over time (3, 5) than
noncyclers. Although both weight and weight gain are known
to increase risk of cancer (6–8), whether weight cycling is
associated with cancer risk, independent of body weight, is
largely unknown (9).
Although few studies have examined the physiological
effects of weight cycling, there is some evidence that it could
affect processes that contribute to carcinogenesis. Studies in
mice have shown that weight cycling increases T-cell accumulation (10) and enhances inflammatory responses (11) in adipose tissue. In a study of postmenopausal women, frequent
intentional weight loss was associated with lower natural killer
cell cytotoxicity (12). Such changes could result in increased
chronic inflammation and decreased immune surveillance for
transformed cells. Animal and human studies suggest that
weight cycling results in increased fat accumulation (13–15),
particularly abdominal or visceral fat (16, 17); alters levels of
glucose, insulin, cholesterol, and some hormones (18–20);
causes impaired systemic glucose tolerance (10); and is associated with shorter telomeres in peripheral blood lymphocytes
(21). However, many of these findings have not been replicated, and at least 2 other studies showed no associations between weight cycling and accumulation of fat mass (22, 23).
The epidemiologic studies of weight cycling in relation to
cancer risk published to date investigated relatively few cancers, and most included only women. French et al. (24) investigated the relationship between weight cycling and several
diseases in the Iowa Women’s Health Study and found no associations with endometrial, breast, colon, or lung cancer.
394
Am J Epidemiol. 2015;182(5):394–404
Weight Cycling and Cancer Incidence 395
Three case-control (25–27) and 1 prospective cohort study
(28) also investigated the association of weight cycling with
endometrial cancer. Two of these studies had null results (25,
28), 1 found a 26% increased risk (26), and the other found
no association in women who had never been obese and an
almost 3-fold increased risk for women who had ever been
obese (27). Two case-control studies of breast cancer found
disparate results, with 1 being null (29) and the other reporting that a fluctuating weight pattern was associated with a
nearly 2-fold increased risk (30). A case-control study of renal
cell cancer found no association for men but a 2.3-fold increased risk for women (31), while a prospective study with
only women found a 2.6-fold increased risk (32). Finally,
weight cycling was not associated with risk of non-Hodgkin
lymphoma, leukemia, or multiple myeloma in a case-control
study of women (33). Thus, the existing evidence does not
clearly indicate whether weight cycling is associated with
risk of any specific cancer (9) and does not address overall
cancer risk.
Because many people try to lose weight (1), understanding
whether weight cycling is associated with cancer risk has important public health implications. If weight cycling is associated with increased cancer risk, then more effort should be
made to encourage the avoidance of weight gain by normal
weight individuals and the maintenance of weight loss by
overweight and obese individuals.
In this study, we investigated the association between weight
cycling and cancer incidence using the American Cancer
Society Cancer Prevention Study II (CPS-II) Nutrition Cohort.
Weight cycling was defined by using self-reported intentional
weight loss and regain episodes collected at baseline in 1992,
and cancer incidence from 1992 to 2009 was used to determine
risk of all cancer and 15 specific cancers in both men and
women.
METHODS
Study population
The participants of this study (86,402 men and 97,785
women) were from the CPS-II Nutrition Cohort, a prospective study of cancer incidence begun in 1992. The Nutrition
Cohort, which is described in detail elsewhere (34), was formed as a subcohort of CPS-II, a prospective study of cancer
mortality involving approximately 1.2 million Americans enrolled in 1982. Nutrition Cohort participants were recruited
from CPS-II participants between the ages of 50 and 74 years
who were living in 21 states. Participants completed a selfadministered questionnaire at baseline in 1992–1993 that included questions on demographic, anthropometric, medical,
and lifestyle information. Follow-up questionnaires to update
exposure information and to ascertain newly diagnosed cancers were sent to living Nutrition Cohort members in 1997 and
every 2 years since. The response rate on all of the follow-up
questionnaires, among those cohort participants who were
mailed surveys, was at least 85%. Follow-up time for this
study was from the date of receipt of the baseline survey in
1992–1993 to the date of diagnosis of cancer, death, date
of the last survey returned, or June 30, 2009, whichever came
first. All aspects of the CPS-II Nutrition Cohort Study were
Am J Epidemiol. 2015;182(5):394–404
approved by the Emory University Institutional Review Board
(Atlanta, Georgia).
CPS-II participants were excluded from this analysis if they
were lost to follow-up (3,141 men and 3,111 women), reported
any prevalent cancer other than nonmelanoma skin cancer
in 1982 or 1992–1993 (9,769 men and 13,094 women), selfreported a cancer that could not be verified (527 men and 354
women), or self-reported a cancer more than 6 months before
the diagnosis date of that cancer (100 men and 48 women).
Also excluded were participants with missing body mass
index (BMI, expressed as weight (kg)/height (m)2) or BMI
<18.5 in 1982 or 1992 (2,161 men and 4,174 women), missing
information on number of times a participant purposely lost
≥10 pounds (4.54 kg), or regained weight that was purposefully lost (1,098 men and 1,866 women), as well as those
who reported never having purposefully lost ≥10 pounds but
reported ≥10 pounds as the most weight ever purposefully lost
(1,070 men and 936 women), and those who reported the
number of times that weight was purposefully lost that differed
from the number of times weight was regained by more than 2
(2,468 men and 3,004 women). The characteristics of the individuals excluded for missing or invalid weight loss and regain
information are shown in Web Table 1 available at http://aje.
oxfordjournals.org/. Finally, because an undiagnosed cancer
at baseline may have influenced weight patterns, the first 2
years of follow-up (1992–1994, 3,276 men and 1,678 women)
were excluded from all analyses. For analyses of endometrial
cancer, women who reported removal of their uterus or had unknown uterine status at baseline were also excluded (27,022
women). Likewise, for ovarian cancer analyses, women who
reported removal of their ovaries or had unknown ovarian status
at baseline were also excluded (15,811 women). Women who
reported removal of their uterus or ovaries during follow-up
were censored from analyses of these cancers at the time of
receipt of the questionnaire on which the procedure was reported. Thus, the final analytical cohort for this study comprised 62,792 men and 69,520 women except for analyses
of endometrial or ovarian cancer, where the analytical cohort
was 42,498 or 53,709 women, respectively.
Identification of cancer cases
A total of 25,317 first cancers occurred in the analytical
cohort between 1994 and 2009. Of these, 24,217 were selfreported on follow-up questionnaires and subsequently verified through medical records (n = 15,911) or linkage with
state cancer registries (n = 8,306). An additional 1,100 cancers
were identified through linkage to the National Death Index.
For analyses of all cancers and specific cancers, all cases
were the first diagnosed cancer for that individual. In situ diseases were not included for breast, prostate, colon, or rectal
cancers. Aggressive prostate cancer was defined as a Gleason
score ≥8, American Joint Committee on Cancer tumor stage
T3 or T4, or lethal prostate cancer of unknown stage.
Characterization of weight cycling
CPS-II Nutrition Cohort participants were asked 2 questions
about weight cycling on the baseline questionnaire: “How
many times in your life have you purposely lost 10 pounds
396 Stevens et al.
Table 1. Age-Standardizeda Frequenciesb of Selected Characteristics of Men by Weight Cycling Status, Cancer
Prevention Study II Nutrition Cohort, 1992
No. of Weight Cycles
Variable
Noncycler
(n = 36,103)
Mean
Age in 1992, years
%
64.5
1–4
(n = 19,466)
Mean
5–9
(n = 3,936)
%
63.1
Mean
%
10–19
(n = 2,430)
Mean
62.0
%
61.6
≥20
(n = 857)
Mean
%
60.9
Race
White
97.3
97.9
98.1
98.3
98.2
Black
1.2
1.0
1.1
0.8
1.4
Other/missing
1.5
1.1
0.8
0.9
0.4
Education
<High school
High school graduate
8.4
6.7
5.8
5.4
6.3
20.2
17.5
15.0
15.5
12.3
Some college
25.2
25.9
24.6
29.1
31.8
≥College graduate
45.6
49.2
53.8
49.3
49.3
0.6
0.7
0.8
0.7
0.3
Never
35.0
29.7
30.0
28.3
31.2
Current
10.6
6.5
6.0
5.9
5.7
Former
53.6
63.0
63.2
65.1
62.4
Missing
0.8
0.8
0.8
0.7
0.7
6.7
11.7
13.9
14.3
15.8
Missing
Smoking status
Diabetes, yes
Alcohol, g/day
Energy, kcal/day
11.5
1,818
11.0
1,776
10.8
1,784
10.8
1,807
11.7
1,872
BMIc at age 18
21.3
22.4
23.3
23.6
24.6
BMIc in 1982
25.0
27.2
28.2
28.8
29.9
BMIc in 1992
25.2
27.7
29.0
29.9
31.0
Exercise, METs/hour
13.2
13.1
13.6
13.9
13.4
Wt Δ between age 18 years
and 1982, poundsd
26.0
33.6
34.9
36.3
37.3
1.7
4.0
5.5
7.7
8.2
Wt Δ between 1982 and
1992, pounds
Abbreviations: BMI, body mass index; METs, physical activity in metabolic equivalent; Wt Δ, weight change.
a
Standardized to the age distribution of the men in the cohort.
b
Values are presented as percentages unless otherwise noted.
c
BMI expressed as weight (kg)/height (m)2.
d
One pound = 0.454 kg.
or more?” and “How many times in your life have you regained as much as 10 pounds that you previously had lost?”
A write-in answer was required for both questions with space
for a range from 0 to 99. A weight cycle was defined as a
combination of both an intentional loss and a subsequent
regain. Thus, someone who reported 3 weight losses and
3 subsequent regains was assigned 3 weight cycles, whereas someone who reported 3 weight losses but only 2 regains
was assigned 2 weight cycles. The individuals who reported weight cycling (“weight cyclers”) were further categorized by the total number of cycles they reported: 1–4, 5–9,
10–19, and 20 or more weight cycles. Weight change pattern
or trajectory was not considered in assigning weight cycling
status or category.
Statistical analyses
BMI in 1982 and 1992 was calculated by using self-reported
body weight on questionnaires completed at those times and
height self-reported on the 1982 questionnaire. BMI at age 18
years was calculated by using weight at age 18 years recalled
and reported on the 1992 questionnaire.
All statistical analyses were conducted with SAS, version
9.3, software (SAS Institute, Inc., Cary, North Carolina). Cox
proportional hazards regression (35) was used to determine
the hazard ratios and 95% confidence intervals for the association of weight cycling with cancer incidence. The time scale
for the analysis was time in study from cohort entry in 1992–
1993. P values for linear trend were estimated by modeling
Am J Epidemiol. 2015;182(5):394–404
Weight Cycling and Cancer Incidence 397
Table 2. Age-Standardizeda Frequenciesb of Selected Characteristics of Women by Weight Cycling Status, Cancer
Prevention Study II Nutrition Cohort, 1992
No. of Weight Cycles
Noncycler
(n = 29,879)
Variable
Mean
Age in 1992, years
%
63.0
1–4
(n = 25,559)
Mean
5–9
(n = 6,663)
%
Mean
61.7
%
60.7
10–19
(n = 4,990)
Mean
%
60.2
≥20
(n = 2,429)
Mean
%
59.7
Race
White
97.1
97.6
97.9
98.0
98.1
Black
1.3
1.4
1.4
1.5
1.1
Other/missing
1.6
1.0
0.7
0.5
0.8
Education
<High school
High school graduate
4.8
4.7
4.6
5.0
5.6
31.7
32.0
32.0
30.7
27.6
Some college
29.6
31.7
32.3
34.1
38.1
≥College graduate
33.3
30.9
30.5
29.6
27.8
0.6
0.7
0.6
0.6
0.9
47.3
Missing
Smoking status
Never
56.8
54.6
52.5
51.1
Current
9.4
7.3
6.6
6.7
7.2
Former
32.3
36.6
39.5
40.9
44.2
Missing
1.5
1.5
1.4
1.3
1.3
4.1
6.6
8.8
10.8
12.9
Never
42.5
42.5
42.5
42.7
42.4
Current ERT
18.8
19.4
19.3
20.6
17.8
Diabetes, yes
PMH use in 1992
Current CHRT
13.7
11.3
10.2
10.2
10.4
Former ERT/CHRT
17.8
19.7
21.1
19.9
22.1
Ever/other/missing
7.2
Alcohol, g/day
Energy, kcal/day
5.4
1,357
7.1
4.4
1,359
6.9
4.0
1,385
6.6
3.7
1,393
7.3
3.5
1,451
BMIc at age 18
19.9
20.9
21.7
22.3
23.0
BMIc in 1982
22.5
25.1
26.9
27.9
29.1
BMIc in 1992
23.3
26.5
28.7
29.9
31.3
Exercise, METs/hour
12.3
11.8
11.6
11.7
12.1
Wt Δ between age 18 and
1982, poundsd
15.2
24.5
30.9
33.3
36.3
Wt Δ between 1982 and
1992, pounds
5.1
8.5
10.5
11.7
13.0
Abbreviations: BMI, body mass index; CHRT, combined hormone replacement therapy; ERT, estrogen replacement
therapy; METs, physical activity in metabolic equivalents; PMH, postmenopausal hormone; Wt Δ, weight change.
a
Standardized to the age distribution of the women in the cohort.
b
Values are presented as percentages unless otherwise noted.
c
BMI expressed as weight (kg)/height (m)2.
d
One pound = 0.454 kg.
weight cycles as a continuous variable and were derived by
use of the Wald test (36). The Cox proportional hazards assumption was tested by modeling multiplicative interaction
terms between weight cycling numbers and time. The statistical significance of the interaction term was assessed by
using the likelihood ratio test (36), and no deviations from
proportional hazards were found. Interactions between either
Am J Epidemiol. 2015;182(5):394–404
BMI or age at baseline and weight cycling were tested by
using the likelihood ratio test in which either variable was
modeled categorically. All P values are 2 sided.
Cox models were adjusted for age by stratifying on exact
year of age in 1992. Additional covariates included in the
multivariable-adjusted models were race (white, black, or
other/missing), education (some high school, high school
398 Stevens et al.
Table 3. Association of Weight Cycling With Cancer Incidence in Men and Women Combined, Cancer Prevention
Study II Nutrition Cohort, 1992–2009
Cancer Type and
No. of Weight Cycles
Total No.
No. of Cases
HRa
95% CI
HRb
0
65,982
13,101
1–4
45,025
8,423
1.00
Referent
1.00
Referent
1.02
0.99, 1.05
0.99
0.96, 1.02
5–9
95% CI
All cancer
10,599
1,930
1.05
1.00, 1.10
1.00
0.95, 1.05
10–19
7,420
1,309
1.07
1.01, 1.13
1.01
0.95, 1.07
≥20
3,286
554
1.07
0.98, 1.16
0.99
0.91, 1.08
P for trend
0.02
0.93
Colon cancerc
0
65,982
872
1.00
Referent
1.00
Referent
1–4
45,025
608
1.12
1.01, 1.24
1.05
0.94, 1.17
5–9
10,599
123
1.04
0.86, 1.26
0.93
0.76, 1.14
10–19
7,420
85
1.08
0.86, 1.35
0.94
0.74, 1.19
≥20
3,286
32
0.95
0.66, 1.35
0.80
0.55, 1.16
P for trend
0.53
0.53
Rectal cancerc
0
65,982
259
1.00
Referent
1.00
Referent
1–4
45,025
146
0.89
0.73, 1.10
0.84
0.67, 1.04
5–9
10,599
44
1.21
0.88, 1.68
1.09
0.77, 1.54
≥10
10,706
29
0.82
0.56, 1.22
0.72
P for trend
0.27
0.47, 1.09
0.11
Pancreatic cancer
0
65,982
263
1.00
Referent
1.00
Referent
1–4
45,025
202
1.23
1.02, 1.49
1.09
0.89, 1.33
5–9
10,599
43
1.20
0.87, 1.67
0.98
0.69, 1.38
≥10
10,706
40
1.18
0.84, 1.67
0.90
P for trend
0.31
062, 1.30
0.65
Renal cancer
0
65,982
207
1.00
Referent
1.00
Referent
1–4
45,025
167
1.25
1.02, 1.54
1.12
0.90, 1.39
5–9
10,599
45
1.51
1.09, 2.10
1.25
0.89, 1.77
≥10
10,706
44
1.55
1.10, 2.16
1.20
P for trend
0.05
0.83, 1.74
0.68
Esophageal cancer
0
65,982
102
1.00
Referent
1.00
Referent
1–4
45,025
68
1.18
0.86, 1.61
1.15
0.83, 1.61
≥5
21,305
28
1.25
0.81, 1.92
1.21
P for trend
0.57
0.75, 1.94
0.76
Table continues
graduate, some college, college graduate, or missing), smoking status (nonsmoker, former <20 years, former ≥20 years,
current <40 years, current ≥40 years, ever/unclassifiable, or
missing), physical activity in metabolic equivalents (<8, 8 to
<17.5, 17.5 to <31.5, ≥31.5/week, or missing), use of aspirin
or other nonsteroidal antiinflammatory drugs (not current user, 1–14 pills/month, 15–29 pills/month, 30–59 pills/month,
≥60 pills/month, or missing/unknown), and history of diabetes
(yes, no). For analyses of women, postmenopausal hormone
use (never, current estrogen, current combined hormone replacement therapy, former estrogen or combined hormone replacement therapy, or ever/other/unknown) was also included
in the model. Included in analyses of prostate, breast, and colon and rectal cancer were history of prostate-specific antigen
testing (starting in 1997), mammography (starting in 1992),
and colorectal endoscopy (starting in 1997), respectively, and
Am J Epidemiol. 2015;182(5):394–404
Weight Cycling and Cancer Incidence 399
Table 3. Continued
Cancer Type and
No. of Weight Cycles
Total No.
No. of Cases
HRa
95% CI
HRb
95% CI
Liver cancer
0
65,982
77
1.00
Referent
1.00
Referent
1–4
45,025
58
1.24
0.88, 1.76
0.99
0.68, 1.42
≥5
21,305
21
1.08
0.66, 1.78
0.69
0.40, 1.20
P for trend
0.82
0.30
Non-Hodgkin lymphoma
0
65,982
744
1.00
Referent
1.00
Referent
1–4
45,025
519
1.08
0.96, 1.21
1.01
0.89, 1.14
5–9
10,599
109
1.00
0.82, 1.23
0.90
0.73, 1.11
10–19
7,420
76
1.04
0.82, 1.32
0.91
0.70, 1.17
≥20
3,286
37
1.19
0.85, 1.66
1.00
0.70, 1.42
P for trend
0.23
0.99
Multiple myeloma
0
65,982
146
1.00
Referent
1.00
Referent
1–4
45,025
102
1.09
0.84, 1.41
1.05
0.80, 1.38
5–9
10,599
26
1.24
0.81, 1.90
1.18
0.76, 1.85
≥10
10,706
20
1.00
0.62, 1.62
0.94
P for trend
0.19
0.56, 1.57
0.31
Lung cancer
0
65,982
1,301
1.00
Referent
1.00
Referent
1–4
45,025
696
0.90
0.82, 0.99
0.94
0.85, 1.04
5–9
10,599
142
0.85
0.72, 1.02
0.91
0.76, 1.10
10–19
7,420
92
0.82
0.66, 1.01
0.89
0.71, 1.11
≥20
3,286
49
1.00
0.75, 1.33
1.11
0.82, 1.50
P for trend
0.30
0.97
Melanoma
0
65,982
813
1.00
Referent
1.00
Referent
1–4
45,025
552
1.03
0.93, 1.15
1.07
0.95, 1.20
5–9
10,599
140
1.14
0.95, 1.36
1.20
0.99, 1.46
10–19
7,420
94
1.15
0.92, 1.42
1.23
0.98, 1.55
≥20
3,286
25
0.72
0.48, 1.07
0.79
0.52, 1.19
P for trend
0.65
0.97
Stomach cancer
0
65,982
134
1.00
Referent
1.00
Referent
1–4
45,025
72
0.97
0.72, 1.30
0.98
0.72, 1.34
≥5
21,305
27
0.92
0.60, 1.42
0.95
0.59, 1.51
P for trend
0.89
0.99
Abbreviations: CI, confidence interval; HR, hazard ratio.
a
Base model: adjusted for age, race, education, smoking status, physical activity, nonsteroidal antiinflammatory
drug use, history of diabetes, and postmenopausal hormone use.
b
Base model plus body mass index expressed as weight (kg)/height (m)2 in 1992 (continuous).
c
Additionally adjusted for colorectal endoscopy starting in 1997 (time dependent).
all were modeled as time-dependent variables. BMI in 1992
(continuous, modeled linearly) was also included in the models as indicated. The inclusion of alcohol intake as a covariate
did not change the associations and was not included in the
final multivariable-adjusted model.
Am J Epidemiol. 2015;182(5):394–404
RESULTS
A large proportion of CPS-II Nutrition Cohort participants
reported some weight cycling. More women (57%) than men
(43%) weight cycled, and women were more likely to report
higher numbers of weight cycles than men.
400 Stevens et al.
A)
B)
HR (95% CI)
Cancer Type No. of Cases
All
Prostate
Aggressive
Prostate
15,333
1.01 (0.99, 1.02)
7,152
0.99 (0.96, 1.02)
1,403
1.02 (0.97, 1.08)
Colon
911
0.99 (0.92, 1.08)
Rectal
288
0.97 (0.83, 1.13)
Pancreatic
294
1.08 (0.99, 1.18)
Renal
288
1.07 (0.97, 1.17)
Esophageal
163
0.96 (0.79, 1.18)
Liver
112
0.92 (0.71, 1.20)
821
1.05 (0.99, 1.12)
176
1.00 (0.86, 1.18)
1,393
0.97 (0.90, 1.04)
Melanoma
986
1.01 (0.94, 1.08)
Stomach
162
1.07 (0.94, 1.22)
Non-Hodgkin
Lymphoma
Multiple
Myeloma
Lung
0.5
1.0
2.0
HR
Cancer Type
No. of Cases
HR (95% CI)
15,333
1.00 (0.98, 1.02)
Prostate
7,152
1.00 (0.97, 1.03)
Aggressive
Prostate
1,403
1.01 (0.96, 1.07)
Colon
911
0.94 (0.86, 1.03)
Rectal
288
0.95 (0.81, 1.12)
Pancreatic
294
1.02 (0.92, 1.14)
Renal
288
1.01 (0.90, 1.13)
Esophageal
163
0.90 (0.71, 1.14)
Liver
112
0.81 (0.59, 1.12)
821
1.03 (0.97, 1.11)
176
0.97 (0.81, 1.16)
1,393
0.98 (0.91, 1.05)
Melanoma
986
1.01 (0.94, 1.08)
Stomach
162
1.08 (0.94, 1.23)
All
Non-Hodgkin
Lymphoma
Multiple
Myeloma
Lung
0.5
1.0
2.0
HR
Figure 1. Association of weight cycling with cancer incidence in men in the Cancer Prevention Study II Nutrition Cohort. Hazard ratios (HRs) and
95% confidence intervals (CIs) are shown for each cancer for analyses adjusted for age, race, education, smoking status, physical activity, nonsteroidal antiinflammatory drug use, and history of diabetes (A) and all these variables plus body mass index expressed as weight (kg)/height (m)2 in
1992 (continuous) (B). The HR is the association per 5 weight cycles determined by using weight cycles as a continuous variable. All analyses for
prostate cancer are additionally adjusted for prostate-specific antigen screening starting in 1997 (time dependent), and all analyses for colon and
rectal cancer are additionally adjusted for colorectal endoscopy starting in 1997 (time dependent).
Weight cyclers tended to be younger, were more likely to be
former smokers and to have a history of diabetes, and were less
likely to be current smokers than noncyclers (Table 1 (men) and
Table 2 (women)). Weight cyclers also had a higher BMI at age
18, in 1982, and in 1992 and gained more weight in the intervals between age 18 and 1982 and between 1982 and 1992 than
did noncyclers. Women weight cyclers were also more likely to
be former postmenopausal hormone users and less likely to be
current postmenopausal hormone users and to have a higher energy intake and a lower alcohol consumption.
The associations of weight cycling with risk of all cancer
and 11 individual cancers for men and women combined are
shown in Table 3. The associations for men and women separately for these cancers plus 4 sex-specific cancers are shown
in Web Table 2. In men and women combined and in women
only, weight cycling was associated with overall cancer risk
before but not after adjustment for BMI at baseline in 1992.
In men alone, weight cycling was not associated with overall
cancer risk in either model.
The association between weight cycling and prostate cancer
risk, which accounted for almost 47% of all cancers in men,
was null, regardless of whether BMI was included in the
model. No association was observed when only aggressive
prostate cancers were examined. The associations with postmenopausal breast and endometrial cancers were similar to
those for overall cancer risk in women, where a positive association was observed before but not after adjustment for BMI.
There was no association between weight cycling and ovarian
cancer risk either prior to or after adjustment for BMI.
Weight cycling was not statistically significantly associated with a number of other cancers, including colon, rectal,
pancreatic, renal, esophageal, liver, lung, and stomach cancers, non-Hodgkin lymphoma, multiple myeloma, and melanoma, in both men and women (Table 2 and Web Table 2).
The associations with cancer risk determined by using
weight cycling as a continuous variable before and after adjustment for BMI in men and women separately are shown in
Figures 1 and 2. The only statistically significant association
after adjustment for BMI was for esophageal cancer in
women. However, this analysis included only 35 cases (20
weight cycling cases).
Because a recent analysis of weight cycling and endometrial cancer found a statistically significant interaction with
history of obesity (BMI <30) (27), we examined associations
stratified by obesity status. “Never obese” individuals were
those whose BMI was <30 at age 18, in 1982, and in 1992,
Am J Epidemiol. 2015;182(5):394–404
Weight Cycling and Cancer Incidence 401
A)
B)
No. of Cases
HR (95% CI)
All
9,984
1.00 (0.98, 1.01)
1.02 (1.00, 1.04)
Breast
3,485
0.99 (0.97, 1.02)
604
1.06 (1.01, 1.10)
Endometrial
604
0.95 (0.89, 1.01)
Ovarian
359
1.02 (0.95, 1.09)
Ovarian
359
1.03 (0.96, 1.11)
Colon
809
1.02 (0.97, 1.07)
Colon
809
1.01 (0.95, 1.06)
Rectal
190
0.93 (0.81, 1.07)
Rectal
190
0.88 (0.75, 1.04)
Pancreatic
254
1.00 (0.91, 1.10)
Pancreatic
254
0.96 (0.85, 1.07)
Renal
175
1.05 (0.97, 1.14)
Renal
175
1.02 (0.92, 1.12)
Esophageal
35
1.11 (0.97, 1.27)
Esophageal
35
1.16 (1.02, 1.31)
Liver
44
4
1.08 (0.93, 1.26)
Liver
44
4
0.97 (0.77, 1.21)
664
1.01 (0.96, 1.06)
664
0.97 (0.91, 1.14)
118
1.08 (0.99, 1.18)
118
1.09 (0.99, 1.19)
Lung
887
0.98 (0.94, 1.04)
Lung
887
1.02 (0.97, 1.07)
Melanoma
638
0.98 (0.92, 1.04)
Melanoma
638
0.99 (0.93, 1.06)
71
0.81 (0.57, 1.13)
Stomach
71
0.82 (0.57, 1.18)
Cancer Type
No. of Cases
HR (95% CI)
All
9,984
1.02 (1.00, 1.03)
Breast
3,485
Endometrial
Non-Hodgkin
Lymphoma
Multiple
Myeloma
Stomach
0.5
1.0
HR
Cancer Type
Non-Hodgkin
Lymphoma
Multiple
Myeloma
2.0
0.5
1.0
HR
2.0
Figure 2. Association of weight cycling with cancer incidence in women in the Cancer Prevention Study II Nutrition Cohort. Hazard ratios (HRs)
and 95% confidence intervals (CIs) are shown for each cancer for analyses adjusted for age, race, education, smoking status, physical activity,
nonsteroidal antiinflammatory drug use, history of diabetes, and postmenopausal hormone use (A) and all these variables plus body mass
index expressed as weight (kg)/height (m)2 in 1992 (continuous) (B). The HR is the association per 5 weight cycles determined by using weight
cycles as a continuous variable. All analyses for breast cancer are additionally adjusted for mammography starting in 1992 (time dependent), and all
analyses for colon and rectal cancer are additionally adjusted for colorectal endoscopy starting in 1997 (time dependent).
whereas “ever obese” individuals reported having a BMI ≥30
at 1 or more of these time points. The results of these analyses
for overall cancer risk, the most common cancers in men and
women ( prostate and postmenopausal breast), kidney cancer
in men, and endometrial cancer in women are shown in Figures 3 (men) and 4 (women). These associations, as well as
those for all other cancers (Web Table 3), were null regardless
of history of obesity. Although the point estimates were
above 1 for some cancers among the ever obese and primarily
less than or equal to 1 for the never obese, the interaction was
not statistically significant for any cancer in either sex.
Whether the association of weight cycling with cancer risk
varied by age was investigated in stratified analyses with either younger or older participants defined relative to the median age of the study population (64 years for men and 62
years for women). In older women aged ≥62 years, there
was a statistically significant positive trend between number
of cycles and ovarian cancer (for 1–4 cycles: hazard ratio
(HR) = 1.02, 95% confidence interval (CI): 0.72, 1.44; for
5–9 cycles: HR = 1.89, 95% CI: 1.15, 3.13; for ≥10 cycles:
HR = 2.07, 95% CI: 1.23, 3.50) (P for trend = 0.03), whereas
Am J Epidemiol. 2015;182(5):394–404
the association was null in younger women aged <62 years
(for 1–4 cycles: HR = 0.88, 95% CI: 0.60, 1.29; for 5–9 cycles: HR = 1.13, 95% CI: 0.66, 1.95; for ≥10 cycles: HR =
0.68, 95% CI: 0.36, 1.28) (P for trend = 0.41). However,
the age-specific associations of weight cycling with ovarian
cancer risk were not statistically significantly different (P for
interaction = 0.12). All other associations with weight cycling in younger and older men and women were null (data
not shown).
Although smoking status was controlled for in analyses of
weight cycling and cancer risk, it is possible that residual
confounding by smoking dose could influence our findings.
Therefore, in sensitivity analyses, the study population was
limited to never and long-term former (≥20 years) smokers.
All associations were null and similar to those obtained with
the full study population (data not shown).
DISCUSSION
This study is the first to comprehensively investigate the
associations of weight cycling with overall and site-specific
402 Stevens et al.
B)
A)
Cancer Type
and No. of
Cycles
No. of
Cases
Total No.
All
0
1–4
5–9
10–19
≥20
8,263
3,644
568
328
73
Prostate
0
1–4
5–9
10–19
≥20
Renal
0
1–4
5–9
≥10
HR (95% CI)
Cancer Type
and No. of
Cycles
No. of
Cases
Total No.
HR (95% CI)
33,589
14,504
2,414
1,281
337
1.00 (Referent)
1.01 (0.97, 1.06)
0.96 (0.88, 1.04)
1.07 (0.96, 1.20)
0.87 (0.69, 1.09)
All
0
1–4
5–9
10–19
≥20
588
1,144
358
282
115
2,514
4,962
1,522
1,149
520
1.00 (Referent)
1.06 (0.95, 1.17)
1.10 (0.96, 1.26)
1.18 (1.02, 1.37)
1.10 (0.90, 1.36)
3,941
1,732
266
164
31
33,589
14,504
2,414
1,281
337
1.00 (Referent)
1.00 (0.94, 1.06)
0.91 (0.80, 1.03)
1.11 (0.94, 1.30)
0.75 (0.52, 1.07)
Prostate
0
1–4
5–9
10–19
≥20
224
461
160
123
50
2,514
4,962
1,522
1,149
520
1.00 (Referent)
1.03 (0.87, 1.21)
1.16 (0.94, 1.43)
1.20 (0.95, 1.51)
1.11 (0.81, 1.53)
136
70
11
8
33,589
14,504
2,414
1,618
1.00 (Referent)
1.03 (0.76, 1.40)
0.99 (0.53, 1.86)
1.04 (0.50, 2.17)
Renal
0
1–4
5–9
≥10
12
33
8
10
2,514
4,962
1,522
1,669
1.00 (Referent)
1.37 (0.70, 2.70)
1.04 (0.41, 2.62)
1.24 (0.51, 3.01)
0.5
1.0
HR
2.0
0.5 1.0 2.0
HR
Figure 3. Association of weight cycling with overall, prostate, and renal cancer risk in men stratified by obesity status in the Cancer Prevention
Study II Nutrition Cohort, 1992. Obesity status: A) never obese, body mass index <30; B) ever obese, body mass index ≥30. Hazard ratios (HRs) and
95% confidence intervals (CIs) were determined in analyses for age, race, education, smoking status, physical activity, nonsteroidal antiinflammatory
drug use, history of diabetes, and body mass index expressed as weight (kg)/height (m)2 in 1992 (continuous).
B)
A)
Cancer Type
and No. of
Cycles
No. of
Cases
Total No.
HR (95% CI)
All
0
1–4
5–9
10–19
≥20
4,089
2,890
591
344
149
28,602
20,793
4,242
2,663
1,043
1.00 (Referent)
0.96 (0.91, 1.01)
0.96 (0.88, 1.06)
0.90 (0.80, 1.01)
0.98 (0.83, 1.16)
Breast
0
1–4
5–9
10–19
≥20
1,399
1,021
196
124
54
28,602
20,793
4,242
2,663
1,043
189
171
28
19
5
18,237
12,687
2,483
1,480
625
Endometrial
0
1–4
5–9
10–19
≥20
0.5
1.0
HR
2.0
Cancer Type
and No. of
Cycles
No. of
Cases
Total No.
HR (95% CI)
All
0
1–4
5–9
10–19
≥20
191
745
413
355
217
1,277
4,766
2,421
2,327
1,386
1.00 (Referent)
1.06 (0.90, 1.25)
1.18 (0.99, 1.40)
1.04 (0.87, 1.25)
1.05 (0.86, 1.29)
1.00 (Referent)
0.95 (0.87, 1.03)
0.87 (0.74, 1.02)
0.87 (0.72, 1.06)
0.97 (0.73, 1.28)
Breast
0
1–4
5–9
10–19
≥20
60
271
142
138
80
1,277
4,766
2,421
2,327
1,386
1.00 (Referent)
1.20 (0.90, 1.59)
1.24 (0.91, 1.68)
1.24 (0.91, 1.70)
1.17 (0.83, 1.65)
1.00 (Referent)
1.14 (0.91, 1.44)
0.86 (0.57, 1.31)
1.03 (0.63, 1.69)
0.65 (0.26, 1.60)
Endometrial
0
1–4
5–9
10–19
≥20
14
72
46
39
21
795
2,744
1,377
1,285
785
1.00 (Referent)
1.39 (0.78, 2.48)
1.65 (0.90, 3.04)
1.44 (0.77, 2.71)
1.27 (0.63, 2.55)
0.5
1.0
3.0
HR
Figure 4. Association of weight cycling with overall, breast, and endometrial cancer risk in women stratified by obesity status in the Cancer Prevention Study II Nutrition Cohort, 1992. Obesity status: A) never obese, body mass index <30; B) ever obese, body mass index ≥30. Hazard ratios
(HRs) and 95% confidence intervals (CIs) were determined in analyses for age, race, education, smoking status, physical activity, nonsteroidal
antiinflammatory drug use, history of diabetes, postmenopausal hormone use, and body mass index expressed as weight (kg)/height (m)2 in 1992
(continuous).
Am J Epidemiol. 2015;182(5):394–404
Weight Cycling and Cancer Incidence 403
cancer risk in men and women. Results of this large prospective study showed that weight cycling was not associated with
overall cancer risk or risk of 15 specific cancers independent
of BMI in men or in women.
Our null findings agree with the results of some (24, 25, 28,
29, 33), but not all (26, 27, 30–32), previous studies of weight
cycling and risk of specific cancers. Two factors that likely
contribute to inconsistencies across studies include variability in how body size was accounted for in statistical models
and differences in how weight cycling was defined. The importance of adjusting for BMI can be seen by comparing our
results for all cancers, endometrial cancer, and renal cancer in
women before and after this adjustment (Figure 2 and Web
Table 2). In these analyses, weight cycling was statistically
significantly associated with increased risk before adjustment
for BMI but not afterwards. Potential confounding by BMI
was not considered in the case-control study that found that
weight cycling was associated with a statistically significantly 2.1-fold higher risk of breast cancer (30). Body size
was adjusted for somewhat differently by 2 of the other
case-control studies that reported significant results (26, 32).
Although BMI as a continuous variable was used in our
study, Trentham-Dietz et al. (26) used quartiles of BMI and
Luo et al. (32) used the waist-to-hip ratio in their analyses of
the associations of weight cycling and risk of endometrial or
renal cancer, respectively. It is possible that residual confounding might account for the positive findings in these
studies.
The lack of a standardized definition of weight cycling has
led to considerable variability in the classification of weight
cyclers among studies. The intentionality of the weight loss
is particularly important to consider because unintentional
weight loss occurs with some unhealthy behaviors such as
smoking and is also an early symptom of many diseases, including several cancers. Therefore, it may be strongly related
to the disease or mortality endpoint. Thus, only intentional
weight loss should be considered when defining weight cyclers. However, many studies defined weight cyclers on the
basis of weight patterns or variability over time, regardless
of intentionality. Such studies have reported significant associations for breast (30) and renal (32) cancer. Other features
of the weight cycling criteria that often differ between studies
are the magnitude of the weight cycle and the period over
which the weight cycling occurred. In the case-control study
that found a significant association between weight cycling
and risk of renal cancer in women but not men (31), a weight
cycle was defined as 3 or more instances of intentional weight
loss of >5 kg for women and >10 kg for men. Lindblad et al.
(31) suggest that this difference may account for their disparate findings by sex. However, we defined a weight cycle as
≥10 pounds, which was similar to the >5 kg criteria for women, and our results in both men and women were null (31).
Whether consideration of weight cycles of a larger magnitude
would alter our findings is not known.
The possibility that the association between weight cycling
and endometrial cancer risk might differ by obesity status was
suggested by an Australian case-control study that found no
association among never-obese women and a statistically significant 2.8-fold higher risk among ever-obese women (27).
In our study, there were modest, but not statistically significant,
Am J Epidemiol. 2015;182(5):394–404
differences between never-obese and ever-obese women and
men for some cancers, including endometrial cancer. Whereas 90% of the weight cycling cases in the Australian casecontrol study fell into the ever-obese category, only 45% of
the endometrial cancer cases in our study were ever obese.
Thus, it is possible that our ability to detect an association
may have been limited by the number of ever-obese individuals. Further research in populations with more obese subjects is needed to better understand the association between
weight cycling and cancer risk in this group.
To our knowledge, this study is the largest and most comprehensive study of weight cycling and cancer risk conducted
to date. The size of the study and the large number of weight
cyclers enabled the examination of dose-response relationships. Other strengths include the prospective nature of the
study and the availability of information about important covariates, which were needed to separate the influence of these
factors on cancer risk from those of weight cycling.
An important limitation to this study was the lack of information on when during adulthood the weight cycling occurred. Also absent was information on the magnitude of
the weight cycling or information on weight cycles of more
than 10 pounds. Thus, we were unable to assess whether
specific exposure periods or amplitude of gain/loss was associated with cancer risk. An additional limitation was the number of ever-obese individuals, which limited the power of our
analyses in this group. Small sample sizes in analyses stratified by either obesity or age for some of the rarer cancers also
limit our ability to determine whether our null results reflect a
true lack of association or difference between associations.
The limited number of esophageal cancer cases also precluded investigation of subtypes of this cancer. Finally, because weight cycling was self-reported, there may be some
misclassification due to incorrect responses.
In summary, the results of this study indicate that weight
cycling, independent of BMI, is not associated with overall
cancer risk or risk of any specific cancer in either men or
women. Therefore, overweight and obese individuals should
be encouraged to attempt to lose weight even though the
weight lost may be regained.
ACKNOWLEDGMENTS
Author affiliations: Epidemiology Research Program, American Cancer Society, Atlanta, Georgia (Victoria L. Stevens, Eric
J. Jacobs, Alpa V. Patel, Juzhong Sun, Marjorie L. McCullough,
Peter T. Campbell, Susan M. Gapstur).
The American Cancer Society funds the creation, maintenance, and updating of the Cancer Prevention Study II cohort.
We thank the Cancer Prevention Study II Management
Group for their invaluable contributions to this research.
We would also like to acknowledge the contribution to this
study from central cancer registries supported through the
National Program of Cancer Registries of the Centers for Disease Control and Prevention and cancer registries supported
by the Surveillance, Epidemiology, and End Results Program
of the National Cancer Institute.
Conflict of interest: none declared.
404 Stevens et al.
REFERENCES
1. Yaemsiri S, Slining MM, Agarwal SK. Perceived weight status,
overweight diagnosis, and weight control among US adults: the
NHANES 2003–2008 Study. Int J Obes (Lond). 2011;35(8):
1063–1070.
2. Kraschnewski JL, Boan J, Esposito J, et al. Long-term weight
loss maintenance in the United States. Int J Obes (Lond). 2010;
34(11):1644–1654.
3. Stevens VL, Jacobs EJ, Sun J, et al. Weight cycling and
mortality in a large prospective US study. Am J Epidemiol.
2012;175(8):785–792.
4. Field AE, Malspeis S, Willett WC. Weight cycling and
mortality among middle-aged or older women. Arch Intern
Med. 2009;169(9):881–886.
5. Kroke A, Liese AD, Schulz M, et al. Recent weight changes and
weight cycling as predictors of subsequent two year weight
change in a middle-aged cohort. Int J Obes Relat Metab Disord.
2002;26(3):403–409.
6. Renehan AG, 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.
7. Bhaskaran K, Douglas I, Forbes H, et al. Body-mass index and
risk of 22 specific cancers: a population-based cohort study of
5.24 million UK adults. Lancet. 2014;384(9945):755–765.
8. Wolin KY, Carson K, Colditz GA. Obesity and cancer.
Oncologist. 2010;15(6):556–565.
9. Thompson HJ, McTiernan A. Weight cycling and cancer:
weighing the evidence of intermittent caloric restriction and
cancer risk. Cancer Prev Res (Phila). 2011;4(11):1736–1742.
10. Anderson EK, Gutierrez DA, Kennedy A, et al. Weight cycling
increases T-cell accumulation in adipose tissue and impairs
systemic glucose tolerance. Diabetes. 2013;62(9):3180–3188.
11. Barbosa-da-Silva S, Fraulob-Aquino JC, Lopes JR, et al.
Weight cycling enhances adipose tissue inflammatory
responses in male mice. PLoS One. 2012;7(7):e39837.
12. Shade ED, Ulrich CM, Wener MH, et al. Frequent intentional
weight loss is associated with lower natural killer cell
cytotoxicity in postmenopausal women: possible long-term
immune effects. J Am Diet Assoc. 2004;104(6):903–912.
13. Dankel SN, Degerud EM, Borkowski K, et al. Weight cycling
promotes fat gain and altered clock gene expression in adipose
tissue in C57BL/6J mice. Am J Physiol Endocrinol Metab.
2014;306(2):E210–E224.
14. Benini ZL, Camilloni MA, Scordato C, et al. Contribution of
weight cycling to serum leptin in human obesity. Int J Obes
Relat Metab Disord. 2001;25(5):721–726.
15. Beavers KM, Lyles MF, Davis CC, et al. Is lost lean mass from
intentional weight loss recovered during weight regain in
postmenopausal women? Am J Clin Nutr. 2011;94(3):767–774.
16. Cereda E, Malavazos AE, Caccialanza R, et al. Weight cycling
is associated with body weight excess and abdominal fat
accumulation: a cross-sectional study. Clin Nutr. 2011;30(6):
718–723.
17. Wallner SJ, Luschnigg N, Schnedl WJ, et al. Body fat
distribution of overweight females with a history of weight
cycling. Int J Obes Relat Metab Disord. 2004;28(9):
1143–1148.
18. Hooper LE, Foster-Schubert KE, Weigle DS, et al. Frequent
intentional weight loss is associated with higher ghrelin and
lower glucose and androgen levels in postmenopausal women.
Nutr Res. 2010;30(3):163–170.
19. Olson MB, Kelsey SF, Bittner V, et al. Weight cycling and
high-density lipoprotein cholesterol in women: evidence of an
adverse effect: a report from the NHLBI-sponsored WISE
Study. Women’s Ischemia Syndrome Evaluation Study Group.
J Am Coll Cardiol. 2000;36(5):1565–1571.
20. Sea MM, Fong WP, Huang Y, et al. Weight cycling-induced
alteration in fatty acid metabolism. Am J Physiol Regul Integr
Comp Physiol. 2000;279(3):R1145–R1155.
21. Kim S, Parks CG, DeRoo LA, et al. Obesity and weight gain in
adulthood and telomere length. Cancer Epidemiol Biomarkers
Prev. 2009;18(3):816–820.
22. Hensrud DD, Weinsier RL, Darnell BE, et al. A prospective
study of weight maintenance in obese subjects reduced to
normal body weight without weight-loss training. Am J Clin
Nutr. 1994;60(5):688–694.
23. Bosy-Westphal A, Schautz B, Lagerpusch M, et al. Effect of
weight loss and regain on adipose tissue distribution,
composition of lean mass and resting energy expenditure in
young overweight and obese adults. Int J Obes (Lond). 2013;
37(10):1371–1377.
24. French SA, Folsom AR, Jeffery RW, et al. Weight variability
and incident disease in older women: the Iowa Women’s Health
Study. Int J Obes Relat Metab Disord. 1997;21(3):217–223.
25. Swanson CA, Potischman N, Wilbanks GD, et al. Relation of
endometrial cancer risk to past and contemporary body size and
body fat distribution. Cancer Epidemiol Biomarkers Prev.
1993;2(4):321–327.
26. Trentham-Dietz A, Nichols HB, Hampton JM, et al. Weight
change and risk of endometrial cancer. Int J Epidemiol. 2006;
35(1):151–158.
27. Nagle CM, Marquart L, Bain CJ, et al. Impact of weight change
and weight cycling on risk of different subtypes of endometrial
cancer. Eur J Cancer. 2013;49(12):2717–2726.
28. Stevens VL, Jacobs EJ, Sun J, et al. Weight cycling and risk of
endometrial cancer. Cancer Epidemiol Biomarkers Prev. 2012;
21(5):747–752.
29. Trentham-Dietz A, Newcomb PA, Egan KM, et al. Weight
change and risk of postmenopausal breast cancer (United
States). Cancer Causes Control. 2000;11(6):533–542.
30. Eng SM, Gammon MD, Terry MB, et al. Body size changes in
relation to postmenopausal breast cancer among women on
Long Island, New York. Am J Epidemiol. 2005;162(3):
229–237.
31. Lindblad P, Wolk A, Bergström R, et al. The role of obesity and
weight fluctuations in the etiology of renal cell cancer: a
population-based case-control study. Cancer Epidemiol
Biomarkers Prev. 1994;3(8):631–639.
32. Luo J, Margolis KL, Adami HO, et al. Body size, weight
cycling, and risk of renal cell carcinoma among
postmenopausal women: the Women’s Health Initiative
(United States). Am J Epidemiol. 2007;166(7):752–759.
33. De Roos AJ, Ulrich CM, Ray RM, et al. Intentional weight loss
and risk of lymphohematopoietic cancers. Cancer Causes
Control. 2010;21(2):223–236.
34. Calle EE, Rodriguez C, Jacobs EJ, et al. The American Cancer
Society Cancer Prevention Study II Nutrition Cohort: rationale,
study design, and baseline characteristics. Cancer. 2002;94(9):
2490–2501.
35. Cox DR. Regression models and life-tables. J R Stat Soc B.
1972;34(2):187–202.
36. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic
Research: Principles and Quantitative Methods. New York,
NY: Van Nostrand Reinhold; 1982.
Am J Epidemiol. 2015;182(5):394–404