introduction
The number of female breast cancer survivors is growing because of longer survival
as a consequence of advances in treatment and early diagnosis. There were ∼2.6 million
female breast cancer survivors in US in 2008 [1], and in the UK, breast cancer accounted
for ∼28% of the 2 million cancer survivors in 2008 [2].
Obesity is a pandemic health concern, with over 500 million adults worldwide estimated
to be obese and 958 million were overweight in 2008 [3]. One of the established risk
factors for breast cancer development in post-menopausal women is obesity [4], which
has further been linked to breast cancer recurrence [5] and poorer survival in pre-
and post-menopausal breast cancer [6, 7]. Preliminary findings from randomised, controlled
trials suggest that lifestyle modifications improved biomarkers associated with breast
cancer progression and overall survival [8].
The biological mechanisms underlying the association between obesity and breast cancer
survival are not established, and could involve interacting mediators of hormones,
adipocytokines, and inflammatory cytokines which link to cell survival or apoptosis,
migration, and proliferation [9]. Higher level of oestradiol produced in postmenopausal
women through aromatisation of androgens in the adipose tissues [10], and higher level
of insulin [11], a condition common in obese women, are linked to poorer prognosis
in breast cancer. A possible interaction between leptin and insulin [12], and obesity-related
markers of inflammation [13] have also been linked to breast cancer outcomes. Non-biological
mechanisms could include chemotherapy under-dosing in obese women, suboptimal treatment,
and obesity-related complications [14].
Numerous studies have examined the relationship between obesity and breast cancer
outcomes, and past reviews have concluded that obesity is linked to a lower survival;
however, when investigated in a meta-analysis of published data, only the results
of obese compared with non-obese or lighter women were summarised [6, 7, 15].
We carried out a systematic literature review and meta-analysis of published studies
to explore the magnitude and the shape of the association between body fatness, as
measured by body mass index (BMI), and the risk of total and cause-specific mortality,
overall and in women with pre- and post-menopausal breast cancer. As body weight may
change close to diagnosis and during primary treatment of breast cancer [16], we examined
BMI in three periods: before diagnosis, <12 months after diagnosis, and ≥12 months
after breast cancer diagnosis.
materials and methods
data sources and search
We carried out a systematic literature search, limited to publications in English,
for articles on BMI and survival in women with breast cancer in OVID MEDLINE and EMBASE
from inception to 30 June 2013 using the search strategy implemented for the WCRF/AICR
Continuous Update Project on breast cancer survival. The search strategy contained
medical subject headings and text words that covered a broad range of factors on diet,
physical activity, and anthropometry. The protocol for the review is available at
http://www.dietandcancerreport.org/index.php [17]. In addition, we hand-searched the
reference lists of relevant articles, reviews, and meta-analysis papers.
study selection
Included were follow-up studies of breast cancer survivors, which reported estimates
of the associations of BMI ascertained before and after breast cancer diagnosis with
total or cause-specific mortality risks. Studies that investigated BMI after diagnosis
were divided into two groups: BMI <12 months after diagnosis (BMI <12 months) and
BMI 12 months or more after diagnosis (BMI ≥12 months). Outcomes included total mortality,
breast cancer mortality, death from cardiovascular disease, and death from causes
other than breast cancer. When multiple publications on the same study population
were found, results based on longer follow-up and more outcomes were selected for
the meta-analysis.
data extraction
DSMC, TN, and DA conducted the search. DSMC, ARV, and DNR extracted the study characteristics,
tumour-related information, cancer treatment, timing and method of weight and height
assessment, BMI levels, number of outcomes and population at-risk, outcome type, estimates
of association and their measure of variance [95% confidence interval (CI) or P value],
and adjustment factors in the analysis.
statistical analysis
Categorical and dose–response meta-analyses were conducted using random-effects models
to account for between-study heterogeneity [18]. Summary relative risks (RRs) were
estimated using the average of the natural logarithm of the RRs of each study weighted
by the inverse of the variance and then unweighted by applying a random-effects variance
component which is derived from the extent of variability of the effect sizes of the
studies. The maximally adjusted RR estimates were used for the meta-analysis except
for the follow-up of randomised, controlled trials [19, 20] where unadjusted results
were also included, as these studies mostly involved a more homogeneous study population.
BMI or Quetelet's Index (QI) measured in units of kg/m2 was used.
We conducted categorical meta-analyses by pooling the categorical results reported
in the studies. The studies used different BMI categories. In some studies, underweight
(BMI <18.5 kg/m2 according to WHO international classification) and normal weight
women (BMI 18.5–<25.0 kg/m2) were classified together but, in some studies, they were
classified separately. Similarly, most studies classified overweight (BMI 25.0–<30.0
kg/m2) and obese (BMI ≥30.0 kg/m2) women separately but, in some studies, overweight
and obese women were combined. The reference category was normal weight or underweight
together with normal weight, depending on the studies. For convenience, the BMI categories
are referred to as underweight, normal weight, overweight, and obese in the present
review. We derived the RRs for overweight and obese women compared with normal weight
women in two studies [19, 21] that had more than four BMI categories using the method
of Hamling et al. [22]. Studies that reported results for obese compared with non-obese
women were analysed separately.
The non-linear dose–response relationship between BMI and mortality was examined using
the best-fitting second-order fractional polynomial regression model [23], defined
as the one with the lowest deviance. Non-linearity was tested using the likelihood
ratio test [24]. In the non-linear meta-analysis, the reference category was the lowest
BMI category in each study and RRs were recalculated using the method of Hamling et
al. [22] when the reference category was not the lowest BMI category in the study.
We also conducted linear dose–response meta-analyses, excluding the category underweight
when reported separately in the studies, by pooling estimates of RR per unit increase
(with its standard error) provided by the studies, or derived by us from categorical
data using generalised least-squares for trend estimation [25]. To estimate the trend,
the numbers of outcomes and population at-risk for at least three BMI categories,
or the information required to derive them using standard methods [26], and means
or medians of the BMI categories, or if not reported in the studies, the estimated
midpoints of the categories had to be available. When the extreme BMI categories were
open-ended, we used the width of the adjacent close-ended category to estimate the
midpoints. Where the RRs were presented by subgroups (age group [27], menopausal status
[28, 29], stage [30] or subtype [31] of breast cancer, or others [32–34]), an overall
estimate for the study was obtained by a fixed-effect model before pooling in the
meta-analysis. We estimated the risk increase of death for an increment of 5 kg/m2
of BMI.
To assess heterogeneity, we computed the Cochran Q test and I
2 statistic [35]. The cut points of 30% and 50% were used for low, moderate, and substantial
level of heterogeneity. Sources of heterogeneity were explored by meta-regression
and subgroup analyses using pre-defined factors, including indicators of study quality
(menopausal status, hormone receptor status, number of outcomes, length of follow-up,
study design, geographic location, BMI assessment, adjustment for confounders, and
others). Small study or publication bias was examined by Egger's test [36] and visual
inspection of the funnel plots. The influence of each individual study on the summary
RR was examined by excluding the study in turn [37]. A P value of <0.05 was considered
statistically significant. All analyses were conducted using Stata version 12.1 (Stata
Statistical Software: Release 12, StataCorp LP, College Station, TX).
results
A total of 124 publications investigating the relationship of body fatness and mortality
in women with breast cancer were identified. We excluded 31 publications, including
four publications on other obesity indices [38–41], 12 publications without a measure
of association [42–53], and 15 publications superseded by publications of the same
study with more outcomes [54–68]. A further 14 publications were excluded because
of insufficient data for the meta-analysis (five publications [69–73]) or unadjusted
results (nine publications [74–82]), from which nine publications reported statistically
significant increased risk of total, breast cancer or non-breast cancer mortality
in obese women (before or <12 months after diagnosis) compared with the reference
BMI [69, 71–74, 76, 77, 79, 82], two publications reported non-significant inverse
associations [75, 80] and three publications reported no association [70, 78, 81]
of BMI with survival after breast cancer. Hence, 79 publications from 82 follow-up
studies with 41 477 deaths (23 182 from breast cancer) in 213 075 breast cancer survivors
were included in the meta-analyses (Figure 1). Supplementary Table S1, available at
Annals of Oncology online shows the characteristics of the studies included in the
meta-analyses and details of the excluded studies are in supplementary Table S2, available
at Annals of Oncology online. Results of the meta-analyses are summarised in Table
1.
Table 1.
Summary of meta-analyses of BMI and survival in women with breast cancera
BMI before diagnosis
BMI <12 months after diagnosis
BMI ≥12 months after diagnosis
N
RR (95% CI)
I
2 (%)
P
h
N
RR (95% CI)
I
2 (%)
P
h
N
RR (95% CI)
I
2 (%)
P
h
Total mortality
Under versus normal weight
10
1.10 (0.92–1.31)
48%0.04
11
1.25 (0.99–0.57)
63%<0.01
3
1.29 (1.02–1.63)
0%0.39
Over versus normal weight
19
1.07 (1.02–1.12)
0%0.88
22
1.07 (1.02–1.12)
21%0.18
4
0.98 (0.86–1.11)
0%0.72
Obese versus normal weight
21
1.41 (1.29–1.53)
38%0.04
24
1.23 (1.12–1.33)
69%<0.01
5
1.21 (1.06–1.38)
0%0.70
Obese versus non-obese
–
–
–
12
1.26 (1.07–1.47)
80%<0.01
–
–
–
Per 5 kg/m2 increase
15
1.17 (1.13–1.21)
7%0.38
12
1.11 (1.06–1.16)
55%0.01
4
1.08 (1.01–1.15)
0%0.52
Breast cancer mortality
Under versus normal weight
8
1.02 (0.85–1.21)
31%0.18
5
1.53 (1.27–1.83)
0%0.59
1
1.10 (0.15–8.08)
–
Over versus normal weight
21
1.11 (1.06–1.17)
0%0.66
12
1.11 (1.03–1.20)
14%0.31
2
1.37 (0.96–1.95)
0%0.90
Obese versus normal weight
22
1.35 (1.24–1.47)
36%0.05
12
1.25 (1.10–1.42)
53%0.02
2
1.68 (0.90–3.15)
67%0.08
Obese versus non-obese
–
–
–
6
1.26 (1.05–1.51)
64%0.02
–
–
–
Per 5 kg/m2 increase
18
1.18 (1.12–1.25)
47%0.01
8
1.14 (1.05–1.24)
66%0.01
2
1.29 (0.97–1.72)
64%0.10
Cardiovascular disease related mortality
Over versus normal weight
2
1.01 (0.80–1.29)
0%0.87
–
–
–
–
–
–
Obese versus normal weight
2
1.60 (0.66–3.87)
78%0.03
–
–
–
–
–
–
Per 5 kg/m2 increase
2
1.21 (0.83–1.77)
80%0.03
–
–
–
–
–
–
Non-breast cancer mortality
Over versus normal weight
–
–
–
5
0.96 (0.83–1.11)
26%0.25
–
–
–
Obese versus normal weight
–
–
–
5
1.29 (0.99–1.68)
72%0.01
–
–
–
aBMI before and after diagnosis (<12 months after, or ≥12 months after diagnosis)
was classified according to the exposure period which the studies referred to in the
BMI assessment; the BMI categories were included in the categorical meta-analyses
as defined by the studies.
P
h, P for heterogeneity between studies.
Figure 1.
Flowchart of search.
Studies were follow-up of women with breast cancer identified in prospective aetiologic
cohort studies (women were free of cancer at enrolment), or cohorts of breast cancer
survivors whose participants were identified in hospitals or through cancer registries,
or follow-up of breast cancer patients enrolled in case–control studies or randomised
clinical trials.
Some studies included only premenopausal women [83–85] or postmenopausal women [21,
27, 86–94], but most studies included both. Menopausal status was usually determined
at time of diagnosis. Year of diagnosis was from 1957–1965 [70] to 2002–2009 [74].
Patient tumour characteristics and stage of disease at diagnosis varied across studies,
and some studies included carcinoma in situ. No all studies provided clinical information
on the tumour, treatment, and co-morbidities.
Most of the studies were based in North America or Europe. There were three studies
from each of Australia [79, 95, 96], Korea [97, 98] and China [99–101]; two studies
from Japan [71, 102]; one study from Tunisia [103] and four international studies
[19, 104–106]. Study size ranged from 96 [107] to 24 698 patients [97]. Total number
of deaths ranged from 56 [93] to 7397 [108], and the proportion of deaths from breast
cancer ranged from 22% [27] to 98% [84] when reported. All but eight studies [30,
93, 94, 98, 99, 109–111] had an average follow-up of more than 5 years.
BMI and total mortality
categorical meta-analysis
For BMI before diagnosis, compared with normal weight women, the summary RRs were
1.41 (95% CI 1.29–1.53, 21 studies) for obese women, 1.07 (95% CI 1.02–1.12, 19 studies)
for overweight women, and 1.10 (95% CI 0.92–1.31, 10 studies) for underweight women
(Figure 2). For BMI <12 months after diagnosis and the same comparisons, the summary
RRs were 1.23 (95% CI 1.12–1.33, 24 studies) for obese women, 1.07 (95% CI 1.02–1.12,
22 studies) for overweight women, and 1.25 (95% CI 0.99–1.57, 11 studies) for underweight
women (supplementary Figure S1, available at Annals of Oncology online). Substantial
heterogeneities were observed between studies of obese women and underweight women
(I
2 = 69%, P < 0.01; I
2 = 63%, P < 0.01, respectively). For BMI ≥12 months after diagnosis, the summary
RRs were 1.21 (95% CI 1.06–1.38, 5 studies) for obese women, 0.98 (95% CI 0.86–1.11,
4 studies) for overweight women, and 1.29 (95% CI 1.02–1.63, 3 studies) for underweight
women (supplementary Figure S2, available at Annals of Oncology online). Twelve additional
studies reported results for obese versus non-obese women <12 months after diagnosis,
and the summary RR was 1.26 (95% CI 1.07–1.47, I
2 = 80%, P < 0.01).
Figure 2.
Categorical meta-analysis of pre-diagnosis BMI and total mortality.
dose–response meta-analysis
There was evidence of a J-shaped association in the non-linear dose–response meta-analyses
of BMI before and after diagnosis with total mortality (all P < 0.01; Figure 3), suggesting
that underweight women may be at slightly increased risk compared with normal weight
women. The curves show linear increasing trends from 20 kg/m2 for BMI before diagnosis
and <12 months after diagnosis, and from 25 kg/m2 for BMI ≥12 months after diagnosis.
When linear models were fitted excluding the underweight category, the summary RRs
of total mortality for each 5 kg/m2 increase in BMI were 1.17 (95% CI 1.13–1.21, 15
studies, 6358 deaths), 1.11 (95% CI 1.06–1.16, 12 studies, 6020 deaths), and 1.08
(95% CI 1.01–1.15, 4 studies, 1703 deaths) for BMI before, <12 months after, and ≥12
months after diagnosis, respectively (Figure 4). Substantial heterogeneity was observed
between studies on BMI <12 months after diagnosis (I
2 = 55%, P = 0.01).
Figure 3.
Non-linear dose–response curves of BMI and mortality.
Figure 4.
Linear dose–response meta-analysis of BMI and total mortality.
BMI and breast cancer mortality
categorical meta-analysis
BMI was significantly associated with breast cancer mortality. Compared with normal
weight women, for BMI before diagnosis, the summary RRs were 1.35 (95% CI 1.24–1.47,
22 studies) for obese women, 1.11 (95% CI 1.06–1.17, 21 studies) for overweight women,
and 1.02 (95% CI 0.85–1.21, 8 studies) for underweight women (Figure 5). For BMI <12
months after diagnosis, the summary RRs were 1.25 (95% CI 1.10–1.42, 12 studies) for
obese women, 1.11 (95% CI 1.03–1.20, 12 studies) for overweight women, and 1.53 (95%
CI 1.27–1.83, 5 studies) for underweight women (supplementary Figure S3, available
at Annals of Oncology online). Substantial heterogeneity was observed between studies
of obese women (I
2 = 53%, P = 0.02). For BMI ≥12 months after diagnosis, the summary RRs of the two
studies identified were 1.68 (95% CI 0.90–3.15) for obese women and 1.37 (95% CI 0.96–1.95)
for overweight women (supplementary Figure S4, available at Annals of Oncology online).
The summary of another six studies that reported RRs for obese versus non-obese <12
months after diagnosis was 1.26 (95% CI 1.05–1.51, I
2 = 64%, P = 0.02).
Figure 5.
Categorical meta-analysis of pre-diagnosis BMI and breast cancer mortality.
dose–response meta-analysis
There was no significant evidence of a non-linear relationship between BMI before,
<12 months after, and ≥12 months after diagnosis and breast cancer mortality (P =
0.21, P = 1.00, P = 0.86, respectively) (Figure 3). When linear models were fitted
excluding data from the underweight category, statistically significant increased
risks of breast cancer mortality with BMI before and <12 months after diagnosis were
observed (Figure 6). The summary RRs for each 5 kg/m2 increase were 1.18 (95% CI 1.12–1.25,
18 studies, 5262 breast cancer deaths) for BMI before diagnosis and 1.14 (95% CI 1.05–1.24,
8 studies, 3857 breast cancer deaths) for BMI <12 months after diagnosis, with moderate
(I
2 = 47%, P = 0.01) and substantial (I
2 = 66%, P = 0.01) heterogeneities between studies, respectively. Only two studies
on BMI ≥12 months after diagnosis and breast cancer mortality (N = 220 deaths) were
identified. The summary RR was 1.29 (95% CI 0.97–1.72).
Figure 6.
Linear dose–response meta-analysis of BMI and breast cancer mortality.
BMI and other mortality outcomes
Only two studies reported results for death from cardiovascular disease (N = 151 deaths)
[27, 112]. The summary RR for obese versus normal weight before diagnosis was 1.60
(95% CI 0.66–3.87). No association was observed for overweight versus normal weight
(summary RR = 1.01, 95% CI 0.80–1.29). For each 5 kg/m2 increase in BMI, the summary
RR was 1.21 (95% CI 0.83–1.77). Five studies reported results for deaths from any
cause other than breast cancer (N = 2704 deaths) [21, 34, 108, 113, 114]. The summary
RRs were 1.29 (95% CI 0.99–1.68, I
2 = 72%, P = 0.01) for obese women, and 0.96 (95% CI 0.83–1.11, I
2 = 26%, P = 0.25) for overweight women compared with normal weight women.
subgroup, meta-regression, and sensitivity analyses
The results of the subgroup and meta-regression analyses are in supplementary Tables
S3 and S4, available at Annals of Oncology online. Subgroup analysis was not carried
out for BMI ≥12 months after diagnosis as the limited number of studies would hinder
any meaningful comparisons.
Increased risks of mortality were observed in the meta-analyses by menopausal status.
While the summary risk estimates seem stronger with premenopausal breast cancer, there
was no significant heterogeneity between pre- and post-menopausal breast cancer as
shown in the meta-regression analyses (P = 0.28–0.89) (supplementary Tables S3 and
S4, available at Annals of Oncology online). For BMI before diagnosis and total mortality,
the summary RRs for obese versus normal weight were 1.75 (95% CI 1.26–2.41, I
2 = 70%, P < 0.01, 7 studies) in women with pre-menopausal breast cancer and 1.34
(95% CI 1.18–1.53, I
2 = 27%, P = 0.20, 9 studies) in women with post-menopausal breast cancer.
Studies with larger number of deaths [105, 115], conducted in Europe [28, 115], or
with weight and height assessed through medical records [28, 104, 115, 116] tended
to report weaker associations for BMI <12 months after diagnosis and total mortality
compared with other studies (meta-regression P = 0.01, 0.02, 0.01, respectively) (supplementary
Table S3, available at Annals of Oncology online); while studies with larger number
of deaths [101], conducted in Asia [101, 102], or adjusted for co-morbidity [101,
102] reported weaker associations for BMI <12 months after diagnosis and breast cancer
mortality (meta-regression P = 0.01, 0.02, 0.01, respectively) (supplementary Table
S4, available at Annals of Oncology online).
Analyses stratified by study designs, or restricted to studies with invasive cases
only, early-stage non-metastatic cases only, or mammography screening detected cases
only, or controlled for previous diseases did not produce results that were materially
different from those obtained in the overall analyses (results not shown). Summary
risk estimates remained statistically significant when each study was omitted in turn,
except for BMI ≥12 months after diagnosis and total mortality. The summary RR was
1.06 (95% CI 0.98–1.15) per 5 kg/m2 increase when Flatt et al. [117] which contributed
315 deaths was omitted.
small studies or publication bias
Asymmetry was only detected in the funnel plots of BMI <12 months after diagnosis
and total mortality, and breast cancer mortality, which suggests that small studies
with an inverse association are missing (plots not shown). Egger's tests were borderline
significant (P = 0.05) or statistically significant (P = 0.03), respectively.
discussion
The present systematic literature review and meta-analysis of follow-up studies clearly
supports that, in breast cancer survivors, higher BMI is consistently associated with
lower overall and breast cancer survival, regardless of when BMI is ascertained. The
limited number of studies on death from cardiovascular disease is also consistent
with a positive association. For before, <12 months after, and 12 months or more after
breast cancer diagnosis, compared with normal weight women, obese women had 41%, 23%,
and 21% higher risk for total mortality, and 35%, 25%, and 68% increased risk for
breast cancer mortality, respectively. The findings were supported by the positive
associations observed in the linear dose–response meta-analysis. All associations
were statistically significant, apart from the relationship between BMI ≥12 months
after diagnosis and breast cancer mortality. This may be due to limited statistical
power, with only 220 breast cancer deaths from two follow-up studies.
Positive associations, in some cases statistically significant, were also observed
in overweight, and underweight women compared with normal weight women. Women with
BMI of 20 kg/m2 before, or <12 months after diagnosis, and of 25 kg/m2 12 months or
more after diagnosis appeared to have the lowest mortality risk in the non-linear
dose–response analysis. Co-morbid conditions may cause the observed increased risk
in underweight women. Thorough investigation within the group and on their contribution
to the shape of the association is hindered, as not all studies in this review reported
results for this group. The increased risk associated with obesity was similar in
pre- or post-menopausal breast cancer. We did not find any evidence of a protective
effect of obesity on survival after pre-menopausal breast cancer, contrary to what
has been observed for the development of breast cancer in pre-menopausal women [4].
A large body of evidence with 41 477 deaths (23 182 from breast cancer) in over 210
000 breast cancer survivors was systematically reviewed in the present study. We carried
out categorical, linear, and non-linear dose–response meta-analyses to examine the
magnitude and the shape of the associations for total and cause-specific mortality
in underweight, overweight, and obese women by time periods before and after diagnosis
that is important in relation to the population-at-risk and breast cancer survivors.
Our findings agree with and further extend the results from previous meta-analyses.
A review published in 2010 reported statistically significant increased risks of 33%
of both total and breast cancer mortality for obesity versus non-obesity around diagnosis
[7]. These estimates are slightly higher than ours, which may be explained by the
different search periods and inclusion criteria for the articles (33 studies and 15
studies included in the analyses, respectively). Another review published in 2012
further reported consistent positive associations of total and breast cancer mortality
with higher versus lower BMI around diagnosis [6]. No significant differences were
observed by menopausal status or hormone receptor status. The After Breast Cancer
Pooling Project of four prospective cohort studies found differential effects of levels
of pre-diagnosis obesity on survival [118]. Compared with normal weight women, significant
or borderline significant increased risks of 81% of total and 40% of breast cancer
mortality were only observed for morbidly obese (≥40 kg/m2) women and not for women
in other obesity categories. We observed statistically significant increased risks
also for overweight women, probably because of a larger number of studies. We were
unable to investigate the associations with severely and morbidly obese women because
only two studies included in this review reported such results [19, 113]. Overall,
our findings are consistent with previous meta-analyses in showing elevated total
and breast cancer mortality associated with higher BMI and support the current guidelines
for breast cancer survivors to stay as lean as possible within the normal range of
body weight [4], for overweight women to avoid weight gain during treatment and for
obese women to lose weight after treatment [119].
The present review is limited by the challenges and flaws encountered by the individual
epidemiological studies evaluating the body fatness–mortality relationship in breast
cancer survivors. Most studies did not adjust for co-morbidities and assess intentional
weight loss. Women with more serious health issues, and especially smokers, may lose
weight but are at an increased risk of mortality, and this might cause an apparent
increased risk in underweight women. Body weight information through the natural history
of the disease and treatment information were usually not complete or available. Increase
of body weight post-diagnosis is common in women with breast cancer, particularly
during chemotherapy [16]. Chemotherapy under-dosing is a common problem in obese women
and may contribute to their increased mortality [120]. Although several studies with
pre-diagnosis BMI adjusted for underlying illnesses or excluded the first few years
of follow-up, reverse causation may have affected the results in studies that assessed
BMI in women with cancer and other illnesses. However, in these studies, the associations
were similar to other studies. Possible survival benefit (subjects with better prognostic
factors survive) may be present in the survival cohorts, in which the range of BMI
could be narrower, and may cause an underestimation of the association.
Follow-up studies with variable characteristics were pooled in the meta-analysis.
Women identified in clinical trials may have had specific tumour subtypes, with fewer
co-morbidities, and were more likely to receive protocol treatments with high treatment
completion rates. Women who were recruited through mammography screening programmes
may have had healthier lifestyles or access to medical facilities, and more likely
to be diagnosed with in situ or early-stage breast cancer. Cancer detection methods,
tumour classifications and treatment regimens change over time, and may vary within
(if follow-up is long) and between studies, and could not be simply examined by using
the diagnosis or treatment date. We cannot rule out the effect of unmeasured or residual
confounding in our analysis. Nevertheless, most results were adjusted for multiple
confounding factors, including tumour stage or other-related variables and stratified
analyses by several key factors showed similar summary risk estimates. Small study
or publication bias was observed in the analyses of BMI <12 months after diagnosis.
However, the overall evidence is supported by large, well-designed studies and is
unlikely to be changed. We did not conduct analyses by race/ethnicity and treatment
types as only limited studies had published results.
Future studies of body fatness and breast cancer outcomes should aim to account for
co-morbidities, separate intended and unintended changes of body weight, and collect
complete treatment information during study follow-up. Randomised clinical trials
are needed to test interventions for weight loss and maintenance on survival in women
with breast cancer.
In conclusion, the present systematic literature review and meta-analysis extends
and confirms the associations of obesity with an unfavourable overall and breast cancer
survival in pre- and post-menopausal breast cancer, regardless of when BMI is ascertained.
Increased risks of mortality in underweight and overweight women were also observed.
Given the comparable elevated risks with obesity in the development (for post-menopausal
women) and prognosis of breast cancer, and the complications with cancer treatment
and other obesity-related co-morbidities, it is prudent to maintain a healthy body
weight (BMI 18.5–<25.0 kg/m2) throughout life.
funding
This work was supported by the World Cancer Research Fund International (grant number:
2007/SP01) (http://www.wcrf-uk.org/). The funder of this study had no role in the
decisions about the design and conduct of the study; collection, management, analysis,
or interpretation of the data; or the preparation, review, or approval of the manuscript.
The views expressed in this review are the opinions of the authors. They may not represent
the views of the World Cancer Research Fund International/American Institute for Cancer
Research and may differ from those in future updates of the evidence related to food,
nutrition, physical activity, and cancer survival.
disclosure
DCG reports personal fees from World Cancer Research Fund/American Institute for Cancer
Research, during the conduct of the study; grants from Danone, and grants from Kelloggs,
outside the submitted work. AM reports personal fees from Metagenics/Metaproteomics,
personal fees from Pfizer, outside the submitted work. All remaining authors have
declared no conflicts of interest.
Supplementary Material
Supplementary Data