Chang et al.

BMC Cancer (2019) 19:543

https://doi.org/10.1186/s12885-019-5642-0

RESEARCH ARTICLE Open Access

Iron intake, body iron status, and risk of

breast cancer: a systematic review and
meta-analysis
Vicky C. Chang1,2* , Michelle Cotterchio1,2 and Edwin Khoo3

Abstract
Background: Iron has been shown to promote breast carcinogenesis in animal models through generation of
oxidative stress and interaction with estrogen. Heme iron, which is found exclusively in animal-sourced foods, is
suggested to have a more detrimental effect. Epidemiological evidence of the association between iron and breast
cancer risk remains inconclusive and has not been comprehensively summarized. This systematic review and meta-
analysis evaluated associations between both iron intake and body iron status and breast cancer risk.
Methods: Four electronic databases (MEDLINE, EMBASE, CINAHL, and Scopus) were searched up to December 2018
for studies assessing iron intake and/or biomarkers of iron status in relation to breast cancer risk. Using random-
effects meta-analyses, pooled relative risks (RRs) and 95% confidence intervals (CIs) were calculated comparing the
highest vs. lowest category of each iron measure. Dose-response meta-analyses were also performed to investigate
linear and nonlinear associations.
Results: A total of 27 studies were included in the review, of which 23 were eligible for meta-analysis of one or more
iron intake/status measures. Comparing the highest vs. lowest category, heme iron intake was significantly associated
with increased breast cancer risk, with a pooled RR of 1.12 (95% CI: 1.04–1.22), whereas no associations were found for
dietary (1.01, 95% CI: 0.89–1.15), supplemental (1.02, 95% CI: 0.91–1.13), or total (0.97, 95% CI: 0.82–1.14) iron intake.
Associations of iron status indicators with breast cancer risk were generally in the positive direction; however, a
significant pooled RR was found only for serum/plasma levels (highest vs. lowest) of iron (1.22, 95% CI: 1.01–1.47), but
not for ferritin (1.13, 95% CI: 0.78–1.62), transferrin saturation (1.16, 95% CI: 0.91–1.47), or total iron-binding capacity
(1.10, 95% CI: 0.97–1.25). In addition, a nonlinear dose-response was observed for heme iron intake and serum iron
(both Pnonlinearity < 0.05).
Conclusions: Heme iron intake and serum iron levels may be positively associated with breast cancer risk. Although
associations were modest, these findings may have public health implications given the widespread consumption of
(heme) iron-rich foods. In light of methodological and research gaps identified, further research is warranted to better
elucidate the relationship between iron and breast cancer risk.
Keywords: Breast cancer, Iron intake, Heme iron, Iron status, Ferritin, Systematic review, Meta-analysis, Dose-response

* Correspondence: vickycd.chang@mail.utoronto.ca
1
Dalla Lana School of Public Health, University of Toronto, 155 College Street,
6th Floor, Toronto, ON M5T 3M7, Canada
2
Prevention and Cancer Control, Cancer Care Ontario, 620 University Avenue,
Toronto, ON M5G 2L7, Canada
Full list of author information is available at the end of the article

© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Chang et al. BMC Cancer (2019) 19:543 Page 2 of 28

Background storage, further amplifying oxidative stress and inducing

Iron is an essential nutrient required for many biological DNA damage [7, 24]. Importantly, the role of iron in
processes in the human body, such as oxygen transport, breast cancer is strongly supported by animal experiments
DNA synthesis, and energy production [1]. However, demonstrating that excess iron through diet or subcutane-
owing to its strong capacity to both accept and donate ous injection promotes the initiation and growth of mam-
electrons, iron also readily participates in reduction-oxi- mary tumours in rodents [25–28].
dation (redox) reactions that lead to the generation of Despite strong biological plausibility and evidence
reactive oxygen species (ROS) and subsequent oxidative from animal studies, epidemiological evidence of the
damage to tissues and cellular components, particularly association between iron and breast cancer risk in
DNA, proteins, and lipids [2, 3]. As such, both high diet- humans is inconsistent, inconclusive, and has not been
ary iron intake and elevated body iron status have been adequately summarized. Although several narrative
hypothesized to increase the risks of several cancers, reviews have discussed iron’s role in breast cancer, they
including breast cancer [4–9]. Notably, the World focused largely on biological mechanisms and only pre-
Health Organization’s International Agency for Research sented selected epidemiologic findings [4–9]. A 2014
on Cancer has identified “iron (in food and as supple- systematic review/meta-analysis on iron and cancer risk
ments)” as one of the “high priority” agents or exposures by Fonseca-Nunes et al. identified a total of 59 studies
to be assessed in relation to cancer risk [10]. published between 1995 and 2012, of which only seven
In the general, non-transfused population, iron is studies assessed iron intake, and none assessed body
obtained almost exclusively from the diet, either in the iron status, in relation to breast cancer risk [29]. Overall,
form of heme or non-heme iron [11, 12]. Heme iron is the review concluded that heme iron intake may be posi-
the organic form of iron derived only from animal tively (albeit not significantly for breast cancer) associ-
source foods, including meat, poultry, and fish/seafood ated with cancer risk, whereas biomarkers of iron status,
[12]. On average, heme iron constitutes approximately such as serum ferritin, may be negatively associated with
40% of the total iron content in cooked meats, with the cancer risk [29]. However, the review was limited in
highest levels found in red meat (e.g., beef, pork), terms of its search strategy (e.g., single database
although concentrations may further vary according to searched, missing relevant search terms), lacked infor-
differences in meat type/cut, cooking or preparation mation on study quality assessment, and provided a
method, and meat doneness level [12–14]. While also primarily qualitative synthesis of findings, with meta-
present in animal sources, non-heme iron constitutes all analysis conducted for heme iron intake only.
of the iron content in plant-based foods, including vege- Given the growing body of literature on iron intake/sta-
tables, fruits, and legumes, as well as iron-fortified prod- tus and breast cancer risk, an updated, comprehensive,
ucts (e.g., cereals) [12]. Heme iron may be of particular and quantitative review focusing specifically on this topic
concern with respect to cancer risk due to its greater is warranted. We conducted a systematic review and
bioavailability and involvement in the formation of car- meta-analysis of epidemiologic studies to evaluate associa-
cinogenic N-nitroso compounds [15, 16]. Once absorbed tions between different types of iron intake, as well as
by intestinal cells and exported into circulation, iron is indicators of body iron status, and risk of breast cancer.
bound by its transport protein transferrin and delivered
to tissues and cells, where it is either used or stored by Methods
binding to ferritin [11]. Common indicators of body iron This systematic review and meta-analysis was con-
status include circulating (serum or plasma) levels of fer- ducted and reported with reference to the Preferred
ritin, iron, transferrin, transferrin receptor (TfR), total Reporting Items for Systematic Reviews and Meta-
iron-binding capacity (TIBC), and transferrin saturation Analyses (PRISMA) [30] and the Meta-analysis Of
(TSAT) [17, 18]. Iron status may also be assessed using Observational Studies in Epidemiology (MOOSE) [31]
nail [19], hair [20], and tissue [17, 21] samples. guidelines and checklists.
Iron is suggested to have a role in breast cancer develop-
ment through its interaction with estrogen in oxidative Data sources and search strategy
stress and other pathways [22]. For example, iron catalyzes Systematic electronic database searches were conducted
the redox cycling of catechol estrogen metabolites to form using MEDLINE, EMBASE, CINAHL, and Scopus to
quinones and semiquinones, which have been shown to identify studies published up to December 31, 2018 that
stimulate ROS production and contribute to breast car- investigated the association between iron intake/status
cinogenesis in cell cultures and in vivo [7, 8, 23]. In and breast cancer risk, without any language restrictions.
addition, superoxide radicals generated from estrogen The search included a combination of Medical Subject
redox cycling may trigger the release of free iron (i.e., the Headings terms, keywords, and variations of text words
more biologically active ferrous [Fe2+] form) from ferritin related to iron (e.g., “iron”, “Fe”, “ferric”, “ferrous”,
Chang et al. BMC Cancer (2019) 19:543 Page 3 of 28

“ferritin”, “transferrin”, “TfR”, or “TIBC”) and breast animal-based foods as described in the original studies,
cancer (e.g., “breast”, “mammary”, or “nipple”, combined e.g., 40% of total iron from meat, literature-based
with “cancer”, “neoplasm”, “tumor”, “carcinoma”, “adeno- meat-specific percentages [13], laboratory-based heme
carcinoma”, or “malignancy”). The full electronic search iron database [14]), and non-heme iron (total dietary
strategy is presented in Additional file 1. To identify iron minus heme iron).
additional potentially eligible studies, reference lists of all The following serum or plasma indicators of body iron
included studies and relevant review articles were also status were included when available: ferritin (marker of
hand-searched. body iron stores), iron (circulating iron bound to trans-
ferrin), transferrin (direct measure of circulating trans-
Eligibility criteria and study selection ferrin available to bind iron), TIBC (total amount of iron
Studies were eligible for inclusion if they: 1) involved that can be bound by circulating transferrin, i.e., indirect
human subjects; 2) were primary research studies; 3) uti- or proxy measure of transferrin), TSAT (percentage of
lized a cohort or case-control design, including trad- iron-binding sites on transferrin that are occupied by
itional case-control, nested case-control, and case-cohort iron, typically calculated as the ratio of serum iron to
studies; 4) assessed any prediagnostic measure of iron TIBC or serum iron to transferrin), and TfR (indicator
intake and/or body iron status as an exposure (see below of balance between cellular iron demand and supply)
section on “Exposure definitions” for details); 5) exam- [17, 18]. In addition, finger/toenail and hair iron, which
ined breast cancer as an outcome in females; and 6) may reflect longer-term exposure [19, 20], as well as
reported (or provided sufficient data to calculate) an tissue (e.g., bone marrow, liver, breast) iron [17, 21] were
odds, risk, or hazard ratio for the association between also considered. Higher levels of each biomarker are
iron intake/status and breast cancer risk. associated with higher iron status, with the exceptions of
Animal and cell culture studies, non-primary studies transferrin, TIBC, and TfR, which are inversely related
(e.g., reviews, editorials, letters to editor), conference to iron status [17, 18].
abstracts without full-text, case reports, case series,
cross-sectional studies, ecological studies, and studies Data extraction
combining female and male breast cancer were excluded. The following information was extracted from each in-
We also excluded studies assessing postdiagnostic levels cluded study: author name, publication year, country of
of iron intake (i.e., studies specifically asking about diet study conduct, study name, study design, study period
or supplement use after diagnosis) or body iron status and setting, duration of follow-up (where applicable),
(i.e., studies where biological samples were collected sample size (number of cases/total number of partici-
after diagnosis), since these measures may be influenced pants for cohort studies; number of cases/controls for
by breast cancer pathogenesis and treatment [32, 33] case-control studies), population characteristics (age and
and are thus less relevant for evaluating the role of iron menopausal status), measure(s) of iron intake/status re-
in relation to breast cancer risk. ported and their methods of assessment, breast cancer
Following removal of duplicate records, titles and case ascertainment, effect estimates and corresponding
abstracts of citations retrieved from the electronic data- 95% confidence intervals (CIs), variables matched or ad-
bases were screened to identify potentially relevant stud- justed for in the analysis, and any information needed
ies. Full-texts of these identified studies were then for study quality assessment. Where available, results
obtained and assessed in detail for inclusion or exclu- stratified by menopausal status (premenopausal and
sion. Both title/abstract screening and full-text eligibility postmenopausal) at breast cancer diagnosis and
assessment were performed independently by two hormone receptor (estrogen receptor [ER]/progesterone
authors (VCC and EK) using the web-based systematic receptor [PR]) tumour subtype were also extracted.
review tool Covidence (Veritas Health Innovation, Data extraction was performed by one author (VCC)
Melbourne, Australia) [34]. Any disagreement was and verified independently by another (EK). For studies
resolved through discussion and consensus, and all with missing information, we referred to related publi-
authors approved the final list of studies included. cations (e.g., detailed reports of study design and popu-
lation characteristics) or contacted the corresponding
Exposure definitions author of the original study for clarification or
In this review, measures of iron intake were classified additional information.
and defined as below: dietary iron (iron from foods
alone), supplemental iron (iron from single-ingredient Quality assessment
iron supplements and/or iron-containing multivitamin/ The quality of included studies was assessed independently
mineral supplements), total iron (sum of dietary and by two authors (VCC and EK) using the Newcastle-Ottawa
supplemental iron), heme iron (iron estimated from Scale (NOS) [35], with any disagreement resolved by
Chang et al. BMC Cancer (2019) 19:543 Page 4 of 28

discussion and consensus. The NOS includes study for “use” (all categories > 0 mg/day) vs. “no use” (0 mg/
design-specific items for cohort and case-control studies day) was estimated using the method described by
and evaluates three broad domains of bias: 1) selection of Hamling et al., which involves the reconstruction of con-
study subjects; 2) comparability of groups (i.e., control for tingency tables to calculate the adjusted effect estimates
potential confounding factors); and 3) ascertainment of the and their CIs [40]. For one iron biomarker study where
exposure or the outcome [35]. If a study examined the the reference category was not the lowest [41], the
association of both iron intake and iron status with breast adjusted RR comparing the highest vs. lowest category
cancer risk, its quality was assessed separately for each type was calculated also using the Hamling method [40]. For
of exposure because of possible differences in confounding one study that reported RR for each 1-standard devi-
control and/or biases related to exposure ascertainment. ation (SD) increase in iron intake [42], we converted the
The NOS yields a score ranging from 0 (lowest) to 9 (high- RR such that it corresponded to a comparison for the
est) [35]. In this review, studies with scores of 7 or greater highest vs. lowest quartile; this was done by multiplying
were considered high-quality, while those scoring below 7 the natural logarithm of the original RR by 2.54 and expo-
were considered low-quality. Detailed NOS coding man- nentiating the product, under the assumption of a stand-
uals are presented in Additional file 2. ard normal distribution where the difference in means
between the highest and lowest quartiles is 2.54 SDs [43].
Statistical analysis To investigate linear and nonlinear dose-response rela-
Meta-analyses of the associations between iron intake/ tions between iron intake/status and breast cancer risk,
status and breast cancer risk were performed separately we further conducted random-effects dose-response
for each subtype of iron intake or iron status indicator meta-analyses using a generalized least-squares method
with at least two available studies. When multiple publi- for trend estimation, as proposed by Greenland and
cations reported data on the same iron measure from Longnecker [44] and Orsini et al. [45, 46]. To prepare
identical or overlapping study populations, only the pub- the data for these analyses, RRs and 95% CIs across at
lication with the largest sample size or longest duration least three categories of the exposure (iron intake or sta-
of follow-up was included in the meta-analysis for the tus) were obtained from each study, along with exposure
specific iron measure. values (i.e., dose) and numbers of cases/non-cases for
For each subtype of iron intake or iron status indica- each category [44, 45]. Whenever reported, the mean or
tor, the pooled relative risk (RR) was used as the sum- median value of iron intake or iron biomarker level for
mary measure of association and was estimated by each category was assigned as the “dose” corresponding
combining odds, risk, and hazard ratios reported by in- to each RR estimate; otherwise, the midpoint (calculated
dividual studies. Odds, risk, and hazard ratios, hereafter as the average of the maximum and minimum values for
all referred to as RRs, were assumed to be equivalent in each category) was used. If a study did not report the
our analyses given that breast cancer is a relatively rare maximum or minimum value for the highest or lowest
disease outcome (i.e., less than 10%) [36]. If a study re- category, respectively, the midpoint was calculated by
ported RRs and 95% CIs from two or more regression assuming the range of that category to be the same as
models with different levels of covariate adjustment, esti- that of the adjacent category. When units of measure-
mates from the most fully adjusted model were used in ment for a specific exposure differed across studies, they
the analyses. To account for within- and between-study were converted to the most commonly reported or con-
variability, pooled RRs and corresponding 95% CIs were ventional unit. For example, when iron intake was re-
computed using the DerSimonian and Laird (DL) ported in mg/1000 kcal, we converted it to mg/day using
random-effects model [37]. Additionally, pooled RRs the mean total energy intake (kcal/day) provided by the
and 95% CIs were also calculated using the profile likeli- study. Similarly, serum iron concentration reported in
hood random-effects model as the DL method has been μmol/L was converted to μg/dL by multiplying by
suggested to overestimate precision when there is a 5.5866 (1 μg/dL = 0.179 μmol/L iron) [47]. If the number
small number of studies [38]; however, since the two of cases/non-cases across exposure categories was not
models yielded very similar results and led to the same available, it was estimated by dividing the total number
conclusions for all iron measures, estimates from the DL of subjects (or person-years; for cohort studies) or con-
model (most common method) were presented. trols (for case-control studies) by the total number of
In our main analysis, the pooled RR comparing the categories (assuming nearly equal distribution across
highest to the lowest category of each iron intake/status quantiles); the number of cases was then estimated ac-
measure was computed. For supplemental iron intake, cordingly based on the RRs. In addition to meta-analysis
we examined the dichotomous measure “use vs. no use” assuming a linear trend (e.g., pooled RR per unit
instead, as only one study reported RRs across doses of increase in iron intake) [45], we examined potential non-
supplemental iron [39]; for that study, the adjusted RR linear associations using restricted cubic splines analyses
Chang et al. BMC Cancer (2019) 19:543 Page 5 of 28

with three knots (located at the 10th, 50th, and 90th per- studies were excluded from all meta-analyses but
centiles), and the presence of nonlinearity was assessed remained in the review, including one assessing adoles-
by testing the significance of the coefficient for the cent intakes of total and heme iron [64] in the same (but
second spline [46]. a smaller subset of ) study cohort as another study asses-
Heterogeneity between studies was assessed using the sing adult intakes of total and heme iron [63], one iron
Cochran’s Q test (P < 0.10 considered statistically signifi- status study where CIs for the RRs were not reported [67],
cant) and the I2 statistic quantifying the proportion of and two studies that were the only ones analyzing toenail
the total variability attributable to heterogeneity [48]; I2 [69] or breast tissue [70] iron. A flow diagram detailing
values of 25, 50, and 75% roughly indicate low, moder- the study selection process is presented in Fig. 1.
ate, and high heterogeneity, respectively [49]. To explore
potential effect modification and sources of heterogen- Study characteristics and quality
eity, subgroup analyses were performed according to Table 1 summarizes the main characteristics and find-
study design (cohort or case-control), geographic loca- ings of included studies. Among all 27 studies reviewed,
tion (North America, Europe, Asia, or Australia), meno- the year of publication ranged from 1990 to 2018, with
pausal status (premenopausal or postmenopausal), study six studies published before 2000 [52–54, 67–69], nine
quality (NOS score ≥ 7 or < 7), dietary assessment studies between 2000 and 2009 [39, 55–61, 70], and 12
method (structured interview or self-administered ques- studies in 2010 or later [41, 42, 62–66, 71–75]. The
tionnaire), biological sample (serum or plasma), and majority of studies were conducted in the United States
adjustments for specific confounders, including body (n = 12) [39, 42, 57, 58, 62–64, 66, 68–70, 73] or Canada
mass index (BMI), physical activity, alcohol intake, oral (n = 1) [59], while the rest were in Europe (n = 9; includ-
contraceptive (OC) and/or hormone replacement ing two in Germany and one in each of Denmark, Italy,
therapy (HRT) use, and family history of breast cancer. the United Kingdom, Switzerland, France, Finland, and
Where at least 10 studies were available, univariable Sweden) [52–56, 65, 67, 72, 75], Asia (n = 4; including
meta-regression was performed on each of the afore- two in China, one in Taiwan, and one in Japan) [41, 60,
mentioned variables to further assess their influence on 61, 71], and Australia (n = 1) [74].
heterogeneity, with P < 0.10 indicating statistical signifi- Of the 17 studies assessing iron intake, seven were co-
cance [48]. Notably, although pre-specified, subgroup hort studies [39, 59, 62–66], with study size ranging
analyses were not conducted by breast cancer tumour from 4646 to 193,742 participants, follow-up ranging
(ER/PR) status, as there were less than two studies from 5.5 to 20 years, and number of breast cancer cases
reporting these results for each iron measure. ranging from 188 to 9305; the remaining ten studies were
Publication bias was evaluated using funnel plots and case-control studies, of which four were hospital-based
Begg’s rank-correlation [50] and Egger’s regression [51] [53–56], three were population-based [42, 52, 60], and
tests (P < 0.10 considered statistically significant). Finally, three were nested within existing cohorts [57, 58, 61], with
influence of individual studies was investigated by recal- case numbers ranging from 220 to 3452. Of the 11
culating the pooled RR and 95% CI each time a single studies assessing body iron status, five were cohort
study was omitted from the analysis. studies [41, 67, 68, 72, 74], with study size ranging
Analyses were performed using Stata/MP, version 14 from 1795 to 164,355 participants, follow-up ranging
(StataCorp LP, College Station, TX, USA). Statistical from 7.1 to 17.6 years, and number of cases ranging
tests were two-sided, with statistical significance evalu- from 80 to 3238; the remaining six studies used a
ated at P < 0.05 unless otherwise specified. nested case-control [61, 69–71, 73] or case-cohort
[75] design, with follow-up (where reported) ranging
from 4 to 15.7 years and case numbers ranging from
Results 107 to 795. Most studies consisted of both pre- and
Search results postmenopausal women across a wide age range at
Our search initially yielded 7589 records. After dupli- baseline and/or time of diagnosis, except for four
cates were removed, titles and abstracts of 4411 articles cohort/nested case-control studies conducted among
were screened, of which 167 full-texts were further postmenopausal women alone [39, 58, 62, 66], one
assessed for eligibility. Twenty-seven studies, including nested case-control study with primarily (86%) post-
17 studies examining iron intake [39, 42, 52–66] and 11 menopausal breast cancer cases [71], and one cohort
studies examining body iron status [41, 61, 67–75] in re- study restricted to women who were premenopausal
lation to breast cancer risk, met the inclusion criteria of at baseline [63]. With respect to outcome ascertain-
our systematic review. Several studies reported data on ment, incident breast cancer cases in prospective
multiple measures of iron intake and/or status and were studies (e.g., cohort, nested case-control) were identi-
included in more than one meta-analysis. Four of the 27 fied either through record linkage to cancer registries
Chang et al. BMC Cancer (2019) 19:543 Page 6 of 28

Fig. 1 Flow diagram of study selection for the systematic review and meta-analysis. *One study reporting on toenail iron [69] and the other on
breast tissue iron [70] as the only iron measure

(and vital statistics) or through self-reports verified by 2 years of follow-up [65], all iron intake studies used a
medical records. Similarly, traditional case-control one-time, self- or interviewer-administered food fre-
studies identified newly diagnosed cases (typically quency questionnaire (FFQ) to assess usual intake at
within 1 year of diagnosis, where reported) from can- baseline (cohort studies) or during a specified period
cer registries and/or hospital records. (e.g., 2 years) before breast cancer diagnosis (case-con-
With the exception of one cohort study where mul- trol studies). Two studies involved more distant recall,
tiple 24-h dietary recalls were completed during the first including one assessing total and heme iron intake
Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Iron intake (n = 17)
Ewertz and Population-based All women Cases: < 70 y, 21-item, self- Iron supplement All: 1.14 Yes Age, place of residence 7
Gill 1990 [52], case-control (1486/1336) 56%; administered FFQ + use: yes vs. no (0.65–2.02)
Chang et al. BMC Cancer

Denmark (1983–1984) Controls: < 70 y, supplement use

59% questions
(pilot-tested against
interviews)
assessing usual
intake in the year
prior to diagnosis
(2019) 19:543

Negri et al. Hospital-based All women Cases: 23–74 y, 78-item, validated, Dietary iron All: 0.85 Yes Age, study centre, 6
1996 [53], case-control (2569/2588) 61%; interviewer- (mg/day, (0.7–1.0) education, parity,
Italy (multicentre study Controls: 20–74 y, administered FFQ energy-adjusted energy intake, alcohol
in six Italian areas, 67% assessing usual using the residual intake
1991–1994) intake in the method): highest
2 years before (> 16.52) vs.
diagnosis lowest (≤10.49)
quintile
Cade et al. Hospital-based All women Cases: 50–65 y, 25-item, interviewer- Dietary iron All: 0.49 Yes Age, age at menarche, 5
1998 [54], UK case-control (220/825) 86%; administered FFQ (mg/day, crude): (0.23–1.01) age at first birth, social
(UK breast Controls: 50–65 y, (completed before highest (NR) vs. class, BMI, smoking,
screening 80% mammogram lowest (NR) components of calorie
programme clinics results were known) quartile intake (alcohol, complex
in southern England, and 141-item, carbohydrates, protein,
1990–1992) validated, self- polyunsaturated fat,
administered FFQ monounsaturated fat,
(completed after saturated fat,
clinic visit) assessing cholesterol, sugar),
usual intake over non-caloric nutrients
the past year (vitamin E)
Levi et al. Hospital-based All women Cases: < 75 y, 69%; 79-item, validated, Dietary iron All: 1.21 Yes Age, education, parity, 6
2001 [55], case-control (289/442) Controls: 23–74 y, interviewer- (mg/day, crude): (0.65–2.26) menopausal status, BMI,
Switzerland (single-centre, 65% administered FFQ highest (median total energy intake,
1993–1999) assessing usual 16.8) vs. lowest alcohol drinking
intake in the 2 years (median 9.0)
before diagnosis tertile
Adzersen Hospital-based All women Cases: 25–75 y, 161-item, validated, Dietary iron All: 0.66 Yes Age, total energy intake 5
et al. 2003 case-control (310/353) 55%; self-administered (mg/day, crude): (0.32–1.33) without alcohol, age at
[56], Germany (single-centre, Controls: 25–75 y, FFQ assessing highest (> 14.3) menarche, age at first
1998–2000) 57% usual intake in the vs. lowest (< 9.0) birth, age at
year before hospital quartile menopause, mother/
admission sister with breast
cancer, current
smoking, history of BBD
and/or operation,
BMI, alcohol intake,
current HRT or HRT
Page 7 of 28

during the past year

Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Michels et al. Nested case-control All women 25–55 y at cohort 30-item, validated, Dietary iron All: 0.79 Yes Age, age at menarche, 6
2006 [57], (Nurses’ Mothers’ (582/1569) enrolment; 27% self-administered (mg/day, energy- (0.55–1.13) parity, age at first birth,
USA Study, nested in postmenopausal FFQ assessing diet adjusted using family history of breast
Chang et al. BMC Cancer

Nurse’s Health at diagnosis (cases) during preschool the residual cancer, adult BMI, total
Study I and II, age (3–5 y); method): highest energy intake
1976–1993) completed by (mean 7.23) vs.
participants’ lowest (mean
mothers after case 2.54) quintile
diagnosis
Hong et al. Nested case-control Postmenopausal 50–74 y at cohort 68-item, validated, Total iron Post: 1.06 Yes Age, family history of 7
(2019) 19:543

2007 [58], (American Cancer (502/505) enrolment; 100% self-administered (mg/day, energy- (0.77–1.47)c breast cancer, HRT,
USA Society Cancer postmenopausal FFQ assessing adjusted using BMI, age at menarche,
Prevention Study at diagnosis usual intake over the residual age at menopause,
II Nutrition the past year; method): highest smoking status, race,
Cohort, completed at (> 22.5) vs. lowest parity (crude RRs
1992–2001); 10 y baseline of the (≤9.6) tertile calculated from raw
original cohort tabulated data)
study Iron-containing Post: 1.13 Yes
multivitamin (0.87–1.49)c
supplement use:
yes vs. no
Kabat et al. Prospective cohort All women 40–59 y and 37% 86-item, validated, Dietary iron All: 0.97 Yes Age, BMI, menopausal 8
2007 [59], (Canadian National (2491/48662) postmenopausal self-administered (mg/day, energy- (0.85–1.10) status, parity, age at
Canada Breast Screening Premenopausal at baseline FFQ assessing usual adjusted using Pre: 1.07 menarche, family
Study, 1982–2000); cases: 1171 intake reported at the residual (0.89–1.30) history of breast cancer
mean 16.4 y Postmenopausal baseline method): highest Post: 0.87 in a first-degree relative,
cases: 993 (≥14.99) vs. lowest (0.71–1.06) history of BBD, OC use,
(< 11.90) quintile HRT, total energy intake,
alcohol intake,
Heme iron All: 1.03 Yes education, study centre,
(mg/day, energy- (0.90–1.18) randomisation group in
adjusted using Pre: 1.03 the original trial
the residual (0.84–1.25)
method): highest Post: 0.97
(> 2.95) vs. lowest (0.78–1.20)
(< 1.58) quintile
Kallianpur Population-based All women Cases: 25–70 y, 76-item, validated, Dietary iron All: 1.31 Yes Age, education, BMI, 8
et al. 2008 case-control (3452/3474) 40%; interviewer- (mg/day, crude): (0.96–1.78) waist-to-hip ratio, age
[60], China (Shanghai Breast Premenopausal Controls: 25–70 y, administered FFQ highest (NR) vs. Pre: 1.30 at menarche, age at
Cancer Study I and (2086/1968) 43% assessing usual lowest (NR) (0.86–1.97) first live birth, family
II, 1996–1998 and Postmenopausal intake over the past quartile Post: 1.33 history of breast
2002–2005) (1366/1506) 5 years, ignoring (0.83–2.14) cancer, regular
any recent changes exercise, total energy
Animal-derived All: 1.50 Yes (as a intake, study phase,
(largely heme) (1.19–1.88) proxy for age at menopause,
iron (mg/day, Pre: 1.61 heme iron) vitamin A, vitamin C,
crude): highest (1.20–2.15) vitamin E, folic acid,
(NR) vs. lowest Post: 1.42 isoflavone intake,
(NR) quartile (0.98–2.04) vitamin supplement
Page 8 of 28

use, saturated fat and

Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Plant-derived All: 0.99 No (only study monounsaturated fat
(non-heme) iron (0.75–1.29) that reported intake
(mg/day, crude): Pre: 0.96 this measure)
Chang et al. BMC Cancer

highest (NR) vs. (0.67–1.36)

lowest (NR) Post: 1.02
quartile (0.67–1.56)
Ferrucci et Prospective cohort Postmenopausal 55–74 y and 100% 124-item, validated, Total iron Post: 1.08 Yes Age, race, education, 7
al. 2009 [39], (Prostate, Lung, (1205/52158) postmenopausal self-administered (mg/day, energy- (0.90–1.30) study centre,
USA Colorectal, and at baseline FFQ assessing usual adjusted using randomisation group,
Ovarian [PLCO] intake over the past the residual family history of
(2019) 19:543

Cancer Screening year; completed at method): highest breast cancer, age at

Trial, 1998–2006); baseline (> 31.2) vs. lowest menarche, age at
mean 5.5 y (≤11.4) quintile menopause, age at
first birth and number
Dietary iron Post: 1.25 Yes of live births, history
(mg/1000 kcal, (1.02–1.52) of BBD, number of
energy-adjusted mammograms during
by nutrient past 3 years,
density): highest menopausal HRT use,
(> 10.3) vs. lowest BMI, alcohol intake,
(≤6.9) quintile total fat intake, total
Heme iron Post: 1.12 Yes energy intake
(mg/1000 kcal, (0.92–1.38)
energy-adjusted
by nutrient
density): highest
(> 0.23) vs. lowest
(≤0.07) quintile
Supplemental iron Post: 1.00 Yes
(mg/day): highest (0.74–1.35)
(21.4–39.4) vs.
lowest (0)
category
Moore et Nested case-control All women 30–63 y at cohort 115-item, validated, Dietary iron All: 0.96 Yes Age, year of interview, 4
al. 2009 [61], (Breast Self (248/1040) enrolment; 35% interviewer- (mg/day, crude): (0.53–1.77) total energy intake,
China Examination Trial postmenopausal administered FFQ highest (> 17.5) dietary vitamin C intake
cohort, 1989–2000) at diagnosis assessing usual intake vs. lowest (≤12.0)
in adult life; quartile
completed prior to
biopsy
Kabat et al. Prospective cohort Postmenopausal 50–71 y and 100% 124-item, validated, Dietary iron Post: 1.02 Yes Age, BMI, age at 7
2010 [62], (National Institutes (3396/116674) postmenopausal self-administered (mg/1000 kcal, (0.90–1.15) menarche, age at first
USA of Health at baseline FFQ assessing usual energy-adjusted live birth, family
[NIH]–AARP Diet intake over the past by nutrient history of breast
and Health Study, year (completed at density): highest cancer, menopausal
1995–2003); 6.5 y baseline) and a (≥10.1) vs. lowest HRT, education, race,
second FFQ with (< 6.8) quintile total energy intake,
meat-cooking total fat intake, total
Page 9 of 28
Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
module Heme iron Post: 1.01 No (analysis fibre intake, alcohol
(within 6 months (μg/1000 kcal, (0.89–1.14) based on a intake, physical
after initial FFQ) energy-adjusted subset of a activity, smoking, age
Chang et al. BMC Cancer

by nutrient larger cohort at menopause,

density): highest [66]) number of breast
(≥216.7) vs. lowest biopsies
(< 62.9) quintile
Bradshaw Population-based All women 20–98 y, 67% 101-item, validated, Total iron All: 1.10 Yes Age, total energy 6
et al. 2013 case-control (1463/1500) postmenopausal self-administered (mg/day, crude): (0.71–1.72); intake, carbohydrates,
[42], USA (Long Island Breast FFQ assessing usual per 1-SD (4.41) 1.27 calcium, fibre,
(2019) 19:543

Cancer Study intake 1 year prior increase; highest (0.42–3.96)c magnesium, zinc,
Project, 1996–1997) to study interview (> 12.3) vs. lowest alpha-carotene,
(< 7.1) quartile beta-carotene,
cryptoxanthin, lutein,
lycopene, oleic acid,
pro-alpha carotenes,
vitamin C, vitamin E,
riboflavin, cobalamin,
pyridoxine, folate,
betaine, free choline,
glycerophosphocholine,
methionine, free
phosphocholine,
phosphotidylcholine,
sphingomyelin,
anthocyanidins,
flavan-3-ols, flavanones,
flavones, flavonols,
isoflavones, lignans
Farvid et al. Prospective cohort All women 26–45 y and 0% 130-item, validated, Total iron All: 0.85 Yes Age, race, family history 7
2014 [63], (Nurses’ Health (2830/88803) postmenopausal self-administered (mg/day, energy- (0.75–0.96) of breast cancer in
USA Study II, Premenopausal (i.e., 100% FFQ assessing usual adjusted using Pre: 0.88 mother or sisters,
1991–2011); 20 y cases: 1511 premenopausal) intake over the past the residual (0.74–1.04) history of BBD, smoking,
Postmenopausal at baseline year; completed at method): Post: 0.83 height, BMI, age at
cases: 918 baseline All: highest (0.68–1.01) menarche, parity and
(median 50.9) vs. age at first birth, OC
lowest (median use, alcohol intake,
10.2) quintile energy intake, HRT use,
Pre: highest menopausal status, age
(median 50.9) vs. at menopause
lowest (median
10.2) quintile
Post: highest
(median 44.4) vs.
lowest (median
10.2) quintile
Heme iron All: 1.12 Yes
(mg/day, energy- (0.99–1.28)
Page 10 of 28

adjusted using Pre: 1.15

Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
the residual (0.97–1.37)
method): Post: 0.96
All: highest (0.79–1.17)
Chang et al. BMC Cancer

(median 1.6) vs.

lowest (median
0.6) quintile
Pre: highest
(median 1.6) vs.
lowest (median
0.6) quintile
Post: highest
(2019) 19:543

(median 1.7) vs.

lowest (median
0.7) quintile
Farvid et al. Prospective cohort All women 33–52 y at baseline 124-item, validated, Total iron (mg/day, All: 0.88 No (analysis Age, race, family history 7
2015 [64], (Nurses’ Health (1132/44231) self-administered energy-adjusted (0.72–1.07) based on a of breast cancer in
USA Study II, 1998–2011); 13 y Premenopausal FFQ assessing diet using the residual Pre: 0.88 subset of a mother or sisters, history
cases: 546 during adolescence method): (0.65–1.18) larger cohort of BBD, smoking, height,
Postmenopausal (i.e., 1960–1980); All: highest (median Post: 0.72 [63]) weight gain since age
cases: 483 completed in 1998 17.5) vs. lowest (0.54–0.97) 18, BMI at age 18, age
(start of follow-up) (median 11.7) quintile at menarche, parity and
Pre: highest age at first birth, OC use,
(median 17.6) vs. adolescent alcohol
lowest (median intake, adult alcohol
11.8) quintile intake, adolescent
Post: highest energy intake, HRT use,
(median 16.7) vs. menopausal status, age
lowest (median at menopause
11.6) quintile
Heme iron All: 1.01 No (analysis
(mg/day, energy- (0.83–1.22) based on a
adjusted using Pre: 1.14 subset of a
the residual (0.86–1.51) larger cohort
method): Post: 0.92 [63])
All: highest (0.69–1.22)
(median 2.6) vs.
lowest (median
1.0) quintile
Pre: highest
(median 2.5) vs.
lowest (median
1.0) quintile
Post: highest
(median 2.6) vs.
lowest (median
1.1) quintile
Page 11 of 28
Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Diallo et al. Prospective cohort All women 35–60 y and 30% Repeated 24-h Dietary iron All: 1.67 Yes Age, energy intake 8
2016 [65], (Supplémentation (188/4646) postmenopausal dietary records (mg/day, crude): (1.02–2.71) without alcohol,
France en Vitamines et Premenopausal at baseline administered via a highest (> 11.9) Pre: 1.39 intervention group of
Chang et al. BMC Cancer

Minéraux cases: 59 telephone-based vs. lowest (< 9.3) (0.58–3.29) the initial SU.VI.MAX
Antioxydants Postmenopausal terminal; completed tertile Post: 1.85 trial, number of 24-h
[SU.VI.MAX] trial, cases: 129 every 2 months (1.02–3.34) dietary records,
1994–2007); median during the first smoking status,
12.6 y 2 years of follow-up Iron from red All: 1.00 Yes (as a education, physical
(intake averaged meat (mg/day, (0.70–1.43) proxy for activity, height, BMI,
from ≥3 valid crude): highest heme iron) alcohol intake, family
records) (NR) vs. lowest history of breast cancer,
(2019) 19:543

(NR) tertile lipid intake, HRT use,

number of children;
additionally adjusted for
OC use, heavy period,
and use of hormonal
intrauterine system in
premenopausal women
Inoue-Choi Prospective cohort Postmenopausal 50–71 y and 100% 124-item, validated, Heme iron Post: 1.11 Yes Age, race, BMI, height, 7
et al. 2016 (NIH-AARP Diet and (9305/193742) postmenopausal self-administered (μg/1000 kcal, (1.03–1.19) education, smoking,
[66], USA Health Study, at baseline FFQ assessing energy-adjusted ER+/PR+: alcohol intake, physical
1995–2006); mean usual intake over by nutrient 1.07 activity, family history
9.4 y the past year; density): highest (0.94–1.21) of breast cancer, age
completed at (median 303.5) vs. ER−/PR–: at menarche, age at
baseline lowest (median 1.06 menopause, age at first
44.2) quintile (0.83–1.37) live birth, number of
live births, HRT use, OC
use, number of
previous breast biopsy,
total energy intake,
total fat intake, fibre
intake
Body iron status (n = 11)
Knekt et al. Prospective cohort All women 20–74 y at baseline Serum; Technicon Serum iron All: 0.85 (NR), No Age, smoking 7
1994 [67], (Finnish Mobile (192/18813) AutoAnalyzer (μg/dL): highest trend not (CI missing)
Finland Clinic Health (colorimetric (> 125) vs. lowest significant
Examination Survey, method) (≤69) quartile
1966–1984); mean
14 y Serum TIBC All: 1.00 (NR), No
(μg/dL): highest trend not (CI missing)
(> 388) vs. lowest significant
(≤313) quartile
Serum TSAT (%): All: 0.78 (NR), No
highest (> 36.3) trend not (CI missing)
vs. lowest (< 20.0) significant
quartile
Page 12 of 28
Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Herrinton Prospective cohort All women 20–84 y (39% aged Serum; Serum TSAT (%): All: 1.10 Yes Age, race 7
et al. 1995 (Kaiser Permanente (900/28150) ≥50 y) at baseline Autochemist highest (≥34.5) (0.88–1.30)
[68], USA Northern California multichannel vs. lowest (≤20.3)
Chang et al. BMC Cancer

Multiphasic Health analyzer quartile

Checkup cohort, (colorimetric
1969–1990); mean method)
17.6 y
Garland et al. Nested case-control All women 36–61 y at baseline Toenail; Toenail iron All: 0.89 No (only Age, date of nail return, 8
1996 [69], (Nurses’ Health (433/459) (toenail collection) instrumental (μg/g): highest (0.56–1.40) study that smoking, age at first
USA Study, 1982–1987); Premenopausal neutron activation (> 50.8) vs. lowest Pre: 0.45 reported this birth, parity, history of
(2019) 19:543

4y (193/190) analysis (< 20.3) quintile (0.21–0.95) measure) BBD, history of breast
Postmenopausal Post: 1.56 cancer in mother,
(208/241) (0.80–3.03) history of breast cancer
in a sister, age at
menarche, menopausal
status, BMI, alcohol
consumption
Cui et al. Nested case-control All women 18–85 y (50% of Benign breast Breast tissue iron All: 1.58 No (only Age, age at BBD 9
2007 [70], (Kaiser Permanente (252/252) cases and 55% of tissue; X-ray (ng/cm2, normalized (1.02–2.44) study that diagnosis, duration of
USA Northwest cohort controls fluorescence by sulfur content): Post: 2.77 reported this Kaiser Permanente
of women diagnosed postmenopausal) at spectroscopy highest (NR) vs. (1.25–6.13) measure) membership, age at
with BBD in baseline (BBD lowest (NR) quintile menarche, parity, age
1970–1994) diagnosis) at first live birth, history
of bilateral
oophorectomy, family
history of breast cancer,
BMI, smoking,
menopausal status, OC
use, HRT, presence of
proliferative changes
in benign breast tissue
Moore et al. Nested case-control All women 30–63 y at cohort Plasma; Plasma ferritin All: 1.77 Yes Age, year of blood draw 5
2009 [61], (Breast Self (248/1040) enrolment; 35% immunoradiometric (μg/L): highest (0.96–3.27)
China Examination Trial postmenopausal assay (> 101.9) vs.
cohort, 1989–2000) at diagnosis lowest (≤18.9)
quartile
Stevens et al. Nested case-control All women Age not specified; Serum; Serum ferritin All: 1.3 No (use Matched by age at 7
2011 [71], (Adult Health Study (107/212) 86% chemiluminescent (μg/L): per log (1.0–1.7) tertile data time of blood collection,
Japan cohort, from the Premenopausal postmenopausal enzyme immunoassay unit increase Pre: 1.0 below) menopausal status,
Life Span Study of (15/29) at diagnosis (0.5–1.9) sample collection year,
atomic bomb Postmenopausal (60% with Post: 1.4 and city; adjusted for
survivors, (92/183) postmenopausal (1.1–1.9) radiation dose
1969–2001); mean serum)
13 y (range: 2 to Premenopausal Post: 1.1 Yes
26 y) serum ferritin (0.4–3.5)
(log [μg/L]): highest
(> 3.5 [> 33]) vs. lowest
(< 2.4 [< 11])
Page 13 of 28

tertile
Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Postmenopausal Post: 2.5 Yes
serum ferritin (1.1–5.7)
(log [μg/L]):
Chang et al. BMC Cancer

highest (> 4.4

[> 81]) vs. lowest
(< 3.8 [< 45]) tertile
Gaur et al. Prospective cohort All women ≥20 y at baseline Serum; Serum iron Pre: 1.00 Yes Age, socioeconomic 8
2013 [72], (Apolipoprotein (3238/105795) colorimetric assay (μmol/L): highest (0.83–1.20) status, history of lung
Sweden Mortality Risk Premenopausal on Technicon DAX (≥22) vs. lowest Post: 1.24 disease, CRP; model for
[AMORIS] Study, cases: 1108 96 multichannel (< 14) quartile (1.05–1.47) serum iron also adjusted
(2019) 19:543

1985–2002); mean Postmenopausal analyzer for TIBC and vice versa

10.57 y cases: 2130 Serum TIBC Pre: 1.06 Yes
(μmol/L): highest (0.87–1.29)
(≥67) vs. lowest Post: 1.13
(< 42) quartile (0.96–1.33)
Graff et al. Nested case-control All women 32–54 y and 24% Plasma; Plasma ferritin All: 1.05 Yes Age at blood draw, race, 9
2014 [73], (Nurses’ Health (795/795) postmenopausal electrochemi- (μg/L): highest (0.77–1.45) menopausal status at
USA Study II, 1996–2009); Premenopausal at baseline luminescence (> 72.0) vs. lowest Pre: 1.21 blood draw and
mean 6.1 y (range: (406/402) (blood draw) immunoassay (≤23.9) quartile (0.77–1.88) diagnosis, month/year
1 month to 13.3 y) Postmenopausal Post: 1.09 of blood draw, luteal
(299/301) (0.65–1.83) day at blood draw, time
ER+/PR+: of day at blood draw,
1.14 fasting status at blood
(0.77–1.69) draw, age at menarche,
ER−/PR–: 0.70 BMI at age 18, weight
(0.35–1.39) change since age 18,
parity and age at first
birth, family history of
breast cancer, history of
BBD
Wen et al. Prospective cohort All women ≥20 y at baseline Serum; Serum iron (μg/dL): All: 1.31 Yes Age, BMI, systolic blood 7
2014 [41], (MJ Health screening (913/164355) colorimetric assay highest (≥140) vs. (1.01–1.70); pressure, total cholesterol,
Taiwan cohort, 1997–2008); on Abbott referent (60–79) 1.62 CRP, hemoglobin,
median 7.07 y Architect C8000 category; highest (1.22–2.14)c smoking, alcohol
automatic analyzer (≥140) vs. lowest drinking, physical activity
(< 60) category
Chua et al. Prospective cohort All women 25–79 y and 57% Serum; Serum ferritin All: 0.97 Yes Age, smoking, alcohol 9
2016 [74], (Busselton Health (80/1795) postmenopausal chemiluminescence (μg/L): highest (0.54–1.74) consumption, BMI,
Australia Survey, 1994–2010); Premenopausal at baseline immunoassay (> 103) vs. lowest Pre: 0.44 waist circumference,
15–16 y (39/775) (ferritin), colorimetric (< 53) tertile (0.16–1.23) systolic blood pressure,
Postmenopausal assay (iron), and Post: 1.74 diastolic blood
(41/1020) immunoturbidimetry (0.64–4.72) pressure, high-density
(transferrin) on an lipoprotein,
automated analyser Serum iron All: 1.64 Yes triglycerides, glucose,
(μmol/L): highest (0.90–2.98) HOMA-IR, CRP, alanine
(≥20) vs. lowest Pre: 1.14 transaminase,
(< 16) tertile (0.48–2.70) γ-glutamyltransferase,
Post: 2.16 bilirubin, albumin,
(0.91–5.12)
Page 14 of 28

menopausal status
Table 1 Summary of studies investigating associations between iron intake and iron status and breast cancer risk (Continued)
Author and Study design and Population Age range and Iron intake or Iron intake or Main resultsb Included Adjusted or matched NOS
year [ref], study period/ (no. of cases/ percent status assessment status measure RR (95% CI) in meta- variables score
country setting; duration total or cases/ postmenopausal (unit) and analysis
of follow-up controls)a comparison
Serum TSAT All: 1.90 Yes
(%): highest (≥30) (1.06–3.38)
vs. lowest (< 23) Pre: 1.27
Chang et al. BMC Cancer

tertile (0.54–2.98)
Post: 2.45
(1.08–5.58)
Quintana Case-cohort All women 35–65 y and 40% Serum; Roche Serum ferritin All: 0.67 Yes Age, waist 9
Pacheco (European (627/1466) postmenopausal Cobas 6000 (μg/L): highest (0.49–0.92) circumference, height,
et al. 2018 Prospective at baseline analytical system (median 193) vs. Pre: 0.66 alcohol consumption,
[75], Germany Investigation into lowest (median (0.41–1.05) CRP, smoking status,
(2019) 19:543

Cancer and Nutrition 22) quartile Post: 0.90 education, menopausal

[EPIC]-Heidelberg (0.53–1.54) status, physical activity,
Study, 1994–2009); current aspirin use,
mean 15.7 y in the Serum iron All: 1.04 Yes fibre intake, total red
subcohort (8.4 y (μmol/L): highest (0.78–1.40) meat intake, energy
among breast (median 25) vs. intake, HRT use,
cancer cases) lowest (median current OC use, parity
11) quartile
Serum transferrin All: 0.92 No (only
(μmol/L): highest (0.70–1.23) study that
(median 41) vs. reported this
lowest (median measure)
29) quartile
Serum TSAT (%): All: 1.03 Yes
highest (median (0.77–1.39)
53.8) vs. lowest
(median 22.4)
quartile
Abbreviations: BBD Benign breast disease, BMI Body mass index, CRP C-reactive protein, CI Confidence interval, ER Estrogen receptor, FFQ Food frequency questionnaire, HRT Hormone replacement
therapy, HOMA-IR Homeostatic model assessment of insulin resistance, NOS Newcastle-Ottawa Scale, NR Not reported, OC Oral contraceptive, Post Postmenopausal, Pre Premenopausal, PR Progesterone
receptor, RR Relative risk, SD Standard deviation, TIBC Total iron-binding capacity, TSAT Transferrin saturation
a
Cohort studies: number of cases/total number of participants; case-control studies: number of cases/number of controls
b
Results presented for pre- and postmenopausal breast cancer combined (All), premenopausal breast cancer (Pre), and/or postmenopausal breast cancer (Post), as well as by ER/PR status, where available
c
Calculated using tabulated raw data or RRs reported for other comparisons in the study
Page 15 of 28
Chang et al. BMC Cancer (2019) 19:543 Page 16 of 28

during adolescence [64] and one assessing dietary iron studies examining iron intake, scores ranged from 4 to 8
intake during preschool age [57], with FFQs completed (mean: 6.5), with 10 studies considered to be of
retrospectively by participants at 33–52 years of age high-quality (NOS ≥7) [39, 52, 58–60, 62–66] and seven
(start of follow-up) or by mothers of participants (after studies to be of low-quality (NOS < 7) [42, 53–57, 61].
case diagnosis), respectively. Although the use of a previ- For the 11 studies examining iron status, NOS scores
ously validated (or pilot-tested [52]) FFQ was noted in ranged from 5 to 9 (mean: 7.7), with only one study
all iron intake studies (Table 1), the validity and repro- scoring below 7 [61].
ducibility of the FFQ have not always been directly
assessed among the population under study. For Iron intake and breast cancer risk
example, several studies utilized an FFQ adapted from Highest vs. lowest analysis
one that was designed and validated for a different study Figure 2 shows forest plots for the associations of
without re-evaluating its performance in the current dietary, supplemental, total, and heme iron intake
study population [42, 54–57, 61]. (yes vs. no for supplemental iron; highest vs. lowest
Of the 11 studies examining iron status, nine assessed intake category for all other measures) with breast
one or more serum/plasma biomarkers (ferritin, iron, cancer risk.
transferrin, TIBC, and/or TSAT) [41, 61, 67, 68, 71–75], A meta-analysis combining estimates from 11 studies
while the other two analyzed iron levels in toenail [69] [39, 53–57, 59–62, 65] did not reveal an association be-
or benign breast tissue [70]. Two of these studies noted tween dietary iron intake and breast cancer risk, with a
that biological samples were collected at more than one pooled RR of 1.01 (95% CI: 0.89–1.15); however, signifi-
time point for a small proportion of participants, includ- cant heterogeneity was detected across studies (I2 = 55%,
ing a cohort study where 23% of women provided serum Pheterogeneity = 0.01). With the exception of the one study
samples at two or three health examinations [68] and a reporting intake during preschool age (highest vs. lowest
nested case-control study where 5% of cases and 2% of quintile, mean intakes of 7.23 and 2.54 mg/day, respect-
controls had both pre- and postmenopausal serum sam- ively) [57], dietary iron intake levels across studies
ples [71]. It can be assumed that all other studies involved (where reported) ranged between > 11.9 and > 17.5 mg/
measurements taken at a single time point (i.e., baseline). day for the highest category and between < 9.0 and ≤ 12.0
Results of almost all included studies were reported as mg/day for the lowest (referent) category. Results did
RRs across quantiles (tertiles, quartiles, or quintiles) of not change appreciably when the study assessing
iron intake or status. While age was matched and/or preschool iron intake [57] was excluded from the
adjusted for in all studies, the level of adjustment of analysis (pooled RR = 1.04, 95% CI: 0.91–1.18; I2 = 56%,
other potential confounders differed across studies Pheterogeneity = 0.02).
(Table 1). Most iron intake studies included total energy Similarly, no associations were found for intakes of
intake as a covariate in the multivariable model, regard- supplemental iron (pooled RR = 1.02, 95% CI: 0.91–1.13;
less of whether the iron intake variable (other than sup- I2 = 0%, Pheterogeneity = 0.61) and total iron (pooled RR =
plemental iron) itself was crude (i.e., absolute intake) 0.97, 95% CI: 0.82–1.14; I2 = 46%, Pheterogeneity = 0.14),
[42, 54–56, 60, 61, 65] or adjusted for energy using the based on results combined from three studies [39, 52,
nutrient density [39, 62, 66] or the residual [39, 53, 57– 58] and four studies [39, 42, 58, 63], respectively.
59, 63, 64] method. Other commonly adjusted variables In contrast, heme iron intake showed a significant
in iron intake studies included BMI, alcohol intake, fam- positive association with breast cancer risk based on six
ily history of breast cancer, and reproductive/hormonal studies [39, 59, 60, 63, 65, 66], with a pooled RR of 1.12
factors, such as age at menarche, parity, age at meno- (95% CI: 1.04–1.22) and low-to-moderate heterogeneity
pause, and OC and/or HRT use. Several studies also (I2 = 39%, Pheterogeneity = 0.15). This association persisted
adjusted for education, smoking, physical activity, history after excluding the two studies where animal [60] or red
of benign breast disease (BBD), and/or dietary factors meat [65] derived iron was used as a proxy measure for
(e.g., fat intake). Iron status studies, especially those heme iron intake (pooled RR = 1.10, 95% CI: 1.04–1.16;
where breast cancer was not the only outcome of inter- I2 = 0%, Pheterogeneity = 0.78), or when restricting to stud-
est, generally had more limited adjustment for estab- ies that used a previously developed laboratory-based
lished breast cancer risk factors (e.g., reproductive database to assess heme iron intake [39, 66] (pooled
history). Notably, four recent iron status studies adjusted RR = 1.11, 95% CI: 1.04–1.19; I2 = 0%, Pheterogeneity =
for C-reactive protein (CRP) as a marker of inflamma- 0.94). A meta-analysis was not conducted for non-heme
tion [41, 72, 74, 75]. iron intake, as only one study reported its association
Details of the quality assessment of individual studies (assessed as plant-derived iron) with breast cancer risk
are presented in Additional file 3: Table S1. Overall, (RR [highest vs. lowest quartile] = 0.99, 95% CI: 0.75–
NOS scores ranged from 4 to 9 (mean: 7.0). For the 17 1.29) [60].
Chang et al. BMC Cancer (2019) 19:543 Page 17 of 28

Fig. 2 Forest plot of associations between iron intake (highest vs. lowest category) and breast cancer risk. The diamonds represent the pooled
relative risks and corresponding 95% confidence intervals obtained from random-effects meta-analyses. The dots and horizontal lines represent
the relative risks and corresponding 95% confidence intervals of individual studies, and the sizes of shaded squares are proportional to the
weight contributed by each study to the pooled estimate. I2 is the proportion of the total variability attributable to between-study heterogeneity,
and P is from Cochran’s Q test evaluating the presence of heterogeneity
Chang et al. BMC Cancer (2019) 19:543 Page 18 of 28

Table 2 presents results from subgroup analyses for diet- Body iron status and breast cancer risk
ary, total, and heme iron intake (not conducted for supple- Highest vs. lowest analysis
mental iron due to limited number of studies). The Figure 4 shows forest plots of associations between each
association between dietary iron intake and breast cancer serum/plasma indicator of body iron status (highest vs.
risk did not differ significantly among subgroups defined lowest category) and breast cancer risk. A meta-analysis
by study design, geographic location, menopausal status, combining five RRs derived from four studies [41, 72,
dietary assessment method, or adjustments for specific 74, 75] (one study reported separate RRs for pre- and
confounders (Pdifference > 0.10 from meta-regression), with postmenopausal breast cancer [72]) revealed a significant
substantial heterogeneity remaining within most sub- positive association between serum iron and breast
groups. However, when stratified by study quality, a sig- cancer risk (pooled RR = 1.22, 95% CI: 1.01–1.47), with
nificant inverse association was observed for low-quality significant heterogeneity (I2 = 61%, Pheterogeneity = 0.04).
studies (pooled RR = 0.84, 95% CI: 0.72–0.96), whereas a The associations were also in the positive direction, but
positive but nonsignificant association was seen for not statistically significant, for ferritin (pooled RR = 1.13,
high-quality studies (pooled RR = 1.12, 95% CI: 0.98–1.29) 95% CI: 0.78–1.62; I2 = 65%, Pheterogeneity = 0.01) and
(Pdifference = 0.03), suggesting study quality may be a con- TSAT (pooled RR = 1.16, 95% CI: 0.91–1.47; I2 = 43%,
tributor to heterogeneity. Furthermore, post-hoc subgroup Pheterogeneity = 0.17), based on six RRs from five stud-
analyses stratifying results by the highest dietary iron in- ies [61, 71, 73–75] and three RRs from three studies
take category (> 15 mg/day, ≤15 mg/day, or not reported) [68, 74, 75], respectively. High levels of TIBC, which
and method of energy adjustment (covariate only, nutrient is indicative of low body iron status, was not associ-
density, or residual method) did not reveal significant dif- ated with breast cancer risk when two RRs from one
ferences (Pdifference = 0.96 and 0.32, respectively; data not study [72] were combined (pooled RR = 1.10, 95% CI:
shown). No notable differences were observed for total 0.97–1.25; I2 = 0%, Pheterogeneity = 0.62). Similarly,
iron intake, which remained unassociated with breast can- serum transferrin was not associated with breast can-
cer risk across subgroups. For heme iron intake, all six cer risk according to the only study reporting this
studies were of high-quality, and significant positive asso- measure (RR = 0.92, 95% CI: 0.70–1.23) [75].
ciations remained among cohort studies and studies con- Meta-analysis combining results for transferrin and TIBC
ducted in North America. In addition, heme iron intake (proxy measure of transferrin) also revealed no significant
showed a slightly stronger association with premenopausal association (pooled RR = 1.07, 95% CI: 0.95–1.20; I2 = 0%,
(pooled RR = 1.21, 95% CI: 0.97–1.51) than postmeno- Pheterogeneity = 0.46; data not shown).
pausal (pooled RR = 1.08, 95% CI: 0.99–1.18) breast can- Table 3 presents results from subgroup analyses for
cer, although statistical significance was not reached in serum/plasma ferritin and iron. Despite the lack of asso-
either subgroup. ciation overall, ferritin was significantly associated with
increased breast cancer risk among studies conducted in
Dose-response analysis Asia (pooled RR = 1.81, 95% CI: 1.16–2.83). In general,
Similar to the highest vs. lowest analysis, linear pooled RRs from ferritin studies adjusting for potential
dose-response meta-analyses (Additional file 4: confounders (e.g., BMI, physical activity, alcohol intake)
Figure S1) revealed no associations between either showed nonsignificant inverse associations, whereas
dietary or total iron intake and breast cancer risk, those not adjusting for confounders showed positive as-
with pooled RRs of 1.00 (95% CI: 0.97–1.03) and sociations; slightly stronger positive associations were
1.00 (95% CI: 0.98–1.01), respectively, per 5-mg/day seen among iron studies adjusting for confounders. For
increase in intake. In contrast, each 1-mg/day in- both ferritin and iron, a stronger positive association
crease in heme iron intake, was associated with a was observed for postmenopausal (ferritin: pooled RR =
statistically significant 8% increase in breast cancer 1.23, 95% CI: 0.87–1.75; iron: pooled RR = 1.39, 95% CI:
risk (pooled RR = 1.08, 95% CI: 1.002–1.17). Based 0.90–2.15) than premenopausal (ferritin: pooled RR =
on nonlinear dose-response meta-analyses, no signifi- 0.79, 95% CI: 0.46–1.35; iron: pooled RR = 1.01, 95% CI:
cant curvilinear associations with breast cancer risk 0.84–1.20) breast cancer; however, statistical significance
were found for intakes of dietary iron (Pnonlinearity = was not reached within subgroups.
0.41; Fig. 3a) and total iron (Pnonlinearity = 0.46; Fig.
3b), although a small decrease in risk nearing statis- Dose-response analysis
tical significance was observed across levels of total No significant linear associations were found between
iron. Meanwhile, there appeared to be a threshold any of the serum/plasma indicators of body iron status
effect in the dose-response curve for heme iron and breast cancer risk (Additional file 4: Figure S2). The
intake (Pnonlinearity = 0.03; Fig. 3c), with risk leveling dose-response curve for ferritin suggested a tendency
off at approximately 1 mg/day. towards a decrease in breast cancer risk with increasing
Chang et al. BMC Cancer (2019) 19:543 Page 19 of 28

Table 2 Subgroup analyses for the associations of dietary, total, and heme iron intake with breast cancer risk
Subgroups Dietary iron intake (highest vs. lowest) Total iron intake (highest vs. lowest) Heme iron intake (highest vs. lowest)
No. RR I2 Pheterogeneityb Pdifferencec No. RR I2 Pheterogeneityb No. RR I2 Pheterogeneityb
of (95% CI) (%)a of (95% CI) (%)a of (95% CI) (%)a
RRs RRs RRs
Overall 11 1.01 55 0.01 4 0.97 46 0.14 6 1.12 39 0.15
(0.89–1.15) (0.82–1.14) (1.04–1.22)
Study design 0.20
Cohort 4 1.10 63 0.05 2 0.95 78 0.03 5 1.10 0 0.86
(0.94–1.27) (0.75–1.20) (1.04–1.16)
Case-control 7 0.91 45 0.10 2 1.08 0 0.76 1 1.50 NA NA
(0.73–1.12) (0.79–1.47) (1.19–1.88)
Geographic 0.65
location
North America 4 1.02 54 0.09 4 0.97 46 0.14 4 1.10 0 0.78
(0.90–1.16) (0.82–1.14) (1.04–1.16)
Europe 5 0.93 63 0.03 0 NA NA NA 1 1.00 NA NA
(0.65–1.34) (0.70–1.43)
Asia 2 1.23 0 0.37 0 NA NA NA 1 1.50 NA NA
(0.93–1.62) (1.19–1.88)
Menopausal 0.78
statusd
Premenopausal 3 1.12 0 0.62 1 0.88 NA NA 3 1.21 68 0.05
(0.94–1.32) (0.74–1.04) (0.97–1.51)
Postmenopausal 5 1.11 64 0.03 3 0.97 50 0.14 5 1.08 21 0.28
(0.92–1.33) (0.81–1.17) (0.99–1.18)
Study quality 0.03
High (NOS 5 1.12 60 0.04 3 0.97 62 0.07 6 1.12 39 0.15
score ≥ 7) (0.98–1.29) (0.81–1.16) (1.04–1.22)
Low (NOS 6 0.84 0 0.53 1 1.27 NA NA 0 NA NA NA
score < 7) (0.72–0.96) (0.41–3.92)
Dietary assessment 0.34
method
Structured 5 1.13 63 0.03 0 NA NA NA 2 1.25 72 0.06
interview (0.85–1.51) (0.85–1.86)
Self- 6 0.98 57 0.04 4 0.97 46 0.14 4 1.10 0 0.78
administered (0.84–1.13) (0.82–1.14) (1.04–1.16)
Adjustments for confounders
BMI 0.33
Yes 9 1.05 56 0.02 2 0.95 78 0.03 6 1.12 39 0.15
(0.91–1.21) (0.75–1.20) (1.04–1.22)
No 2 0.86 0 0.70 2 1.08 0 0.76 0 NA NA NA
(0.72–1.02) (0.79–1.47)
Physical activity 0.13
Yes 3 1.22 63 0.07 0 NA NA NA 3 1.20 70 0.04
(0.93–1.59) (0.96–1.48)
No 8 0.94 52 0.04 4 0.97 46 0.14 3 1.08 0 0.64
(0.80–1.10) (0.82–1.14) (1.00–1.18)
Alcohol intake 0.95
Yes 8 1.01 61 0.01 2 0.95 78 0.03 5 1.10 0 0.86
(0.88–1.17) (0.75–1.20) (1.04–1.16)
No 3 1.02 55 0.11 2 1.08 0 0.76 1 1.50 NA NA
(0.72–1.44) (0.79–1.47) (1.19–1.88)
OC and/or 0.35
HRT use
Chang et al. BMC Cancer (2019) 19:543 Page 20 of 28

Table 2 Subgroup analyses for the associations of dietary, total, and heme iron intake with breast cancer risk (Continued)
Subgroups Dietary iron intake (highest vs. lowest) Total iron intake (highest vs. lowest) Heme iron intake (highest vs. lowest)
No. RR I2 Pheterogeneityb Pdifferencec No. RR I2 Pheterogeneityb No. RR I2 Pheterogeneityb
of (95% CI) (%)a of (95% CI) (%)a of (95% CI) (%)a
RRs RRs RRs
Yes 5 1.08 59 0.05 2 0.95 78 0.03 5 1.10 0 0.86
(0.93–1.25) (0.75–1.20) (1.04–1.16)
No 6 0.93 50 0.08 2 1.08 0 0.76 1 1.50 NA NA
(0.74–1.17) (0.79–1.47) (1.19–1.88)
Family history of 0.20
breast cancer
Yes 7 1.07 57 0.03 2 0.95 78 0.03 6 1.12 39 0.15
(0.93–1.23) (0.75–1.20) (1.04–1.22)
No 4 0.86 15 0.32 2 1.08 0 0.76 0 NA NA NA
(0.69–1.07) (0.79–1.47)
Abbreviations: BMI Body mass index, CI Confidence interval, HRT Hormone replacement therapy, NA Not applicable, NOS Newcastle-Ottawa Scale, OC Oral
contraceptive, RR Relative risk
a 2
I statistics indicating the proportion of the total variability attributable to between-study heterogeneity
b
P values from Cochran’s Q test evaluating the presence of heterogeneity across studies
c
P values for difference between subgroups calculated from meta-regression, conducted only for dietary iron intake (i.e., at least 10 studies available)
d
Pooled estimates were calculated only from studies providing menopausal status-specific results

concentration; however, the CIs were wide due to het- Egger’s P = 0.91), supplemental iron (Begg’s P = 0.60; Egger’s
erogeneous results and included the null value across all P = 0.35), total iron (Begg’s P > 0.99; Egger’s P = 0.39), and
ferritin levels, and no departure from linearity was de- heme iron (Begg’s P = 0.85; Egger’s P = 0.64) intake, or for
tected (Pnonlinearity = 0.70) (Fig. 5a). On the other hand, serum/plasma ferritin (Begg’s P = 0.19; Egger’s P = 0.17),
serum iron exhibited a J-shaped dose-response relation- iron (Begg’s P = 0.62; Egger’s P = 0.47), or TSAT (Begg’s
ship with breast cancer risk, with strong evidence of a P = 0.60; Egger’s P = 0.41), whereas publication bias was
nonlinear effect (Pnonlinearity < 0.001) (Fig. 5b). Specific- detected for the combined analysis of TIBC and trans-
ally, a steady increase in risk was noted for serum iron ferrin (Begg’s P = 0.12; Egger’s P = 0.02). Visual inspec-
levels above ~ 100 μg/dL, with the association becoming tion of the funnel plots (Additional file 5: Figures S3
statistically significant at just beyond ~ 125 μg/dL. No and S4) indicated some asymmetry for total iron intake
evidence of curvilinear associations was found for TIBC and serum/plasma indicators of iron status (i.e., ferritin,
or TSAT (data not shown). iron, and TSAT), where smaller studies with inverse
associations may have been excluded; however, this
Other iron biomarkers was based only on a limited number of studies.
Two nested case-control studies assessed iron biomarkers There was no statistical evidence of publication bias
in samples other than serum or plasma [69, 70]. In the in the dose-response meta-analyses (Begg’s and
only study assessing toenail iron in relation to breast can- Egger’s P > 0.10 for all).
cer risk, no overall association was observed (RR [highest No notable changes in the pooled estimates were
vs. lowest quintile] = 0.89, 95% CI: 0.56–1.40) [69]. How- observed when individual studies were omitted one at
ever, when stratified by menopausal status, toenail iron time in the sensitivity analyses (Additional file 6: Figures
was inversely associated with premenopausal (RR [highest S5–S8), although the association between the highest
vs. lowest quintile] = 0.45, 95% CI: 0.21–0.95) and posi- (vs. lowest) level of serum iron and breast cancer risk
tively associated with postmenopausal (RR [highest vs. lost statistical significance in some cases given the small
lowest quintile] = 1.56, 95% CI: 0.80–3.03) breast cancer number of studies.
(Pinteraction = 0.08). In another study where iron levels were
measured in benign breast tissue among women with Discussion
BBD, an elevated breast cancer risk was observed overall The results of our systematic review and meta-analysis
(RR [highest vs. lowest quintile] = 1.58, 95% CI: 1.02–2.44) suggest that heme iron intake is positively associated
and in postmenopausal women (RR [highest vs. lowest with breast cancer risk, with a statistically significant
quintile] = 2.77, 95% CI: 1.25–6.13) [70]. 12% increase in risk when comparing the highest vs.
lowest level of intake and 8% increase in risk for each
Publication bias and sensitivity analysis 1-mg/day increase in intake. In contrast, no associations
No publication bias was detected in the highest vs. were found for dietary, supplemental, total, or non-heme
lowest meta-analyses for dietary iron (Begg’s P = 0.48; iron intake. Among serum/plasma indicators of body
Chang et al. BMC Cancer (2019) 19:543 Page 21 of 28

Fig. 3 Dose-response curves for intakes of (a) dietary iron; (b) total iron; and (c) heme iron in relation to breast cancer risk. Data were modeled
using random-effects restricted cubic spline models with three knots fixed at the 10th, 50th, and 90th percentiles. The solid lines represent the
fitted relative risks for the nonlinear trend, and the dashed lines represent pointwise 95% confidence intervals

iron status, the highest (vs. lowest) level of iron, but not specific iron biomarker terms, such as “ferritin” and
ferritin, transferrin, TIBC, or TSAT, also showed a statis- “transferrin”), may also explain the considerably smaller
tically significant association with increased breast number of studies identified in the previous review.
cancer risk (22%). Furthermore, dose-response meta- In contrast to our finding of a positive association
analyses indicated a nonlinear threshold effect for heme between heme iron intake and breast cancer risk, the
iron intake and a J-shaped pattern for serum iron in previous meta-analysis reported a lack of association
relation to breast cancer risk. between heme iron intake and breast cancer risk (pooled
This is the first systematic review and meta-analysis RR [per 1-mg/day] = 1.03, 95% CI: 0.97–1.09) based on
specifically assessing breast cancer risk in relation to only three studies [29]. The inclusion of recent
various measures of iron intake and body iron status. additional studies in our analysis, including larger cohort
Our review identified many additional studies not in- studies with longer follow-up [63, 65, 66], likely in-
cluded in the previous systematic review/meta-analysis creased statistical power to detect the relatively modest
on iron and cancer risk [29], which only identified seven association. Our results were, however, consistent with
studies assessing iron intake [39, 54–56, 59, 60, 62] and meta-analyses evaluating heme iron intake in relation to
zero studies assessing body iron status (versus 17 and 11 colorectal cancer risk [29, 76, 77]. The catalytic effects
studies, respectively, in our review), in relation to breast of heme iron on endogenous N-nitrosation and lipid
cancer risk. While this discrepancy is partly due to the peroxidation, and subsequent oxidative damage to cellu-
narrower range of publication year (1995–2012) [29] lar biomolecules, have been suggested to contribute to
compared to the current review (up to 2018), the use of the development of both colorectal and breast cancer
only one electronic database (versus four databases plus [76, 78]. Furthermore, differences in bioavailability may
manual search of reference lists in our review), as well explain why an association with breast cancer risk was
as a limited set of relevant search terms (e.g., missing found only for heme, and not for non-heme (or overall
Chang et al. BMC Cancer (2019) 19:543 Page 22 of 28

Fig. 4 Forest plot of associations between serum/plasma indicators of body iron status (highest vs. lowest category) and breast cancer risk. The
diamonds represent the pooled relative risks and corresponding 95% confidence intervals obtained from random-effects meta-analyses. The dots and
horizontal lines represent the relative risks and corresponding 95% confidence intervals of individual studies, and the sizes of shaded squares are
proportional to the weight contributed by each study to the pooled estimate. I2 is the proportion of the total variability attributable to between-study
heterogeneity, and P is from Cochran’s Q test evaluating the presence of heterogeneity. *Stevens et al. 2011 [71] reported separate estimates for
premenopausal (pre/post) and postmenopausal (post/post) ferritin levels in relation to postmenopausal breast cancer risk; Gaur et al. 2013 [72]
reported separate estimates for premenopausal (pre) and postmenopausal (post) breast cancer

dietary), iron intake [79]. Surrounded by a water-soluble Interestingly, heme iron intake exhibited a nonlinear
porphyrin ring, heme iron is more efficiently absorbed threshold effect in our dose-response meta-analysis,
by intestinal cells [79] and is a stronger predictor of although absolute intake values should be interpreted
body iron status [80–82] compared to non-heme iron. with caution given differences in methods used to assess
Heme iron absorption is also less influenced by the heme iron levels across studies. For example, while
body’s iron requirements or the presence of other dietary literature-based meat-specific percentages (e.g., 69% in
components known to enhance (e.g., vitamin C) or beef, 39% in pork/ham/luncheon meats, 26% in chicken
inhibit (e.g., phytate) non-heme iron uptake [83]. and fish, 21% in liver) [13, 14] were applied in some
Chang et al. BMC Cancer (2019) 19:543 Page 23 of 28

Table 3 Subgroup analyses for the associations of serum/plasma ferritin and iron with breast cancer risk
Subgroups Serum/plasma ferritin (highest vs. lowest) Serum/plasma iron (highest vs. lowest)
No. of RRs RR (95% CI) I2 (%)a Pheterogeneityb No. of RRs RR (95% CI) I2 (%)a Pheterogeneityb
Overall 6 1.13 (0.78–1.62) 65 0.01 5 1.22 (1.01–1.47) 61 0.04
Study design
Cohort 1 0.97 (0.54–1.74) NA NA 4 1.27 (1.02–1.59) 68 0.03
Nested case-control/case-cohort 5 1.18 (0.76–1.84) 72 0.01 1 1.04 (0.78–1.40) NA NA
Geographic location
North America 1 1.05 (0.77–1.45) NA NA 0 NA NA NA
Europe 1 0.67 (0.49–0.92) NA NA 3 1.10 (0.95–1.28) 35 0.21
Asia 3 1.81 (1.16–2.83) 0 0.50 1 1.62 (1.22–2.14) NA NA
Australia 1 0.97 (0.54–1.74) NA NA 1 1.64 (0.90–2.98) NA NA
c
Menopausal status
Premenopausal 3 0.79 (0.46–1.35) 61 0.08 2 1.01 (0.84–1.20) 0 0.78
Postmenopausal 5 1.23 (0.87–1.75) 18 0.30 2 1.39 (0.90–2.15) 35 0.21
Study quality
High (NOS score ≥ 7) 5 1.02 (0.70–1.48) 61 0.04 5 1.22 (1.01–1.47) 61 0.04
Low (NOS score < 7) 1 1.77 (0.96–3.27) NA NA 0 NA NA NA
Biological sample
Serum 4 1.06 (0.61–1.85) 67 0.03 5 1.22 (1.01–1.47) 61 0.04
Plasma 2 1.27 (0.78–2.09) 55 0.14 0 NA NA NA
Adjustments for confounders
BMI
Yes 3 0.86 (0.63–1.19) 51 0.13 3 1.36 (0.98–1.90) 61 0.08
No 3 1.81 (1.16–2.83) 0 0.50 2 1.12 (0.91–1.38) 65 0.09
Physical activity
Yes 1 0.67 (0.49–0.92) NA NA 2 1.30 (0.84–2.00) 78 0.03
No 5 1.28 (0.93–1.76) 31 0.21 3 1.16 (0.94–1.43) 54 0.12
Alcohol intake
Yes 2 0.74 (0.54–1.03) 18 0.27 3 1.36 (0.98–1.90) 61 0.08
No 4 1.41 (0.93–2.13) 42 0.16 2 1.12 (0.91–1.38) 65 0.09
OC and/or HRT use
Yes 1 0.67 (0.49–0.92) NA NA 1 1.04 (0.78–1.40) NA NA
No 5 1.28 (0.93–1.76) 31 0.21 4 1.27 (1.02–1.59) 68 0.03
Family history of breast cancer
Yes 1 1.05 (0.77–1.45) NA NA 0 NA NA NA
No 5 1.19 (0.71–2.00) 72 0.01 5 1.22 (1.01–1.47) 61 0.04
Abbreviations: BMI Body mass index, CI Confidence interval, HRT Hormone replacement therapy, NA Not applicable, NOS Newcastle-Ottawa Scale, OC Oral
contraceptive, RR Relative risk
a 2
I statistics indicating the proportion of the total variability attributable to between-study heterogeneity
b
P values from Cochran’s Q test evaluating the presence of heterogeneity across studies
c
Pooled estimates were calculated only from studies providing menopausal status-specific results

studies [59, 63], others [39, 62, 66] used a with breast cancer risk was observed in one or more of
laboratory-based heme iron database (restricted to certain the middle heme iron intake quantiles and leveled off (or
meats) that accounts for meat type, cooking method, and became weaker and lost statistical significance) in the
doneness level [14]. Nevertheless, regardless of heme iron highest quantile [39, 59, 62, 63]. Furthermore, since heme
assessment method, this threshold effect was also evident iron is derived only from animal source foods, with par-
in several individual studies where a significant association ticularly high content in (and hence highly correlated
Chang et al. BMC Cancer (2019) 19:543 Page 24 of 28

reporting of dietary intake between cases and controls,

as well as potential biases related to hospital-based con-
trol selection, may have contributed to the inverse asso-
ciation. Even among higher-quality studies (all cohort or
population-based case-control) where the association
was in the positive direction, substantial heterogeneity
was present. Given the association observed between
heme iron intake and breast cancer risk, the relative
contribution of heme and non-heme iron to overall diet-
ary iron intake among different study populations may
be another possible source of heterogeneity. This also
highlights the importance of considering different sub-
types of dietary iron in future studies, especially since
our review only identified one study assessing non-heme
(plant-derived) iron in addition to heme iron intake [60].
Our meta-analysis provided no evidence of an associ-
ation between supplemental or total iron intake and
breast cancer risk; however, the dose-response curve for
total iron intake (Fig. 3b) suggested a weakly protective
effect nearing statistical significance. Residual confound-
ing by health behaviours, which are likely associated
with supplement use (and hence high levels of total iron
intake), may have partly contributed to this inverse
trend. Nevertheless, only a few studies examined these
associations, and the assessment of supplemental iron
was not always comprehensive (e.g., did not include both
single-ingredient iron supplements and iron-containing
multivitamin/mineral products) or clearly described. Al-
Fig. 5 Dose-response curves for serum/plasma (a) ferritin and (b) though iron supplements typically contain non-heme
iron in relation to breast cancer risk. Data were modeled using iron (e.g., ferrous sulfate), they deliver high doses of iron
random-effects restricted cubic spline models with three knots fixed that account for a major proportion of total iron intake
at the 10th, 50th, and 90th percentiles. The solid lines represent the
among users [88]. Moreover, use of iron-containing sup-
fitted relative risks for the nonlinear trend, and the dashed lines
represent pointwise 95% confidence intervals plements (both single and multi-ingredient products) is
especially common among female populations [89, 90],
suggesting that dietary iron alone would substantially
with) red meat, the possibility that other red meat compo- underestimate total iron intake. Thus, additional studies
nents (e.g., fat, meat mutagens) contributed to the associ- with detailed assessments of supplemental (and total)
ation observed cannot be precluded [39, 84]. However, the iron intake, including dosage and frequency/duration of
association between red meat intake (and meat mutagens) use, are warranted before reaching firm conclusions
and breast cancer risk is not strongly supported by current about their associations with breast cancer risk.
epidemiologic literature [39, 84–87], and most studies In contrast to the lack of association observed for total
assessing heme iron in our review adjusted for fat intake and dietary iron intake, high levels of serum iron were
as a potential confounder. found to be associated with increased breast cancer risk.
No overall association was observed between dietary Similarly, although based only on single studies, iron
iron intake and breast cancer risk, although results were measured in toenail [69] and benign breast tissue [70]
heterogeneous, with studies reporting positive [39, 65], may also exhibit a positive association with breast cancer
inverse [53, 54], or null [55–57, 59–62] associations. risk, especially in postmenopausal women. Given limita-
While subgroup and meta-regression analyses suggested tions of dietary assessment, as well as inter-individual
that study quality may partly explain the heterogeneity, a variation in iron absorption and metabolism, biomarkers
closer examination of the lower-quality studies (where may better reflect exposure to biologically available iron
an inverse association was found) indicated that they than dietary intake [91]. Thus, our results provide some
were either hospital-based [53–56] or nested [57, 61] support for the role of biologically available iron in
case-control studies with FFQs administered after breast breast cancer etiology. Conversely, the review by
cancer diagnosis or just before biopsy. Thus, differential Fonseca-Nunes et al. suggested an inverse association
Chang et al. BMC Cancer (2019) 19:543 Page 25 of 28

between body iron status and gastrointestinal cancers intake [66] and one assessing ferritin [73]) investigated as-
[29], while several studies reported sex differences in the sociations according to tumour hormone receptor sub-
associations of iron biomarkers (e.g., positive for women type. Although neither of these studies reported
and inverse or null for men) with overall cancer risk [74, significant differences, additional studies assessing ER/PR
92]. It is possible that iron exerts different effects on dif- status are warranted to explore potential etiologic hetero-
ferent cancer sites and in women (vs. men), among whom geneity, especially given in vitro evidence suggesting a
iron-induced carcinogenesis likely involves a complex stronger role of iron in ER-positive breast carcinogenesis
interplay with reproductive/hormonal factors [7, 93]. [98].
Furthermore, the J-shaped dose-response we observed This is the first systematic review and meta-analysis
between serum iron and breast cancer risk is similar to a specifically assessing the associations between various
study assessing serum iron in relation to overall cancer measures of iron intake, as well as body iron status, and
risk [41]. Individuals with very low body iron levels, such risk of breast cancer. A major strength is the extensive
as those with iron-deficiency anemia, may be distinct search strategy, allowing us to identify many additional
from others (e.g., altered immune function) with re- studies not included in the 2014 review on iron and can-
spect to cancer risk [94], suggesting the need to con- cer risk [29], especially those evaluating body iron status
sider these individuals as a separate group or to assess in relation to breast cancer risk. Importantly, our review
iron levels as a continuous variable without assuming a included detailed assessments of study quality and pro-
linear dose-response. vided a comprehensive and quantitative synthesis of
The significant positive association observed for serum findings, including subgroup analyses to explore sources
iron but not ferritin in our meta-analysis has also been of heterogeneity. Furthermore, in addition to category-
reported by one study examining multiple iron bio- based (highest vs. lowest) analyses, we also performed
markers within a single population [74], suggesting that dose-response meta-analyses to examine linear and non-
circulating iron may be more relevant to breast carcino- linear relationships.
genesis than stored iron (ferritin); however, this warrants Several limitations should be considered when inter-
additional investigation, given the significant heterogen- preting the findings of this review. First, the meta-ana-
eity detected for both ferritin and iron and the small lysis for some iron measures was based only on a small
number of studies assessing these measures. Serum iron number of studies, which could have resulted in limited
has been suggested as a poorer indicator of iron status statistical power for the overall or subgroup analyses and
and is subject to greater within-person variability (30%) the assessments of publication bias, as well as greater in-
compared to ferritin (10–25%) [18]. In addition, serum fluence of single studies. Nevertheless, sensitivity ana-
biomarkers may not be reliable indicators of iron status lyses with individual studies omitted one at a time
in the presence of inflammation, where ferritin levels are generally led to no notable changes in the pooled esti-
elevated and iron and transferrin are decreased [95]. mates, indicating the robustness of our results. Second,
These limitations highlight the need to measure iron our restricted cubic spline dose-response analyses were
biomarkers at multiple time points, to explore use of limited to the range of exposure values derived from in-
novel or more stable indicators of iron status, and to dividual studies, and the trends observed may be driven
evaluate the potential impact of inflammation on iron by single studies with more extreme exposure values.
status measures in future studies. Thus, a larger number of homogeneous studies across a
Although stratified analyses by menopausal status gen- wide range of exposure values are needed to confirm
erally revealed no significant associations due to limited these results in the future. Third, as an inherent issue in
statistical power, several indicators of iron status (serum/ meta-analyses, our analyses combined risk estimates
plasma ferritin and iron, toenail iron, and breast tissue across studies with different designs, populations,
iron) appeared to be more strongly associated with in- settings, statistical adjustments of covariates, etc. (some
creased postmenopausal breast cancer risk. A possible of which were explored in our subgroup analyses),
explanation is age-related dysregulation of iron metabol- which likely contributed to heterogeneity in our re-
ism and declines in antioxidant defense mechanisms sults. Finally, genetic association studies were not con-
[96]. Conversely, the associations for iron intake did not sidered. For example, mutations in the HFE gene
differ by menopausal status, except for a slightly stron- underlying hereditary hemochromatosis (iron overload)
ger association between heme iron intake and premeno- have been implicated in several cancers, including
pausal breast cancer risk; this is unexpected since breast cancer [99]. Although it was beyond the scope
postmenopausal women, who are no longer losing iron of this review, inclusion of such studies, as well as a
through menstruation, are more likely to accumulate closer examination of possible iron-gene interactions,
iron in their body [97]. Furthermore, it was surprising to may provide a more complete picture of iron’s role in
find that only two studies (one assessing heme iron breast cancer etiology.
Chang et al. BMC Cancer (2019) 19:543 Page 26 of 28

Conclusions Authors’ contributions

This systematic review and meta-analysis suggests that VCC contributed to the conception and design of the systematic review and
meta-analysis, conducted the literature search, screened studies for inclusion,
heme iron intake and serum iron levels may be positively extracted data from individual studies, evaluated study quality, performed
associated with breast cancer risk, whereas no associa- the statistical analyses and interpreted the data, and wrote the first draft of
tions were found for intakes of dietary, supplemental, or the manuscript. MC supervised and contributed to the conception and de-
sign of the systematic review and meta-analysis, interpreted the data, and
total iron, or serum/plasma levels of ferritin, TIBC, or critically revised the manuscript for its intellectual content. EK contributed to
TSAT. Although the increases in risk were modest, our the screening of studies for inclusion, verified extracted data from individual
findings may have public health implications given the studies, evaluated study quality, provided guidance on statistical analyses,
interpreted the data, and critically revised the manuscript for its intellectual
widespread consumption of (heme) iron-rich foods. In content. All authors read and approved the final manuscript.
light of the methodological and research gaps identified
in our review (e.g., consideration of different sources/ Ethics approval and consent to participate
subtypes of dietary iron intake, comprehensive assess- Not applicable.
ment of supplemental and total iron intake, repeated
Consent for publication
measures of iron intake/status at multiple time points, Not applicable.
exploration of nonlinear trends, stratification of results
by menopausal and hormone receptor status), further Competing interests
research is needed to better elucidate the association The authors declare that they have no competing interests.
between iron intake/status and risk of breast cancer.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in
Additional files published maps and institutional affiliations.

Additional file 1: Electronic database search strategy. (DOCX 17 kb) Author details
1
Dalla Lana School of Public Health, University of Toronto, 155 College Street,
Additional file 2: NOS coding manuals for study quality assessment. 6th Floor, Toronto, ON M5T 3M7, Canada. 2Prevention and Cancer Control,
(DOCX 21 kb) Cancer Care Ontario, 620 University Avenue, Toronto, ON M5G 2L7, Canada.
3
Additional file 3: Table S1. Quality of included studies assessed using Analytics and Informatics, Cancer Care Ontario, Toronto, ON, Canada.
the NOS. (DOCX 29 kb)
Additional file 4: Figures S1 and S2. Linear dose-response analyses Received: 26 November 2018 Accepted: 26 April 2019
of associations between iron intake/status and breast cancer risk.
(PDF 298 kb)
Additional file 5: Figures S3 and S4. Funnel plots for the evaluation of References
publication bias. (PDF 307 kb) 1. Abbaspour N, Hurrell R, Kelishadi R. Review on iron and its importance for
human health. J Res Med Sci. 2014;19(2):164–74.
Additional file 6: Figures S5–S8. Sensitivity analyses investigating the
2. Galaris D, Pantopoulos K. Oxidative stress and iron homeostasis: mechanistic
influence of individual studies. (PDF 210 kb)
and health aspects. Crit Rev Clin Lab Sci. 2008;45(1):1–23.
3. Toyokuni S. Role of iron in carcinogenesis: cancer as a ferrotoxic disease.
Cancer Sci. 2009;100(1):9–16.
Abbreviations 4. Torti SV, Torti FM. Iron and cancer: more ore to be mined. Nat Rev Cancer.
BBD: Benign breast disease; BMI: Body mass index; CI: Confidence interval; 2013;13(5):342–55.
CRP: C-reactive protein; DL: DerSimonian and Laird; ER: Estrogen receptor; 5. Beguin Y, Aapro M, Ludwig H, Mizzen L, Osterborg A. Epidemiological and
FFQ: Food frequency questionnaire; HRT: Hormone replacement therapy; nonclinical studies investigating effects of iron in carcinogenesis – a critical
MOOSE: Meta-analysis of observational studies in epidemiology; review. Crit Rev Oncol Hematol. 2014;89(1):1–15.
NOS: Newcastle-Ottawa Scale; OC: Oral contraceptive; PR: Progesterone 6. Davoodi SH, Jamshidi-Naeini Y, Esmaeili S, Sohrabvandi S, Mortazavian AM.
receptor; PRISMA: Preferred reporting items for systematic reviews and meta- The dual nature of iron in relation to cancer: a review. Iran J Cancer Prev.
analyses; ROS: Reactive oxygen species; RR: Relative risk; SD: Standard 2016;9(6):e5494.
deviation; TfR: Transferrin receptor; TIBC: Total iron-binding capacity; 7. Kabat GC, Rohan TE. Does excess iron play a role in breast carcinogenesis?
TSAT: Transferrin saturation An unresolved hypothesis. Cancer Causes Control. 2007;18(10):1047–53.
8. Huang X. Does iron have a role in breast cancer? Lancet Oncol. 2008;9(8):
803–7.
Acknowledgements
9. Marques O, da Silva BM, Porto G, Lopes C. Iron homeostasis in breast
The authors would like to thank Dr. Susan Bondy and Dr. Joanne
cancer. Cancer Lett. 2014;347(1):1–14.
Kotsopoulos for their insightful comments on this work.
10. Straif K, Loomis D, Guyton K, Grosse Y, Lauby-Secretan B, El Ghissassi F, et al.
Future priorities for the IARC Monographs. Lancet Oncol. 2014;15(7):683–4.
Funding 11. Andrews NC. Forging a field: the golden age of iron biology. Blood. 2008;
This work was supported by the Canadian Institutes of Health Research 112(2):219–30.
(CIHR) Doctoral Research Award (#146236) to VCC. The funding body had no 12. Carpenter CE, Mahoney AW. Contributions of heme and nonheme iron to
role in the design of the study, collection, analysis, or interpretation of the human nutrition. Crit Rev Food Sci Nutr. 1992;31(4):333–67.
data, or in the writing of the manuscript. 13. Balder HF, Vogel J, Jansen MC, Weijenberg MP, van den Brandt PA,
Westenbrink S, et al. Heme and chlorophyll intake and risk of colorectal
cancer in the Netherlands cohort study. Cancer Epidemiol Biomarkers Prev.
Availability of data and materials 2006;15(4):717–25.
All data generated or analyzed during this study are derived from previously 14. Cross AJ, Harnly JM, Ferrucci LM, Risch A, Mayne ST, Sinha R. Developing a
published original research articles. Details are available from the heme iron database for meats according to meat type, cooking method
corresponding author on reasonable request. and doneness level. Food Nutr Sci. 2012;3(7):905–13.
Chang et al. BMC Cancer (2019) 19:543 Page 27 of 28

15. Hooda J, Shah A, Zhang L. Heme, an essential nutrient from dietary 38. Cornell JE, Mulrow CD, Localio R, Stack CB, Meibohm AR, Guallar E, et al.
proteins, critically impacts diverse physiological and pathological processes. Random-effects meta-analysis of inconsistent effects: a time for change.
Nutrients. 2014;6(3):1080–102. Ann Intern Med. 2014;160(4):267–70.
16. Cross AJ, Pollock JR, Bingham SA. Haem, not protein or inorganic iron, is 39. Ferrucci LM, Cross AJ, Graubard BI, Brinton LA, McCarty CA, Ziegler RG, et al.
responsible for endogenous intestinal N-nitrosation arising from red meat. Intake of meat, meat mutagens, and iron and the risk of breast cancer in
Cancer Res. 2003;63(10):2358–60. the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. Br J
17. Joint World Health Organization/Centers for Disease Control and Prevention Cancer. 2009;101(1):178–84.
Technical Consultation on the Assessment of Iron Status at the Population 40. Hamling J, Lee P, Weitkunat R, Ambühl M. Facilitating meta-analyses by
Level (2004: Geneva, Switzerland). Assessing the iron status of populations: deriving relative effect and precision estimates for alternative comparisons
including literature reviews: report of a Joint World Health Organization/ from a set of estimates presented by exposure level or disease category.
Centers for Disease Control and Prevention Technical Consultation on the Stat Med. 2008;27(7):954–70.
Assessment of Iron Status at the Population Level, Geneva, Switzerland, 6–8 41. Wen CP, Lee LH, Tai YP, Wen C, Wu SB, Tsai MK, et al. High serum iron is
April 2004, 2nd ed. Geneva: World Health Organization; 2007. http://www. associated with increased cancer risk. Cancer Res. 2014;74(22):6589–97.
who.int/iris/handle/10665/75368. Accessed 27 Feb 2019. 42. Bradshaw PT, Khankari NK, Teitelbaum SL, Xu X, Fink BN, Steck SE, et al.
18. Pfeiffer CM, Looker AC. Laboratory methodologies for indicators of iron Nutrient pathways and breast cancer risk: the Long Island Breast Cancer
status: strengths, limitations, and analytical challenges. Am J Clin Nutr. 2017; Study Project. Nutr Cancer. 2013;65(3):345–54.
106(Suppl 6):1606S–14S. 43. Danesh J, Collins R, Appleby P, Peto R. Association of fibrinogen, C-reactive
19. He K. Trace elements in nails as biomarkers in clinical research. Eur J Clin protein, albumin, or leukocyte count with coronary heart disease: meta-
Invest. 2011;41(1):98–102. analyses of prospective studies. JAMA. 1998;279(18):1477–82.
20. Sahin C, Pala C, Kaynar L, Torun YA, Cetin A, Kurnaz F, et al. Measurement of 44. Greenland S, Longnecker MP. Methods for trend estimation from
hair iron concentration as a marker of body iron content. Biomed Rep. summarized dose-response data, with applications to meta-analysis. Am J
2015;3(3):383–7. Epidemiol. 1992;135(11):1301–9.
21. Wood JC. Estimating tissue iron burden: current status and future prospects. 45. Orsini N, Bellocco R, Greenland S. Generalized least squares for trend
Br J Haematol. 2015;170(1):15–28. estimation of summarized dose-response data. Stata J. 2006;6(1):40–57.
22. Torti SV, Torti FM. Cellular iron metabolism in prognosis and therapy of 46. Orsini N, Li R, Wolk A, Khudyakov P, Spiegelman D. Meta-analysis for linear
breast cancer. Crit Rev Oncog. 2013;18(5):435–48. and nonlinear dose-response relations: examples, an evaluation of
approximations, and software. Am J Epidemiol. 2012;175(1):66–73.
23. Liehr JG, Jones JS. Role of iron in estrogen-induced cancer. Curr Med Chem.
47. Centers for Disease Control and Prevention. National report on biochemical
2001;8(7):839–49.
indicators of diet and nutrition in the U.S. population 1999–2002. Atlanta:
24. Wyllie S, Liehr JG. Release of iron from ferritin storage by redox cycling of
National Center for Environmental Health, Centers for Disease Control and
stilbene and steroid estrogen metabolites: a mechanism of induction of free
Prevention; 2008. https://www.cdc.gov/nutritionreport/99-02/pdf/nutrition_
radical damage by estrogen. Arch Biochem Biophys. 1997;346(2):180–6.
report.pdf. Accessed 28 Mar 2018.
25. Hann HW, Stahlhut MW, Menduke H. Iron enhances tumor growth. 48. Higgins JPT, Green S. Cochrane Handbook for Systematic Reviews of
Observations on spontaneous mammary tumors in mice. Cancer. 1991; Interventions, version 5.1.0. The Cochrane Collaboration; 2011. www.
68(11):2407–10. handbook.cochrane.org. Accessed 5 Mar 2018.
26. Thompson HJ, Kennedy K, Witt M, Juzefyk J. Effect of dietary iron deficiency 49. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in
or excess on the induction of mammary carcinogenesis by 1-methyl-1- meta-analyses. BMJ. 2003;327(7414):557–60.
nitrosourea. Carcinogenesis. 1991;12(1):111–4. 50. Begg CB, Mazumdar M. Operating characteristics of a rank correlation test
27. Singh M, Lu J, Briggs SP, McGinley JN, Haegele AD, Thompson HJ. Effect of for publication bias. Biometrics. 1994;50(4):1088–101.
excess dietary iron on the promotion stage of 1-methyl-1-nitrosourea- 51. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis
induced mammary carcinogenesis: pathogenetic characteristics and detected by a simple, graphical test. BMJ. 1997;315(7109):629–34.
distribution of iron. Carcinogenesis. 1994;15(8):1567–70. 52. Ewertz M, Gill C. Dietary factors and breast-cancer risk in Denmark. Int J
28. Diwan BA, Kasprzak KS, Anderson LM. Promotion of Cancer. 1990;46(5):779–84.
dimethylbenz[a]anthracene-initiated mammary carcinogenesis by iron in 53. Negri E, La Vecchia C, Franceschi S, D’Avanzo B, Talamini R, Parpinel M, et al.
female Sprague-Dawley rats. Carcinogenesis. 1997;18(9):1757–62. Intake of selected micronutrients and the risk of breast cancer. Int J Cancer.
29. Fonseca-Nunes A, Jakszyn P, Agudo A. Iron and cancer risk – a systematic 1996;65(2):140–4.
review and meta-analysis of the epidemiological evidence. Cancer 54. Cade J, Thomas E, Vail A. Case-control study of breast cancer in south
Epidemiol Biomarkers Prev. 2014;23(1):12–31. east England: nutritional factors. J Epidemiol Community Health. 1998;
30. Moher D, Liberati A, Tetzlaff J, Altman DG, The PRISMA Group. Preferred 52(2):105–10.
reporting items for systematic reviews and meta-analyses: the PRISMA 55. Levi F, Pasche C, Lucchini F, La Vecchia C. Dietary intake of selected
statement. PLoS Med. 2009;6(7):e1000097. micronutrients and breast-cancer risk. Int J Cancer. 2001;91(2):260–3.
31. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al. 56. Adzersen KH, Jess P, Freivogel KW, Gerhard I, Bastert G. Raw and cooked
Meta-analysis of observational studies in epidemiology: a proposal for vegetables, fruits, selected micronutrients, and breast cancer risk: a case-
reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) control study in Germany. Nutr Cancer. 2003;46(2):131–7.
group. JAMA. 2000;283(15):2008–12. 57. Michels KB, Rosner BA, Chumlea WC, Colditz GA, Willett WC. Preschool diet
32. Ludwig H, Evstatiev R, Kornek G, Aapro M, Bauernofer T, Buxhofer-Ausch V, and adult risk of breast cancer. Int J Cancer. 2006;118(3):749–54.
et al. Iron metabolism and iron supplementation in cancer patients. Wien 58. Hong CC, Ambrosone CB, Ahn J, Choi JY, McCullough ML, Stevens VL, et al.
Klin Wochenschr. 2015;127(23–24):907–19. Genetic variability in iron-related oxidative stress pathways (Nrf2, NQ01,
33. Custódio ID, Marinho Eda C, Gontijo CA, Pereira TS, Paiva CE, Maia YC. NOS3, and HO-1), iron intake, and risk of postmenopausal breast cancer.
Impact of chemotherapy on diet and nutritional status of women with Cancer Epidemiol Biomarkers Prev. 2007;16(9):1784–94.
breast cancer: a prospective study. PLoS One. 2016;11(6):e0157113. 59. Kabat GC, Miller AB, Jain M, Rohan TE. Dietary iron and heme iron intake
34. Covidence Systematic Review Software. Melbourne, Australia: Veritas Health and risk of breast cancer: a prospective cohort study. Cancer Epidemiol
Innovation. www.covidence.org. Biomarkers Prev. 2007;16(6):1306–8.
35. Wells GA, Shea B, O’Connell D, Peterson J, Welch V, Losos M, et al. The 60. Kallianpur AR, Lee SA, Gao YT, Lu W, Zheng Y, Ruan ZX, et al. Dietary
Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomised animal-derived iron and fat intake and breast cancer risk in the Shanghai
studies in meta-analyses. http://www.ohri.ca/programs/clinical_epidemiology/ Breast Cancer Study. Breast Cancer Res Treat. 2008;107(1):123–32.
oxford.asp. Accessed 15 Feb 2018. 61. Moore AB, Shannon J, Chen C, Lampe JW, Ray RM, Lewis SK, et al. Dietary
36. Greenland S. Quantitative methods in the review of epidemiologic and stored iron as predictors of breast cancer risk: a nested case-control
literature. Epidemiol Rev. 1987;9(1):1–30. study in Shanghai. Int J Cancer. 2009;125(5):1110–7.
37. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 62. Kabat GC, Cross AJ, Park Y, Schatzkin A, Hollenbeck AR, Rohan TE, et al.
1986;7(3):177–88. Intakes of dietary iron and heme-iron and risk of postmenopausal breast
Chang et al. BMC Cancer (2019) 19:543 Page 28 of 28

cancer in the National Institutes of Health-AARP Diet and Health Study. Am systematic review and meta-analysis of prospective studies. Int J Cancer.
J Clin Nutr. 2010;92(6):1478–83. 2018;143(11):2787–99.
63. Farvid MS, Cho E, Chen WY, Eliassen AH, Willett WC. Dietary protein sources 88. National Institutes of Health. Iron: Fact Sheet for Health Professionals. Bethesda:
in early adulthood and breast cancer incidence: prospective cohort study. Office of Dietary Supplements, National Institutes of Health; 2018. https://ods.
BMJ. 2014;348:g3437. od.nih.gov/factsheets/Iron-HealthProfessional/. Accessed 23 Jul 2018.
64. Farvid MS, Cho E, Chen WY, Eliassen AH, Willett WC. Adolescent meat intake 89. Cogswell ME, Kettel-Khan L, Ramakrishnan U. Iron supplement use among
and breast cancer risk. Int J Cancer. 2015;136(8):1909–20. women in the United States: science, policy and practice. J Nutr. 2003;
65. Diallo A, Deschasaux M, Partula V, Latino-Martel P, Srour B, Hercberg S, et al. 133(6):1974S–7S.
Dietary iron intake and breast cancer risk: modulation by an antioxidant 90. Shakur YA, Tarasuk V, Corey P, O’Connor DL. A comparison of micronutrient
supplementation. Oncotarget. 2016;7(48):79008–16. inadequacy and risk of high micronutrient intakes among vitamin and mineral
66. Inoue-Choi M, Sinha R, Gierach GL, Ward MH. Red and processed meat, supplement users and nonusers in Canada. J Nutr. 2012;142(3):534–40.
nitrite, and heme iron intakes and postmenopausal breast cancer risk in the 91. Ahn J, Abnet CC, Cross AJ, Sinha R. Dietary intake and nutritional status.
NIH-AARP Diet and Health Study. Int J Cancer. 2016;138(7):1609–18. IARC Sci Publ. 2011;163:189–98.
67. Knekt P, Reunanen A, Takkunen H, Aromaa A, Heliövaara M, Hakulinen T. 92. Hercberg S, Estaquio C, Czernichow S, Mennen L, Noisette N, Bertrais S, et
Body iron stores and risk of cancer. Int J Cancer. 1994;56(3):379–82. al. Iron status and risk of cancers in the SU.VI.MAX cohort. J Nutr. 2005;
68. Herrinton LJ, Friedman GD, Baer D, Selby JV. Transferrin saturation and risk 135(11):2664–8.
of cancer. Am J Epidemiol. 1995;142(7):692–8. 93. Miller EM. Iron status and reproduction in US women: National Health and
69. Garland M, Morris JS, Colditz GA, Stampfer MJ, Spate VL, Baskett CK, et al. Nutrition Examination Survey, 1999–2006. PLoS One. 2014;9(11):e112216.
Toenail trace element levels and breast cancer: a prospective study. Am J 94. Hung N, Shen CC, Hu YW, Yeh CM, Teng CJ, Kuan AS, et al. Risk of cancer in
Epidemiol. 1996;144(7):653–60. patients with iron deficiency anemia: a nationwide population-based study.
70. Cui Y, Vogt S, Olson N, Glass AG, Rohan TE. Levels of zinc, selenium, calcium, PLoS One. 2015;10(3):e0119647.
and iron in benign breast tissue and risk of subsequent breast cancer. 95. Suchdev PS, Williams AM, Mei Z, Flores-Ayala R, Pasricha SR, Rogers LM, et
Cancer Epidemiol Biomarkers Prev. 2007;16(8):1682–5. al. Assessment of iron status in settings of inflammation: challenges and
71. Stevens RG, Cologne JB, Nakachi K, Grant EJ, Neriishi K. Body iron stores and potential approaches. Am J Clin Nutr. 2017;106(Suppl 6):1626S–33S.
breast cancer risk in female atomic bomb survivors. Cancer Sci. 2011; 96. Xu J, Jia Z, Knutson MD, Leeuwenburgh C. Impaired iron status in aging
102(12):2236–40. research. Int J Mol Sci. 2012;13(2):2368–86.
72. Gaur A, Collins H, Wulaningsih W, Holmberg L, Garmo H, Hammar N, et al. 97. Jian J, Pelle E, Huang X. Iron and menopause: does increased iron affect the
Iron metabolism and risk of cancer in the Swedish AMORIS study. Cancer health of postmenopausal women? Antioxid Redox Signal. 2009;11(12):
Causes Control. 2013;24(7):1393–402. 2939–43.
73. Graff RE, Cho E, Lindström S, Kraft P, Willett WC, Eliassen AH. Premenopausal 98. Dai J, Jian J, Bosland M, Frenkel K, Bernhardt G, Huang X. Roles of hormone
plasma ferritin levels, HFE polymorphisms, and risk of breast cancer in the replacement therapy and iron in proliferation of breast epithelial cells with
Nurses’ Health Study II. Cancer Epidemiol Biomarkers Prev. 2014;23(3):516–24. different estrogen and progesterone receptor status. Breast. 2008;17(2):172–9.
74. Chua AC, Knuiman MW, Trinder D, Divitini ML, Olynyk JK. Higher 99. Osborne NJ, Gurrin LC, Allen KJ, Constantine CC, Delatycki MB, McLaren CE,
concentrations of serum iron and transferrin saturation but not serum et al. HFE C282Y homozygotes are at increased risk of breast and colorectal
ferritin are associated with cancer outcomes. Am J Clin Nutr. 2016;104(3): cancer. Hepatology. 2010;51(4):1311–8.
736–42.
75. Quintana Pacheco DA, Sookthai D, Graf ME, Schübel R, Johnson T, Katzke
VA, et al. Iron status in relation to cancer risk and mortality: findings from a
population-based prospective study. Int J Cancer. 2018;143(3):561–9.
76. Bastide NM, Pierre FH, Corpet DE. Heme iron from meat and risk of
colorectal cancer: a meta-analysis and a review of the mechanisms involved.
Cancer Prev Res (Phila). 2011;4(2):177–84.
77. Qiao L, Feng Y. Intakes of heme iron and zinc and colorectal cancer
incidence: a meta-analysis of prospective studies. Cancer Causes Control.
2013;24(6):1175–83.
78. Tappel A. Heme of consumed red meat can act as a catalyst of oxidative
damage and could initiate colon, breast and prostate cancers, heart disease
and other diseases. Med Hypotheses. 2007;68(3):562–4.
79. West AR, Oates PS. Mechanisms of heme iron absorption: current questions
and controversies. World J Gastroenterol. 2008;14(26):4101–10.
80. Young I, Parker HM, Rangan A, Prvan T, Cook RL, Donges CE, et al.
Association between haem and non-haem iron intake and serum ferritin in
healthy young women. Nutrients. 2018;10(1):81.
81. Cade JE, Moreton JA, O’Hara B, Greenwood DC, Moor J, Burley VJ, et al. Diet
and genetic factors associated with iron status in middle-aged women. Am
J Clin Nutr. 2005;82(4):813–20.
82. Liu JM, Hankinson SE, Stampfer MJ, Rifai N, Willett WC, Ma J. Body iron
stores and their determinants in healthy postmenopausal US women. Am J
Clin Nutr. 2003;78(6):1160–7.
83. Hurrell R, Egli I. Iron bioavailability and dietary reference values. Am J Clin
Nutr. 2010;91(5):1461S–7S.
84. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans.
Red meat and processed meat. IARC Monogr Eval Carcinog Risks Hum.
2018;114:1–502.
85. Lippi G, Mattiuzzi C, Cervellin G. Meat consumption and cancer risk: a critical
review of published meta-analyses. Crit Rev Oncol Hematol. 2016;97:1–4.
86. Anderson JJ, Darwis NDM, Mackay DF, Celis-Morales CA, Lyall DM, Sattar N,
et al. Red and processed meat consumption and breast cancer: UK Biobank
cohort study and meta-analysis. Eur J Cancer. 2018;90:73–82.
87. Farvid MS, Stern MC, Norat T, Sasazuki S, Vineis P, Weijenberg MP, et al.
Consumption of red and processed meat and breast cancer incidence: a