Metabolic Syndrome Is Associated with Increased Breast Cancer Risk: A Systematic Review with Meta-Analysis

Background. Although individual metabolic risk factors are reported to be associated with breast cancer risk, controversy surrounds risk of breast cancer from metabolic syndrome (MS). We report the first systematic review and meta-analysis of the association between MS and breast cancer risk in all adult females. Methods. Studies were retrieved by searching four electronic reference databases [PubMed, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science, and ProQuest through June 30, 2012] and cross-referencing retrieved articles. Eligible for inclusion were longitudinal studies reporting associations between MS and breast cancer risk among females aged 18 years and older. Relative risks and 95% confidence intervals were calculated for each study and pooled using random-effects models. Publication bias was assessed quantitatively (Trim and Fill) and qualitatively (funnel plots). Heterogeneity was examined using Q and I 2 statistics. Results. Representing nine independent cohorts and 97,277 adult females, eight studies met the inclusion criteria. A modest, positive association was observed between MS and breast cancer risk (RR: 1.47, 95% CI, 1.15–1.87; z = 3.13; p = 0.002; Q = 26.28, p = 0.001; I 2 = 69.55%). No publication bias was observed. Conclusions. MS is associated with increased breast cancer risk in adult women.


Introduction
Breast cancer, the most common cancer in women worldwide, accounted for 1.7 million new cases in 2012, comprising a quarter of all new cancer cases [1]. While traditional risk factors for breast cancer include age, family history of cancer, and reproductive and menstrual history, the National Cancer Institute also recognizes overweight, lack of physical activity, and consumption of alcohol as risk factors [2]. Several of these risk factors are associated with metabolic syndrome [3].
Previous epidemiologic studies on MS and breast cancer risk show contrary results. For example, only four [13,14,43,51] of eight studies [13,14,43,48,51,[62][63][64] reported a statistically significant association between MS and risk of breast cancer. This might invite a conclusion that the association between MS and breast cancer risk is unknown. However, such an inference would be based on the votecounting approach, an approach that ignores the magnitude of the association [65].
A recent systematic review and meta-analysis of MS and postmenopausal breast cancer found that MS was moderately associated with the risk of postmenopausal breast cancer [10]. However, to the best of our knowledge, no metaanalytic research has addressed the conflicting results from individual studies of MS and breast cancer risk in all adult women. Therefore, the purpose of this study was to use the aggregate data meta-analytic approach to examine the association between MS and breast cancer risk in women.

Study Eligibility.
The a priori inclusion criteria for this study were as follows: (1) observational studies using cohort (both prospective and retrospective), case-control, or nested case-control study designs; (2) studies examining the association between MS (presence of a cluster of three or more metabolic abnormalities) and breast cancer incidence, as defined by the authors; (3) studies with adult females ≥ 18 years of age as participants; (4) English-language studies published as journal articles, doctoral dissertations, or masters' theses; (5) published and indexed studies up to June 30, 2012; and (6) studies reporting sufficient data (e.g., rate ratios, risk ratios, odds ratios, standardized incidence ratios, hazard ratios, or frequencies) for calculating a common effect size. Neither lobular carcinoma in situ nor ductal carcinoma in situ breast cancer cases were excluded from the study.
Studies not meeting all inclusion criteria were excluded from this review. Excluded studies were those that (1) were not published as full reports, such as conference abstracts and letters to the editors; (2) only examined individual components of MS; (3) measured the MS variables at time of cancer diagnosis; (4) used cancer mortality, rather than incidence, as the outcome; and (5) were published in a language other than English.

Data Sources.
A comprehensive and systematic search was conducted using four electronic databases: PubMed, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science, and ProQuest (from their commencement to June 30, 2012). Since the term MS dates back to the late 1950s, with variations in use as early as the 1920s, the start dates of each of the databases were used as the commencement date for study search: Web of Science (1900), CINAHL (1952), PubMed (1966), and ProQuest (1861). In addition, cross-referencing from retrieved studies was also performed. Major keywords used in the search for potentially eligible studies included "metabolic syndrome" ("insulin resistance syndrome, " "syndrome x") and "breast cancer" ("neoplasm and breast"). Using the most recent publication, trials published as duplicate reports (parallel publications) were only included once. All electronic searches were conducted using the graphical user interface for each database. The last search was conducted on June 30, 2012. An initial cut-off point for the inclusion of studies was not used given the difficulty in establishing such a point, as well as our concern about the potential loss of studies that met our eligibility criteria.

Study Selection.
At the first screening, one author (RB) screened all abstracts and selected articles for full-text examination. At the second level of the study selection process, two of the authors (RB and TH) examined the full-text articles and then selected the included studies following mutual discussion and consensus.

Data Extraction.
Two of the authors (RB and TH) reviewed every study selected and independently extracted data from studies onto electronic coding forms. These forms could hold up to 52 items per study. Attempts were made to contact authors of three of the original studies for missing information [13, 62,64], but only one provided the requested information [13]. After initial coding, the two coders (RB and TH) reviewed each item for agreement. Discrepancies were resolved by consensus. Using Cohen's kappa ( ) statistic [66], the overall interrater agreement rate prior to correcting discrepant items was 0.96 for all included studies.
2.5. Risk of Bias Assessment. Risk of bias was assessed using a modified version of Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist [67]. The items assessed included (1) study design, (2) adjustments for confounders, (3) selection of participants and their eligibility criteria, (4) measurement of predictor variables, (5) breast cancer diagnosis, (6) study size, (7) handling of missing data, and (8) reasons for nonparticipation of individuals at each stage of the study. A description of the criteria for risk of bias assessment is shown in Table 1. Two of the authors (RB and TH) conducted all assessments, independently of each other. Disagreements were resolved through discussion. No 2.6. Statistical Analysis 2.6.1. Calculation of Study-Level Effect Sizes. Risk estimates were used to examine the association between MS and risk of breast cancer. These were derived from reported relative risks, odds ratios, hazard ratios, incident rate ratios, or standardized incidence ratios, together with corresponding 95% confidence intervals (CIs), from the original studies. Where necessary and possible, all metrics were converted to risk ratios (RRs). Adjusted risk estimates were pooled for analysis from multivariable models in the original studies. However, for two case-control studies that were included [14,51], adjusted odds ratios were used because of the lack of the requisite data to convert odds ratios to RRs.
2.6.2. Effect Size Pooling. All RR results were pooled using a random-effects model, an approach that incorporates between-study heterogeneity into the model [68]. A -score two-tailed alpha value ≤ 0.05 was considered to be statistically significant. In addition, 95% CIs were calculated for each result from each study as well as for pooled estimates. Heterogeneity was calculated using the [69] and 2 statistics [70]. An alpha level ≤ 0.10 for the statistic was considered to be evidence of statistically significant heterogeneity. While somewhat arbitrary, 2 values of 25%, 50%, and 75% were considered to represent low, moderate, and high amounts of heterogeneity [70]. Publication bias was assessed using the Trim and Fill approach of Duval and Tweedie [71]. In addition, Rosenthal's Fail-Safe test was used to compute the number of missing null studies that would be needed to nullify the overall pooled RR as being statistically significant [72]. Statistically significant standardized residuals ( ≤ 0.05) were considered to be outliers.

Sensitivity Analyses.
Influence analysis was conducted with each study result deleted from the model once, in order to examine the effects of each on the overall pooled results. Cumulative meta-analysis, ranked by year, was also conducted in order to examine the accumulation of results over time. A separate pooled analysis, limited to postmenopausal women, was conducted because studies show that MS in postmenopausal women increases the risk of breast cancer [13,14,43,48,51,62]. In addition, pooled analyses were conducted with the following caveats post hoc: (1) deletion of results from two case-control studies because odds ratios were used instead of RR [14, 51], (2) deletion of results from studies that were not prospective cohort designs [13, 14, 51], and (3) limiting the results to studies that controlled for four or more of the important confounders (as listed in Table 1) [14, 43,48,51]. Given the potential for diabetes and diabetes medications to affect breast cancer risk, post hoc data analysis was also conducted with studies that included participants with diabetes and/or taking medications for diabetes, deleted from the model [14, 43,64]. All analyses were performed using Comprehensive Meta-Analysis, Version 2.2 [73].  A general description of the included studies is shown in Table 2. Studies were published between 2008 and 2012 and from five different countries. The study designs included four prospective cohorts [48,[62][63][64], one retrospective cohort [13], one prospective nested case-control study [43], and two case-control studies [14,51]. The baseline year for cohort inception ranged from 1983 to 2004, with average followup ranging between 2.7 and 13.5 years. Sample sizes ranged from 792 to 49,172 (total 97,277) adult females, excluding one study that did not report these data [64]. The ages of the participants ranged from 21 to 86 years. Six studies conducted analyses on postmenopausal women [13,14,43,48,51,62]. The results of each cohort or case-control study were initially reported as a hazard ratio [13, 48,63], incidence rate ratio [43,62], standardized incidence ratio [64], or odds ratio [14,51]. Methods for exposure assessment, cancer identification, and control of confounders varied across the eight included studies (Table 3). Seven studies identified the outcome (breast cancer) through histological reports or medical reports or from a cancer registry [13,14,43,48,51,62,64], while one used self-report [63]. Only three studies examined invasive breast cancer cases [43,48,64]. One study also reported on the in situ breast cancer cases but there were only seven such cases in that study [43]. Another study analyzed all breast cancer cases (in situ and invasive) as well as invasive cancers separately, and results remained unchanged [48]. Table 4. All of the studies were considered to be at low risk for selection of participants and meeting eligibility criteria in addition to providing adequately powered sample sizes. Out of eight studies, a majority were also considered low risk with respect to study design (six studies) and measurement of the outcome variable (seven studies). In terms of handling potential confounders, half the studies were low risk, three were high risk, and one was unclear risk. Missing confounding variables included education, smoking status, alcohol use, family history of cancer, contraceptive use, or hormonal history. Similarly, half the studies had objective measurements of predictor variables, while the remainder relied on self-report, and were consequently considered high risk. Four studies deleted the participants with missing variables in their analyses (high risk), while two did not report how they handled missing data. Lastly, six studies were considered high risk because they did not report the reasons for nonparticipation of subjects at each stage of follow-up.  (Figure 2). With the exception of one study [63], all other studies had RR in the direction of increased risk [13,14,43,48,51,62,64]. Funnel plot results for potential publication bias are shown in Figure 3. Using the Trim and Fill approach that resulted in two imputations, the risk decreased by 16% but remained significant (RR: 1.31, 95% CI, 1.01-1.70). The Fail-Safe N was 69, implying that 69 "null" studies would be needed to nullify the statistically International Journal of Breast Cancer 5

Sensitivity Analyses.
With each study deleted from the model once, results remained positive and statistically significant (Figure 4). The pooled RR fell within a range of 20% (RR = 1.36-1.56) and none of the CIs for the point estimates was less than 1.0. Cumulative meta-analysis, ranked by year, revealed that results have been statistically significant since 2011 ( Figure 5)

Discussion
The purpose of this aggregate data meta-analysis was to examine the association between MS and the risk for breast cancer in adult females. Overall, the results suggest that there was a modest positive association between MS and risk of breast cancer. This finding is strengthened by the robustness of results from other analyses. These include (1) examination for publication bias, (2) influence analysis with each study being deleted from the model once, (3) deletion of the two case-control studies with odds ratios from the overall model, (4) limiting the analysis to prospective designs, (5) including only postmenopausal women in the analysis, and (6) limiting the results to studies that controlled for four or more of the important confounders. In addition, the results from cumulative meta-analysis, ranked by year, indicate an increasingly statistically significant association since 2011. In contrast, despite a slightly increased mean RR, overlapping CIs were observed when studies that included participants with diabetes or taking medications for diabetes were deleted from the model [14, 43,64]. However, whether this reduced precision is the result of these specific characteristics or some other factors, for example, loss of power with a reduced number of studies, is not known. Assessment for risk of bias indicated that a majority of studies were at low risk regarding study design, cancer assessment, and sample size. However, a majority were at high risk or unclear risk in terms of handling of missing data and nonparticipation of subjects at each stage of follow-up. It is suggested that future studies provide complete information on the handling of missing data and on the nonparticipation of subjects at each stage of follow-up.
When limited to postmenopausal women, a stronger association between MS and breast cancer was observed. This association was stronger in case-control and retrospective cohort study designs compared to prospective cohort study    designs. These findings concur with those from a recent meta-analysis on MS and breast cancer risk in postmenopausal women [10]. Several studies have shown that MS in this group increases the risk of breast cancer [43,46,102], suggesting that the etiology of breast cancer may differ among pre-and postmenopausal women. There are several potential mechanisms linking MS with an increased risk of breast cancer. First, obese postmenopausal women produce higher levels of estrogens, which in turn increase the biologically available fraction of circulating estradiol by reducing plasma concentration of sex hormone binding globulin (SHBG) [103]. Low plasma SHBG levels are associated with insulin resistance [104,105] and other components of MS [106,107]. Second, adipose tissue produces two adipokines (cytokine-like factors), leptin and adiponectin, that affect breast cancer biology [108]. Higher plasma leptin levels are associated with obesity [54,57,109], insulin resistance [110,111], and MS [112,113]. Leptin stimulates human breast cancer cell lines, whereas adiponectin acts protectively, inhibiting the growth of these cell lines [57,108,114]. Obesity is associated with reduced adiponectin levels [115]. Third, insulin has been shown to have a mitogenic effect upon breast cancer cells in vitro through several mechanisms [57]. It can act synergistically with estradiol and stimulate proliferation of the cell line [116]. Insulin can also lower SHBG production [117], thereby increasing biologically available estradiol. Moreover, low serum HDL-C concentrations indicate higher circulating bioactive estrogen levels, which in turn may stimulate target breast tissue [77].
The increasing prevalence of MS and its association with breast cancer, among other comorbidities, point toward the critical need to develop public health strategies to manage MS. Given the increasingly large global burden of metabolic risk factors, even a small association with breast cancer can have a substantial public health impact. Risk assessment tools can be developed which incorporate MS as a risk factor for breast cancer. Healthcare providers will then be better equipped to identify high-risk women for primary and secondary prevention.
This study has several strengths. First, to the best of our knowledge, this is the first systematic review and metaanalysis examining the association between MS and risk of breast cancer in all adult women. The analysis incorporates all women and a subanalysis of postmenopausal women. The overlapping meta-analysis on metabolic syndrome and breast cancer was confined to postmenopausal women only [10]. Second, a number of other analyses were performed which strengthened the robustness of findings. Third, the results of this study provide direction for future research on this topic.
This study also has several potential limitations. These include (1) the different methods used to assess exposure, identify cancer, control for confounders, and define MS, (2) limiting studies to those published in English, which may have inflated the results [118], (3) the relatively small number of studies that met the inclusion criteria, (4) the inability of some studies to provide raw data for calculating the RR, (5) the different study designs employed, and (6) the varied populations studied, including those with diabetes and/or taking medications for diabetes. Most notably and with respect to controlling for adiposity, a potential confounder, two of the included studies controlled for BMI [48,62] but no information was available from the other studies with respect to controlling for BMI or any other obesity-related measures, including such measures of central obesity as waist circumference and waist-to-hip ratio [13,14,43,51,63,64]. Given the potential association between breast cancer and adiposity, it may be prudent for future studies to control for this potential confounder. This may be especially true for measures of central adiposity. To this point, Kabat et al. suggested that some, but not all, studies have reported an association between increased central adiposity and an increased risk for postmenopausal breast cancer [48]. Another limitation was a lack of information on tumor subtypes. The inclusion of such information in future studies may be important, given the potential differences in risk according to exposure and disease subtype.
In order to inform and undergird a biological rationale for the observed positive association between MS and breast cancer risk in adult females, future research should comprise analyses based on a standard definition of MS and employ objective and standard biomarkers for assessing each MS component. In addition, adjustments for all important potential confounders need to be made. It would be helpful if future studies examined the relationship between MS and breast cancer risk separately in perimenopausal and premenopausal women since breast cancer in women may be estrogenindependent. Along those lines, not all studies adjusted for hormone replacement therapy, a potential confounder. Future studies should report this information. Furthermore, they need to examine in situ and invasive cancers separately in relation to metabolic syndrome. Finally, a focus on obese women with respect to MS and breast cancer seems appropriate.
In conclusion, the overall results of this meta-analysis suggest that there is a modest positive association between MS and risk of breast cancer in adult females.