Diabetes mellitus and risk of breast cancer: A meta-analysis

Abstract

Diabetes mellitus has been associated with an increased risk of several types of cancers, but its relationship with breast cancer remains unclear. We conducted a meta-analysis of case–control and cohort studies to assess the evidence regarding the association between diabetes and risk of breast cancer. Studies were identified by searching MEDLINE (1966–February 2007) and the references of retrieved articles. We identified 20 studies (5 case–control and 15 cohort studies) that reported relative risk (RR) estimates (odds ratio, rate ratio/hazard ratio, or standardized incidence ratio) with 95% confidence intervals (CIs) for the relation between diabetes (largely Type II diabetes) and breast cancer incidence. Summary RRs were calculated using a random-effects model. Analysis of all 20 studies showed that women with (versus without) diabetes had a statistically significant 20% increased risk of breast cancer (RR, 1.20; 95% CI, 1.12–1.28). The summary estimates were similar for case–control studies (RR, 1.18; 95% CI, 1.05–1.32) and cohort studies (RR, 1.20; 95% CI, 1.11–1.30). Meta-analysis of 5 cohort studies on diabetes and mortality from breast cancer yielded a summary RR of 1.24 (95% CI, 0.95–1.62) for women with (versus without) diabetes. Findings from this meta-analysis indicate that diabetes is associated with an increased risk of breast cancer. © 2007 Wiley-Liss, Inc.

Diabetes mellitus is a serious and growing health problem worldwide.1 Type 2 diabetes accounts for ∼90–95% of all diagnosed cases of diabetes2 and is characterized by insulin resistance and hyperinsulinemia in the early phases of the disease.3 It has been hypothesized that hyperinsulinemia may increase the risk of breast cancer through direct effects on breast tissue or indirectly by increasing circulating concentrations of estrogens, testosterone and insulin-like growth factors.4, 5 Thus, Type 2 diabetes may confer an excess risk of breast cancer. Diabetes has been related to an elevated risk of several cancers. Meta-analyses have indicated that diabetes is associated with a 1.2-fold increased risk of bladder cancer,6 1.3-fold increased risk of colorectal cancer,7 1.7-fold increased risk of pancreatic cancer8 and 2.5-fold increased risk of hepatocellular carcinoma.9 Wolf et al. combined the results of 4 case–control and 6 cohort studies and found that diabetes was associated with a 13 and 25% increased risk of breast cancer in case–control and cohort studies, respectively.10

The purpose of the present study was to summarize all available evidence from case–control and cohort studies on the relationship between diabetes and breast cancer incidence and mortality following the meta-analysis of observational studies in epidemiology (MOOSE) guidelines for meta-analyses of observational studies.11 This meta-analysis includes a total of 23 studies, thus providing more precise risk estimates than the previous analysis by Wolf et al.10 that was based on only 10 studies. In this study, we also examined whether the association between diabetes and breast cancer incidence differs according to various study characteristics and menopausal status.

Material and methods

Search strategy

We searched the MEDLINE database (from 1966 to February 2007) using the search terms diabetes and breast cancer or breast neoplasm. We also reviewed the reference lists of retrieved articles to search for more studies. No language restrictions were imposed.

Inclusion and exclusion criteria

Studies were eligible for inclusion in the meta-analysis if they met the following criteria: (i) had a case–control or prospective study design; (ii) the exposure of interest was diabetes mellitus; (iii) the outcome was breast cancer incidence or mortality; and (iv) reported relative risk estimates with 95% confidence intervals (CIs) or provided sufficient information to calculate them. Studies of Type 1 diabetes, defined as diagnosis before age 30, were not included. To avoid violating independent assumptions, studies were included only once. We therefore decided, a priori, on the following hierarchy: when there were multiple publications from the same study population, the study that controlled for the most appropriate confounders were included; otherwise, the study that had the largest number of case subjects was used.

Data extraction

The following information was extracted from each study: first author's last name, the year of publication, study design, control source (in case–control studies), study location, age of subjects, sample size (cases and controls or cohort size), diabetes assessment (self-report, blood glucose, hospitalized diabetic patients), adjustment factors and relative risk estimates with 95% CIs for breast cancer associated with diabetes. From each study, we extracted the relative risk estimate that was adjusted for the greatest number of potential confounders.

Statistical analysis

We included in this meta-analysis studies reporting different measures of relative risk (RR): case–control studies (odds ratio), prospective cohort studies (rate ratio/hazard ratio) and cohort studies of hospitalized diabetic patients with an external population comparison group (standardized incidence/mortality ratio). Because the absolute risk of breast cancer is low, the 3 measures of association yield similar estimates of relative risk; we present all results as relative risk for simplicity. Studies of breast cancer incidence and breast cancer mortality were analyzed separately.

Summary relative risks with 95% CIs were calculated with the method of DerSimonian and Laird by use of the assumptions of a random effects model, which considers both within-study and between-study variation.12 We examined statistical heterogeneity in results across studies using the Q test (p < 0.1 was considered representative of statistically significant heterogeneity) and the I² statistic.13I² is the proportion of total variation contributed by between-study variation.13 To assess for publication bias, we constructed a funnel plot, and applied regression methods to determine funnel plot asymmetry as suggested by Egger et al.14 We performed subgroup analyses and used meta-regression models to evaluate potential sources of heterogeneity. Study characteristics examined included study design, diabetes assessment, geographic region, publication year, and adjustment for body mass index, physical activity and alcohol consumption. We also conducted analyses stratified by menopausal status. All statistical analyses were carried out with Stata, version 9.0 (StataCorp, College Station, TX).

Results

Breast cancer incidence

We identified 6 case–control studies15, 16, 17, 18, 19, 20 and 16 cohort studies21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36 that presented results on diabetes and breast cancer incidence. Two studies (1 case–control17 and 1 cohort23) were excluded because of duplicate publications from the same study population. Of the 20 eligible studies, 9 were from North America, 7 from Europe and 4 from Asia (Table I).

Table I. Characteristics of Case–Control and Cohort Studies of Diabetes and Incidence of Breast Cancer1

Study and country	Age, year	Years of follow-up	No. of cases	Controls or cohort size	Diabetes assessment (age at diagnosis)	Breast cancer ascertainment	Adjusted RR2 (95% CI)	Adjustments
1 BMI, body mass index; CI, confidence interval; DM, diabetes mellitus; NA, information not available; OC, oral contraceptive; PMH, postmenopausal hormone; RR, relative risk. 2 The measure of relative risk is an odds ratio (case–control studies),15, 16, 18, 19, 20 rate ratio/hazard ratio,24, 25, 26, 30, 31, 33, 34, 35, 36 or a standardized incidence ratio.21, 22, 27, 28, 29, 32 3 Type of controls in parentheses (H = hospital-based; P = population-based). 4 The confidence interval was estimated based on data provided in the article. 5 The relative risk (and its 95% confidence interval) was calculated from observed and expected number of cases reported in the article.
Case–control studies
O'Mara et al., 198515; United States	30–89	–	1883	2420 (H)3	Self-report (≥29 years)	Medical records	1.2 (0.9–1.6)4	Age
Franceschi et al., 199016; Italy	25–74	–	2663	2344 (H)	Self-report (all ages)	Medical records	1.0 (0.8–1.3)	Age, study area, education, age at first birth, menopausal status, BMI
Talamini et al., 199718; Italy	23–74	–	2569	2588 (H)	Self-report (all ages)	Medical records	1.4 (1.0–1.8)	Age, study area, education, parity, menopausal status, BMI
Weiss et al., 199919; United States	20–54	–	2158	1980 (P)	Self-report (≥30 years)	Cancer registries	1.13 (0.70–1.90)	Age, race, previous breast biopsy, family history, parity, age at first birth, menopausal status, BMI, alcohol
Baron et al., 200120; United States	50–75	–	5101	5430 (P)	Self-report (≥35 years)	State-wide tumor registries	1.2 (1.0–1.4)	Age, state, family history, parity, age at first birth, menopausal status, age at menopause, OC use, PMH use, BMI, alcohol
Cohort studies
de Waard and Baanders-van Halewijn, 197421; Netherlands	55–75	8.1	70	7259	Self-report (all ages)	Medical records	1.5 (0.6–3.3)5	Age
Raggozzino et al., 198222; United States	NA	25	14	1135	Blood glucose levels	Medical records	1.3 (0.7–2.2)5	Age
Sellers et al., 199424; United States	55–69	5	611	41 837	Self-report (≥30 years)	State Health Registry of Iowa	0.96 (0.68–1.36)	Age
Steenland et al., 199525; United States	25–74	7.7	163	NA	Self-report (all ages)	Medical records	1.40 (0.70–2.78)	Age, income, menopausal status, smoking, BMI, physical activity, alcohol
Goodman et al., 199726; Japan	NA	8.3	161	22 200	Self-report (all ages)	Cancer registry	2.06 (1.27–3.34)	Age, city, age at the time of the bombings, radiation dose
Weiderpass et al., 199727; Sweden	≥40	24	1145	70 110	Discharge diagnosis (hospitalized DM patients; ≥40 years)	Cancer registry	1.3 (1.2–1.4)	Age, calendar year
Hjalgrim et al., 199728; Denmark	≥30	20	7	402	Hospitalized DM patients (≥30 years)	Cancer registry	0.72 (0.29–1.48)	Age, calendar year
Wideroff et al., 199729; Denmark	≥50	17	493	109 581	Discharge diagnosis (hospitalized DM patients; ≥50 years)	Cancer registry	1.2 (1.1–1.2)	Age, calendar year
Mink et al., 200230; United States	45–64	7.1	187	7894	Self-report (all ages)	Cancer registry and medical records	1.39 (0.86–2.23)	Age, race, study center, family history, age at menarche, age at first birth, age at menopause, smoking, BMI, alcohol
Michels et al., 200331; United States	30–55	22	5605	116 488	Self-report (≥30 years)	Self-report, verified by medical records and pathology reports	1.17 (1.01–1.35)	Age, family history, history of benign breast disease, age at menarche, parity, age at first birth, menopausal status, age at menopause, PMH use, height, BMI, physical activity, alcohol
Swerdlow et al., 200532; United Kingdom	30–49	30	41	5066	Hospitalized insulin-treated DM patients (30–49 years)	Cancer registry	0.87 (0.62–1.17)	Age, calendar year, country of residence
Jee et al., 200533; Korea	30–95	10	NA	468 615	Blood glucose levels and medication use	Cancer registry and medial records	1.51 (1.26–1.80)	Age, smoking, alcohol
Khan et al., 200634; Japan	40–79	9	1139	33 503	Self-report (all ages)	Cancer registry	1.27 (0.51–3.14)	Age, smoking, BMI, drinking
Lipscombe et al., 200635; Canada	55–79	8	6107	465 510	Discharge diagnosis (hospitalized DM patients)	Cancer registry	1.08 (1.01–1.16)	Age, income
Inoue et al., 200636; Japan	40–69	14	451	51 223	Self-report (all ages)	Cancer registry	0.83 (0.44–1.57)	Age, study area, history of cardiovascular disease, smoking, BMI, physical activity, alcohol, dietary factors

Of the 20 studies, 15 found an increased risk of breast cancer in women with diabetes, and in 8 studies the relationship was statistically significant (Fig. 1). In analysis of all studies, the summary relative risk of breast cancer was 1.20 (95% CI, 1.12–1.28) for women with diabetes compared with women with no diabetes. There was statistically significant heterogeneity among studies (Q = 36.53, p = 0.01, I² = 48.0%). Three studies, all based on a discharge diagnosis of diabetes,27, 29, 35 contributed significantly to the summary estimate. In a sensitivity analysis excluding these 3 studies, the summary relative risk remained unchanged but the CI slightly widened (RR, 1.20; 95% CI, 1.09–1.33) and heterogeneity among studies was reduced (Q = 23.60, p = 0.10, I² = 32.2%). Stratification by study design showed that diabetes was associated with a statistically significant 18 and 20% increased risk of breast cancer in case–control and cohort studies, respectively (Fig. 1 and Table II). The relation between diabetes and breast cancer was similar in population-based case–control studies and hospital-based case–control studies (Table II).

Figure 1. Relative risks for the association between diabetes and incidence of breast cancer in case–control and cohort studies. Studies are ordered by study design and year of publication. Squares represent study-specific relative risks (the square sizes are proportional to the weight of each study in the summary estimate); horizontal lines represent 95% confidence intervals (CIs); diamonds represent summary relative risks with 95% CIs. Test for heterogeneity among studies: Q = 36.53; p = 0.01; I² = 48.0%.

Table II. Summary Relative Risks and 95% Confidence Intervals for the Association between Diabetes and Breast Cancer Incidenceby Study Characteristics

	No. of studies	RR (95% CI)	Heterogeneity tests
			Q	p	I^2 (%)
Study design
Case–control studies	5	1.18 (1.05–1.32)	3.16	0.53	0
Population-based	2	1.19 (1.02–1.40)	0.05	0.82	0
Hospital-based	3	1.17 (0.96–1.43)	3.06	0.22	34.7
Cohort studies	15	1.20 (1.11–1.30)	33.31	0.003	58.0
Diabetes assessment
Self-report	13	1.19 (1.09–1.29)	11.66	0.47	0
Blood glucose levels	2	1.49 (1.26–1.77)	0.25	0.62	0
Patients with diabetes	5	1.16 (1.05–1.27)	17.84	0.001	77.6
Geographic region
North America	9	1.12 (1.06–1.18)	4.37	0.82	0
Europe	7	1.19 (1.08–1.31)	12.68	0.05	52.7
Asia	4	1.45 (1.07–1.97)	5.11	0.16	41.3
Publication year
1970–1994	5	1.08 (0.93–1.26)	2.21	0.70	0
1995–1999	7	1.27 (1.16–1.38)	10.24	0.12	41.4
2000–2006	8	1.17 (1.04–1.32)	16.79	0.02	58.3
Adjustment for body mass index
No	10	1.19 (1.09–1.30)	26.47	0.002	66.0
Yes	10	1.20 (1.09–1.33)	10.06	0.35	10.5
Adjustment for physical activity
No	17	1.20 (1.12–1.29)	35.00	0.004	54.3
Yes	3	1.16 (1.01–1.33)	1.36	0.51	0
Adjustment for alcohol intake
No	13	1.17 (1.08–1.27)	27.67	0.006	56.5
Yes	7	1.26 (1.13–1.40)	7.37	0.29	18.6
Adjustment for postmenopausal hormone use
No	18	1.20 (1.11–1.30)	36.46	0.004	53.4
Yes	2	1.18 (1.06–1.32)	0.05	0.82	0
Menopausal status
Premenopausal	3	0.91 (0.62–1.34)	0.35	0.84	0
Postmenopausal	8	1.16 (1.08–1.23)	11.56	0.12	39.4

To examine whether differences in diabetes assessment accounted for the heterogeneity in results, we stratified studies by diabetes assessment. Summary estimates were similar for studies where diabetes was defined based on self-report, blood glucose levels or a discharge diagnosis (patients with diabetes); there was no statistically significant heterogeneity only among studies based on self-reported diabetes (Table II). Stratifying study results by control for potential confounders, we found that there was no statistical significant heterogeneity within strata of studies that adjusted for body mass index, physical activity, alcohol intake or use of postmenopausal hormones (Table II). In meta-regression analysis, the association between diabetes and breast cancer varied statistically significantly by geographic region (stronger association in Asia than in North America; p = 0.02), but not by study design, diabetes assessment, publication year or control for potential confounders.

Three studies provided results stratified by menopausal status,16, 18, 31 and 5 studies included only or predominantly postmenopausal women.20, 21, 24, 29, 35 In stratified analysis by menopausal status, diabetes was associated with an increased risk of breast cancer in postmenopausal women but not in premenopausal women (Table II). The difference in summary relative risks across strata of menopausal status was not statistically significant (p = 0.27).

There was no indication of publication bias on the funnel plot (data not shown) or by Egger's test (p = 0.94).

Breast cancer mortality

We identified 5 cohort studies that reported results on diabetes and mortality from breast cancer (Table III).32, 33, 37, 38, 39 Combining the results from these studies yielded a summary relative risk of 1.24 (95% CI, 0.95–1.62) for women with (versus without) diabetes. There was statistically significant heterogeneity among studies (Q = 20.54, p < 0.001, I² = 80.5%). There was no evidence of publication bias (Egger's test: p = 0.89).

Table III. Characteristics of Cohort Studies of Diabetes and Mortality from Breast Cancer1

Study and country	Years of follow-up	Age, year	No. of deaths2	Cohort size	Diabetes assessment (age at diagnosis)	Adjusted RR3 (95% CI)	Adjustments
1 BMI, body mass index; CI, confidence interval; DM, diabetes mellitus; NA, information not available; RR, relative risk. 2 Deaths from breast cancer. 3 The measure of relative risk is a rate ratio/hazard ratio33, 39 or a standardized mortality ratio32, 37, 38 4 The confidence interval was calculated from observed and expected number of deaths reported in the article.
Kessler, 197037; United States	26	NA	21	21 447	Blood glucose level	0.88 (0.70–1.08)4	Age
Verlato et al., 200338; Italy	10	NA	28	7148	Diabetes clinics, drug prescriptions, or family physicians	1.40 (1.06–1.81)	Age
Coughlin et al., 200439; United States	16	≥30	4346	588 321	Self-report (all ages)	1.27 (1.11–1.45)	Age, education, race, smoking, physical activity, BMI, alcohol, dietary factors
Swerdlow et al., 200532; United Kingdom	30	30–49	17	5066	Hospitalized insulin-treated DM patients (30–49 y)	0.86 (0.50–1.38)	Age, calendar year, country of residence
Jee et al., 200533; Korea	10	30–95	NA	468 615	Blood glucose levels or medication use	2.23 (1.49–3.33)	Age, smoking, alcohol

Discussion

Findings from this meta-analysis support a positive association between diabetes (largely Type II diabetes) and breast cancer risk. Summary results showed that women with diabetes may have ∼20% increased risk of breast cancer. The association between diabetes and breast cancer was consistent for case–control and cohort studies and for studies carried out in North America, Europe and Asia.

There are several potential limitations that should be considered when interpreting the results of this meta-analysis. First, most of the studies did not distinguish between Type I and Type II diabetes. Because Type I diabetes (which accounts for 5–10% of all diagnosed cases of diabetes2) may not be a risk factor for breast cancer,28, 32, 40 the magnitude of the relationship between diabetes and breast cancer may have been somewhat underestimated. Furthermore, because diabetes is an underdiagnosed disease, some misclassification of exposure is likely to have occurred and this misclassification would tend to attenuate any true association between diabetes and breast cancer risk. Based on data from a population-based study in the United States, about 3–4% of adults (aged ≥40 years) have undiagnosed diabetes.41 In cohort studies of hospitalized diabetic patients, the comparison group (i.e., the general population) includes individuals with diabetes; this would also lead to attenuated relative risk estimates. A second limitation is methodologic issues related to study design. Case–control studies are susceptible to recall and selection biases. The association between diabetes and breast cancer risk was similar in case–control and cohort studies, which argues against the possibility that the observed association was the result of recall or selection bias. Third, as our analyses are based on observational studies, uncontrolled confounding cannot be entirely excluded as a potential explanation for the observed association. Nevertheless, many studies adjusted for potential confounders and the association was similar for studies that controlled for body mass index, physical activity and alcohol consumption, and for studies that did not adjust for these variables. It is remarkable that only 2 studies20, 31 adjusted for postmenopausal hormone use, a risk factor for breast cancer. In the study by Michels et al.,31 women with diabetes were less likely to use postmenopausal hormones. Hence, failure to adjust for postmenopausal hormone use may result in an underestimation of the true relationship between diabetes and risk of breast cancer. Finally, inherent in meta-analysis of published studies is the possibility of publication bias. However, we found no evidence of such bias in this meta-analysis.

None of the studies included in this meta-analysis provided results stratified by insulin-dependent versus noninsulin-dependent Type II diabetes. We therefore could not examine whether risks may differ in these 2 types of Type II diabetics. One cohort study consisted entirely of patients with insulin-treated diabetes, and in this study no association was found between insulin-dependent Type II diabetes and breast cancer incidence or mortality.32

The mechanisms underlying the relation between diabetes and breast cancer risk may be related to alterations in circulating concentrations of insulin, insulin-like growth factors (IGFs) and endogenous sex hormones. Type II diabetes is usually associated with insulin resistance and increased pancreatic insulin secretion for long periods both before and after disease onset. Insulin has been demonstrated to have mitogenic effects on breast tissue,42, 43 and insulin receptors are frequently over-expressed in breast cancer cells.44, 45 A positive association between circulating concentrations of insulin or C-peptide (a marker of insulin secretion) and breast cancer risk has been observed in several,46, 47, 48, 49, 50 but not in all epidemiologic studies.30, 51, 52 High fasting insulin concentrations53 and Type II diabetes54 have also been associated with adverse prognostic factors and poor outcomes in women with breast cancer. Elevated insulin concentrations may also stimulate tumor growth by increasing bioavailable IGF-I.5 High circulating IGF-I concentrations have been shown to predict premenopausal breast cancer risk.55 Insulin inhibits the production of sex hormone-binding globulin,56 which leads to an increase in bioavailable estradiol and testosterone.57 Compared to healthy women, diabetes patients have been found to have higher concentrations of circulating estrogens and androgens.58, 59 Epidemiologic studies have generally indicated positive relationships between estrogen and testosterone concentrations and risk of breast cancer in postmenopausal women.60, 61 Some studies have found positive relations between estrogen and testosterone concentrations and breast cancer risk in premenopausal women62, 63 but the associations seem to be weaker than in postmenopausal women. In this meta-analysis, the relation between diabetes and breast cancer appeared to be confined to postmenopausal women, but this finding was based on a limited number of studies of premenopausal breast cancer and a test for difference in association by menopausal status was not statistically significant. Thus, this may be a chance finding.

In summary, this meta-analysis supports the hypothesis that women with diabetes may have an increased risk of breast cancer. Whether this association varies by menopausal status or type of Type II diabetes (insulin-dependent versus noninsulin-dependent) warrants further investigation.