|Home | About | Journals | Submit | Contact Us | Français|
Emerging evidence suggests that metformin may reduce breast cancer incidence, but reports are mixed and few provide information on tumor characteristics. Therefore, we assessed associations among diabetes, metformin use, and breast cancer in postmenopausal women participating in Women's Health Initiative clinical trials.
In all, 68,019 postmenopausal women, including 3,401 with diabetes at study entry, were observed over a mean of 11.8 years with 3,273 invasive breast cancers diagnosed. Diabetes incidence status was collected throughout follow-up, with medication information collected at baseline and years 1, 3, 6, and 9. Breast cancers were confirmed by review of central medical records and pathology reports. Cox proportional hazards regression, adjusted for breast cancer risk factors, compared breast cancer incidence in women with diabetes who were metformin users or nonusers with breast cancer incidence in women without diabetes.
Compared with that in women without diabetes, breast cancer incidence in women with diabetes differed by diabetes medication type (P = .04). Women with diabetes receiving medications other than metformin had a slightly higher incidence of breast cancer (hazard ratio [HR], 1.16; 95% CI, 0.93 to 1.45), and women with diabetes who were given metformin had lower breast cancer incidence (HR, 0.75; 95% CI, 0.57 to 0.99). The association was observed for cancers positive for both estrogen receptor and progesterone receptor and those that were negative for human epidermal growth factor receptor 2.
Metformin use in postmenopausal women with diabetes was associated with lower incidence of invasive breast cancer. These results can inform future studies evaluating metformin use in breast cancer management and prevention.
Diabetes mellitus is a common condition that has been associated with increased incidence of breast cancer,1–3 a finding recently challenged by reports from large population-based studies.4,5 Nonetheless, diabetes is linked to adverse breast cancer outcome.6 A meta-analysis7 and two recent studies6,8 found that patients with breast cancer and diabetes had significantly higher all-cause mortality than those without diabetes. In addition, all patients with breast cancer with higher fasting levels of insulin,9 higher levels of C-peptide (a marker of insulin production),10 or increased insulin resistance by homeostasis model assessment (HOMA) score11 are at increased risk of death from breast cancer.
Metformin is a biguanide, commonly used for treating type 2 diabetes, that increases insulin sensitivity and improves glycemic control.12,13 On the basis of these properties and preclinical evidence of its inhibition of breast cancer growth,14,15 metformin has been studied for anticancer effects.
Following early reports,16 a meta-analysis of observational studies in individuals with type 2 diabetes associated metformin use with a 30% lower overall incidence of cancer compared with patients with diabetes receiving other therapies17 and, in a cohort study,18 with 44% lower cancer mortality compared with an age-matched general population. Although some observational studies19–21 describe lower incidence of breast cancer with metformin use, results are mixed,22,23 and information on breast cancer subtypes is sparse.24 Consequently, we examined relationships among diabetes, metformin use, and breast cancer in the Women's Health Initiative (WHI) clinical trials.
The WHI program includes four clinical trials and an observational study.25 General eligibility required age between 50 and 79 years, being accessible for follow-up, and estimated survival of ≥ 3 years. The clinical trials excluded women with prior breast cancer and had additional eligibility requirements largely related to medical history. The study population included all clinical trials participants (N = 68,132), excluding only those reporting bilateral mastectomy, diabetic coma, or diabetes diagnosed at younger than age 21 (to exclude likely type 1 diabetes), or those with missing baseline diabetes information, leaving 68,019 women for these analyses.
Study implementation details have been published.25,26 Protocols had institutional review board approval from the clinical centers, and all participants provided written informed consent. At baseline, participants completed questionnaires that collected information on demographics, medical history, and breast cancer risk. All reported medications were matched to the Master Drug Data Base (MDDB; Medi-Span, Indianapolis, IN), a procedure repeated in years 1, 3, 6, and 9. The initial study period ended on March 31, 2005, with reconsenting participants observed for subsequent clinical outcomes.
At baseline, participants were asked “Did a doctor ever say that you had sugar diabetes or high blood sugar when you were not pregnant?” Medical history was updated semiannually during the initial study period and annually thereafter when participants were asked “Since the date given on this form, has a doctor prescribed any of the following pills or treatments?” Choices included “pills for diabetes” and “insulin shots for diabetes.” Women with diabetes were defined as those reporting targeted diabetes at baseline or at subsequent medical history update or reporting use of antidiabetic medications at any time.
This approach to identifying women with diabetes has been evaluated.27 Fasting glucose levels were determined in a random sample of 5,884 baseline specimens from the entire WHI population. Glucose levels ≥ 140 mg/dL (the pre-1997 diabetes diagnostic threshold prevalent during most of the WHI recruitment) were seen in 4.7% of women without reported diabetes and in 95.3% of women with reported diabetes.27
Baseline mammogram and clinical breast examinations not suggestive of cancer were eligibility requirements. Mammograms and breast examinations were mandated annually in the hormone therapy trials and biennially in the dietary trial. Breast cancers were initially verified by pathology report review at the local clinical center by trained physician adjudicators.28 Final adjudication and coding for stage and hormone receptor and human epidermal growth factor receptor 2 (HER2) status (both by local laboratory criteria) were performed at the Clinical Coordinating Center by using Surveillance, Epidemiology, and End Results (SEER) criteria.29
Baseline characteristics of women with diabetes were compared with those of women without diabetes by χ2 test of association. In addition, baseline characteristics of women with treated diabetes using metformin either as monotherapy or with other diabetes medications were compared with women with treated diabetes not using metformin and with women without diabetes. The 3,407 women with diabetes identified at entry were included in these comparisons.
Cox regression models with a time-dependent categorical exposure variable that incorporated all collected information on diabetes diagnoses and medication use were used to compute hazard ratios (HRs) and 95% CIs for breast cancer incidence. The two exposures of interest were (1) a two-level categorical variable that classified women as either having or not having diabetes and (2) a four-level categorical variable that further classified women with diabetes as metformin users, users of other antidiabetic medication without metformin, or having unknown diabetes treatment, with women without diabetes as comparators. If treated diabetes was reported but medication data were not available at that time, the exposure was coded as “unknown treatment” and was later updated to metformin users or users of other diabetes medication when medication data became available. For example, a woman with no diabetes on entry who reported treated diabetes after 26 months and metformin use beginning at year 3 would be analyzed in the “no diabetes” group for time 0 to 26 months, the “diabetes therapy unknown” group for 26 to 36 months, and the “metformin use” group after 36 months. A 2 df test of significance was used to compare women who used metformin with women without diabetes and to compare users of other diabetes medication with women without diabetes.
The Cox proportional hazard analyses were adjusted for the baseline covariates of age (linear), first-degree relatives with breast cancer, prior breast biopsy, age at menarche, age at menopause, age at parity, age at first live birth, number of months of breastfeeding, education, smoking, alcohol use, body mass index (BMI; linear), physical activity, duration of prior use of estrogen alone and duration of use of estrogen plus progestin (considered separately), and bilateral oophorectomy. The baseline hazard functions were allowed to vary by age (10-year group), the four hormone therapy trial randomization arms, the dietary trial randomization arms, race/ethnicity, and enrollment in the WHI extension study. The strategy of using both covariate adjustment and allowing the hazard functions to vary by group provides a more refined adjustment for age and has been used by others.30 A variant of Kaplan-Meier incidence curves that allows for time-dependent exposures31 was used to display cumulative hazard functions.
Separate analyses were conducted by breast cancer subtypes. Tumor characteristics and stage and hormone receptor and HER2 status were compared in groups defined by diabetes and associated medication use.
Subgroup comparisons involved age, BMI (as a linear continuous variable), physical activity, alcohol use, use of estrogen alone, use of estrogen plus progestin, and recency of diabetes diagnosis (prevalent v incident cases). The 14 tests (both exposures times seven subgroups) were planned with statistical significance based on the nominal P value for interaction from a multivariable Cox regression model. With regard to multiplicity,32 less than one would be expected to be significant (P < .05) by chance alone.
Subgroups were time-dependent variables with BMI updated annually; physical activity updated at years 1, 3, 6, and 9; and age continuously adjusted by summing age at baseline with length of follow-up.
Post hoc analyses were conducted to assess whether metformin association was mediated through weight loss. Change in weight (year 1 minus baseline) was examined since medication data were available at both visits and was assessed among only recent users (those beginning use < 6 months before baseline) by use of linear regression adjusted for age, height, race/ethnicity, physical activity, and self-reported health. The least squares regression model incorporated change in weight as the response variable and metformin use as the exposure variable of interest.
All analyses were conducted by using SAS software, version 9.1 (SAS Institute, Cary, NC) and R software version 2.11 (R Foundation for Statistical Computing; http://www.r-project.org/). All statistical tests were two-sided.
Women with diabetes were older and more likely to be black, to engage in less recreational physical activity, and to be obese (Table 1). Use of medication for diabetes at baseline is outlined in Table 1. Reflecting general practice trends, the percentage of women with diabetes who used metformin increased year by year from 20.3% at baseline to 30.6%, 41.0%, 51.6%, and 55.0% in years 1, 3, 6, and 9, respectively. Women with diabetes who used metformin were similar to those who used other medications except they were somewhat less likely to be black, to never have smoked, and to be older; they were less likely to use sulfonylureas and were not using insulin (Table 2).
During the study, the annual frequency of mammography was similar in women with and without diabetes (annualized rates of 62.0% and 61.2%, respectively) and, in women with diabetes, the frequency was somewhat greater in metformin users (65.0%) compared with nonusers (59.0%).
During 801,066 person-years over a mean of 11.8 years (standard deviation [SD], 3.1 years), a cumulative total of 11,290 women were diagnosed with diabetes, 3,273 with invasive breast cancer, and 754 with ductal carcinoma in situ (DCIS; Table 3). There was no difference in the incidence of invasive breast cancer (HR, 0.99; 95% CI, 0.85 to 1.14) or DCIS (HR, 0.99; 95% CI, 0.73 to 1.36) between all women with diabetes and women without diabetes. However, when women with diabetes were compared with women without diabetes, incidence of invasive breast cancer was associated with diabetes medication type (P = .04). The incidence of invasive breast cancer was lower in women with diabetes who used metformin (HR, 0.75; 95% CI, 0.57 to 0.99). In contrast, the incidence was slightly higher in women with diabetes who used other medications (HR, 1.16; 95% CI, 0.93 to 1.45; Table 4 and Fig 1). Incidence of DCIS was not associated with metformin use (Table 4).
In women with diabetes, metformin use was associated with a lower incidence of breast cancers positive for both estrogen receptor (ER) and progesterone receptor (PR; HR, 0.64; 95% CI, 0.45 to 0.92) and negative for HER2 overexpression (Table 4). Although the analysis was limited by few patients, metformin use was also associated with more HER2-positive cancers.
In exploratory analyses, the other diabetes medication group was further divided into insulin users and nonusers. In comparison with women without diabetes, HRs were 1.09 (95% CI, 0.83 to 1.42) for nonusers of insulin and 1.34 (95% CI, 0.92 to 1.95) for users of insulin. The number of women using metformin alone was insufficient to provide a reliable association estimate.
Breast cancers in metformin users were somewhat more likely to be ductal and less likely to be poorly differentiated, but none of the differences were statistically significant. In metformin users, the incidence of localized cancers was substantially lower than that in nonusers (65 [0.25%] v 90 [0.31%] cases [annualized %], respectively) and the incidence of regional or metastatic cancers was closely comparable (37 [0.14%] v 36 [0.13%] cases [annualized %], respectively; Table 5).
Subgroup analysis results were null. The metformin association with breast cancer was not modified by age, with a P value for the interaction term (P-int) of 0.47, physical activity (P-int = 0.65), use of estrogen alone (P-int = 0.73), use of estrogen plus progestin (P-int = 0.36), alcohol use (P-int = 0.59), or recency of diabetes diagnosis (P-int = 0.68). There was no significant interaction with BMI (P-int = 0.74) despite an HR for metformin of 1.01 (95% CI, 0.64 to 1.58) for BMI less than 30; HR, 0.43 (95% CI, 0.24 to 0.78) for BMI 30 to less than 35; and HR, 0.86 (95% CI, 0.57 to 1.30) for BMI ≥ 35.
Use of metformin was associated with weight loss (mean, −1.4 kg for baseline to year 1 [95% CI, −2.6 to −0.1) compared with use of other medications for diabetes or in women without diabetes (P = .02). However, adjusting for weight loss did not appreciably change the metformin association with breast cancer (HR, 0.75; 95% CI, 0.57 to 0.99).
In this prospective cohort of postmenopausal women, the incidence of invasive breast cancer was lower in women with diabetes treated with metformin compared with women without diabetes. Fewer cancers that were positive for both ER and PR and fewer cancers that were HER2 negative were diagnosed in metformin users.
The predominance of prior observational studies2,3,33 have associated diabetes with higher incidence of breast cancer. However, this association has recently been challenged in two large population-based cohort analyses.4,5 In the first, the British Columbia Linked Health Database covering 99% of the British Columbia population of about 4.8 million residents was used to generate a retrospective cohort. With 2,381 patients, incidence of breast cancer was not associated with diabetes status (HR, 1.01; 95% CI, 0.92 to 1.10; P = .88).5,34 In the second, the Danish National Diabetes Register and Cancer Registry were linked to perform a cohort analyses of the entire Danish population. In that setting, although incidence of several cancers, including those of the liver, pancreas, and lung, were significantly associated with diabetes, incidence of breast cancer was not (P = .37).4 In addition, time-varying analyses found evidence of potential detection bias, suggestive of increased cancer surveillance soon after diabetes detection.4,5,34 Similar to women in these recent contemporary cohort reports, women with diabetes in our WHI cohort did not have a higher incidence of breast cancer compared with women without diabetes, a finding potentially influenced by the relatively high frequency of metformin use in these women.
Until recently, observational studies19,20 examining metformin use and incidence of breast cancer were limited and were mixed with two of four studies22,23 reporting significantly lower incidence of breast cancer for women with diabetes who used metformin. Ruitter et al21 have just reported findings from analyses that used the Dutch National Medical Register (a drug-dispensing database) to generate a cohort of 85,289 women. In this setting, a statistically significantly lower incidence of breast cancer was seen in women who used metformin compared with those who used sulfonylurea derivatives (HR, 0.95; 95% CI, 0.91 to 0.98).21 The metformin results in this study add to such analyses by comprehensively adjusting for breast cancer risk factors, including BMI, physical activity, smoking, and frequency of mammography.
Although some preclinical work suggested predominant metformin influence on triple-negative cancers,35 our review found only one prior study that examined metformin influence on breast cancer subtypes. In 90 women with diabetes and breast cancer, the incidence of PR-positive tumors was higher in metformin users.24 Further studies are clearly needed in this area.
Women with diabetes present with a more advanced stage of breast cancer.36–38 However, many39–41 but not all42 studies find lower mammographic screening rates in women with diabetes. In the WHI clinical trials, mammograms were mandated by protocol. Consequently, mammogram frequency was comparable in women with and without diabetes, and no significant difference in cancer stage was seen comparing women with diabetes who were not users of metformin with women without diabetes. Women with diabetes who were metformin users had a somewhat higher frequency of mammography compared with nonusers. Metformin nonusers were more commonly receiving insulin and therefore more likely to be under subspecialty care. Because providers of subspecialty care are described as less likely to order screening procedures,43 a difference in screening could result. In any event, analyses were adjusted for mammography frequency.
Clinical studies also support a metformin influence on cancer. In a preoperative study,44 women without diabetes with invasive breast cancer randomly assigned to metformin for 2 weeks had reduced Ki-67, a measure of tumor proliferation,45 compared with nonusers. In a retrospective neoadjuvant therapy analysis, patients with breast cancer who had diabetes and used metformin had a higher frequency of complete response (24%) than patients who had diabetes and did not use metformin (8%) and patients without diabetes (16%; P = .02).46 Finally, in retrospective analyses, patients with diabetes and HER2-positive breast cancer who used metformin had better clinical outcome that nonusers.47 In post hoc analyses in an adjuvant setting, use of metformin was not associated with improved survival in patients with triple-negative breast cancer, but there was a trend for decreased distant recurrence compared with women without diabetes.48
An inhibitory influence of metformin on breast cancer is biologically plausible but the potential mediating mechanism is not understood. Proposed mechanisms include indirect insulin-mediated effects and direct effects on cancer cells via influence on the AMPK pathway with resultant inhibition of the mammalian target of rapamycin (mTOR) pathway.15,49
We have previously presented metformin findings with similar trends but with less strong associations that were based on the entire WHI population, including participants in both clinical trials and observational studies.50 Our analyses included only participants in the clinical trials because medication information in the observational study was available only through year 3 compared with through year 9 in the clinical trials.
Diagnoses of diabetes were not based on medical record review; rather, they were determined by ongoing direct query and review of the use of antidiabetic medication. As described in Patients and Methods, this approach has been evaluated,27 and the associations seen compare favorably with the commonly used International Classification of Disease, Ninth Revision (ICD-9) clinical modification codes51; confirmation studies suggest 72% sensitivity and 96% specificity for identification of diabetes.52
The strengths of our study include the prospective cohort design; the large, diverse population well characterized for risk of breast cancer; serial assessment of mammography; breast cancer verification via review of pathology reports; and information on diabetes and the use of diabetes medication updated throughout. This database allowed for time-dependent, exposure-related analyses of use of diabetes medication and risk of breast cancer. Limitations of our study include lack of information on the severity of diabetes and reliance on local laboratory assessment of hormone receptor and HER2 status, precluding information on quality control. Especially noteworthy are the substantial differences in baseline characteristics of women with and without diabetes for factors including obesity and physical activity that could result in residual confounding despite adjustment in analyses for many breast cancer risk factors.
Our findings are of most direct relevance to women with diabetes, most of whom were overweight or obese. However, consideration of the totality of available evidence does provide support for the ongoing clinical studies of metformin,53 including a prospective, full-scale, multicenter adjuvant trial54 and proof of principal studies in prevention settings.55
In a large population of postmenopausal women, use of oral metformin was associated with lower incidence of invasive breast cancer. The influence of metformin on breast cancer subtypes requires further study. Our results inform future studies evaluating use of metformin in the management and prevention of breast cancer.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
See accompanying editorial on page 2812
Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.
Employment or Leadership Position: None Consultant or Advisory Role: Anne McTiernan, Pfizer (C), Novartis (C), Procter & Gamble (C), ZymoGenetics (C), Metagenics (C) Stock Ownership: Anne McTiernan, Merck Honoraria: None Research Funding: None Expert Testimony: None Other Remuneration: None
Conception and design: Rowan T. Chlebowski, Jean Wactawski-Wende, JoAnn E. Manson, Robert Wallace
Administrative support: Rowan T. Chlebowski
Provision of study materials or patients: Lawrence Phillips
Collection and assembly of data: Rowan T. Chlebowski, Jean Wactawski-Wende, JoAnn E. Manson, Aaron K. Aragaki, Lawrence Phillips, Karen Margolis
Data analysis and interpretation: All authors
Manuscript writing: All authors
Final approval of manuscript: All authors