Type 2 Diabetes and Subsequent Incidence of Breast Cancer in the Nurses’ Health Study

RESEARCH DESIGN AND METHODS

Population

The Nurses’ Health Study was established in 1976 when 121,700 female registered nurses 30–55 years of age completed a mailed questionnaire on their health status and on various potential risk factors for cancer, cardiovascular disease, and other major illnesses. Participants receive follow-up questionnaires biennially to update information on demographic, anthropometric, and lifestyle factors and on newly diagnosed diseases including diabetes and breast cancer. The response rate still exceeds 90%. The Nurses’ Health Study was approved by the institutional review board of the Brigham and Women’s Hospital (Boston, MA).

Diabetes confirmation and validation

All women who report a physician diagnosis of diabetes on the biennial questionnaire are mailed a supplemental questionnaire requesting detailed information on diagnosis, laboratory results, and treatment. Participants who confirmed a diagnosis of diabetes on this supplementary questionnaire were considered to have “definite” type 2 (non-insulin-dependent) diabetes if they met the National Diabetes Data Group criteria for diabetes (15), did not meet criteria for type 1 (insulin-dependent and ketosis-prone) diabetes, and were diagnosed at age ≥30 years. No weight criteria were used in the type 2 classification. Only women with definite type 2 diabetes were classified as diabetic in this analysis; women who reported a diagnosis of diabetes on the main questionnaire but were classified as probable or unlikely on the basis of supplementary questionnaire information and women who were classified as having type 1 diabetes were excluded from this analysis.

The validity of supplementary questionnaires to confirm and characterize diabetes type was evaluated in a random subsample of Nurses’ Health Study participants with self-reported type 2 diabetes (16). Of 84 women contacted, 71 gave permission for medical record review; records were obtained for 62 women. Self-reports of type 2 diabetes were confirmed by medical record review by an endocrinologist for 61 (98%) of the cases.

The diagnostic criteria for diabetes were changed by the American Diabetes Association in 1997 (17). Because we assessed diabetes only up to 1996 for this analysis, the change in criteria did not affect our diabetes definition.

Information on diabetes medication (insulin and sulfonylureas) was obtained from the main and supplementary questionnaires.

Identification of breast cancer

On each biennial questionnaire, participants were asked whether they had been newly diagnosed with breast cancer during the previous 2 years, and if so, what was the date of diagnosis. The National Death Index is also routinely searched for deaths among women who do not respond to the questionnaires. All women who reported breast cancer (or the next of kin for those who had died) were asked for permission to review the relevant medical records to confirm the diagnosis. Pathology reports, obtained for 93% of the cases, confirmed breast cancer in >99% of women whose reports were reviewed. Although medical records could not be obtained for 7% of the cases, analyses were based on all reports of newly diagnosed breast cancer, because the degree of accuracy of the participants’ reports was extremely high among those for whom records were obtained. Cases of breast carcinoma in situ (n = 612) were censored from this analysis because we do not follow them further for the occurrence of invasive breast cancer in our cohort. Only a fraction of in situ breast cancers progresses to become invasive. We excluded ductal carcinoma in situ as end point because differential use of preventive services could lead to higher detection of carcinoma in situ and lead to spurious associations. Diabetic women see providers more often than healthy women and therefore might have greater access to screening.

Population for analysis

Women who reported cancer (except for nonmelanoma skin cancer) at baseline in 1976 were excluded from the analyses (n = 3,302), as were those who reported type 1 diabetes (n = 497) or type 2 diabetes that was not confirmed by the supplementary questionnaire (n = 1,091) or if their date of diagnosis of diabetes was missing (n = 4). Women who reported onset of diabetes before age 30 years were excluded from the study population, as they were more likely to have type 1 diabetes (n = 112). Participants also were excluded if their date of birth was missing (n = 27), if they died shortly after agreeing to participate in the study (n = 2), if they did not report their height (n = 149), or if they developed breast cancer during follow-up but their date of diagnosis was not available (n = 29). This left a study population of 116,488 women. During follow-up, women were censored from the analysis if they developed breast cancer or any other cancer, if they died, or if they were lost to follow-up.

Statistical analysis

Women were followed prospectively for the occurrence of diabetes from 1976 to 1996 and for invasive breast cancer from 1976 to 1998. Women with prevalent diabetes in 1976 were included in the analyses. When duration of diabetes was considered, self-reported date of diagnosis was used also for women who reported onset before 1976. Person-years of follow-up were calculated as the time from completion of the 1976 questionnaire at baseline to the date of return of the 1998 questionnaire or to the date of diagnosis of invasive breast cancer, breast carcinoma in situ, other cancer, death, or loss to follow-up, whichever occurred first. Women contributed person-years as diabetic subjects (exposed) from the time of diagnosis of type 2 diabetes until they reached one of the above-listed censoring points. Women with prevalent diabetes in 1976 contributed person-years as diabetic subjects from the start of follow-up. Women contributed person-years as nondiabetic subjects (nonexposed) during the time they were free of type 2 diabetes. A Cox proportional hazards model was used to calculate the hazard of developing invasive breast cancer given a history of type 2 diabetes (18). The proportional hazards model allows us to adjust simultaneously for multiple potential confounders of this association. Regression models were adjusted for age (in months), family history of breast cancer in first-degree relative(s) (dichotomous), history of benign breast disease (dichotomous), height (continuous), BMI at age 18 years (continuous), current BMI (continuous), waist-to-hip ratio (continuous), age at menarche (≤11, 12, 13, 14, ≥15 years), parity (0, 1, 2, 3, ≥4 children), age at birth of first child (<25, 25–29.9, 30–34.9, ≥35 years), menopausal status (premenopausal, postmenopausal, unknown), age at menopause (continuous), use of postmenopausal hormones (never, past user for <5 years, past user for ≥5 years, current user for <5 years, current user for ≥5 years), moderate and vigorous physical activity (<1, 1–1.9, 2–3.9, 4–6.9, ≥7 h per week), and alcohol consumption (none, 0.1–4.9, 5.0–14.9, ≥15 g per day). Covariate values were updated in the analysis whenever new information was obtained from the biennial questionnaire between 1976 and 1996. Information on alcohol consumption was first obtained in 1980 and assumed constant between 1976 and 1980. Because waist-to-hip ratio was assessed only in 1986 and 1996, we did not include this variable as a covariate in all our analytic models, but in secondary analyses we tested whether inclusion would alter the estimates. Analyses were stratified by BMI (<25, 25–29.9, ≥30 kg/m²), menopausal status (pre- and postmenopausal), both of the above, postmenopausal hormone use, family history of breast cancer, mammography screening, use of diabetes medication (insulin, oral hypoglycemic medication, or other oral diabetes medication), and by duration of diabetes (≤5, 5.1–10, 10.1–15, 15.1–20, ≥20 years) before a diagnosis of breast cancer. Separate analyses were conducted to explore the association between a history of type 2 diabetes and estrogen receptor-positive and -negative breast cancer and fatal breast cancer.

All tests of statistical significance are two sided.

RESULTS

During 2.3 million person-years of follow-up over 22 years, 6,120 women who were cancer-free reported type 2 diabetes and 5,605 women were newly diagnosed with invasive breast cancer. Of the women with breast cancer for whom estrogen receptor status was known, 2,915 had estrogen receptor-positive and 989 had estrogen receptor-negative breast cancer. A total of 1,100 women had died from breast cancer as of June 1998.

Table 1 shows the distribution of characteristics during follow-up among participants who reported diabetes at baseline or during follow-up and among those who remained free of the disease. Women who reported type 2 diabetes had a substantially higher BMI than nondiabetic women during adult life and at age 18 years. Women with type 2 diabetes also had a somewhat higher waist-to-hip ratio, were less physically active, drank considerably less alcohol, and were less likely to use postmenopausal hormones than nondiabetic women (Table 1).

The age-adjusted hazard ratio (HR) for developing invasive breast cancer was 1.11 (95% CI 0.96–1.28) for women with type 2 diabetes compared with nondiabetic women (Table 2). After adjustment for potential confounders, family history of breast cancer, a history of benign breast disease, height, BMI at age 18 years, current BMI, age at menarche, parity, age at first child’s birth, menopausal status, age at menopause, postmenopausal hormone use, physical activity, and alcohol consumption, the association between a history of type 2 diabetes and the risk of invasive breast cancer was somewhat strengthened and was modest but significant (HR = 1.17; 95% CI 1.01–1.35). BMI at age 18 years and current BMI were entered into the regression model as continuous variables to provide the most complete control for confounding by obesity possible given the observed data. Both diabetes and obesity were independently related to breast cancer in this model. The slight increase in the HR after adjustment for covariates was due to negative confounding by a number of variables, in particular menopausal status, age at menopause, and alcohol consumption. This negative confounding was stronger than the positive confounding by current BMI. A model adjusting for all covariates except current BMI resulted in an HR = 1.23 (95% CI 1.07–1.42). Adjusting for all covariates, including current BMI, but not for the menopausal variables (menopausal status, age at menopause, and postmenopausal hormone use) produced an HR for the association of diabetes and breast cancer of 1.13 (95% CI 0.98–1.31). A model, fully adjusted except for alcohol consumption, resulted in an HR = 1.15 (95% CI 0.99–1.33).

Additional adjustment for waist-to-hip ratio did not materially alter the estimate (HR = 1.16; 95% CI 1.00–1.34). Because data on waist-to-hip ratio were not available at baseline and not available for a considerable number of participants during follow-up, results are presented without considering waist-to-hip ratio.

The association between type 2 diabetes and breast cancer was significant among women with estrogen receptor-positive tumors but not among women with estrogen receptor-negative tumors (Table 3). There were, however, fewer cases in the latter group. There was no association between a history of diabetes and fatal breast cancer (Table 3).

The association between type 2 diabetes and breast cancer incidence was not appreciably modified by BMI (Table 4). Among premenopausal women, breast cancer risk was slightly lower for diabetic than nondiabetic women, but numbers were small. The positive association between type 2 diabetes and breast cancer was restricted to postmenopausal women and was independent of their BMI (Table 4). There was no statistically significant interaction between type 2 diabetes and menopausal status (test for interaction, P = 0.16). The association between type 2 diabetes and breast cancer incidence was not substantially modified by a family history of breast cancer or the use of insulin or other diabetes medication. Similarly, restricting the analysis to women who never used postmenopausal hormones or to women who ever had a mammogram did not appreciably change the observed association.

There was no consistent relation between duration of diabetes and breast cancer incidence (Table 5). The hazard for breast cancer was slightly higher among women who were diagnosed with diabetes between 10 and 15 years before a diagnosis of breast cancer than for women who had diabetes for <10 years or >15 years. Women who had diabetes for <20 years had a lower HR for breast cancer than women with no diabetes, but there were few cases in this group. The relation between duration of diabetes and subsequent breast cancer risk was similar when the analysis was restricted to postmenopausal women (data not shown).

CONCLUSIONS

These data from the Nurses’ Health Study indicate that women with type 2 diabetes have a slightly but significantly higher risk of developing breast cancer than women with no diabetes. The elevated risk was apparent among postmenopausal women but not among premenopausal women. Although the association between type 2 diabetes and breast cancer did not appear to be explained by adiposity, residual confounding by adiposity is likely. We controlled for obesity in our analyses as well as possible, but measurement error in self-reported weight precludes achieving complete control of confounding by adiposity.

As early as 1885, hyperglycemia was described in patients with cancer (19). In the early 1920s, Warburg (20) demonstrated that tumor slices sustained higher rates of glucose utilization and lactic acid production than normal tissue sections. Women with breast cancer were noted decades ago to have higher rates of diabetes than women with benign breast disease or healthy women (21–23). Of the systematic prospective and case-control studies that followed, some reported elevated risk ratios for breast cancer among women with type 2 diabetes (24–27), but other studies indicated no association (28–36). Most studies, however, were limited in size and therefore underpowered to reveal a modest association. Furthermore, only two studies (35,36) accounted for general and/or central obesity, which is associated with both type 2 diabetes and breast cancer. Weiss et al. (35), in a large population-based case-control study in the U.S., found no significant association between a history of adult-onset diabetes and risk of breast cancer after adjusting for BMI (odds ratio = 1.13; 95% CI 0.7–1.90). The Iowa Women’s Study did not reveal an overall association between type 2 diabetes and breast cancer incidence, adjusting for both BMI and waist-to-hip ratio (36).

The data from the Nurses’ Health Study provide the largest population with the longest follow-up in which the association between diabetes and breast cancer has been studied. Our findings of a modest association are consistent with several previous reports of a slightly elevated risk (21–27,29,33). Most previous studies, however, did not have sufficient power to provide stable estimates.

We found the modest association between diabetes and breast cancer to be largely independent of self-reported adiposity and body fat distribution. We attempted to adjust for confounding by obesity as closely as possible using the observed data on weight and height by entering height, BMI at age 18 years, and current BMI as continuous variables into the regression model (37). Neither current adiposity nor central obesity appeared to be an important confounder of the diabetes-breast cancer association. Central obesity, which is an even stronger predictor of insulin resistance and type 2 diabetes than BMI, has been found to increase breast cancer risk among both pre- and postmenopausal women (38).

In the Nurses’ Health Study, the association between type 2 diabetes and breast cancer was only apparent among postmenopausal women. Our finding that the risk for breast cancer was nonsignificantly lower in premenopausal women with diabetes than in nondiabetic women parallels the observation that obese premenopausal women have a lower risk of breast cancer than women of normal weight (39). Whereas in postmenopausal women estrogen synthesis takes place in adipose tissue, in premenopausal women estrogen synthesis is primarily gonadal and obesity is associated with reduced endogenous estrogens (40). Insulin may elevate free plasma estrogen levels through SHBG inhibition (5–7), which may be particularly relevant in women with low estrogen levels, e.g., postmenopausal women with no exogenous estrogen replacement.

We could not confirm any effect modification of the association between diabetes and breast cancer by a family history of breast cancer, as was reported from the Iowa Women’s Health Study cohort (41). In that analysis, however, the only confounders considered were BMI and waist-to-hip ratio (41).

It is possible that women who are diagnosed with diabetes are more likely to be screened for breast cancer. We examined a possible detection bias by restricting our cohort to women who received mammography screening. The estimates remained unchanged.

Given the misclassification and confounding inherent in an observational study, we have to consider the possibility that the observed association was a result of such biases. Furthermore, waist-to-hip ratio measurements were not available for the entire cohort, and thus, our final results were not adjusted for central obesity. Generally, however, associations in our data were strengthened by adjustment for potential confounders, indicating that the true association was more likely under- rather than overestimated. Nevertheless, residual or unmeasured confounding has to be considered as a possible explanation for the modest increase in risk observed. One-third or more of persons with diabetes go undetected in the population and are therefore misclassified as nondiabetic (42). Diagnostic criteria for diabetes changed after the women included in this analysis were diagnosed, and some women classified as free of type 2 diabetes would be classified as having type 2 diabetes under the new criteria (17). A reduction in nondifferential misclassification would strengthen the association between diabetes and breast cancer observed.

Between 1990 and 1998, the prevalence of diagnosed diabetes increased by 33% in the U.S.(43), reflecting the rapid increase in the prevalence of obesity (44). This phenomenon may have an impact on possible late induction of tumorigenesis and on tumor growth. In 1998, the total number of U.S. adults who suffered from diabetes was estimated at 16 million (43). Thus, even a slight increase in breast cancer risk in women with diabetes would be of public health concern. If hyperinsulinemia plays a role in breast cancer pathogenesis, interventions that improve insulin sensitivity such as exercise and dietary modifications may be expected to lower associated breast cancer incidence (45), but this assumption requires further study.