Introduction

Converging epidemiological evidence indicates that individuals with type 2 diabetes mellitus are at an increased risk of developing cancer [1, 2]. Associations with cancers of the pancreas [3], liver [4], colon [5], and endometrium [6–8] are well established.

Results from meta-analyses also suggest that diabetes is associated with an increased risk of breast cancer [9–11]. Most studies of ovarian cancer have on the other hand found no association [12–15].

Impaired glucose control and type 2 diabetes are common disorders affecting an increasing part of the adult population. Several studies have presented results suggesting that not only diabetes but also elevated glucose levels below the diagnostic threshold for diabetes may be associated with an elevated risk for cancer [16–23]. This could be expected if the risk increase associated with diabetes is due to consequences of long-standing disturbances of glucose metabolism without clinically manifest disease. Results from previous studies to date on possible associations between glucose values in the upper normal or impaired range and breast cancer risk are conflicting [16, 19–26], and studies on endometrial and ovarian cancer are scarce [17, 20, 22, 23, 27].

To improve the understanding in this area, we investigated and contrasted in detail possible associations between glucose levels and the risk of the three major female sex hormone dependent cancers in a large cohort of more than 230,000 women, for which information on outcome (breast, endometrial and ovarian cancer) and potential confounders was obtained by means of record linkage to Swedish population-based registers.

Methods

Study population and data sources

The present study was based on the AMORIS (Apolipoprotein-related MOrtality RISk) population [28] recruited in the greater Stockholm area between 1985 and 1996. The cohort consists of individuals that took part in routine health checkups and primary care patients referred for laboratory testing. In total, the AMORIS population encompasses 689,588 individuals of which 338,101 are women. Of these women, 267,223 (79%) were classified with regard to glucose levels. Following exclusion of women younger than 25 years at index examination (n = 27,636), with a previous malignancy (n = 7,551) or with a cancer diagnosis within 6 months from the date of the index visit (n = 1,299), the final study population encompassed 230,737 women.

To obtain information about outcomes and potential confounders, the AMORIS database was linked to several other registers, including the Swedish Cancer Register, the Patient Register before and after the index date, the Multi-Generation Register, the National Population Register, and national Censuses using the individually unique national registration number, assigned to all Swedish residents at birth or at time of first permanent residency.

The study was approved by the Ethics Research Committee at Karolinska Institutet.

The Swedish cancer register

The national Swedish Cancer Register was established in 1958 and is updated yearly. Clinicians and pathologists/cytologists are mandated by law to report all newly diagnosed malignant tumors. In 1998, the reporting rate to the Cancer Register was estimated to be about 96% of all diagnosed cases [29].

The Swedish patient register

The Swedish Patient Register, established in 1964 and with complete national coverage from 1987 onward, includes data on individual hospital discharges. Each record corresponds to one hospital admission and contains the dates of admission and discharge. The main and as many as seven additional discharge diagnoses are recorded for each hospitalization.

The multi-generation register

The Swedish Multi-Generation Register includes as index persons all Swedish residents born in 1932 or later who were alive in 1961 and all those born thereafter, with links to their parents, siblings, and offspring. Adoptions and nonbiological relations are flagged and can be identified.

The national population register

The National Population Register contains information on current place of residence and vital status of all residents in Sweden. The database is continuously updated with information on dates of immigration or emigration which, for the purpose of the present study, allowed censoring of women that emigrated from Sweden during follow-up.

Census data

The Census system is based on a mandatory questionnaire, to be completed and returned by every household in Sweden. The 1980 and 1990 Census were utilized in the present study and includes information on a socioeconomic index based on occupation.

Classification of exposures and outcomes

All laboratory analyses were automated and carried out at the same clinical laboratory (CALAB laboratories, Stockholm, Sweden [30]). Information is available on a broad range of biomedical factors, including glucose levels. Glucose was measured enzymatically by means of a glucose-oxidase/peroxidase method. Subjects were classified with regard to glucose levels at the first contact examination (index examination) where a glucose level below 6.1 mmol/L was classified as normal glucose metabolism (NGM), 6.1–6.9 mmol/L as impaired glucose metabolism (IGM), and 7.0 mmol/L or higher as diabetes mellitus. In addition, if a subject prior to the index examination had information about hospitalization with diabetes as discharge diagnosis in the Patient Register, this was also regarded as manifest diabetes irrespective of the recorded glucose value. The data at hand did not allow for identification of type of diabetes.

Height and weight were available for 38,065 women and were used to compute body mass index (BMI).

Information on age at first child birth and number of children (including nulliparity) was obtained from the Swedish Multi-Generation Register.

Information on socioeconomic standing was retrieved from the most recent Census prior to the index examination and was categorized into high or low based on a commonly used classification system that distinguishes between manual workers, lower nonmanual employees, and intermediate or high nonmanual employees [31].

Incident cases of breast, endometrial, and ovarian cancer in the AMORIS population were identified in the national Swedish Cancer Register.

Statistical analysis

Proportional hazards regression analysis was used to estimate relative risk of developing cancer (hazard ratios), comparing subjects with different glucose levels at the index examination. In all analyses, adjustment was made for age. Additional analyses were made to investigate possible confounding from fasting status and socioeconomic standing. In a separate analysis restricted to parous women, adjustments were also done for reproductive history (age at first birth and parity). In the subcohort of 38,065 women with BMI-measurements, the possible confounding role of obesity was assessed. Separate analyses were also performed in women known to be fasting at the time of blood sampling, as well as in subjects who we judged to have provided glucose measurements as part of a regular health check up based on the profile of the test panel.

In the analyses of breast cancer risk, stratification was made for age at diagnosis as a proxy for menopausal status and a split of follow-up was made at age 50. In the analysis of premenopausal women, individuals were followed to age 50 after which they were censored. In the assessment of postmenopausal risk individuals with glucose measurement before being 50 years of age entered the study at age 50 by means of delayed entry.

In the assessment of endometrial cancer risk, women who had a hysterectomy prior to the index examination (n = 8474) were excluded. Similarly, for analyses of ovarian cancer risk, women who had undergone resection of at least one ovary were excluded (n = 7494).

Follow-up started at the index examination and terminated at the date of the cancer diagnosis of interest (breast, endometrial, and ovarian cancer in separate analyses), date of hysterectomy (in analyses of endometrial cancer), date of oophorectomy (in analyses of ovarian cancer), death, emigration out of Sweden or 31 December 2002, whichever came first.

Results

In the study population of 230,737 women, 218,279 (94.6%) had normal glucose metabolism (NGM) at the index examination, 6,843 (3.0%) had an impaired glucose metabolism (IGM), and 5,615 (2.4%) had glucose levels above the threshold for diabetes or had a hospital discharge diagnosis of diabetes mellitus prior to the index examination. The mean follow-up time was 11.7 years. A total of 127,112 (55%) women were recorded as fasting at the time of the index examination. The proportion of glucose measurements labeled as fasting was highest in the NGM group (55.8%) compared with (43.0%) in the IGM group and (42.5%) in the group with diabetes. Younger age and nulliparity at index examination were more common among individuals with NGM. Among parous women, age at first child birth was comparable between the three groups of glucose levels. In the subcohort with information available on anthropometric measurements, the mean BMI was lowest in the NGM group (23.8 kg/m2) and highest in the diabetes group (26.7 kg/m2). During follow-up, 6,070 cases of breast cancer, 1,070 cases of endometrial cancer, and 783 cases of ovarian cancer were identified (Table 1).

Table 1 Characteristics of women in the AMORIS population stratified on glucose levels at index examination
Full size table

In all analyses, adjustments were made for age at the index examination. Further adjustments for socioeconomic status did not change the results (not shown in Table), and this variable was, therefore, not kept in the models. In separate analyses restricted to subjects who provided glucose measurements as part of a regular health control estimates were practically unchanged (not shown in Table). In analyses restricted to women who were known to be fasting associations remained virtually unchanged. If anything, associations between IGM and risks of endometrial and postmenopausal breast cancer were slightly more pronounced (Tables 2, 3). To address more in detail, possible reverse causation separate analyses were also performed with lag times of 1 and 3 years, respectively. With this approach, results remained unchanged (not shown in Table).

Table 2 Glucose levels and risk of breast, endometrial, and ovarian cancer in the AMORIS population
Full size table
Table 3 Glucose levels and risk of breast cancer in pre and postmenopausal women in the AMORIS population
Full size table

Breast cancer

The overall risk of breast cancer in patients with diabetes was elevated (HR = 1.15, 95% CI 0.99–1.34), but no association was found for impaired glucose metabolism (Table 2). However, in an analysis stratified on age at cancer diagnosis as a proxy for menopausal status, an increased risk associated with diabetes was confined to postmenopausal breast cancer (HR = 1.22, 95% CI 1.04–1.43). A slightly elevated risk was also suggested in postmenopausal women with impaired glucose metabolism (HR = 1.11, 95% CI 0.96–1.28) (Table 3). In a subcohort of parous women adjustments for parity and age at birth of first child did not change the results (Tables 2, 3). In the subcohort of women in which BMI-measurements were available, relative risk estimates were virtually the same as in the full cohort and adjustments for BMI did not markedly change the point estimates (Table 3).

Endometrial cancer

Both diabetes (HR = 1.46, 95% CI 1.09–1.96) and impaired glucose metabolism (HR = 1.41, 95% CI 1.08–1.85) were associated with an increased risk of endometrial cancer. Adjustments for parity and age at birth of first child did not change the relative risk estimates (Table 2). In the subcohort of women with BMI-measurements, relative risk estimates were slightly more marked than in the full cohort. Following adjustment for BMI, point estimates were somewhat attenuated (Table 2).

Ovarian cancer

No clear associations were found between glucose levels and ovarian cancer risk. There were very few cases in the subcohort of women where BMI-measurements were available yielding estimates of relative risk with wide confidence intervals. Adjustments for BMI did not change the results (Table 2).

Discussion

In this large cohort of women, diabetes was associated with an increased risk of endometrial cancer and postmenopausal breast cancer. Also, increased risks were observed for endometrial cancer at glucose levels below the diagnostic threshold for diabetes. For postmenopausal breast cancer, an increased risk was suggested in women with impaired glucose metabolism, but this association did not reach statistical significance. Adjustments for possible confounding factors such as age, socioeconomic standing, BMI, parity, and age at birth of first child did not materially change the estimates. No associations were seen between glucose levels and the risk of ovarian cancer.

Several mechanisms have been postulated to contribute to an association between adult onset diabetes and cancer risk: activation of the insulin pathway, activation of the insulin-like growth factor pathway, and impaired regulation of endogenous sex hormones [9]. Both insulin and insulin-like growth factor I (IGF-I) have been shown to promote tumor development by stimulating cell proliferation, while inhibiting apoptosis [32]. Also, glucose may increase risk directly by stimulating proliferation of tumor cells acting as energy source [33]. The relative contribution from glucose, insulin, or IGF-I to the risk of cancer is unclear.

Breast cancer

The analysis of breast cancer was stratified into two age groups as a proxy for menopausal status (<50 years and ≥50 years at the time of diagnosis). Our findings of an increased risk of post, but not premenopausal breast cancer in women with diabetes are supported by the results of two meta-analyses [10, 11].

Furthermore, our results indicate an association between impaired glucose metabolism and an increased risk of postmenopausal breast cancer. As the risk associated with diabetes was only modestly increased, the risks associated with glucose levels below this threshold can be assumed to be slightly lower and thereby also more difficult to detect. Results from previous studies on nondiabetic glucose levels and breast cancer have been contradictory. Four out of nine studies reported no association [19–21, 24]. For example, one large Korean study reported an increased risk of breast cancer associated with diabetes, but found no association for fasting glucose levels below the diabetic level [19]. Five studies found evidence of an association with glucose levels below the threshold for diabetes, but with divergent results for menopausal status. Muti et al. [16] reported an association between hyperglycemia and risk of premenopausal breast cancer. Similarly, Stattin et al. [22] found elevated fasting glucose to be associated with an increased risk in women <49 years at baseline, but not in older women. In a study where the majority of cases (136 out of 187) were older than 50 years of age at the start of follow-up, Mink et al. [25] reported an association between glucose levels above the normal range and breast cancer risk, even though the associations were attenuated when BMI was considered. A large recent study based on six European cohorts found glucose levels below the diabetic level to be associated with an increased risk of breast cancer among women ≥60 years of age [34]. Based on a random sample of women from the Women’s Health Initiative study, Kabat et al. [26] found no association between glucose levels and postmenopausal breast cancer risk in white women, but an increased risk for African-American women. Heterogeneity in findings across studies may at least to some extent be explained by differences in cut points for glucose levels, follow-up times, whether or not menopausal status was considered in the analyses, and statistical power issues.

Endometrial cancer

Diabetes has previously been reported to be associated with approximately a two times increased risk of endometrial cancer [6–8], findings which our results corroborate. In the present study, an increased risk was also present for individuals with impaired glucose metabolism. Similarly, three out of four earlier studies with data available on fasting glucose levels have found evidence of increased risks associated with increasing glucose levels [17, 22, 23, 27].

Ovarian cancer

Consistent with findings from earlier studies, we found no associations between diabetes [12–15] or between impaired glucose metabolism [20, 23] and risk of ovarian cancer. However, Hemminki et al. [35] observed an increased risk of ovarian cancer among patients hospitalized for type 2 diabetes.

Strengths of the present study included the size and the prospective study design with almost 100% complete coverage of incident cancers in the Swedish Cancer Register. The number of cancer cases available for analyses was considerably higher than in earlier studies in this area conducted in Western populations. The data at hand also allowed adjustments for known reproductive risk factors. Another important strength was that recorded glucose values were based on the same automated analytic methods performed in one laboratory. No patients were hospitalized at the time of blood draw.

There are some limitations that warrant consideration. First, no information was available on lifestyle factors such as physical activity, alcohol consumption, and smoking. To some extent, adjustment for BMI may reduce confounding from low physical activity. In addition, if impaired fasting glucose in part is caused by low physical activity, adjustment for this factor in the analyses could be overly conservative. While smoking is not an established risk factor for breast cancer [36, 37], it has been suggested to be associated with an increased risk of mucinous ovarian cancer [38] and a decreased risk of endometrial cancer [39]. Second, we had no information on use of exogenous female hormones or antidiabetic medication. There are reports of associations between use of antidiabetics and cancer risk and mortality, with metformin associated with a decreased risk and sulfonylurea and insulin with an increased risk [40–42]. A third potential limitation was misclassification of glucose levels, particularly in individuals with unknown fasting status. This notion was supported by our finding of a larger proportion of individuals with unknown fasting status observed in the IGM and diabetes groups. Finally, information on glucose levels was based on one blood sample only. However, it is unlikely that any such misclassification is related to the likelihood of developing cancer. If anything, a nondifferential misclassification of glucose levels would tend to underestimate the strength of observed associations. Adjustments for fasting status at the index examination did not change results. Also, when the analysis was restricted to women known to be fasting, the associations remained virtually the same.

No detailed information was available on the primary reason for the measuring of glucose. However, an estimated 80% of subjects provided measurements as part of a regular health control as indicated by the profile of the test panel. Following separate analyses of this group estimates remained virtually unchanged.

BMI-measurements were only available for a subcohort of women; nevertheless, the number of cases available for analysis in this subgroup was larger than in most previous studies. We can not exclude that this subcohort was slightly different from the full cohort, but this should not affect the relation between glucose levels and cancer incidence within the cohort, nor influence the possibility to evaluate the potential confounding effect from obesity. Following adjustments for BMI, estimates for endometrial cancer risk were somewhat attenuated, while estimates for breast and ovarian cancer remained unchanged.

In conclusion, our results based on a very large cohort of women confirm previous reports of associations not only between diabetes mellitus and endometrial cancer but also between diabetes and postmenopausal breast cancer risk. More importantly, we found evidence of risk modifying effects of glucose levels below the diagnostic threshold for diabetes. Given an expected continued increase in the prevalence of impaired fasting glucose in many populations, even small excess risks for breast cancer have important public health implications. Impaired glucose metabolism is a potentially modifiable risk factor by means of lifestyle changes such as increased physical activity, dietary habits, and weight control [43]. In a public health perspective, such changes may not only reduce the risk of developing adult onset diabetes but may also prove to be of importance to reduce the burden of cancer in women.