European Journal of Nutrition

, Volume 52, Issue 5, pp 1533–1540

Blood glucose concentrations and breast cancer risk in women without diabetes: a meta-analysis

Authors

  • Peter Boyle
    • International Prevention Research Institute
  • Alice Koechlin
    • International Prevention Research Institute
  • Cécile Pizot
    • International Prevention Research Institute
  • Mathieu Boniol
    • International Prevention Research Institute
  • Chris Robertson
    • International Prevention Research Institute
    • Department of Mathematics and StatisticsUniversity of Strathclyde
  • Patrick Mullie
    • International Prevention Research Institute
  • Geremia Bolli
    • Department of Internal Medicine and Oncology, S.M. della Misericordia HospitalUniversity of Perugia
  • Julio Rosenstock
    • Dallas Diabetes and Endocrine Center at Medical City
    • International Prevention Research Institute
Original Contribution

DOI: 10.1007/s00394-012-0460-z

Cite this article as:
Boyle, P., Koechlin, A., Pizot, C. et al. Eur J Nutr (2013) 52: 1533. doi:10.1007/s00394-012-0460-z

Abstract

Purpose

Some studies have suggested an increased risk of breast cancer associated with elevated fasting serum glucose in nondiabetic subjects. Given how common both breast cancer and impaired glucose tolerance are in our aging societies, this is an important issue for public health.

Methods

We performed a systematic review of prospective cohort studies that examined the association between elevated serum glucose levels in nondiabetic subjects (levels below 7.0 mml/L) and the subsequent risk of breast cancer. We performed a systematic literature search and extracted relevant data in a standard way. We then computed summary relative risks (SRR) and 95 % confidence intervals using a random effects model applied on the risk of highest versus lowest quantile of serum glucose concentrations.

Results

Ten cohort studies were retrieved. The SRR for all studies was 1.11 (1.00–1.23), with no evidence of heterogeneity or publication bias. The SRR was not affected when the analysis was restricted to the 8 studies that reported results for fasting subjects (SRR = 1.11; 95 % CI 0.98–1.25). Three studies provided BMI-unadjusted and BMI-adjusted SRRs of 1.24 (95 % CI 0.60–2.56) and 1.20 (95 % CI 0.63–2.27), respectively. Similar magnitudes of associations were observed in sensitivity analyses, but statistical significance was lost.

Conclusion

In nondiabetic subjects, the risk of breast cancer associated with fasting serum glucose levels seems to be small. Potential limitations to this meta-analysis include the fact that not all studies reported risks adjusted for adiposity and that serum glucose levels of comparison groups were variable across studies.

Keywords

Fasting glucoseDiabetesBreast cancerMeta-analysis

Introduction

Diabetes is a growing condition worldwide. Approximately 7 % of the global population aged between 20 and 79 currently have diabetes, and this prevalence seems set to increase and will affect all regions and countries of the world [1]. Observational studies have shown that diabetes is associated with increasing risk of cancer occurrence and death [2]. Case–control and cohort studies have found a 10–20 % increased breast cancer risk in women with diabetes [3], and a recent meta-analysis of prospective studies found a significant 27 % increase in postmenopausal breast cancer risk associated with type 2 diabetes mellitus [4]. However, less is known about breast cancer risk among subjects with moderately elevated glucose levels who do not have diabetes.

Glycemia is the amount of glucose present in the serum. Normally, the human body tightly regulates glycemia as a part of metabolic homeostasis. Serum glucose levels persistently outside the normal range may be an indicator of impaired glucose tolerance. Diabetes mellitus is the most prominent disease related to failure of blood sugar regulation that is characterized by persistent hyperglycemia during fasting periods. The fasting serum glucose level, which is measured after a fast of (usually) 8 h, is the most commonly used indication of glucose homeostasis. Raised fasting serum glucose levels are an early indicator of impaired glucose tolerance that is widely used as a screening test for diabetes, and type 2 diabetes mellitus is established when fasting serum glycemia is above 7.0 mmol/L [5]. Adiposity is associated with both increased risk of glucose intolerance or diabetes, and breast cancer risk [4]. A challenge for observational studies is thus to figure out whether associations between disturbance of the glucose metabolism and breast cancer would not be in fact due to the confounding effect of adiposity.

We undertook a meta-analysis to investigate the potential association, and its strength and consistency, between fasting glucose in women without diabetes and the risk of breast cancer.

Materials and methods

Pre-defined protocol

A systematic literature search and quantitative analysis were planned, conducted, and reported following MOOSE guidelines regarding meta-analysis of observational studies [6].

Study selection

Studies’ published reports until May 31, 2011, were identified in the following databases: Ovid MEDLINE database; ISI Web of Science, Science Citation Index Expanded (SCI Expanded); and PubMed (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi). All searches used MeSH index terms: “breast neoplasm” or “neoplasm,” each combined with “fasting glucose” and “prospective studies.” Other sources were found in the reference lists of the retrieved articles and preceding reviews on the topic.

Articles eligible for this review had to meet three main criteria: (1) to report data on incident cases of breast cancer, being in situ or invasive; and (2) to report measurement of serum glucose levels, with reporting of data for levels below 7.0 mmol/L, the threshold for diagnosis of diabetes mellitus [5]; and (3) to have a prospective design. The search was restricted to articles published in English. There was no restriction on geographical location of studies.

We restricted our meta-analysis to cohort studies in which blood samples had been taken years before the diagnosis. Analysis of data from cohort studies is usually of two sorts: first, the case-cohort design, in which the exposure history of cancer cases is compared with the exposure history of the rest of the cohort subjects. Secondly, the nested case–control built into a cohort compares the exposure history of cancer cases to that of a sample taken at random among all potential control subjects. The latter procedure allows using a subsample of serums stored for control subjects, which is economical while respecting the prospective nature of the cohort design that entails measuring exposure to hyperglycemia years before the diagnosis of cancer.

We screened titles and looked at abstracts when the title suggested a study possibly meeting the three main criteria. If the abstract content was relevant, full copies of articles were retrieved and fully read by at least two co-authors. Articles reporting on prognosis after breast cancer diagnosis were excluded. Review articles not reporting original data were also excluded but checked for references. Figure 1 summarizes the process of literature search of the present meta-analysis.
https://static-content.springer.com/image/art%3A10.1007%2Fs00394-012-0460-z/MediaObjects/394_2012_460_Fig1_HTML.gif
Fig. 1

Flowchart of the literature search strategy to identify studies on breast cancer and fasting glucose levels

Data extraction from studies

Each study contained a slightly different approach regarding the methodology employed and the presentation of the findings. The serum glucose levels at baseline delimiting the referent category for the calculation of relative risks associated with higher glycemia were not consistent across studies. We took the referent category elicited by each study. For higher serum glucose levels, we extracted data for categories up to 7.0 mmol/L and left out data for higher levels as these were most probably related to established diabetes. Whenever possible, we selected data related to fasting glycemia and discarded data related to categories mixing fasting and nonfasting subjects.

Data extraction was done by one co-author in a pre-defined database. The resulting table was verified by another author and by the statistician who performed the meta-analysis. We extracted the most adjusted risk estimates as well as risk estimates not adjusted for body mass index (BMI). Data extraction also involved key study characteristics including the type of stratified analysis that had been done (e.g., menopausal status), the confounding factors that were adjusted for. Some studies reported the menopausal status according to the period (or age) during which the cancer had been diagnosed, while others took the period (or age) when the blood draw was done.

Statistical methods

The aim was to evaluate the potential relationship between fasting glucose and the risk of developing breast cancer in women. Various risk estimates and their confidence intervals were transformed into log (RR). From the transformed data, summary relative risk (SRR) estimates of highest versus lowest quantile of blood glucose concentrations were calculated using a random effects model [7]. Heterogeneity across studies was evaluated by I2, which represents the percentage of total variation across studies that is attributable to heterogeneity rather than to chance [8]. Since the number of studies included was low, SRRs were obtained with restricted maximum likelihood (REML method in R package) estimates, and confidence intervals were based on t-distribution. Sensitivity analyses were carried out to evaluate the influence of each study on the overall estimate from the meta-analysis.

The impact of whether publication bias might affect the validity of the estimates was initially investigated graphically with a funnel plot. In order to complete this visual analysis, since the sensitivity of the two main publication bias tests Begg [9] test and Egger test [10] is low in meta-analyses based on less than 20 studies, a regression of log (RR) on sample size, weighted by the inverse of the pooled variance (Macaskill test), is the preferred approach [11].

Results

The literature search is summarized in Fig. 1. In total, 527 unique studies were found in the consulted databases, all of which were part of the PubMed database (other databases were less complete). Five hundred and fifteen studies were eliminated because of inappropriate study design. Twelve prospective cohort studies were identified, which fulfilled the established criteria for the meta-analysis. Two studies were excluded: one from the ORDET cohort in Italy [12] that was reanalyzed in 2012 [13] and a study in Israel that did not report risk estimates [14]. The remaining ten studies included in the meta-analysis are summarized in Table 1 [13, 1523]. Large proportions of subjects included in two studies were not fasting at the time of blood draw [19, 22].
Table 1

Main characteristics of cohort studies included in the meta-analysis, ranked by year of publication

First author

Years

Country

Study name

Study type

Menopausal statusa,b

Age (mean and/or range)

Fasting status at blood draw

Glucose level limits of most extreme quantiles (mmol/L)

No. of breast cancer cases

Risk

95 % CI

Adjustments

Manjer et al. [15]

2001

Sweden

Malmö Preventive Project (MPP)

Cohort

Pre-menopausala

43

Fasting

>5.0 versus ≤4.4

110

1.03

0.60–1.75

 

Postmenopausala

54

 

>5.1 versus ≤4.4

157

1.05

0.67–1.63

 

Mink et al. [16]

2002

United States

The Atherosclerosis Risk in Communities Study (ARIC)

Cohort

Pre- and postmenopausal

45–64

Fasting

[5.5–7.0] versus <5.5

161

1.23

0.88–1.71

Age, race, BMI, reproductive factors, family history, alcohol, study center, smoking, number of sisters

Jee et al. [17]

2005

Korea

Korean Cancer Prevention Study (KCPS)

Cohort

Pre- and postmenopausal

30–95

Fasting

[6.1–7.0] versus <5.0

279

1.12

0.92–1.34

Age, alcohol, smoking

Rapp et al. [18]

2006

Austria

Vorarlberg Health Monitoring and Promotion Programme (VHM&PP)

Cohort

Pre- and postmenopausal

19–95 (mean: 43)

Fasting

[6.1–6.9] versus [4.2–5.2]

619

0.91

0.67–1.24

Age, SES, BMI, smoking

<50 yearsb,f

125

0.67

0.29–1.53

50–65 yearsb

261

0.80

0.47–1.36

≥65 yearsb

234

1.03

0.67–1.58

Stattin et al. [19]

2007

Sweden

Västerbotten Intervention Project

Case-cohort

Pre- and postmenopausal

29–61 (mean 46.1)

Variable

Quartile limits not reported

510

1.06

0.82–1.37

Age, no fasting, fasting for less than 8 h and fasting for 8 h or more, BMI, calendar year, smoking

     

<49a

   

92

2.13

1.20–4.10

 
     

≥49a

   

418

0.90

0.68–1.20

 

Gunter et al. [20]

2009

United States

Women’s Health Initiative -Observational Study (WHI-OS)

Nested case–control

Postmenopausal

50–79

Fasting

≥5.5 versus <4.8

836

0.92

0.65–1.29

Age, race, BMI, reproductive factors, family history, alcohol, OC use, HRT, physical activity, smoking, educational attainment, NSAID use, endogenous estradiol levels (in nonusers of HRT)

Kabat et al. [21]

2009

United States

Women’s Health Initiative (WHI)

Nested case–control

Postmenopausal

50–79

Fasting

≥5.5 versus <5.0

79

1.28

0.61–2.66

Age, education, ethnicity, BMI, waist circumference, OC use, HRT, reproductive factors, alcohol, history of breast biopsy, physical activity, family history, total energy intake

Bjorge et al. [22]

2010

Sweden

Metabolic Syndrome and Cancer (Me-Can)e

Pooled analysis of 6 cohorts

Pre- and postmenopausal

29–70+

48 % fasting ≥8 h and 52 % fasting <4 hc,d

Mean 6.6 versus Mean 4.1

4,853

1.30

0.93–1.83

Age, BMI, smoking, year of birth

     

<50b

 

Mean 6.6 versus Mean 4.2

NR

1.28

0.69–2.38

 
     

≥60b

 

Mean 6.6 versus Mean 4.3

NR

1.61

0.95–2.72

 

Lambe et al. [23]

2011

Sweden

Apolipoprotein-related Mortality Risk (AMORIS)

Cohort

Pre- and postmenopausal

Mean 46.6 (SD: 13.6)

Fasting

[6.1–6.9] versus <6.1

3576

1.11

0.91–1.34

Age

     

<50b

  

[6.1–6.9] versus <6.2

826

0.57

0.24–1.37

 
     

≥50b

  

[6.1–6.9] versus <6.3

2750

1.16

0.95–1.42

 

Sieri et al. [13]

2012

Italy

The Hormones and Diet in the Etiology of Breast Cancer Prospective Cohort (ORDET)

Nested case–control

Pre- and postmenopausal

35–69

Fasting

Quartile median 5.3 versus 4.1

356

1.63

1.14–2.32

Age, education, BMI, reproductive factors, family history, alcohol, OC use, smoking

     

<56b

<56

 

Quartile median 5.3 versus 4.1

148

1.14

0.65–1.98

Age, education, reproductive factors, family history, alcohol, OC use, smoking, HRT use and years from menopause

     

>55b

>55

 

Quartile median 5.3 versus 4.2

205

2.19

1.33–3.62

SI conversion: divide by 0.0555 to convert glucose level in mmol/L to mg/gL; BMI: body mass index; SES: socioeconomic status; HRT: hormone replacement therapy; OC: oral contraceptive; NSAID: nonsteroidal anti-inflammatory drugs

aMenopausal status at blood draw

bMenopausal status at breast cancer diagnosis

cFrom Stocks et al. 2010

dFasting status in the Västerbotten Intervention Project was uncertain (Stattin et al. [19])

eIncludes the Västerbotten Intervention Project (Stattin et al. [19]), the Vorarlberg Health Monitoring and Promotion Programme (Rapp et al. [18]), and the Malmö Preventive Project (Manjer et al. [15])

f6 women less than 50 years of age with levels 6.1–6.9 or ≥7.0 mmol/L hence probably including cases of diabetes mellitus

When all ten studies were analyzed together, the SRR for breast cancer in the highest category of fasting glucose recorded was 1.11 (95 % CI 1.00–1.23) (Fig. 2). The I2 was 0 %, indicating that there was no heterogeneity in the risk across studies. There was no evidence of publication bias as indicated by the Begg, Egger, or Macaskill tests. The sensitivity analysis (leave-one-out) also indicated that no single study was unduly influencing the overall finding (data not shown). The SRR was not affected when the analysis was restricted to the 8 studies in which all subjects had fasting blood draw (SRR = 1.11; 95 % CI 0.98–1.25) (Table 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs00394-012-0460-z/MediaObjects/394_2012_460_Fig2_HTML.gif
Fig. 2

Forest plot of meta-analysis of breast cancer in women according to fasting glucose levels. Individual studies are represented by relative risk and 95 % confidence interval (highest versus lowest category of fasting glucose). The size of the square is proportional to the variance of the relative risk. The percent contribution informs on the relative weight each study has in the summary relative risk computed by the meta-analysis

Table 2

Summary of meta-analysis and sensitivity analysis for highest versus lowest fasting glucose category and breast cancer risk

Analysis

No. studies

SRR

95 % CI

I2 (%)

95 % CI (%)

Macaskill test

All studies

10

1.11

1.00

1.23

0

0

56

0.80

All but Bjørge and Stattin

8

1.11

0.98

1.25

0

0

66

0.99

Heterogeneity analysis

Adjustment for BMI

Not adjusted for BMIb

3

1.10

0.84

1.46

0

0

0

0.53

Adjusted for BMI

7

1.13

0.94

1.36

19

0

63

0.72

Adjusted for BMI, excluding Bjørge and Stattin

5

1.12

0.82

1.54

39

0

77

0.31

Not adjusted for BMIc

3

1.24

0.60

2.56

70

0

91

0.80

Adjusted for BMIc

3

1.20

0.63

2.27

53

0

86

0.91

Menopausal status

Pre-menopausala

6

1.11

0.72

1.71

39

0

76

0.93

Postmenopausala

8

1.15

0.92

1.44

45

0

76

0.41

Pre-menopausal, excluding Bjørge and Stattin

4

0.92

0.54

1.56

0

0

81

0.01

Postmenopausal, excluding Bjørge and Stattin

6

1.17

0.89

1.53

41

0

77

0.94

SRR: summary relative risk; RR: relative risk

aAt blood draw or at breast cancer diagnosis (see Table 1)

bThese studies did not report a RR adjusted for BMI

cThese studies (Mink et al. 2002 [16], Gunter et al. 2009 [20], Sieri et al. [13]) reported RR not adjusted and adjusted for BMI (and other factors)

SRRs were not materially different in studies that adjusted and did not adjust for body mass index (BMI) even when the analysis was restricted to studies that reported both unadjusted and adjusted SRRs.

SRRs including all studies were quite similar in pre- and in postmenopausal women. Exclusion of Stattin and Bjørge studies led to sharper differences in SRRs between pre- and postmenopausal women, although these remained statistically nonsignificant.

Conclusion

In late-stage type 2 diabetes, malfunction of pancreatic beta cells leads to an absolute insulin deficiency. However, the onset of type 2 diabetes is usually preceded by a long asymptomatic phase characterized by a number of factors including insulin resistance and hyperinsulinemia, and impaired fasting glucose. Our meta-analysis suggests that the breast cancer risk might be slightly increased in nondiabetic women with impaired fasting glucose.

To the best of our knowledge, it is the first time such a systematic review is conducted on fasting serum glucose levels and breast cancer risk. We restricted our meta-analysis to prospective studies and did not include case–control studies for the following reasons. Studies labeled as having a “case–control design” had rather a cross-sectional design, because blood was drawn after cancer diagnosis was made and not some time before. Hence, the possibility of reverse causality could not be dismissed, that is, women with cancer had higher levels of serum glucose because of their diseases or of conditions associated with treatment or psychological impact of the disease.

Potential limitations to this meta-analysis are first those of published studies. Reported relative risks were adjusted for different factors, and in three studies, no adjustment was done for adiposity. This is however not likely to invalidate the findings because sensitivity analyses did not demonstrate variations in SRR in function of possible confounding factors like adiposity or fasting status at blood draw. Another limitation was that serum glucose levels of comparison groups were variable across studies.

Risk of breast cancer is increased in postmenopausal women with type 2 diabetes (SRR = 1.27; 95 % CI 1.16–1.39), whereas in diabetic pre-menopausal women, the SRR was 0.86 (95 % CI 0.66–1.12) [4]. Our analysis by menopausal status agrees with the finding in studies including diabetic women, provided that studies including nonfasting blood draws are excluded.

There is no clear biological explanation of a role for elevated fasting blood (or serum) glucose and an increased risk of breast cancer, although there are several competing concepts. An independent role for glucose in carcinogenic processes has been proposed with a variety of potential mechanisms, ranging from the generation of free radicals to the induction of damage to DNA repair enzymes [24]. It has been speculated for over half a century that the increased availability of glucose in the blood could encourage the growth of potentially malignant cells which require glucose for progression [25]. However, the mechanism is still a source of speculation rather than certainty. A meta-analysis of randomized trials on intense versus standard glycemic control of type 2 diabetic patients did not provide evidence for reduced cancer risk in the group of patients with improved serum glucose levels glycemia [26]. The risk level we found is somewhat lower than that found for diabetes and breast cancer [4]. In that meta-analysis, studies that adjusted for adiposity (i.e., through inclusion of body mass index in multivariate models) had an SRR of 1.16 (95 % CI 1.08–1.24) for breast cancer, whereas the SRR for studies that did not adjust for adiposity was 1.33 (95 % CI 1.18–1.51) [4]. In this study, adjustments for adiposity did not affect SRRs, suggesting that glucose metabolism disturbances would not just be a marker of adiposity and could play a genuine causal role in breast cancer occurrence.

However, only three studies displayed breast cancer risks with and without adjustment for adiposity, which substantially restricts the ability of observational studies to inform on whether chronic hyperglycemia itself or adiposity or both factors are associated with increased breast cancer risk. Impaired glucose tolerance is associated with hyperinsulinemia, and chronically increased circulating levels of insulin could be the real culprit as insulin is thought to be involved in breast cancer occurrence [27, 28]. Insulin resistance is characterized by a resistance of cells to the action of insulin with slow cellular uptake of glucose. The direct consequence of the insulin resistance is a raise in serum glucose concentration that stimulates insulin secretion. Insulin is a mitogenic agent, and chronic hyperinsulinemia has been related to down-regulation of insulin-like growth factor binding protein-1 and protein-2, with an increase in insulin-like growth factor 1 (IGF-1) as consequence [29]. This increase in bioavailable insulin and IGF-1 could induce mutagenic changes in the cellular environment.

In conclusion, outside the context of established diabetes, our analysis supports a moderate increased risk of breast cancer associated with elevated fasting glucose. However, many individuals with impaired glucose tolerance and/or elevated fasting glucose levels are at high risk of progressing to type 2 diabetes. Given the prevalence of each of these diseases, this could represent a major public health problem. An increased risk of breast cancer in women who have not been diagnosed with diabetes but who have high fasting glucose could represent an even greater problem for global public health. In this respect, early detection of disturbances in the glucose metabolism may contribute to the prevention of many cases of diabetes as well as of breast cancer.

Future researches should focus on metabolic disturbances associated with adiposity and impaired glucose tolerance, with use of potent technologies like nuclear magnetic resonance and mass spectrometry for exploring the disturbances in physiological pathways possibly involved in breast cancer occurrence and in progress to diabetes. Such research may help in discovering new methods for both diabetes and breast cancer prevention.

Acknowledgments

This study was part of the research activities of the International Prevention Research Institute (iPRI) Research Group on Diabetes, Metabolic Disorders and Cancer, whose members are P. Autier, A. Koechlin, C. Pizot, M. Boniol, P. Mullie, P. Boyle, F. Valentini, K. Coppens, L-L. Fairley, M. Boniol-Rech, M. Pasterk, M. Smans, M-P. Curado, M. Bota (iPRI, Lyon, France); S. Gandini (European Institute of Oncology, Milan, Italy); Chris Robertson (Department of Mathematics and Statistics, University of Strathclyde, Glasgow, Scotland); Tongzhang Zheng, Yawei Zhang (Yale University School of Public Health, New Haven, Connecticut, United States of America); Geremia Bolli (Department of Internal Medicine and Oncology, S.M. Misericordia Hospital, University of Perugia, Perugia, Italy); J. Rosenstock (Dallas Diabetes and Endocrine Center, Dallas, United States of America).

Conflict of interest

This study was part of works associated with an unrestricted research grant from Sanofi. The authors declare that they have no conflict of interest.

Copyright information

© Springer-Verlag Berlin Heidelberg 2012