Personal History of Diabetes, Genetic Susceptibility to Diabetes, and Risk of Brain Glioma: A Pooled Analysis of Observational Studies

This article is featured in Highlights of This Issue, p. 1

Glioma, the most common type of malignant brain tumor in adulthood, is a relatively uncommon and highly fatal malignancy, with a 5-year survival rate of only 34% (1). The most common and aggressive subtype of glioma, glioblastoma, has a 5-year survival of less than 5% (1). Currently, there are no effective means of glioma prevention apart from avoiding unnecessary exposure to ionizing radiation, particularly during childhood (2).

Several studies (3–9), although not all (10–13), have suggested that elevated blood glucose levels or a previous history of diabetes are inversely associated with risk of glioma or other brain tumors. Diabetes, particularly type 2, has been associated with an increased risk of several other types of malignancies, possibly mediated by inflammatory markers or elevated levels of glucose, insulin, or insulin-like growth factor-1 (14). The potential biological mechanisms that could explain an inverse association between diabetes and risk of glioma remain unclear, although immune response may play a role (15). Similar to diabetes, a personal history of allergies has consistently been associated with an approximately 40% reduced risk of glioma (16).

To further our understanding of the relationship between diabetes and glioma and possible underlying biological mechanisms, we combined questionnaire-based and genotyping data from 5 cohort and 2 case–control studies to examine the association between self-reported history of diabetes and subsequent risk of glioma and possible effect modification by 7 glioma-related single nucleotide polymorphisms (SNP). We additionally examined whether an association between diabetes and glioma risk might be explained by genetic mechanisms using genotyping data from 13 diabetes risk-associated SNPs.

The study population included glioma cases and controls from the previously conducted genome-wide association study (GWAS) in the GliomaScan Consortium (17). This collaborative project included 1,856 incident cases and 4,955 controls from 3 case–control studies, 1 population-based case–only study, and 14 case–control studies nested within cohort studies. For the current analysis, we restricted the study population to the 5 nested case–control studies [Agricultural Health Study (AHS), Physicians' Health Study I and II (PHS), Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO), Shanghai Men's and Women's Health Study (SMWHS), Vitamins and Lifestyle Study (VITAL)] and 2 case–control studies [Multicenter Study of Environment and Health-National Cancer Institute- Adult Brain Tumor Study (NCI-BTS) and National Institute for Occupational Safety and Health-Upper Midwest Health Study (NIOSH-UMHS)] in which one or more glioma cases and controls reported a history of diabetes (18–26). Six of these studies were conducted in the United States, and one in China (SMWHS). Four studies collected information on age at diabetes diagnosis (NCI-BTS, AHS, PHS, and SMWHS). We excluded 1 case and 2 controls under the age of 18, 13 cases and 31 controls with missing questionnaire data on history of diabetes, and 1 case and 1 control from SMWHS whose self-reported ages at diagnosis of diabetes occurred after the age or referent age at glioma diagnosis. The final study population consisted of 962 glioma cases (557 males and 405 females) and 2,195 controls (1,196 males and 999 females). Brief descriptions of studies and participant characteristics are shown in Table 1.

Glioma cases were defined as primary malignant brain tumors with ICD-O-3 histology codes 9380–9480 (or equivalent). We further classified the following mutually exclusive subgroups of glioma according to histology: glioblastoma including gliosarcoma (ICD-O-3 codes 9440–9442), astrocytoma (ICD-O-3 codes 9384, 9400–9401, 9410–9411, 9420–9421, 9424), and oligodendroglioma/oligoastrocytoma/oligodendroblastoma (ICD-O-3 codes 9382, 9450, 9451, and 9460). Cases with histology codes outside of these ranges were classified as “other and unspecified gliomas.” Controls from cohort studies were originally chosen from a subset of the full cohort for the purposes of previous GWAS, or were matched to cases on age or year of birth, sex, race, and/or other factors (Table 1), and were glioma-free at the time of glioma diagnosis. Controls from case–control studies were frequency-matched by age, sex, and other factors, and were glioma-free at the time of interview.

All variables included in this analysis were combined across studies using standardized definitions and categories. Demographic characteristics, including dates of birth, sex, race/ethnicity (White, Black, American Indian/Alaskan Native, Asian/Pacific Islander, other), education (less than high school graduate, high school graduate or equivalent, some college or vocational school post-high school, college graduate or post-graduate), personal history of allergies including asthma, hay fever, eczema, other allergy, or allergic skin condition (yes, no), hormone replacement therapy (ever, never), and smoking status (never, former, current) were self-reported in study-specific questionnaires and provided by all 7 studies. Height and weight were measured by study personnel in the SMWHS study and self-reported in all other studies. Questionnaire data on personal history of diabetes (yes, no) and other factors were ascertained at study enrollment in most of the cohort studies (AHS, PHS, PLCO, VITAL), apart from one (SMWHS) that assigned diabetes status using questionnaire information from both baseline and follow-up. Case–control studies (NCI-BTS, NIOSH-UMHS) ascertained this information soon after diagnosis among cases and at the corresponding referent age among controls (i.e., the age at time of interview). Missing data for any of the covariates (<1% for race, education, smoking, and body mass index; 14% for history of allergies; 34% for hormone replacement therapy use in women) were handled by including missing indicator variables in the models.

Blood, saliva, or buccal samples were provided by each of the studies (apart from SMWHS, which did not contribute to the genetic analyses) to use for the extraction of DNA for genotyping. Genotyping was performed using the Illumina 660-W Human BeadChip for all glioma cases. Genotyping for controls was conducted using various platforms (Table 1). To account for differences in the assays used, SNP data for the control samples were imputed if necessary, as described by Marchini and colleagues (27). A total of 13 SNPs that have been associated with diabetes and 7 SNPs that have been associated with glioma, based on statistical significance after correction for multiple testing in previous GWAS (17, 28–35), were selected to evaluate the association for diabetes-related genetic polymorphisms on risk of glioma and the potential for effect modification of the diabetes–glioma association by genetic susceptibility to glioma, respectively. We computed a genetic risk score for the diabetes and glioma SNPs by assigning 0, 1, or 2 for homozygous for the nonsusceptible allele, heterozygous, or homozygous for the susceptible allele, respectively, and summing these values.

We calculated pooled odds ratios (OR) and 95% confidence intervals (CI) for glioma using unconditional logistic regression models adjusted for study, age (at baseline for cohort studies or at time of diagnosis/pseudodiagnosis for case–control studies), and birth year (<1930, 1930–1939, 1940–1949, 1950–1959, ≥1960). Models that combined male and female subjects were further adjusted for sex. Adjustment for other potential confounders, including race, education, body mass index (weight in kg divided by height in meters squared), smoking status, menopausal hormone replacement therapy, and history of allergies did not meaningfully change the β coefficients for diabetes (<10% change), so these covariates were not retained in the final models.

We also examined the possible effect modification by sex, current and early-adulthood BMI, history of allergies, and 7 recently identified glioma risk-associated variants by comparing the magnitude of the associations for diabetes and glioma risk across categories of these potential effect modifiers. Tests for interaction were conducted by comparing the fit of a model with an additional cross-product term for diabetes and the potential effect modifier to a model without this term using the likelihood ratio test. Between-study heterogeneity was evaluated using the Mantel–Haenszel χ² test.

Pooled ORs were additionally estimated for 13 diabetes-associated SNPs using the most common genotype among controls as the referent genotype. Trend tests were conducted by modeling the genotypes (categorized as 0, 1, and 2, corresponding to the number of variant alleles) as continuous variables, and statistical significance was evaluated by comparing the fit of a model with the continuous term to one without using the likelihood ratio test.

All statistical analyses were conducted using Stata/SE 11.0 (StataCorp).

Of the 962 total glioma cases, 517 (311 male and 206 female) were classified as glioblastoma, 200 (119 male and 81 female) as astrocytoma, and 152 (74 male and 78 female) as oligodendroglioma/oligoastrocytoma/oligodendroblastoma. The median age at glioma diagnosis was 57 years (interquartile range: 42–69; range: 18–91).

We observed a 42% reduced risk of glioma (OR = 0.58; 95% CI, 0.40–0.84) for individuals with a history of diabetes versus those without (Table 2). Similar ORs were observed in men and women (P_interaction = 0.91) and by case–control and nested case–control design (P_{heterogeneity} = 0.87). Results were similar after excluding each individual study. The association between diabetes history and glioma risk was more strongly inverse among participants with current BMI values of 18.5 to 24.9 (OR = 0.48; 95% CI, 0.24–0.96) and 25.0 to 29.9 (OR = 0.43; 95% CI, 0.22–0.85) compared with those with BMI ≥30 (OR = 1.16; 95% CI, 0.61–2.20), but these differences were not statistically significant (P_interaction = 0.12). Among 210 glioma cases and 1,027 controls with data on BMI in early-adulthood (age 18–21), there was no clear variation in ORs for diabetes history within strata of early-adult BMI (P_interaction = 0.72). We found no evidence of a multiplicative interaction by history of allergies (P_interaction = 0.21). The inverse association was statistically significant for diabetes diagnosed before age 50 but not for diabetes diagnosed at older ages (Table 2). After additionally excluding 2 cases and 7 controls diagnosed with diabetes within 2 years of glioma diagnosis, the ORs for history of diabetes versus no history (OR = 0.59; 95% CI, 0.40–0.86) and for history of diabetes by age at diagnosis (OR = 0.34, 0.55, and 0.52 for diabetes diagnosis at <50, 50–59, and ≥60 years old, respectively; P_trend = 0.24) did not change materially. Similar results were observed when we restricted the analysis to glioblastoma while keeping the same controls (based on 25 cases with diabetes and 492 without; OR = 0.58; 95% CI, 0.37–0.92). Although the number of cases was small, we also observed reduced risks associated with a diabetes history for astrocytoma (based on 6 cases with diabetes and 194 without; OR = 0.56; 95% CI, 0.24–1.31) and oligodendroglioma (based on 2 cases with diabetes and 157 without; OR = 0.28; 95% CI, 0.07–1.18).

Within the population subset with genetic data, we did not observe any significant per-allele trends among the 13 diabetes-related SNPs examined in relation to glioma risk (Table 3). Nonetheless, the TT genotype for rs2237892 on the KCNQ1 gene (OR = 4.46; 95% CI, 1.19–16.70) was associated with significantly increased risk of glioma compared with the CC genotype. Among controls, however, only rs7480010 and rs1113132 were significantly associated with self-reported history of diabetes. The diabetes genetic risk score was not significantly associated with risk of glioma [ORs (95% CI) for risk scores of ≥11–<13, ≥13–<15, and ≥15 vs. <11 = 1.00 (0.72–1.38), 1.13 (0.83–1.55), and 1.18 (0.86–1.62), respectively; P_trend = 0.22], nor with self-reported diabetes among controls [OR (95% CI) = 0.95 (0.41–2.20), 0.79 (0.35–1.80), and 1.60 (0.75–3.42); P_trend = 0.27]. Additional adjustment for the diabetes genetic risk score had little influence on the OR for history of diabetes (OR = 0.57 and 0.55 before and after adjustment, respectively). None of the 7 glioma risk-associated SNPs individually modified the association between history of diabetes and glioma risk (Supplementary Table S1); however, the inverse association for diabetes was nonsignificantly greater among individuals with a lower versus higher glioma genetic risk score [OR (95% CI) for risk scores of 0–<7, 7–<9, and ≥9–14 were 0.39 (0.18–0.86), 0.59 (0.30–1.18), and 0.96 (0.38–2.42), respectively; P_interaction = 0.11].

We observed a 42% reduced risk of glioma for individuals with a history of diabetes versus those without (OR = 0.58; 95% CI, 0.40–0.84). Results were similar by sex, study design (case–control vs. nested case–control), and after restricting the outcome to glioblastoma, the most common histological subtype. The risks for other glioma subtypes were also reduced for those with versus without diabetes, albeit nonsignificantly. Diabetes-associated SNPs identified from GWAS were generally not associated with glioma risk, nor did these SNPs explain the inverse association we observed between glioma risk and diabetes history.

A lower risk of glioma associated with diabetes was first reported in a 1965 study, reporting a lower frequency of intracranial gliomas among diabetics versus nondiabetics but no difference in the frequency of meningiomas and pituitary neoplasms (36). Our finding of an inverse association between diabetes and glioma risk is consistent with results from previous case–control studies on glioma, all reporting relative risks below 1.0 for individuals with before diagnosis versus those without (range: 0.3–0.82; refs. 3–7). Like our study, these studies did not distinguish between type 1 versus type 2 diabetes. One case–control study showed a stronger inverse association for diabetes diagnosed at younger versus older ages (6), which could act as a proxy for type 1 versus type 2 diabetes, but this study was among those included in the current pooled analysis. Results from prospective cohort studies, which have lower potential for biases because of the reliance on proxy respondents or differential recall or selection of cases and controls, have been lacking. The California Seven-Day Adventists cohort, which included only 21 incident cases of glioma, did not show a clear association between self-reported history of diabetes and glioma risk (relative risk = 1.84; 95% CI, 0.28–7.38; ref. 10). A larger record-linkage study in Denmark found no association between hospital discharge records for diabetes and subsequent brain and central nervous system cancers in either men [standardized incidence ratio (SIR) = 1.1; 95% CI, 0.9–1.4] or women (SIR = 1.1; 95% CI, 0.8–1.3), based on 80 and 79 cases, respectively (11).

An inverse association between history of diabetes and risk of adult glioma may reflect a protective effect of heightened immune response, as has been hypothesized to underlie the relationship between allergies and glioma risk (16). History of allergies was also inversely associated with risk of glioma in this population (data not shown), but this association was independent of diabetes. Studies with more specific information on diabetes history, including type 1 versus type 2, and specific biomarkers of immune function and insulin resistance could provide additional insight on biological mechanisms that may underlie the inverse association between diabetes and glioma risk. Alternatively, our results could be explained by long-term suppression of diabetes-related symptoms, particularly those related to immune response, during glioma development (37). However, assuming that glioma (and, particularly, glioblastoma) has a relatively short natural history because of its relatively aggressive nature, reverse causation is unlikely to explain the significant inverse association we observed for diabetes diagnosed at younger ages.

Our pooled analysis combined original data across a diverse set of studies, which allowed an examination of the consistency of the association between men and women and case–control and case–control studies nested within prospective cohort studies. We applied a standard format for the categorization of variables and standard methods for statistical analysis, which reduced the potential for methodological heterogeneity across studies. We also had information on potential confounding factors, including race/ethnicity, education, and history of allergies, as well as data to assess potential effect modification by genetic susceptibility to glioma. Nevertheless, our results were based on a relatively small number of cases reporting a previous diagnosis of diabetes. Self-reported diabetes diagnoses were not validated in any of the studies included in the current analysis, and the number of individuals with a history of diabetes may have been underestimated. Assuming random misclassification of this exposure, our results may have been biased toward the null. Furthermore, apart from age at diabetes diagnosis, we lacked sufficient data to more accurately distinguish between type 1 and type 2 diabetes. Participants from nested case–control studies were originally selected from the full cohorts for the purposes of genetic analyses. This could have biased our results if the controls had been selected on factors associated with diabetes history; however, the pooled ORs did not differ after individually excluding each nested case–control study. Thus, our findings should be confirmed in larger studies, preferably full prospective cohort studies, with medically validated data on diabetes history. Although we found no significant interactions between diabetes and 7 glioma risk-associated SNPs on glioma risk, statistical power to detect such interactions was limited. Of the 13 SNPs previously strongly associated with diabetes, we found that the TT genotype for rs2237892 on the KCNQ1 gene was associated with a more than 4-fold increased risk of glioma compared with the CC genotype, an association opposite to those observed for risk of type 2 diabetes, higher fasting glucose levels, and reduced β-cell function (31). This finding, should it replicate in other studies, may provide some clues about the role of insulin resistance in the etiology of glioma. Although we did not observe significant associations with glioma in log-additive models for this SNP or 12 other diabetes-associated SNPs, it is possible that as yet undetermined genetic loci may contribute to the protective effect on glioma.

In summary, our international pooled study revealed an inverse association between diabetes and glioma risk, which is consistent with previous case–control studies. The role of genetic susceptibility cannot be excluded and should be pursued in future studies together with other factors that might be responsible for the diabetes–glioma association. Identification of these factors may help to improve our understanding of glioma pathogenesis and identify possible avenues for prevention of this highly fatal disease.

No potential conflicts of interest were disclosed.

Conception and design: C.M. Kitahara, A.V. Brenner, B. Melin, M. Waters, N. Chatterjee, S.J. Chanock, P. Hartge, P. Rajaraman

Development of methodology: C.M. Kitahara, A.V. Brenner, J.M. Gaziano, M. Waters, P. Hartge, P. Rajaraman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Melin, P.D. Inskip, L.B. Freeman, T. Carreon, J.M. Gaziano, U. Peters, M.P. Purdue, A. Ruder, H.D. Sesso, X.-O. Shu, M. Waters, E. White, W. Zheng, R.N. Hoover, M. Yeager, S.J. Chanock, P. Rajaraman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.M. Kitahara, A.V. Brenner, Z. Wang, P.D. Inskip, M. Feychting, H.D. Sesso, N. Chatterjee, S.J. Chanock, P. Rajaraman

Writing, review, and/or revision of the manuscript: C.M. Kitahara, M. Linet, A.V. Brenner, S.S. Wang, B. Melin, Z. Wang, Peter D. Inskip, L.B. Freeman, M.Z. Braganza, T. Carreon, M. Feychting, J.M. Gaziano, U. Peters, M.P. Purdue, A. Ruder, H.D. Sesso, X.-O. Shu, M. Waters, E. White, R.N. Hoover, J.F. Fraumeni Jr, P. Hartge, P. Rajaraman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.D. Sesso, X.-O. Shu, M. Yeager, P. Rajaraman

Study supervision: X.-O. Shu, R.N. Hoover, M. Yeager, P. Rajaraman

The authors thank all investigators who contributed data for the GliomaScan study.

For PHS: The PHS is supported by grants CA 97193, CA 34944, CA 40360, HL 26490, and HL 34595 from the National Institutes of Health.

For PLCO: This research was supported by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics and by contracts from the Division of Cancer Prevention, National Cancer Institute, NIH, DHHS.

For NIOSH: This research was supported by NIOSH operating funds. The contributions of former laboratory director Mary Ann Butler are gratefully acknowledged. The findings and conclusions in this report are those of the author(s) and do not necessarily represent the views of the National Institute for Occupational Safety and Health.

For VITAL: Grant K05CA154337 is funded by the National Cancer Institute (NCI) and the NIH Office of Dietary Supplements (ODS).

This work was supported in part by the Intramural Research Program of the National Cancer Institute, National Institutes of Health.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.