Diabetes Mellitus and Risk of Developing Alzheimer Disease:  Results From the Framingham Study FREE

Diabetes mellitus (DM) has been associated with cognitive decline,^1- 8 anatomical measures of brain aging (whole-brain atrophy and hippocampal atrophy),^9- 11 and with an increased risk of stroke and vascular dementia (VaD).^12- 16 The relationship between DM and Alzheimer disease (AD) is less clear; while some prior studies did find an association,^1,17- 20 others did not.^{13- 14,21- 22} The varying results may be due to differences in the age, ethnicity, sex, and risk factor profile of different study populations; variations in the study designs, criteria used to define DM, dementia, VaD, and AD; and varying lengths of follow-up. Prior studies relating DM to the risk of AD have either had a relatively short follow-up period of less than 6 years or substantial loss to follow-up.¹²

We had previously observed that the risk of developing dementia following stroke was greatest in younger, apolipoprotein E (APOE) ε4–negative participants, that is, in persons at the lowest a priori risk of developing dementia.²³ Furthermore, insulin resistance increased the risk of developing AD only among persons who were APOE ε4 negative.²⁴ We hypothesized that the association between DM and AD would also be best demonstrated in a long-term follow-up of subjects without other major known risk factors for AD (APOE ε4 genotype and an elevated plasma homocysteine [tHcy] level).

Framingham Original cohort participants (n = 2611) were enrolled in a dementia-free cohort between 1976 and 1978, based on their performance on a standardized neuropsychological test battery,²⁵ subsequent biennial Mini-Mental State Examinations,²⁶ and, in persons suspected to have cognitive impairment, further evaluations by a neurologist. At the 16th biennial examination (1979-1982), 2210 participants (1325 women; mean age 70 ± 7.0 years) from this cohort remained free of dementia and attended the examination; they constituted our study sample. Informed consent was obtained from all participants using a consent form approved by the Institutional Review Board at the Boston University School of Medicine.

Methods used for dementia screening and follow-up have been previously described.^27- 28 Participants identified as demented met Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria²⁹ and had definite dementia (Clinical Dementia Rating scale score ≥1)³⁰ for a period of 6 months or longer. Participants diagnosed with AD met National Institute of Neurological and Communicative Diseases and Stroke/Alzheimer's Disease and Related Disorders Association criteria for definite or probable AD,³¹ while those with VaD met Alzheimer Disease Diagnostic and Treatment Center criteria for probable VaD.³² Persons with more than 1 possible cause for their dementia were included only in the all-dementia analysis.

We defined DM as the documented use of insulin or an oral hypoglycemic drug or a casual (nonfasting) plasma glucose level greater than or equal to 200 mg/dL (11.1 mmol/L), recorded at the 16th biennial examination or at an earlier examination.³³

The APOE genotype data were available for only 1139 persons (51%), as other participants either died before DNA could be collected for genotyping, or they refused consent.³⁴ Participants were dichotomized into those with and those without an APOE ε4 allele. Information on all other covariates—plasma tHcy levels (dichotomized as within or below the top age and sex quartile), educational status (dichotomized at high school completion), current cigarette smoking, alcohol intake (0, <2, or ≥2 drinks per day), systolic blood pressure, body mass index, baseline cardiovascular disease, and stroke—was available for most participants.

In our primary analysis, the exposure variable was the presence or absence of DM at the 16th biennial examination, which we related to the subsequent development of all-cause dementia, AD, and VaD (until December 2003) using Cox proportional hazards regression models (age- and sex-adjusted, and multivariable models). Alzheimer disease and VaD were treated as competing risks. In subjects for whom APOE genotype data were available, we further adjusted for APOE ε4 status. We examined effect modification by age, sex, baseline blood pressure, diabetes treatment, a priori risk status, and availability of APOE ε4 genotyping by incorporating corresponding interaction terms in the analyses. We further estimated the risks of developing all dementia and AD within subgroups of subjects at a lower a priori risk of developing dementia and AD. In secondary analyses we modeled diabetes as a time-dependent covariate. All statistical analyses were performed using SAS software (SAS Institute, Cary, NC).

At baseline, 202 (9.1%) of the 2210 study participants had DM; 98 subjects (4.4%) had a documented blood glucose level greater than 198 mg/dL (11 mmol/L), and 176 subjects (8%) were receiving insulin or oral hypoglycemic drugs. Demographic and baseline characteristics of the diabetic and nondiabetic subjects are compared in Table 1.

During a 20-year follow-up period (range, 1-20 years; mean, 12.7 years), 319 of the 2210 participants became demented (237 had definite or probable AD; 32 had probable VaD). Low-risk subjects who did not have either a plasma tHcy level in the top quartile or an APOE ε4 gene had a relative risk (RR) for AD of 0.32 (95% confidence interval, 0.23-0.47; P<.001), and very low-risk subjects with these traits who were also aged less than 75 years had a RR for AD of 0.24 (95% confidence interval, 0.15-0.39; P<.001), compared with all subjects.

Using either the standard or the time-dependent Cox models, we found that overall, DM did not increase the risk of all-cause dementia and AD, but in the standard Cox models, there was an increased risk in the subgroup of low-risk and very low-risk subjects (Table 2). This difference in risk was seen throughout the period of follow-up (Figure 1 and Figure 2). We found effect modification by age (higher risk in younger subjects; P<.05) and a priori risk status, but did not find effect modification by sex, diabetes treatment, baseline blood pressure, or availability of APOE genotyping.

Our numbers are too small to permit a definitive conclusion regarding the risk of developing VaD.

Diabetes mellitus was not an independent risk factor for AD in the overall Framingham Study sample, but it was a strong independent risk factor for AD among persons at a relatively lower initial risk for developing the disease (those who were APOE ε4 negative and did not have a plasma tHcy level in the highest age- and sex-specific quartile), raising the RR for AD by 3- to 5- fold in this group. The RR was greatest in younger subjects (<75 years of age) in whom the population-attributable risk from DM was 25%. Our findings suggest that an influence of DM on AD risk may be evident in subgroups of study populations in which there is no evidence for an overall association between these traits.

The strengths of our study are its community-based sample, prospective study design, and the direct ascertainment of exposure and outcome status. Selection bias was minimized by population enrollment well before the onset of diabetes or dementia, and intensive, ongoing follow-up of survival and cognitive status. The ongoing determination of diabetic status at each biennial examination, starting well before the onset of dementia, minimized the possibility of an ascertainment bias. Finally, subjects were observed for up to 2 decades.

An important limitation is the predominantly white study population. Further, we used casual (nonfasting) rather than fasting blood glucose levels and hence, we may have misclassified patients with impaired glucose tolerance as nondiabetic, which would bias the results of our overall and subgroup analyses toward the null hypothesis, but would not explain our findings of increased risk in some subgroups.

Our findings are consistent with the results of several prior studies,^{13- 14,17,21- 22} which either did not show an association between DM and the subsequent risk of developing AD or described such an association only in specific subgroups (men, persons with systolic blood pressure >180 mm Hg, or APOE ε4 genotype positive) .^13,17,19 They differ from the findings of a few large population-based studies that did report an overall association.^1,18,20,35 One possible explanation for this discrepancy is variation in the definition and prevalence of DM between the studies.

We found that the RR of developing AD attributable to DM was highest in subjects without the 3 known AD risk factors: a tHcy level in the top quartile, an APOE ε4 gene, or age 75 years or older. No prior studies have addressed the risk in this specific subgroup, but a higher RR of AD associated with DM has been described in subjects younger than 75 years compared with subjects older than 85 years.¹⁸ The Kungsholmen project¹³ reported no interaction between DM and the APOE ε4 genotype, while the Honolulu-Asia Aging study¹⁹ observed a higher RR among diabetic patients with an APOE ε4 genotype. These differences could be due to chance or differences in ethnicity and sex distribution in the populations. Prior studies have suggested that the brains of AD subjects without an APOE ε4 gene may be more susceptible to the adverse effects of hyperinsulinemia; this provides a possible biological rationale for our epidemiological findings.³⁶

There are several biologically plausible, nonvascular pathways that could mediate an association between DM and AD. Diabetes mellitus promotes the formation of advanced glycation end products,³⁷ inflammation,³⁸ and neurofibrillary tangle formations.³⁹ Insulin resistance and the resulting hyperinsulinemia appear to increase the risk of AD.⁴⁰ Finally, common genetic pathways may underlie the development of both DM and AD; as one example, mutations in the insulin-degrading enzyme that induce DM⁴¹ also impair the degradation of β-amyloid protein.⁴²

Our findings emphasize the need to consider DM as a potential risk factor for cognitive impairment, dementia, and AD. The utility of cognitive testing and improved diabetic control in reducing the risk of AD in persons with diabetes need to be addressed in further studies.

Correspondence: Sudha Seshadri, MD, Department of Neurology, Boston University School of Medicine, 715 Albany St, B-608, Boston, MA 02118-2526 (suseshad@bu.edu).

Accepted for Publication: May 24, 2006.

Author Contributions: Drs Beiser and Seshadri had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Akomolafe, Beiser, Green, Wolf, and Seshadri. Acquisition of data: Beiser, Au, Wolf, and Seshadri. Analysis and interpretation of data: Beiser, Meigs, Green, Farrer, Wolf, and Seshadri. Drafting of the manuscript: Akomolafe, Green, and Seshadri. Critical revision of the manuscript for important intellectual content: Beiser, Meigs, Au, Green, Farrer, Wolf, and Seshadri. Statistical analysis: Beiser and Farrer. Obtained funding: Akomolafe, Meigs, and Wolf. Administrative, technical, and material support: Au, Farrer, Wolf, and Seshadri. Study supervision: Au, Green, Farrer, and Seshadri.

Financial Disclosure: None reported.

Funding/Support: Supported by the National Heart, Lung, and Blood Institute's Framingham Heart Study (NIH/NHLBI Contract #N01-HC-25195) and grants from the National Institute of Neurological Disorders and Stroke (5R01-NS17950), the National Institute of Aging (5R01-AG08122, 5R01-AG16495, 3R01-AG09029), and the Boston University Alzheimer Disease Center (P30 AG13846). Dr Meigs is supported by an American Diabetes Association Career Development Award.