Type 2 diabetes may be a risk factor for dementia, but the associated pathological mechanisms remains unclear. We evaluated the association of diabetes alone or combined with the apolipoprotein E (APOE) gene with incident dementia and neuropathological outcomes in a population-based cohort of 2,574 Japanese-American men enrolled in the Honolulu-Asia Aging Study, including 216 subjects who underwent autopsy. Type 2 diabetes was ascertained by interview and direct glucose testing. Dementia was assessed in 1991 and 1994 by clinical examination and magnetic resonance imaging and was diagnosed according to international guidelines. Logistic regression was used to assess the RR of developing dementia, and log-linear regression was used to estimate the incident rate ratio (IRR) of neuropathological outcomes. Diabetes was associated with total dementia (RR 1.5 [95% CI 1.01–2.2]), Alzheimer’s disease (AD; 1.8 [1.1–2.9]), and vascular dementia (VsD; 2.3 [1.1–5.0]). Individuals with both type 2 diabetes and the APOE ε4 allele had an RR of 5.5 (CI 2.2–13.7) for AD compared with those with neither risk factor. Participants with type 2 diabetes and the ε4 allele had a higher number of hippocampal neuritic plaques (IRR 3.0 [CI 1.2–7.3]) and neurofibrillary tangles in the cortex (IRR 3.5 [1.6–7.5]) and hippocampus (IRR 2.5 [1.5–3.7]), and they had a higher risk of cerebral amyloid angiopathy (RR 6.6, 1.5–29.6). Type 2 diabetes is a risk factor for AD and VsD. The association between diabetes and AD is particularly strong among carriers of the APOE ε4 allele. The neuropathological data are consistent with the clinical results.
Type 2 diabetes is one of the most common metabolic disorders, and its prevalence increases with age. Pathological complications of diabetes affect several organs, including the brain (1). Diabetes is associated with atherosclerosis of the cerebral arteries (2) and leads to important cerebral vascular changes that cause a decrease in cerebral blood flow (3). Furthermore, hyperglycemia is accompanied by an accelerated rate of advanced glycation end product (AGE) formation, which is associated with increased amyloid deposition, tau formation, and oxidative stress (4). Longitudinal population-based studies (5–7) have shown that diabetes is associated with dementia as well as with the subtypes vascular dementia (VsD) and Alzheimer’s disease (AD). However, results are not consistent, and several questions need to be addressed. Two lines of evidence would help to clarify the association of diabetes and dementia.
First, there may be a modification of the risk in diabetic individuals with a particular genetic susceptibility. The major gene associated with AD is the apolipoprotein E (APOE) gene; in particular, the ε4 allelic variant is associated with an increased risk for cognitive impairment and AD (8), whereas its role in VsD is unclear (9). Apolipoprotein E (apoE) plays a critical role in lipid metabolism and brain physiology (10).
Second, investigations of the association of diabetes with neuropathological markers that are common to AD would help to support findings based on clinical diagnosis and to identify possible mechanisms underlying the association. Although there is substantial neuropathological evidence that diabetes is associated with cerebral infarct lesions and stroke (11,12), there are few data on the association of diabetes to neuropathological markers common in AD (13). The Honolulu-Asia Aging Study (HAAS) provides the unique opportunity to examine the association of diabetes to clinically diagnosed subtypes of dementia. In addition, an autopsy study on a subsample of the same cohort provides the basis to examine the association between diabetes, apoE, and neuropathological data.
RESEARCH DESIGN AND METHODS
The Honolulu Heart Program (HHP) is a longitudinal study on heart disease and stroke based on a cohort of 8,006 Japanese-American men born between 1900 and 1919 who were living in Oahu, Hawaii, when the study began in 1965. Respondents were identified using selective service registration records from World War II. The baseline exam was followed by two other exams in 1967–1970 and 1971–1974 (14).
Research on dementia began in 1991 at the fourth examination, when the HAAS was established (15) to investigate the risk factors associated with neurodegenerative disorders and aging (Fig. 1). A total of 3,734 surviving individuals (80% participation rate) were examined for cognitive function at the fourth exam (1991–1993, prevalent phase) and again at the fifth exam (1994–1996, incident phase). During the prevalent phase, a total of 226 cases of dementia were found. Before the incident phase, 521 (14%) subjects died and 395 (10%) subjects could not be reexamined. Therefore, 2,592 individuals were reexamined. As a part of the study, subjects were asked to participate in an autopsy program. To date, autopsy specimens have been collected on ∼20% of the fourth exam cohort, and complete information for the analysis was available on 216 men.
The study was approved by Institutional Review Board of the Kuakini Medical Center and the Honolulu Department of Veterans Affairs, and informed consent for the HAAS and the autopsy study were collected from the participants. At the time of death, consent for the autopsy was also obtained from a family representative.
Diagnosis of dementia.
In the prevalent and incident phases, all subjects were administrated the 100-point Cognitive Abilities Screening Instrument (CASI) (16), a well recognized instrument to assess cognitive function validated among Japanese and Western sample populations (17). In the prevalent phase, CASI score and age were used to identify a subgroup for dementia evaluation (15). At the follow-up exam, subjects with a CASI score less than an education-adjusted cutoff (77 for those with low education and 79 for those with high education) or an absolute drop ≥9 CASI points (n = 749) underwent a specific dementia examination (18). At each exam, evaluation of clinical dementia included a proxy interview, detailed neuropsychological assessment, neurological examination, and neuroimaging. Final diagnosis of clinical dementia was determined by a consensus committee that included the study neurologist and at least two other physicians expert in geriatric medicine and dementia.
Dementia was diagnosed according to DSM-III R criteria (19). Probable and possible AD were diagnosed following the criteria of the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (20), and diagnosis of VsD was based on the California Alzheimer’s Disease Diagnostic and Treatment Centers guidelines (21). Diagnosis of dementia due to multiple etiologies included AD with cerebrovascular disease (CVD) for which probable or possible AD was either the primary or secondary cause of dementia and that was accompanied by probable or possible VsD. For these diagnoses, clinical criteria and neruoimaging data were used, as suggested by the DSM-IV criteria (22). Diagnosis of VsD without apparent AD included individuals with probable VsD and additional non-AD disease (depression, alcohol abuse, B12 deficiency, or subdural hematoma). Dementia caused by other medical conditions included: Parkinson’s disease (n = 10), dementia with Lewy bodies (DLB; n = 5), Pick’s disease, head trauma, B12 deficiency, hypothyroidism, progressive supranuclear palsy, and other unknown causes (total number of subjects = 13). Clinical diagnosis of DLB was based on the guidelines from the consortium on DLB (23,24).
Incidence of dementia was evaluated after an average (means ± SD) period of 2.91 ± 0.30 years. For the analysis, dementia was grouped as total dementia, which included all causes of dementia; AD, with and without apparent CVD (n = 76), which included AD without apparent CVD (n = 51); and VsD without apparent AD (n = 33). AD with apparent CVD included 19 subjects (76%) with AD as the primary cause of dementia and 6 (24%) with VsD as the primary cause.
Using established neuropathological criteria (25), 25% of clinical AD cases met the criteria for definite AD and 65% for definite or probable AD. Among all the VsD cases, 68% had neuropathological evidence of atherosclerosis and did not have substantial AD-related pathologies. Lewy bodies (LBs) in the cortical area were found in ∼10% of the autopsy group; however, there was only one autopsied case diagnosed with VsD with cortical LB, and there were no diagnosed AD cases with cortical LB (24).
Baseline characteristics of the autopsy group were comparable to those of the whole cohort, except the autopsied sample was older and by design included a higher percentage of individuals clinically demented before death (data not shown) (24). A detailed description of neuropathological methods and clinical pathological correlations has been published (26). Briefly, evaluation of brain infarcts was performed on the parenchymal and meningeal tissue from neocortical areas (frontal, temporal, parietal, and occipital lobes). Brain infarcts were classified by the extent of tissue damage and labeled as “large infarcts” for lesions larger than 1 cm, or as “lacunes” for lesions 1 cm or smaller. Infarcts and lacunes were identified during macroscopic examination. Cerebral amyloid angiopathy (CAA) in the parenchyma of the neocortical tissue was evaluated by βA4 amyloid immunostaining (27) and graded from “mild” to “severe” depending on the number of positive vessels per area (28). To improve the statistical power, CAA was categorized as present or absent. The autopsy rates were similar for the major clinically diagnosed subtypes of dementia in this study (24).
Assessment of diabetes.
Diabetes was assessed on the basis of self-report of a doctor’s diagnosis of diabetes, use of oral hypoglycemic medications or insulin intake, or fasting and postchallenge glucose levels measured at the fourth exam. At the time of the study, only subjects who did not report diabetes (with no gastrectomy, no active peptic ulcer, and no stomach cancer) received a 75-g glucose drink (n = 1,729). In these subjects, diabetes was defined according to the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus (29) and included individuals with fasting blood glucose ≥126 mg/dl or with 2-h postload glucose ≥200 mg/dl. A total of 551 subjects without previously reported diabetes were not tested for 2-h postload glucose level and were classified based on the fasting glucose level. However, among older adults, there are a significant number of individuals with normal fasting glucose and isolated postload hyperglycemia with vascular risk factors as high as those with diabetes (30). To evaluate the effect of the lack of information on the 2-h postload glucose level, we performed a sensitivity analysis. We tested the effect of diabetes on dementia under the hypothesis that nondiabetic individuals with normal fasting glucose level and missing data for the 2-h postload glucose were all diabetic; therefore, 1,451 (56.4%) subjects were classified as diabetic. We also evaluated isolated 2-h hyperglycemia separately to test whether elevated postload glucose was a sign of incipient AD (31). There were 202 subjects (7.8%) with normal fasting glucose and isolated elevated 2-h glucose.
Measure of covariates.
APOE genotyping was performed by PCR amplification followed by restriction enzyme digestion (32) at the Bryan Alzheimer’s Disease Research Center at Duke University. Participants were categorized as APOE ε4–positive if they carried at least one copy of the ε4 allele and ε4-negative otherwise. As a result of possibly opposing effects of ε2 and ε4 alleles on dementia (33), individuals with genotype ε2ε4 (n = 36) were excluded from the analyses.
We measured several covariates as possible confounders. At the fourth exam, we controlled the analysis for age, education, and midlife systolic and diastolic blood pressure, cholesterol, BMI, smoking status, and alcohol intake. We used midlife values because they are less influenced by preclinical dementia status. Blood pressure was the mean of the three exams; at each exam the blood pressure represented the average of three measures made 5 min apart. Total cholesterol levels were determined with an Autoanalyzer 1 N24B cholesterol method (34). BMI (kg/m2) was calculated from participants’ height and weight at each exam and averaged. Midlife self-reported smoking was categorized by smoking status (never, former, or current) and by pack/years of cigarette exposure (packs of cigarettes a year × years of smoking). Alcohol intake was recorded as grams of alcohol per day (g/day) and recoded as drinks per day (none, <1 drink/day [13.2 g], 1–2 drinks/day, and ≥3 drinks/day). History of diabetes medication and treatment with insulin were self-reported at each exam up to the fourth exam and included in the analysis as an indicator of the severity of the underlying metabolic impairment associated with diabetes. Use of antihypertensive medications was self-reported by the subjects from exam 1–3 and included the presentation of medication vials at exam 4; the variable was dichotomized (no/yes). Ankle-brachial index (ABI) was measured at the fourth exam, and the values were dichotomized, with a cutoff of 0.9; values below this point were interpreted as indicators of generalized atherosclerosis (35). Stroke and coronary heart disease (CHD) history were assessed at baseline in 1965 and throughout the entire follow-up, up to 1996. Stroke was ascertained as to certainty and type by a neurologist, who reviewed hospital and study records.
The incident cohort included participants who reached the 1994 exam without previously diagnosed dementia. Of these, 2,574 had complete information on dementia and major risk factors. Cohort characteristics were compared by diabetes status using age-adjusted general linear models for continuous variables and age-adjusted logistic regression models for dichotomous outcomes. Logistic regression analysis was used to calculate the estimated RR and 95% CI of diabetes on incidence of dementia and dementia subtypes. Analyses were adjusted for potential confounders. Two different models were examined: in the first, age, education, APOE ε4, diabetes medications, alcohol, and smoking status were included; in the second we added midlife systolic blood pressure, cholesterol, BMI, ABI, stroke, and CHD.
Based on the hypothesis that APOE ε4 modifies the risk of diabetes for dementia, we stratified the analysis based on ε4 and diabetes status (36). Nondiabetic individuals with no APOE ε4 allele were used as the reference group and were compared with those with diabetes and APOE ε4 allele alone or combined. Participants were assigned to four groups: (no diabetes and no APOE ε4 allele, diabetes, ε4 allele, and diabetes with ε4 allele). The significance of the interaction term between diabetes and APOE ε4 was tested by comparing the likelihood ratio of the full regression model, including diabetes and APOE ε4 cross-product, versus the reduced model without the cross-product term.
Neuropathological data were available for 216 subjects, with complete information on risk factors. Multiple logistic regression was performed to compare the presence of CAA among groups with different APOE ε4 and diabetes status. To evaluate the association of APOE ε4 and diabetes to brain infarct, neuritic plaques (NPs), and neurofibrillary tangles (NFTs) counts we used a log-linear regression model (Poisson regression). This approach assumes that the counts of each outcome variable follow an overdispersed Poisson distribution. Overdispersion of the data were assessed by comparing the mean and the variance of the different outcomes (37). All Poisson models were adjusted for the same variables included in the previous analysis plus the time interval between the fourth exam and death. Statistical analysis was performed by using Stata software, version 6.0 (Stata, College Station, TX).
Among the 2,574 participants with complete information, 900 (35%) were classified as diabetic. Compared with nondiabetic subjects, those with diabetes had significantly higher levels of blood pressure, BMI, and cholesterol; had more CHD; and more frequently took antihypertensive medications (Table 1).
After adjustment, compared with nondiabetic subjects, those with diabetes had a significantly increased risk for total dementia, AD, and VsD (Table 2); the risk associated with AD without CVD was higher but not statistically significant. When all subjects with missed 2-h glucose data were considered as diabetic, the association between diabetes and dementia maintained the same direction and magnitude as the main analysis. Diabetic subjects had an RR of 1.6 (95% CI 1.0–2.3) for dementia, 2.0 (1.0–3.5) for VsD, 1.5 (1.0–2.4) for AD, and 1.5 (0.8–2.7) for AD without CVD. Compared with the nondiabetic group, the RR for those with isolated 2-h hyperglycemia was 1.1 (0.5–2.1) for total dementia, 0.9 (0.2–4.1) for VsD, 1.2 (0.5–2.9) for AD, and 0.8 (0.2–2.8) for AD without CVD (Table 3). Furthermore, by eliminating these subjects from the diabetes group, the RR for those with diabetes was 1.6 (1.1–2.4) for total dementia, 2.9 (1.3–6.5) for VsD, 1.9 (1.1–3.2) for AD, and 1.8 (0.9–3.5) for AD without CVD.
We then stratified diabetes on APOE ε4 status. In the first adjusted model, the combined effect of diabetes and the ε4 allele was significantly stronger than in the reference group for total dementia and for the three dementia subtypes (Table 4). After adjusting for confounders, the diabetes and ε4 combined RR for VsD was reduced and became nonsignificant (RR 2.8 [95% CI 0.8–9.8]). The risk for all AD and AD without CVD increased (4.4 [1.9–10.0] and 5.5 [2.2–13.7], respectively).
We investigated the independent and combined effect of diabetes and the ε4 allele on different pathological outcomes. Diabetes alone was significantly associated with the number of brain infarcts, whereas APOE ε4 alone was associated with the number of NFTs in the cortex and in the hippocampus and with the risk for CAA (Table 5). After full adjustment, the group with both diabetes and the ε4 allele had a higher number of hippocampal NPs and hippocampal and cortical NFTs. The risk for CAA was significantly higher for those with both risk factors.
The present study provides important evidence that type 2 diabetes is a risk factor for the development of both VsD and AD. Our data show that association between diabetes and AD is particularly strong among carriers of the APOE ε4 allele. The association between the two risk factors and neuropathological findings support our results on clinical outcomes and provide clues for the pathological mechanism of dementia.
These findings are different from those previously published on the same cohort (38), when no association between midlife diabetes and prevalence of dementia and dementia subtypes identified in 1991 was found. In the previous study, fewer participants had self-reported diabetes (n = 259, 7%). The prevalence of diabetes is known to increase with age (39). In this cohort, there was a more than twofold increase of self-reported cases of diabetes between the third and fourth exams, which were 20 years apart. The small number of diabetic subjects included in the previous analyses could have limited the power to find an association between diabetes and subtypes of dementia.
Our results are based on a large number of participants and were population-based. Covariates were well characterized over the time and often based on multiple measurements. Diabetes diagnosis included the direct assessment of diabetes status during the examination, reducing the possibility of misclassification.
Our study included a large number of autopsies. In the present cohort, clinical diagnosis and autopsy data showed a good agreement (24). However, a recent study found that vascular and AD-type pathologies are related to the degree of cognitive decline without a clear cut-point for the diagnosis of dementia (40). Therefore, we evaluated multiple pathological outcomes to assess the effect of diabetes on the brain independently from the diagnosis of dementia. Autopsy data were available for only 20% of the participants. A selection bias could have been introduced if this group included more demented diabetic subjects compared with those without autopsy. However, the comparison between the two groups with and without autopsy indicated that the frequency of diabetic subjects with dementia was similar.
Previous prospective studies have shown an association between diabetes and VsD and between diabetes and AD (5–7,10). The Canadian Study on Health and Aging had the largest number of VsD cases. The study showed an RR associated to diabetes that was comparable to the one found in our study. Two major population-based studies in different ethnic groups have shown an association between diabetes and incidence of AD, providing similar results (6,7).
This is the first study to examine the interaction between diabetes and the APOE gene in relation to incident dementia and subtypes. Diabetes is associated with multiple metabolic and hemodynamic defects that cause micro- and macrovascular damage (3), leading to a reduced cerebral blood flow and an impaired vascular reactivity (3). These pathological changes may cause hypoxic and ischemic conditions.
VsD is characterized by small and large brain infarcts usually associated with vascular changes (41), and although ischemic lesions are not a unique trait of VsD (42), they might be a good indicator of CVD pathogenesis. Diabetes is a well-known risk factor for lacunar infarction (11) and stroke (12); therefore, its association to VsD and brain infarcts was not surprising. Interestingly, the combined effect of diabetes and ε4 allele was associated with a slightly higher risk for VsD. Diabetes and the apoE ε4 isoform may interact by worsening the atherosclerotic risk profile. This could explain why the combined effect of the two risk factors was not significant after adjusting for cholesterol, systolic blood pressure, stroke, CHD, BMI, and ABI in our study.
The present study also evaluated the role of diabetes as a risk factor for AD. AD is characterized by altered glucose metabolism (4). Some studies have reported higher 2-h glucose levels among patients with impending AD (31) compared with control subjects. Therefore, the observation that hyperglycemia is associated with incident AD could reflect, rather than affect, a preclinical AD status. To examine this question further, we estimated the risk for AD in people with isolated elevated 2-h glucose levels. We found that their risk for AD did not differ from nondiabetic participants. In addition, by removing those subjects from the diabetes group, the association between diabetes and AD become stronger, suggesting that our approach for the analysis of diabetes and AD relationship was conservative.
The association of diabetes with AD, which includes AD cases with or without CVD, was stronger than the association of diabetes with AD and no CVD. This result would suggest that the association between diabetes and AD could also be mediated by the vascular changes caused by the diabetes, similar to VsD. The fact that there were no changes in the association after adjustment for vascular risk factors suggest that other diabetes-dependent mechanisms may be involved. Hyperglycemia in diabetes is usually associated with accelerated AGE formation. Different groups have demonstrated that AGE-related modifications decrease protein solubility and increase protease resistance (43) of several proteins present in the pathological lesions associated with AD (44). AGE product accumulation has been demonstrated in NPs and NFTs in different brain areas of patients with AD (45). Indeed, it appears that aggregation of amyloid-β is accelerated by AGE-mediated cross-linking of extracellular proteins. Furthermore, AGEs may contribute to the microtubule-associated tau protein stabilization and tangle formation (46).
APOE ε4 significantly modified the risk for AD in diabetic subjects. The joint effect of diabetes and APOE ε4 was synergistic, causing a more than fivefold increase in the risk for AD for subjects with diabetes and APOE ε4 compared with those without the two risk factors. Colocalization of AGE and apoE has been reported in senile plaques, NFTs, and CAA of patients with AD and other types of dementia (47). The APOE ε4 allele has a reduced ability to repair neuronal damage and a decreased antioxidant activity compared with the ε2 and ε3 forms (48). ε4 contributes to stabilize β-amyloid deposits, and autopsy data have confirmed that ε4 is related to a higher degree of insoluble amyloid accumulation compared with the other two common polymorphisms (49). Furthermore, the apoE ε4 isoform shows a threefold greater AGE-binding activity than the ε3 isoform, suggesting that AGE-ε4 interaction may contribute to the pathogenesis of the dense amyloid deposits and extracellular NFT (50). We found that among ε4 carriers, there was a significant association between diabetes and the number of NPs in the hippocampus, the number of NFTs in the cortex and the hippocampus, and the risk of having CAA. In most of the subjects, the number of NPs and NFTs associated with the combined presence of the ε4 allele and diabetes seems to be the result of the independent contribution of the two risk factors (additive effect). In the case of the NFTs in the hippocampus, a moderate but significant interaction between the two risk factors seems to be present. Most of all, the risk of having CAA is higher for ε4 carriers with diabetes than those with only diabetes or ε4, indicating a synergistic interaction between the two risk factors. This suggests that the pathological link between APOE ε4, diabetes, and AD may be partially due to an increased risk of CAA formation. Based on these results, we could speculate that a higher prevalence of CAA associated with ε4 and diabetes may affect the clinical expression of AD pathology, in agreement with the findings on this same cohort in which subjects with AD and CAA had a worst cognitive function than those with AD without CAA (27).
In conclusion, our results showed an association between diabetes and both VsD and AD. The study analyzed for the first time the role of diabetes and APOE ε4 allele combined presence in the pathogenesis of dementia, especially for AD. The pathological data in the autopsy subgroup confirm the results on clinical outcomes.
This work has been supported by the National Institute of Aging (N01-AG-4-2149) and the National Heart, Lung and Blood Institute (N01-HC-O5102), National Institutes of Health.
Address correspondence and reprint requests to Rita Peila, Laboratory of Epidemiology, Demography and Biometry Program, National Institute on Aging, National Institutes of Health, Room 3C-309 Gateway Building, 7201 Wisconsin Ave., Bethesda, Maryland 20892. E-mail: email@example.com.
Received for publication 27 August 2001 and accepted in revised form 4 January 2002.
ABI, ankle-brachial index; AD, Alzheimer’s disease; AGE, advanced glycation end product; apoE, apolipoprotein E; CAA, cerebral amyloid angiopathy; CASI, Cognitive Abilities Screening Instrument; CHD, coronary heart disease; CVD, cerebrovascular disease; DLB, dementia with Lewy bodies; HAAS, Honolulu-Asia Aging Study; IRR, incident rate ratio; LB, Lewy body; NFT, neurofibrillary tangle; NP, neuritic plaque; VsD, vascular dementia.