Abstract
Objective: To evaluate the association of fasting insulin level to incident dementia in a cohort of elderly men.
Methods: Data are from the Honolulu-Asia Aging Study, a community-based study of Japanese-American men, aged 71 to 91 years in 1991. Serum insulin was measured in 1991 and participants were grouped based on their insulin levels. Dementia was ascertained in 1991, 1994, and 1996 according to international guidelines. The 2,568 men dementia-free in 1991 were reexamined in 1994 and 1996; 244 new cases of dementia were diagnosed. Survival analysis with age as the time scale was used to estimate the risk (hazard ratio [HR] and 95% CI) for incident dementia associated with levels of insulin.
Results: The risk of dementia was increased at the two extremes of the insulin distribution (lower and upper 15th percentiles). Compared to the rest of the cohort subjects in the lowest 15th percentile and highest 15th percentile had an increased risk for dementia (HR = 1.54, CI 1.11 to 2.11 and HR = 1.54, CI 1.05 to 2.26). In men with insulin levels <22.2 mIU/L the risk for dementia decreased with increased levels of insulin (HR = 0.76, CI 0.72 to 0.79 for each increase of one logarithmic unit −2.72 mIU/L of insulin). In men with insulin levels ≥22.2 mIU/L the risk for dementia increased with increasing levels of insulin (HR = 1.64, CI = 1.07 to 2.52 for each 2.72 mIU/L).
Conclusions: Both low and high levels of insulin are associated with an increased risk of developing dementia.
Hyperinsulinemia is associated with an adverse risk profile for coronary heart disease and atherosclerosis.1 Abnormalities of insulin metabolism have also been associated with dementia.2 In cross-sectional studies peripheral hyperinsulinemia and reduced insulin sensitivity have been found in cognitively impaired subjects and demented patients.3,4⇓ CNS disturbance of insulin metabolism has also been shown to be associated with the severity of dementia.5
However, more recent evidence suggests that insulin has important neurotrophic properties in the brain that can improve cognitive function and memory processes.6,7⇓ High levels of insulin and insulin receptors are present in several areas of the brain, with a particularly high density in those regions involved in memory.8 In normal and demented subjects, insulin administration significantly improved memory.9,10⇓ In animals, direct administration of insulin to the brain protected against metabolic and structural damage that followed experimental cerebral ischemia.11
The possible role of insulin on the risk of developing dementia has not been tested in a prospective study design. Because previous findings suggest both a negative and a positive effect of insulin in the brain, we hypothesized a nonlinear association between insulin levels and the risk for dementia. This hypothesis was tested in a prospective study of a community-based cohort of Japanese-American men.
Subjects and methods.
Study population.
The Honolulu-Asia Aging Study (HAAS) began in 1991 as a continuation of the Honolulu Heart program (HHP), a population-based longitudinal study of 8,006 Japanese-American men born between 1900 and 1919 and living in Oahu, HI, when the study began.12 Participants were seen at baseline (1965–1968) and again in 1968–1970 (examination 2), 1971–1974 (examination 3), 1991–1993 (examination 4), 1994–1996 (examination 5), and 1997–1999 (examination 6). Physical measurements and demographic and medical information were collected at each examination. The study was approved by the Institutional Review Board of the Kuakini Medical Center and the Honolulu Department of Veterans Affairs and informed consent was signed by the study participants. If subjects were demented at the time of consent, approval was obtained from a proxy.
Dementia-case finding.
Case finding for dementia started in 1991 during examination 4 and has been previously described.13 At that examination, 3,734 men (80% of surviving cohort members) were tested for cognitive function and a subsample underwent a detailed assessment for dementia. Prevalent cases (n = 226) were identified and excluded from the incident cohort. Participation rate was 84% and 90% for the 5th and the 6th examinations.14 Participant cognitive status was measured with the 100-point Cognitive Abilities Screening Instrument (CASI), a combination of the Hasegawa Dementia Screening scale, the Folstein Mini-Mental State Examination, and the Modified Mini-Mental State test.15 The CASI is a well-recognized instrument for the assessment of cognitive function validated among Japanese and Western sample populations.16 Subjects with a CASI score less than an education-adjusted CASI cutoff point (CASI = 77 for those with low education and CASI = 79 for those with high education) or an absolute drop of 9 or more CASI points from the previous examination underwent a specific dementia evaluation. Evaluation of clinical dementia included a neurologic examination, additional neuropsychologic testing, an MRI, and a proxy interview. Diagnoses were made in a consensus conference attended by the neurologist and two other study physicians. Dementia was diagnosed according to Diagnostic and Statistical Manual of Mental Disorders–III R criteria,17 probable and possible Alzheimer disease (AD) were diagnosed following the criteria of the National Institute of Neurologic and Communicative Disorders and Stroke and the AD and Related Disorders Association,18 and the diagnosis of vascular dementia (VaD) was based on the California AD Diagnostic and Treatment Centers guidelines.19 Based on these criteria, 244 cases of incident dementia were identified at examinations 5 (132 cases, 54.1%) and 6 (112 cases, 45.9%), including 148 (64.2%) cases of AD and 39 (16.0%) cases of VaD.
Insulin and diabetes measurements.
In 1991, glucose and insulin levels were measured after an overnight fast. Insulin was measured by a double-antibody radioimmunoassay method at the University of Washington (Diabetes Endocrinology Research Center Core Radioimmunoassay Laboratory) after storage at −70 °C.20 For the present study we used fasting insulin in both non-diabetics and type II diabetics.21 Because of large samples, epidemiologic studies often rely on single measurement of fasting insulin as this is considered a good marker of insulin resistance.21,22⇓
To estimate 2-hour post-load glucose a 75 g glucose drink was administrated to subjects (n = 1,742) unaware of their diabetes status, with no gastrectomy, or active peptic ulcer or stomach cancer. Fasting and 2-hour glucose levels were measured by the glucose oxidase method (University of Vermont).23 Type II diabetes was defined according to World Health Organization criteria24 and included individuals with previously diagnosed type II diabetes, those taking oral hypoglycemic agents or insulin, and those with a fasting blood glucose ≥7.0 mmol/L (126 mg/dL) or with a 2-hour post-load glucose ≥11.1 mmol/L (200 mg/dL). Individuals with a fasting glucose between 6.1 (110 mg/dL) and 7.0 mmol/L or 2-hour post-load glucose levels between 7.8 (140 mg/dL) and 11.1 mmol/L were classified as glucose intolerant and included in the nondiabetic group (n = 323).
Confounding variables and covariates.
Demographic and health related information was collected at each examination for the entire follow-up of the study. In these analyses, we controlled for potentially confounding demographic factors, such as age and education, and cardiovascular risk factors. Smoking behavior, alcohol intake, blood pressure, and cholesterol were measured at mid-life, from examinations 1 to 3, when their values were less influenced by a possible predementia status. Smoking was categorized by status (never, former, current) and by pack/years of cigarette exposure (packs of cigarettes a year × years of smoking). Alcohol consumption was recorded as grams of alcohol per day (g/day) and recoded as drinks per day (none, <1 drink [13.2 g], 1 to 2 drinks, and ≥3 drinks/day). Systolic and diastolic blood pressures were the means of the measures taken at each examination; within an examination, three measures were made 5 minutes apart and then averaged. History of antihypertensive treatment was self-reported from examinations 1 to 3 and obtained from the drug vials presented at examination 4. Mid-life total cholesterol level was measured at examinations 1 and 3 and averaged; values were determined with Autoanalyzer 1 N24B cholesterol method.12 Stroke and coronary heart disease (CHD) history were assessed at baseline in 1965 and, subsequently, via surveillance of hospital records that has been carried out through the entire follow-up period. Ankle-brachial index (ABI) was measured at the fourth examination and the values were dichotomized with a cutoff of 0.9; values below this point were interpreted as an indicator of generalized atherosclerosis.25 The 11-item version of the Center for Epidemiology Studies Depression Scale (CES-D) was used to measure depression symptoms during the previous week.26 Scores were transformed to correspond to the standard 20-item scale; subjects with a score >15 were considered depressed. Because insulin levels may decrease with weight loss, and dementia is associated with weight loss, we also controlled for percentage of weight change from middle to late life [(examination 4 weight − mid-life weight)/mid-life weight]. Finally, we controlled for apolipoprotein E allele (ApoE). ApoE genotyping was performed by PCR amplification and restriction enzyme digestion27 at the Bryan AD Research Center at Duke University (NC). Participants with one or two copies of the APOE ε4 allele were considered ε4 positive, and were considered ε4 negative otherwise. For all variables with missing data, a separate category within each variable was used so that the observations remained in the analysis.
Statistical analysis.
The 2,568 subjects who had at least one follow-up examination for dementia status and complete data on fasting insulin constituted the study population for the analysis.
The distribution of fasting insulin was highly skewed. To compare the age and education adjusted insulin levels between demented and nondemented subjects while accounting for this non-normality we fit a generalized linear model (GLM) with a log-link and a gamma distribution.28 This GLM model provides easily interpreted results because the original scale of the variable— in this case insulin—is preserved.
For subsequent analyses, fasting insulin was log-transformed to improve the approximation to normality. The mean log insulin level was 2.56 (12.93 mIU/L) ± 0.57 (1.77 mIU/L). For the analysis, the study sample was divided in three groups based on the insulin levels—low (insulin <7.20 mIU/L), medium (insulin between 7.20 mIU/L and 23.0 mIU/L), and high (insulin >23.0 mIU/L)—so that the cutoffs of the groups were the 15th and the 85th percentiles of the insulin distribution, corresponding to one SD from the mean. Characteristics of the three groups of participants were compared using age-adjusted logistic regression models for binary outcomes and age-adjusted general linear models for continuous dependent variables. To test the trend across the three insulin groups we assigned them the values of 1, 2, and 3 and included them in the model as continuous variables.
Visual inspection.
Because the literature suggests both a positive and negative effect of insulin levels in the brain, we visually inspected the data to look for nonlinear relationships. We divided the insulin range in small strata and evaluated the frequency of dementia by stratum (figure). A smoothed curve describing the probability of dementia over the distribution of insulin levels was calculated using the lowess procedure (locally weighted scatter plot smoothing).29 The lowess is a non-parametric regression method that fits a weighted least-squares curve. The smoothed values for incident dementia are obtained as the log of the odds (lod) and then transformed to the probability of dementia (probability = 1/1+ elod). On the figure the smoothed values for dementia probability are represented by filled circles.
Figure. The distribution of dementia cases by insulin level.
Multivariate analysis.
We examined the association of fasting insulin level to dementia with a Cox-proportional-hazard regression model using age as the time scale30 and with the nonparametric log-rank test for equality of survivor functions. Age at dementia onset was assigned at the midpoint date between the last examination without dementia and the first in which the subject was diagnosed with dementia. Subjects who died or did not return to the next follow-up were censored as of the time of the last evaluation. The group with a middle level of insulin was the reference category. To confirm the findings based on the three strata, we used a spline function and treated log-insulin as continuous variable.31 The spline approach identifies points in the distribution where the relationship between the independent and dependent variables significantly changes. Based on visual inspection of the lowess plot we identified the cut-off point. This cut-off was located at 3.1 logarithm units; the exact point selection was based on the likelihood test results and two separate continuous variables were included in the model. With this approach we estimated separate hazard ratios of dementia for log-insulin values <3.1 and log-insulin values ≥3.1 (22.20 mIU/L).
The survival analysis was adjusted for age, education, fasting glucose level, and ApoE ε4 status (model 1). Model 2 included the same variables as model 1 plus mid-life blood pressure, cholesterol level, smoking status, alcohol intake, percentage of weight change from mid-life to late-life, late-life depression, ABI, history of type II diabetes, stroke, and coronary artery disease.
Because high levels of insulin are prevalent in type II diabetics, we tested whether the association of insulin to dementia differed by diabetes status. The interaction term between insulin both as continuous and as categorical (low, medium, and high) variable and diabetes was not significant so the analyses were not stratified by diabetes status; instead, an indicator variable for diabetes was included in both models as confounder.
Subjects with cognitive impairment were included in the nondemented group. Subjects diagnosed with VaD and other types of dementia were excluded from the analysis of the association between fasting insulin and incidence of AD; similarly, AD subjects were excluded from the analysis of the relationship between fasting insulin and VaD.
The statistical analysis was performed using STATA 7 statistical software (STATA Corporation, College Station, TX).
Results.
Subjects included in the analysis were followed for an average of 5.1 years (±1.6 years). During this time, 244 new cases of dementia developed. Mean age at the follow-up entry and exit was 77.4 and 82.5 years.
Compared to the medium and high groups, lower insulin levels were associated with older age, and lower prevalence of glucose dysregulation, cardiovascular-related risk factors and a greater reduction of weight from mid-life (table 1). Compared to the other two groups, subjects with high insulin levels had a poorer cardiovascular risk profile.
Table 1 Characteristics of the Honolulu-Asia Aging Study cohort by fasting insulin levels
The age and education-adjusted mean level of insulin was not different between all demented and nondemented participants (insulin level = 17.93 mIU/L for incident demented and 15.81 mIU/L for nondemented, p = 0.10) and between participants with AD and those without any dementia (insulin level = 17.11 mIU/L for AD compared to nondemented participants, p = 0.32). A difference was found between participants with VaD and those without any dementia (insulin level = 24.13 mIU/L for VaD, compared to nondemented participants, p = 0.02). The distribution of dementia cases by insulin level is graphed in the figure. The log-rank test comparing the survival functions among low, medium, and high groups indicated a significant difference among survival curves (χ2 p value < 0.005). Compared to those with a medium level of insulin, the risk for dementia was higher in those in the low and high groups (table 2). Comparable results were observed for AD and VaD. Increasing values of insulin levels up to 22.2 mIU/L (3.1 log-insulin units) were inversely associated with the risk of total dementia (hazard ratio [HR] = 0.76, 95% CI = 0.72 to 0.79 for an increment of one insulin logarithmic unit) (table 3); higher levels of fasting insulin were directly associated with dementia (HR = 1.64, CI = 1.07 to 2.52 associated with each insulin logarithmic unit). A similar trend of association was observed for AD and VaD. If we excluded the incident cases from the first follow-up, which was closer to the measurement of insulin level, low levels of insulin (≤3.1 log-insulin units) were still inversely related to the risk for dementia (HR = 0.73, CI = 0.67 to 0.79) and high insulin levels were marginally associated with an increased risk for dementia (HR = 1.91, CI = 0.99 to 3.77).
Table 2 Cox-proportional hazard model for the risk of dementia and dementia subtypes by fasting insulin groups: the Honolulu-Asia Aging Study
Table 3 Cox proportional hazard model for the risk of dementia and dementia subtypes by logarithm levels of fasting insulin: the Honolulu-Asia Aging Study
Discussion.
This is the first study that evaluates prospectively the potential role of insulin as a risk factor for the development of dementia in a community-based sample. We found a nonlinear association such that high and low levels of insulin increased the risk for dementia. The associations had a similar trend for AD and VaD.
Over 50% of the men in the high insulin group were diabetic. The analysis was adjusted for fasting glucose and diabetes suggesting that the association reflects an independent effect of high levels of insulin on the brain. In this same group the results did not change after removing the cases that were diagnosed within 2 years from the insulin measure. Further, the association between low levels of insulin and an increased risk for dementia could reflect the possibility that low insulin is a marker of preclinical dementia. We examined this possibility using two different approaches. We adjusted for weight loss, a possible marker for prodromal dementia, between the examinations prior to and concurrent with the insulin measure.32 Further, we performed a second analysis only on cases that occurred at least 2.5 years after the insulin measurements. The results showed that low levels of insulin were still inversely related to the risk for dementia.
The findings of a nonlinear relationship are consistent with data from experimental and clinical studies. CNS insulin is derived mainly from the peripheral circulation.33 In normal conditions insulin passage into the CNS is regulated by a specialized saturable transport system localized in the blood–brain barrier endothelium.34 The system regulates insulin passage into the brain proportionally to the plasma insulin level. Under physiologic conditions, a higher serum insulin level may reflect a higher level of brain insulin and hypoinsulinemia could reflect an insufficient insulinization of the brain. However, the transport of insulin into the CNS is altered in hyperinsulinemic conditions.35 Hyperinsulinemia usually corresponds to a higher level of insulin resistance. Insulin resistance could result in a modification of insulin transport into the brain and a reduction of insulin sensitivity at the cellular level. Thus, it is possible that both hypoinsulinemia and hyperinsulinemia lead to insufficient insulinization in the brain, including those areas associated with cognition. A similar hypothesis was suggested in the Bruneck study36 to explain their finding of a U-shaped relationship between fasting serum insulin levels and coronary heart disease. The authors suggested that reduced insulinization might affect the function of specific cell types involved in the atherogenetic process.
Several studies have shown that the insulin degrading enzyme (IDE) is an important regulator of the amyloid-beta protein (Aβ),37 as well as of insulin. High levels of insulin in the CNS may reflect IDE activity and contribute to increased levels of Aβ42 in the cerebral spinal fluid38 and to the accumulations of Aβ in the plaques.
Our results of an inverse association between insulin and dementia are consistent with the insulin attributed neurotropic role in the brain. Rise of insulin level in the brain is followed shortly by improved cognitive function, particularly memory performance.9,10,39⇓⇓ The insulin receptor is widely expressed in several areas of the brain, including choroid plexus, olfactory bulb, pyriform cortex, amygdaloid nucleus, dentate gyrus, hypothalamic nucleus, and the cerebellar cortex. It is also found in the hippocampus including the CA1, CA3 regions, which are involved in spatial memory and learning processes.6 Insulin increases glucose utilization and regulates its metabolism in the glia and neurons, essential functions for cellular survival and cerebral activity.40 Insulin can exert a variety of actions in the brain independent from glucose. At the cellular level, insulin has a direct effect on brain neuronal function by activating several tyrosine kinases and mitogen-activated protein kinases,41 which appear to be involved in learning and memory processes.42 Insulin also modulates norepinephrine and dopamine metabolism,43 and inhibits γ-aminobutyric acid uptake44 indicating a critical neuromodulatory action in the CNS. Normal insulin levels may also reduce the neuropathologies associated with AD, by decreasing the β-amyloid toxicity45 and tau phosphorylation.46
Although the evidence that insulin may have neuroprotective properties is recent, it has long been known that hyperinsulinemia adversely affects the vascular system and is associated with cardiovascular disease.1 Hyperinsulinemia is characterized by prediabetic status and may partly explain the increased risk of AD in type II diabetics. Possibly, peripheral and cerebral pathologic abnormalities related to hyperinsulinemia status limit or eliminate the positive effect of insulin on the brain.47 A recent study showed that insulin treatment could prevent but not reverse the learning deficits associated with streptozotocin-induced diabetes status.48
Although the frequency of AD in this cohort is similar to that reported in white samples from the United States and Europe, the frequencies of VaD as well as glucose impairment and type II diabetes are higher, as is the case in other Japanese samples.49 Cultural, environmental, and genetic factors may be different in the Japanese-American male population. Additional studies are needed to clarify the role of insulin in the development of dementia especially among older persons in other ethnic groups and in women.
Future studies should investigate this relationship with dementia, with a focus on better measurement of insulin resistance and disturbance of insulin signaling in the brain.
Acknowledgments
Supported by the National Institute of Aging contract N01-AG-4-2149, and National Heart, Lung and Blood Institute contract N01-HC-O5102.
The authors thank Dr. Qian-Li Xue for statistical advice.
- Received January 30, 2003.
- Accepted March 22, 2004.
References
Disputes & Debates: Rapid online correspondence
NOTE: All authors' disclosures must be entered and current in our database before comments can be posted. Enter and update disclosures at http://submit.neurology.org. Exception: replies to comments concerning an article you originally authored do not require updated disclosures.
- Stay timely. Submit only on articles published within the last 8 weeks.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- 200 words maximum.
- 5 references maximum. Reference 1 must be the article on which you are commenting.
- 5 authors maximum. Exception: replies can include all original authors of the article.
- Submitted comments are subject to editing and editor review prior to posting.