Hyperinsulinemia and Cognitive Decline in a Middle-Aged Cohort

The ARIC cohort is a large, multiethnic, multisite longitudinal observational study of risk factors for vascular diseases that was initiated in 1987. This study is based on the publicly available ARIC dataset. Initially, 15,732 men and women aged 45–64 years were recruited from area sampling of four locations using population-wide lists for probability sampling (23). Four U.S. communities were sampled; three of the communities in the cohort were probability sampled. The fourth community (Jackson, MS) only sampled nonwhites.

The initial visit, occurring between 1987 and 1989, included comprehensive clinical and laboratory examination including insulin level in 15,027 subjects who fasted ≥8 h. Subsequent examinations of the cohort have been every 3 years. At the second visit, occurring between 1990 and 1992, individuals underwent an initial cognitive evaluation. At the fourth visit, 6 years after initial cognitive evaluation, subjects had repeat cognitive evaluation.

For most of the current analysis, 451 subjects were excluded because they gave a self-reported history of stroke or transient ischemic attack at visit 1, and 936 subjects were excluded because they had type 2 diabetes defined by fasting glucose of >126 mg/dl, nonfasting glucose of >200 mg/dl, a self-reported history of diabetes, or treatment for diabetes at visit 1. Ninety-three subjects with dementia, based on a delayed word recall (DWR) test score <3 when assessed at visit 2, were censored from analysis. A total of 1,409 individuals were excluded from further analysis based on dementia, stroke/transient ischemic attack, or type 2 diabetes. The 7,148 ARIC cohort subjects included in this analysis had information on fasting insulin and control variables at the first visit, control variables at the second visit, and cognitive test scores at the second and fourth visits. The full interview, laboratory, and examination methodology can be found on the ARIC Web site (http://www.cscc.unc.edu/aric/pubuse/).

Cognitive testing in the ARIC included serial assessments of the DWR test, the digit symbol subtest (DSS) of the Wechsler Adult Intelligence Scale-Revised, and the first letter word fluency (WF) test. The DWR is a test of verbal learning and recent memory in which the participant learns 10 common nouns by constructing sentences incorporating the words and then following a 5-min interval attempts to recall the nouns (24). With a cutoff of <3 for dementia and >3 for normal, the DWR test has an overall predictive accuracy of 95.2% for possible or probable Alzheimer’s dementia and is not correlated with education or age in normal subjects (25). The DSS score is the number of correctly drawn symbols completed within the 90-s time limit. The DSS is considered a more sensitive measure of dementia than the often-used Mini-Mental State Exam. Components of cognitive functioning and psychomotor performance evaluated by the DSS include response speed, sustained attention, visual spatial skills, associative learning, and memory (24). The WF test measures mental agility in retrieving words starting with a particular letter; the score is based on the total number of words given by the participant. The WF test is particularly sensitive to damage to the frontal lobes of the brain (24).

Hyperinsulinemia was defined by two clinical and two empirical measures of insulin resistance. Fasting insulin and homeostasis model assessment (HOMA) have been used to predict insulin resistance. HOMA is a measure of insulin sensitivity that compensates for potential fasting hyperglycemia; HOMA is calculated by dividing the product of fasting concentrations of glucose and insulin by a constant. The clinically relevant cutoffs defining hyperinsulinemia used in this current study are fasting insulin >12.2 mU/l and HOMA >2.6 (26). It is unclear what level of insulin resistance is associated with risk of cognitive decline or dementia; thus, we empirically defined hyperinsulinemia at the 75th percentile based on all subjects in the ARIC cohort who had fasting insulin measured at the initial visit. The 75th percentile was 14.0 mU/l for fasting insulin and 3.9 for HOMA.

In an effort to determine the independent relationship between hyperinsulinemia and results of a test of cognitive function, control variables included age, sex, education level, marital status, smoking status, alcohol use, depression score, history of hypertension, and history of hyperlipidemia to control for factors that are known to have an impact on cognitive decline and dementia. Depressive symptom items from the Maastricht Questionnaire at visit 2 were used to determine the depression score (27). Hypertension was considered present if systolic blood pressure was ≥140 mmHg, if diastolic blood pressure was ≥90 mmHg, or if antihypertensive medications were being used at baseline. Hyperlipidemia was considered present if LDL was ≥140 mg/dl or if cholesterol-lowering agents were being used at baseline. To evaluate the potential effects of change in diabetes status and treatment status that may have occurred in subjects by the fourth visit, an additional control variable for later development of diabetes was defined by visit 4: fasting glucose ≥126 mg/dl, nonfasting glucose ≥200 mg/dl, self-reported history of diabetes, or treatment for diabetes at visit 4.

Descriptive statistics and Student’s t tests to compare mean baseline cognitive test scores and cognitive test change scores between those with and without hyperinsulinemia were computed. Change scores for cognitive tests were calculated as visit 4 minus visit 2 values for each subject. SAS statistical software (SAS Institute, Cary, NC) was used; results were considered statistically significant if P < 0.05.

Multiple variable linear regression models with the dependent variable of baseline cognitive test score or cognitive test change score were computed to evaluate associations with measures of hyperinsulinemia. Control variables, as described above, were used in all linear regressions to adjust for potential confounders. In these forced inclusion models, the dependent variable of either baseline cognitive test score or cognitive test change score was a continuous variable. Age, alcohol use, and depression score were treated as continuous variables in linear regressions. Additional multiple linear regression models included the control variables as described above as well as a control variable for later development of diabetes to assess the effects of change over time in diabetes status and treatment status.

The corresponding percentiles for the clinically informed cutoffs of fasting insulin >12.2 mU/l and HOMA >2.6 were 68.2 and 56.8, respectively, in all subjects in the ARIC cohort who had fasting insulin measured. Characteristics of ARIC participants without type 2 diabetes, dementia, or stroke/transient ischemic attack at baseline are presented in Table 1. DWR at baseline, available in the 7,148 individuals included in subsequent analyses, was 6.78 ± 1.43 (means ± SD). Minimum baseline DWR was 3, and maximum DWR was 10. The greatest decline in DWR score at 6-year follow-up was −7, and 5 was the greatest improvement. At follow-up cognitive testing, 72 participants had developed DWR scores <3. DWR change score, available in 7,008 individuals, was −0.153 ± 1.54. DSS at baseline, available in the 7,136 individuals included in subsequent analyses, was 46.9 ± 13.4. Minimum baseline DSS was 0, and maximum DSS was 93. The greatest decline in DSS score at the 6-year follow-up was −59, and 63 was the greatest improvement. DSS change score, available in 6,986 individuals, had a mean of −2.55 ± 6.94. WF at baseline, available in the 7,142 individuals included in subsequent analyses, was 34.6 ± 12.2. Minimum baseline WF was 0, and maximum WF was 86. The greatest decline in WF score at the 6-year follow-up was −44, and 66 was the greatest improvement. WF change score, available in 6,991 individuals, was −0.657 ± 7.91.

In unadjusted analyses, utilizing both clinical cutoffs and highest quartiles of both fasting insulin and HOMA, hyperinsulinemia at baseline was associated with significantly lower baseline DWR, DSS, and WF scores. DWR change scores for participants with hyperinsulinemia based on highest quartile fasting insulin, highest quartile HOMA, and HOMA with a cutoff of 2.6 had significantly greater decline than those without hyperinsulinemia (Table 2). WF change scores for participants with hyperinsulinemia based on highest quartile fasting insulin, fasting insulin with a cutoff of 12.2 mU/l, and highest quartile HOMA had significantly greater decline than those without hyperinsulinemia. DSS change scores were not significantly different between those with and without hyperinsulinemia. Variables for age, sex, race, and education level were statistically significant in most of the fully adjusted multiple linear regression models and explained a majority of the variance in cognitive test scores and change scores. R² values for the linear regression models of baseline cognitive function tests with all control variables, but before addition of any hyperinsulinemia variable, were 0.125 for DWR, 0.492 for DSS, and 0.224 for WF. Multiple variable linear regression models adjusted for potential confounders showed lower baseline DWR scores for those with hyperinsulinemia, although only in the model utilizing fasting insulin with 12.2 mU/l as the cutoff was the hyperinsulinemia variable statistically significant (Table 3). Similarly, adjusted models showed lower baseline DSS scores for those with hyperinsulinemia, although the hyperinsulinemia variable was statistically significant only when utilizing highest quartile fasting insulin or highest quartile HOMA. Adjusted multiple variable linear regression models showed lower baseline WF scores using both empirical and both clinical cutoffs for hyperinsulinemia, and all hyperinsulinemia variables were statistically significant in their models of baseline WF scores.

In adjusted multiple variable linear regression models, greater decline in DWR over 6 years was found in those with hyperinsulinemia, although only in the model utilizing the highest HOMA quartile as the cutoff was the hyperinsulinemia variable statistically significant. Adjusted models showed greater decline in WF change scores for those with hyperinsulinemia, although only in the model utilizing the highest fasting insulin quartile as the cutoff was the hyperinsulinemia variable statistically significant. In adjusted models for DSS change scores, none of the hyperinsulinemia variables were statistically significantly.

In terms of the strength of the relationship of hyperinsulinemia and cognitive tests in addition to the statistical significance previously shown, we conducted several additional analyses. The Spearman’s rank correlation coefficient, summarizing the strength of the relationship between hyperinsulinemia by highest quartile fasting insulin and cognitive testing, was r = 0.144 (P < 0.0001) for baseline DSS score, r = 0.070 (P < 0.0001) for baseline DWR, and r = 0.084 (P < 0.0001) for baseline WF. Following adjustment for control variables in linear regression analyses of cognitive tests, the relationships remained significant; however, there was an indication of substantial shared variance with some of the control variables. For example, the partial correlation between baseline DSS score and hyperinsulinemia by highest quartile fasting insulin, net of the influence of the other control variables, was significant but had dropped to r = 0.029 (P = 0.0154). With the addition of a variable controlling for later development of diabetes to the previously described multiple linear regression models exploring change in cognitive function scores, none of the previously described statistically significant hyperinsulinemia variables became nonsignificant. The variable controlling for later development of diabetes was not statistically significant in any of the additional models.

Cognitive decline and dementia are typically considered diseases of the elderly, although prevention of cognitive decline may require intercession earlier in midlife. We have described the relationship between hyperinsulinemia and change in cognitive function over time independent of diabetes over 6 years in a middle-aged cohort. This study provides evidence that decline in cognitive function in a middle-aged adult cohort without preexisting dementia, stroke/transient ischemic attack, or type 2 diabetes was greater in participants with higher measures of insulin resistance.

With such a large cohort, there is often more than sufficient power to detect significant differences between groups. More relevant to future interpretation is the minimum detectable difference between each group’s mean cognitive changes and whether these differences are clinically significant. These performance decrements were small and probably not clinically significant to participants but offer evidence for initiation of cognitive impairment and cognitive decline associated with insulin resistance in midlife. In considering clinical significance, a previous study (22) using the ARIC cohort showed that diabetic subjects did not have a significant decline in DWR compared with nondiabetic subjects, though diabetic subjects had a greater decline in two other tests of cognitive function. This current study does not address the mechanisms by which hyperinsulinemia causes impaired cognition and cognitive decline. Given our findings of significant differences in decline in DWR and WF between those with and without hyperinsulinemia (excluding individuals with type 2 diabetes and after controlling for later development of diabetes), insulin resistance may increase risk of dementia and progression of dementia by mechanisms discrete to insulin resistance independent of the pathogenic course of type 2 diabetes. Assessing what incremental predictive value novel risk markers add to existing models of risk has been a challenge even in research of coronary heart disease prevention (28,29). The utility of this strategy in dementia prevention, particularly with assessment of potentially modifiable risk factors to identify individuals at risk for dementia toward providing preventive action, is in its infancy, and further exploration of the predictive utility of novel risk factors that have modest though independent statistical associations with cognitive decline is warranted.

The strengths of this study include its large, population-based, multiethnic, longitudinal nature with repeat cognitive testing and its younger subject age than studies that have previously investigated longitudinal associations between hyperinsulinemia and dementia (14,15). Additional strengths are the use of two measures of insulin resistance, which we explored at both empirical and clinical cutoff levels.

Use of the ARIC cohort allowed for examination of the relationship between insulin and cognitive function in both men and women and allowed for assessment of potentially preclinical cognitive decline, unlike the study by Okereke et al. (15), which included only older women in the Nurses’ Health Study, and Peila et al. (14), which included only elderly Japanese-American men. A limitation of this study is that we were unable to directly link time to development of decline; instead, we were limited to using a 6-year time frame. While some decline may have happened earlier, this time frame for an intermediate indicator of dementia seems like a reasonable assessment.

The results of this study lead us to propose that insulin resistance is a potentially modifiable midlife risk factor for dementia. Hopefully, the results of this study will lead to assessments and interventions in middle age that can prevent or reduce the burden of dementia in the aging population. Middle-aged subjects without dementia, type 2 diabetes, or stroke at baseline who have hyperinsulinemia have a greater decline in performance on tests of cognitive function over a 6-year period compared with subjects without hyperinsulinemia.

This study was supported in part by grants 1D12HP00023 and 1D14HP00161 from the Health Resources and Services Administration and grant 1PS0AG021677 from the National Institute of Aging.