Association of diabetes duration (x axis) with odds of cognitive impairment at baseline (y axis, figures in left column) and cognitive decline (y axis, figures in right column) over 4 to 6 years on each of 3 cognitive tests (Digit Symbol test, Trails B test, and the modified Mini-Mental State Examination [m-MMSE]). Vertical lines represent 95% confidence intervals surrounding odds ratios, which are controlled for age, depression, visual impairment, stroke, and education, hypertension, heart disease, estrogen use, walking for exercise, and smoking. DM indicates diabetes mellitus.
Gregg EW, Yaffe K, Cauley JA, Rolka DB, Blackwell TL, Narayan KMV, Cummings SR, for the Study of Osteoporotic Fractures Research Group. Is Diabetes Associated With Cognitive Impairment and Cognitive Decline Among Older Women?. Arch Intern Med. 2000;160(2):174–180. doi:10.1001/archinte.160.2.174
The long-term effect of type 2 diabetes on cognitive function is uncertain.
To determine whether older women with diabetes have an increased risk of cognitive impairment and cognitive decline.
Prospective cohort study.
Four research centers in the United States (Baltimore, Md; Portland, Ore; Minneapolis, Minn; and the Monongahela Valley, Pennsylvania).
Community-dwelling white women 65 years and older (n = 9679).
Physician-diagnosed diabetes and other aspects of health history were assessed by interview. Three tests of cognitive function, the Digit Symbol test, the Trails B test, and a modified version of the Mini-Mental State Examination (m-MMSE), were administered at baseline and 3 to 6 years later. Change in cognitive function was defined by the change in the score for each test. Major cognitive decline was defined as the worst 10th percentile change in the score for each test.
Women with diabetes (n = 682 [7.0%]) had lower baseline scores than those without diabetes on all 3 tests of cognitive function (Digit Symbol and Trials B tests, P<.01; m-MMSE, P = .03) and experienced an accelerated cognitive decline as measured by the Digit Symbol test (P<.01) and m-MMSE (P = .03). Diabetes was also associated with increased odds of major cognitive decline as determined by scores on the Digit Symbol (odds ratio = 1.63; 95% confidence interval, 1.20-2.23) and Trails B (odds ratio, 1.74; 95% confidence interval, 1.27-2.39) tests when controlled for age, education, depression, stroke, visual impairment, heart disease, hypertension, physical activity, estrogen use, and smoking. Women who had diabetes for more than 15 years had a 57% to 114% greater risk of major cognitive decline than women without diabetes.
Diabetes is associated with lower levels of cognitive function and greater cognitive decline among older women.
OVER 33% of women and 16% of men 65 years and older will develop dementia during their lives, with many cases leading to functional decline, comorbid conditions, and the extensive need for health services.1- 3 Although the etiology of dementia is complex and some cross-sectional studies suggest that diabetes may lead to cognitive decline caused by hypoglycemia, hyperglycemia, hyperinsulinemia, or vasculopathy,4- 13 the results from the few prospective studies examining diabetes and change in cognitive function or incidence of dementia have been inconsistent.14- 18 Additionally, no large, population-based, prospective studies have evaluated the relationship between diabetes and change in cognitive function using repeated neuropsychologic tests over time. Similarly, few studies have addressed the relationship between diabetes duration and long-term cognitive decline. Because diabetes affects up to 18% of women and 20% of men 65 years and older,19 a significant relationship between diabetes and cognitive impairment would have important clinical and public health implications. We examined the relationship of diabetes and diabetes duration to cognitive decline during up to a 6-year follow-up in a prospective study of more than 9000 older women.
The study population consisted of 9679 community-dwelling white women aged 65 to 99 years (mean ± SD, 71.7 ± 5.3 years) participating in the Study of Osteoporotic Fractures (SOF), a prospective study of risk factors for fractures and physical and cognitive decline among older women.20- 22 Women were recruited between 1986 and 1988 from population-based lists in Baltimore, Md; Minneapolis, Minn; Portland, Ore; and the Monongahela Valley, Pennsylvania (near Pittsburgh). Women were excluded if they were institutionalized or unable to walk without the assistance of another person. Questionnaires, interviews, and examinations were conducted at baseline, and follow-up visits took place over the subsequent 3 to 6 years.
Participants were asked whether a physician had ever told them that they had diabetes and, if so, their age at diagnosis. We assessed current medication usage, including insulin, hormone replacement therapy, and thiazide diuretics, and histories of cardiovascular events (eg, myocardial infarction, congestive heart failure, and angina), stroke, and arthritis by questionnaire and interview. Women were asked to rate their overall health compared with others on a 5-point scale ranging from excellent to very poor. Walking for exercise, smoking, and years of education were assessed by interview.21 Hypertension was defined as receiving thiazide diuretic medications or having a blood pressure above 160/90 mm Hg. Depression was assessed using a shortened Geriatric Depression Scale, with a score of 6 or higher used as a cutpoint for depression.23 Visual impairment was considered to be corrected visual acuity of 20/40 or worse as assessed by Bailey-Lovie charts.24 All risk factors were assessed at the baseline study visit except depression and heart disease, which were initiated at the second study visit. Stroke, visual impairment, and heart disease were also assessed at the follow-up visit.
Examiners were trained to assess cognitive function using 3 tests.22 A modified version of the Mini-Mental State Examination (m-MMSE), in which questions assessing orientation were dropped, was administered during the baseline study visit to 9627 women and then repeated 6 years later for 6675 women.25,26 The m-MMSE is a brief, general cognitive function test with concentration, language, and memory components designed to screen for cognitive impairment. The m-MMSE scale ranges from 0 to 26, with higher numbers indicating better performance.22 The Digit Symbol test, a timed test that measures attention, psychomotor performance, and perceptual organization, was initiated at the second clinical visit (2 years into the study) for 7425 women and repeated an average of 3.6 years later for 6613 women; its score reflects the number correct within the timed trial, and higher numbers indicate better performance.27 The Trails B test, a timed test of attention, visual scanning and sequential abilities, and executive function, was also initiated at the second clinical visit for 7909 women and an average of 3.6 years later for 6023 women; scores are in seconds, with higher scores indicating poorer performance.28 Among survivors who had baseline measures, 80%, 83%, and 76% had a follow-up m-MMSE, Digit Symbol test, and Trails B test, respectively. The shortened Geriatric Depression Scale also includes 2 questions on cognitive function, the first asking whether the participant thought she had more memory problems than most people her age and the second asking whether she had poor memory during the previous week. All examiners were certified by the study coordinating center, and cross-clinic visits were conducted to assure adherence to protocol.
The χ2 and t tests were used to compare potential confounders, such as health status, health behaviors, education, and visual impairment by diabetes status. Repeated-measures regression analyses (Proc MIXED; SAS Institute, Cary, NC) were employed to simultaneously evaluate cross-sectional and longitudinal associations between diabetes status and performance on the Digit Symbol and Trails B tests while controlling for age, education (less than high school, high school graduate, some college, or college graduate), visual impairment, depression, history of stroke, hypertension, heart disease, walking for exercise, estrogen use (current or never/past), alcohol use (drinks per week), smoking (current or never/past), and self-rated health (5 levels ranging from excellent to poor). All regression models included age, education, visual acuity, and depression. Comorbidity variables (stroke, hypertension, heart disease) and behavioral confounders (physical activity, estrogen use, alcohol use, and smoking status) were entered into regression models in separate steps, with best-fitting models determined for each outcome using backward elimination. Depression, stroke, and heart disease, which were assessed at both baseline and follow-up, were treated as time-dependent covariates in these models. For this analysis, the Trails B test was transformed to its logarithmic form, which resulted in a more normal distribution. Because m-MMSE cross-sectional distributions were nonnormal, nonparametric Wilcoxon tests were used to compare baseline and follow-up scores on this test between women with and without diabetes. Multiple linear regression was used to compare changes from baseline to follow-up on the m-MMSE between women with and without diabetes, controlling for baseline level and other covariates.
In addition to comparing scores on the 3 cognitive tests, we also examined cognitive impairment (defined as the lowest 10th percentile) and major cognitive decline (defined as the greatest 10th percentile reduction in performance from initial to follow-up score).22 This corresponded to an 11-point or greater decrease in scores on the Digit Symbol test, an 81-point or greater increase in scores on the Trails B test, and a 3-point or greater decrease in scores on the m-MMSE. This approach has been used in other studies, and these cutpoints are similar to those used in definitions of clinically important change selected by a Delphi panel.29,30 We used multiple logistic regression to determine the odds ratios (ORs) and 95% confidence intervals (CIs) of major cognitive decline associated with diabetes, controlling for baseline level and potential confounders. The variable selection approach used for logistic regression was similar to that used for repeated-measures regression.
To examine the relationship between diabetes duration and cognitive impairment and major cognitive decline, we grouped women with diabetes according to duration of diabetes (<5 [n = 240], 5-14 [n = 259], and >15 [n = 179] years). We conducted multiple logistic regression and evaluated both tests for trend and odds of cognitive impairment and major cognitive decline across diabetes duration groups, with nondiabetic women serving as the reference group. Several plausible interactions with diabetes were also evaluated in the regression models, including age, education, hypertension, depression, and cardiovascular disease.
Diabetes was reported by 682 women (7.0%), with a mean ± SD duration of disease of 10.2 ± 9.5 years (Table 1). Women with diabetes had fewer years of school, had higher scores on the depression inventory, were more likely to have visual impairment and hypertension, and were less likely to take estrogen, drink alcohol, or take walks for exercise. Women with diabetes were also more likely to report cardiovascular disease, stroke, and fair or poor health status.
Women with diabetes had poorer scores on all 3 tests at both baseline and follow-up (Digit Symbol and Trails B tests, P<.001; m-MMSE, P = .03) (Table 2). Having diabetes was also associated with significantly greater decreases over time on Digit Symbol test (P = .01) and m-MMSE (P = .03) scores, and a decline in scores on the Trails B test that approached significance (P = .06) (Table 2). These differences remained significant after controlling for age, education, depression, stroke, hypertension, visual impairment, heart disease, estrogen use, smoking, alcohol use, walking, and self-rated health. When impairment was defined as performance in the lowest 10th percentile, women with diabetes had significantly increased odds of baseline cognitive impairment as measured by the Digit Symbol (104%) and Trails B (58%) tests, but not as measured by the m-MMSE (Table 3). Defin ing major cognitive decline as the greatest 10th percentile decline, women with diabetes also had a 63% (95% CI, 20%-123%) greater risk of major decline as assessed by the Digit Symbol test and 74% (95% CI, 27%-139%) greater risk of decline as assessed by the Trails B test, after controlling for potential confounders and baseline score (Table 3). Diabetes was not associated with the risk of major decline according to m-MMSE assessments.
Compared with women without diabetes, diabetic women reporting insulin use (n = 113) had an appreciably higher risk of major cognitive decline as determined by the Trails B test (OR, 2.65; 95% CI, 1.26-5.54) than those who were not taking insulin (OR, 1.62; 95% CI, 1.15-2.28) when controlling for confounders. Similarly, insulin use was associated with a somewhat greater risk of major cognitive decline according to scores on the Digit Symbol test (OR, 1.86; 95% CI, 0.88-3.95 [insulin users]; OR, 1.60; 95% CI, 1.14-2.24 [non–insulin users]) compared with women without diabetes. No significant associations with cognitive decline according to scores on the m-MMSE were observed among either insulin-using or non–insulin-using women with diabetes.
Compared with women without diabetes, those with diabetes were also more likely to report that they had more memory problems in general (18.3% among diabetic women vs 11.6% among nondiabetic women; adjusted OR, 1.38; 95% CI, 1.05-1.82]) and during the last week (26.5% among diabetic women vs 21.2% among nondiabetic women; adjusted OR, 1.17; 95% CI, 0.75-2.00).
There were no statistically significant interactions of diabetes with age, education, depression, hypertension, or cardiovascular disease. However, separate analyses conducted after stratification according to age revealed a trend wherein the odds of major cognitive decline as identified by scores on the Trails B test were greater among women below the median (<71 years; OR, 2.39; 95% CI, 1.59-3.59) than above the median (≥71 years; OR, 1.19; 95% CI, 0.73-1.95) age. However, stratification according to age and education did not yield appreciable differences in the relationships between diabetes and cognitive decline on any of the other outcomes.
The odds of cognitive impairment and major cognitive decline increased with the duration of diabetes as assessed by scores on the Digit Symbol and Trails B tests but not as assessed by scores on the m-MMSE (P for trend <.01 for each [Figure 1]). Women with diabetes for less than 5 years were not at significantly increased risk of baseline impairment compared with those without diabetes. They had significantly greater cognitive decline as determined by scores on the Trails B test, but not as determined by scores on the Digit Symbol test or m-MMSE. Women with a 5- to 14-year history of diabetes were more likely to have both cognitive impairment (Digit Symbol test, P<.01; Trials B test, P = .04) and cognitive decline (Digit Symbol test, P = .02; Trials B test, P = .07) than those without diabetes. The risk of both cognitive impairment and decline was the greatest for women with diabetes duration of at least 15 years, who had a 1.4- to 3.2-fold increased risk of baseline impairment and a 1.6- to 2.1-fold increased risk of cognitive decline when controlling for potential confounders (Figure 1).
This large prospective study indicates that older women with self-reported diabetes have poorer cognitive functioning and a more rapid decline in cognitive performance than other women. Our findings of up to a 2-fold increased risk of cognitive impairment and 74% increased risk of cognitive decline as measured by repeated neuropsychologic tests are consistent with 2 longitudinal studies that observed up to a 60% increased risk of clinically ascertained dementia among people with diabetes.16,17 Two other longitudinal studies, however, have not found an association between diabetes and cognitive decline.15,18 One of these studies18 observed nonsignificantly higher rates of decline among people with diabetes, but it had a much smaller sample size and thus less statistical power to observe significant effects. To our knowledge, this is the first large population-based study to evaluate the relationship of diabetes to change in cognitive function using repeated tests of cognitive function.
Our study also differed from previous reports by assessing the relationship of diabetes duration to cognitive function. We found that duration was related to diminished cognition in a dose-response manner that was independent of age and comorbid conditions, such that women with diabetes for 15 years or longer had up to a 3-fold increased risk of baseline cognitive impairment and more than a doubling of risk for major cognitive decline. These observations, when combined with longitudinal associations between diabetes and cognitive decline as assessed with neuropsychologic tests, suggest a causal relationship between diabetes and cognitive function.
While this study did not attempt to identify the mechanism of an effect of diabetes on cognitive function, multiple regression analyses suggested that our findings are not explained by overt stroke or cardiovascular disease, hypertension, or depression. In addition, we controlled for potential behavioral confounders, such as smoking, alcohol use, walking for exercise, and estrogen use, and confounders of the testing process, including visual impairment and education. Previous cross-sectional studies have related hyperglycemia, hyperinsulinemia, and insulin resistance to cognitive function.5,10,12,14,31- 35 Additionally, one uncontrolled intervention found improved cognitive function and one randomized controlled trial found improved quality of life and perceived cognitive function following stricter glycemic control, indicating that acute variation in glycemia may influence cognitive function.35,36 Taken as a whole, the effects of diabetes on cognitive function appear to be caused in part by factors intrinsic to diabetes, such as glycemia and insulin, and may be further compounded by diabetes-related complications, such as stroke. Hyperglycemia and hyperinsulinemia may influence cognitive function directly or through microvascular effects and subclinical vascular disease.4,37 The latter pathways are also consistent with epidemiologic studies relating diabetes particularly to vascular dementia and possibly Alzheimer disease.8,16- 18
In contrast to most previous investigations, our study repeated neuropsychologic tests over several years in a large community-based sample. Because the Digit Symbol and Trails B tests are more sensitive to cognitive decline than the m-MMSE, which was designed to screen for dementia rather than measure cognitive decline, our study was more likely to detect preclinical cognitive decline and early dementia than previous studies.25,38,39 Thus, it is possible that ceiling effects of the m-MMSE and reduced sensitivity explain why we observed more consistent effects of diabetes as determined by scores on the Digit Symbol and Trails B tests than by scores on the m-MMSE. We also evaluated the risk of experiencing the greatest decline (10%), a level of decline that is likely to represent clinically recognizable decline.29,30 Cognitive impairment represents a broad spectrum, ranging from preclinical and mild impairment to more profound dementia and severe Alzheimer disease. Our results suggest that diabetes is related to the increased risk of cognitive impairment as well as to more profound cognitive decline among a large population of community-dwelling women.
Our study was limited by the fact that diabetes was based on self-report. Accuracy of age of diagnosis could also be affected by cognitive function. However, previous studies have suggested that self-reported diabetes has strong reliability, at least as a measure of physician-diagnosed diabetes, with κ coefficients of agreement between self-report and medical record review typically ranging from 85% to 95%.40- 42 Nevertheless, these limitations could have biased our results if women with undiagnosed diabetes had different patterns of cognitive decline than those without diabetes or those with self-reported diabetes.
There was a 22%, 28%, and 31% rate of attrition between initial and follow-up tests for the Digit Symbol and Trails B tests and the m-MMSE, respectively, as a result of mortality or the inability to attend the visit. Women with diabetes were more likely to not attend follow-up visits (OR, 1.26; 95% CI, 1.01-1.56) and to have died during the follow-up period (relative risk [RR], 2.12; 95% CI, 1.84-2.45). Similarly, women with poor cognitive function were more likely to not attend follow-up visits (OR, 1.50; 95% CI, 1.34-1.68) and to have died during the follow-up period (RR, 1.30; 95% CI, 1.17-1.46). However, since diabetes was related to baseline cognitive impairment, it is more likely that misclassification of diabetes led to an underestimation of effects on major cognitive decline, and our findings are probably conservative. We also lacked assessment of oral hypoglycemic medications, glycemic control, diabetic neuropathy and nephropathy, and had limited power to assess the influence of insulin on cognitive function. Our study was also limited to white women, so it is also unclear whether our findings can be generalized to other groups.
Despite these limitations, this was the largest population-based study of older women evaluating the relationship of diabetes and disease duration to cognitive function using repeated clinical neuropsychologic tests. Diabetes prevalence appears to be increasing both in the United States and worldwide, largely as a result of aging and lifestyle changes.43,44 Similarly, the overall prevalence of cognitive impairment is increasing and will continue to do so in future decades.45,46 In light of these parallel trends, our finding of an increased risk of cognitive decline associated with diabetes has important clinical, public health, and societal ramifications. Cognitive impairment may be considered a potential long-term outcome of diabetes that clinicians should be aware of while taking care of older adults with diabetes. Similarly, it is possible that optimal glycemic treatment of older adults with diabetes may reduce the risk of cognitive impairment or perhaps improve cognitive function,35,36 but more research is needed to understand how the treatment of people with diabetes influences cognitive function. Ultimately, prevention of cognitive impairment may be enhanced if future research determines the underlying mechanism for the association between diabetes and cognitive decline.
Accepted for publication April 26, 1999.
This study was supported in part by grants AR35582, AR35583, AM35584, AG05395, and AG05407 from the Public Health Service, National Institutes of Health, Bethesda, Md.
Corresponding author: Edward W. Gregg, PhD, Division of Diabetes Translation, Centers for Disease Control and Prevention, 4770 Buford Hwy NE, Mailstop K-68, Atlanta, GA 30341.