Change in Cognitive Function by Glucose Tolerance Status in Older Adults: A 4-Year Prospective Study of the Rancho Bernardo Study Cohort

Background  A few studies have examined change in cognitive performance by diabetes status with disparate results. We examined the 4-year change in cognitive performance among older adults according to glucose tolerance status.

Methods  Three cognitive tests (Mini-Mental State Examination, Verbal Fluency [VF] test, and Trail-Making Test B) were measured 4 years apart in 999 white men and women aged 42 to 89 years, who were enrolled in the Rancho Bernardo Study. Participants were classified with normal (NGT), impaired (IGT) or diabetic glucose tolerance. Sex-specific linear regression models adjusted for age, education, depression score, apolipoprotein E ϵ4 allele, and current estrogen use. We checked for mediation by further adjusting for total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, and triglyceride levels; blood pressure; glycohemoglobin level; and microalbuminuria, retinopathy, stroke, or coronary heart disease.

Results  At baseline, mean cognitive function scores did not differ between glucose tolerance groups. Women with diabetes mellitus had a 4-fold increased risk of a major cognitive decline on the VF test after 4 years compared with nondiabetic women. After multivariate adjustment, VF test scores at follow-up for women were 15.2 ± 0.6 for those with diabetes, 16.7 ± 0.4 for those with IGT, and 17.2 ± 0.2 for those with NGT (P = .007). Glycohemoglobin attenuated this effect, but lipid levels, blood pressure, and microvascular or macrovascular disease did not. Performance on Mini-Mental State Examination and Trail-Making Test B did not differ by baseline glucose status.

Conclusions  Elderly white women with diabetes had a more rapid decline in performance on the VF test compared with women with IGT or NGT. Better glucose control might ameliorate this decline.

Several studies have reported significant cognitive impairment among adults with diabetes mellitus compared with nondiabetic individuals. However, only 4 studies have measured change in cognitive function, and several have inadequately controlled for depression and education, which are major confounders in the relationship between the predictor variables and cognitive function. Based on a comprehensive review of the published literature in 1999, Stewart and Liolitsa¹ concluded there were cross-sectional and prospective associations between type 2 diabetes and cognitive impairment, affecting both memory and executive function. In an earlier review, the most commonly affected test of cognitive ability for those with diabetes was verbal memory.²

We sought to determine whether older men or women with diabetes or impaired glucose tolerance (IGT) have a greater 4-year decline on cognitive function tests results compared with those with normal glucose tolerance (NGT). Further, we examined the potential effect of glycemic control, blood pressure, lipid levels, and existing microalbuminuria, retinopathy, or coronary heart disease on any observed change in cognitive function among those with diabetes.

The Rancho Bernardo Study has followed a cohort of white, middle or upper-middle class, community-dwelling adults in a southern California suburb since 1972. The present study consists of 999 men and women who participated in the 1984-1987 and the 1988-1991 clinical evaluations at the Rancho Bernardo Study research clinic. The study sample was limited to participants who had been administered 3 tests of cognitive function at both of these visits and who either self-reported diabetes or had a 2-hour glucose tolerance test at the first visit.

Participants were classified into 3 groups by glucose tolerance status using the World Health Organization 1998 classification criteria.³ Those with NGT had a fasting plasma glucose level less than 126 mg/dL (<7.0 mmol/L) and a 2-hour postchallenge glucose level less than 140 mg/dL (<7.8 mmol/L). Those with IGT had a 2-hour postchallenge result between 140 and 199 mg/dL (7.8-11.0 mmol/L). Participants were classified with diabetes mellitus if they reported a physician's diagnosis of diabetes or had a fasting plasma glucose level of 126 mg/dL or greater (≥7.0 mmol/L) or a 2-hour postchallenge glucose level of 200 mg/dL or greater (≥11.1 mmol/L).

All included participants had 3 tests of cognitive function performed at the first and 4-year follow-up visit: the Mini-Mental State Examination (MMSE), the Verbal Fluency (VF) test, and the Trail-Making Test B (Trails B). The MMSE is a global test with components of orientation, attention, calculation, language, and recall.⁴ This test has limited sensitivity for change in cognitive function and is used primarily to screen for incipient dementia. The MMSE is scored on a scale of 0 to 30, with dementia typically suspected for results lower than 24. The VF test⁵ is a test of semantic memory in which participants are asked to name as many animals as possible in 1 minute. This test is scored with the number of correctly named animals; repetitions and variations of words are not counted. Lastly, the Trails B (from the Halstead-Reitan Neuropsychological Test Battery)⁶ is a test of visuomotor tracking and attention in which a participant identifies alternating patterns of letters and numbers in sequence over 300 seconds. This test is scored by the time required to complete the test. For the MMSE and VF test, lower test results imply worse cognitive function, while for the Trails B, a lower test result suggests better cognitive function.

Other data collected at the 1984-1987 examination included demographic information (age, sex, and education), health-related behaviors (smoking history and number of alcoholic drinks per week), clinical history (diagnosis of diabetes and stroke), and medication use (diabetes medications and postmenopausal hormone therapy). During the follow-up visit (1988-1991), participants completed the Rose questionnaire for cardiovascular disease⁷ and underwent a resting 12-lead electrocardiogram. A diagnosis of coronary heart disease was based on Rose questionnaire criteria, a history of myocardial infarction, electrocardiographic criteria for myocardial infarction, or a history of percutaneous transluminal coronary angiography or coronary artery bypass graft surgery.

Weight and height were measured using standard protocols, and body mass index was calculated as weight in kilograms divided by the square of height in meters. Each subject had 2 resting systolic and diastolic blood pressure measurements performed with a mercury sphygmomanometer. The means of the systolic and diastolic blood pressures were used for analyses. Information on depressed mood was obtained using 18 of the 21 items of the Beck Depression Inventory (BDI) at the first examination. Scores from the BDI were computed by summing the responses to the 18 questions and proportionately adjusting the results to correspond to the previously established 21-item scale. The reliability of this reduced scale has been reported previously.⁸

Morning venous blood was collected after a requested 12-hour fast and 2 hours later after a 75-g oral glucose load. Plasma glucose was measured using a glucose oxidase method. Glycohemoglobin was measured using high-performance liquid chromatography. Fasting cholesterol and triglyceride levels were measured by enzymatic methods with an ABA-200 biochromatic analyzer (Abbott Laboratories, Abbott Park, Ill); high-density cholesterol level was assayed by precipitation using a Lipid Research Clinic protocol,⁹ and low-density cholesterol level was calculated by the Friedewald formula.¹⁰ Apolipoprotein E ϵ4 (APOE ϵ4) allele was assessed using standard techniques (Puregene; Gentra, Minneapolis, Minn) from genomic DNA extracted from whole blood samples. APOE genotype was determined by gel electrophoresis following polymerase chain reaction amplification around diagnostic polymorphic sites. A participant was classified with APOE if they had 1 or more ϵ4 alleles present. Morning urine was collected, and protein concentration was measured by a Quan-T (Quantimetrix, Hawthorne, Calif) assay. Urine creatinine levels were measured by the Jaffe (Beckman, Brea, Calif) method. Microproteinuria was defined as a urine protein-creatinine ratio of 0.20 or greater. Retinopathy was detected by grading fundus photographs obtained using a Topcon (Paramus, NJ) nonmydriatic camera at the baseline diabetes visit. The grading was done by the Wisconsin Epidemiologic Study of Diabetic Retinopathy Center using a modified Wisconsin 191 classification system in evaluating the retinal photographs.¹¹

We performed sex-specific separate analyses stratified by glucose tolerance subgroups. Means and standard deviations of the baseline characteristics were compared by analysis of variance (ANOVA) or χ² test when appropriate. We also evaluated the correlation between age-adjusted glycohemoglobin level at baseline and performance on cognitive function tests at baseline and follow-up using partial Pearson correlation. Mean scores for each test of cognitive function at the first and follow-up visits were calculated by linear regression models that adjusted for age alone or in subsequent models that adjusted for age, education, BDI score, presence of the APOE ϵ4 allele, and current estrogen use (for women). Scores were compared by glucose tolerance status using ANOVA. We also calculated an adjusted change score between the first and follow-up visits. We tested for interaction with presence of an APOE ϵ4 allele.

We performed analyses of major cognitive decline for each cognitive test. Major cognitive decline was defined as the greatest 10th-percentile reduction in performance from the initial to follow-up score or the change score. This corresponded to a 2-point or greater decrease in scores on the MMSE, a 7-point or greater decrease in scores on the VF test, and a 79-second or greater increase in score on the Trails B. This approach has been used previously by other studies.¹² We performed multivariate logistic regression models adjusting for age, education, BDI score, presence of an APOE ϵ4 allele, baseline cognitive test score, and current estrogen use in women to determine the odds of major cognitive decline for each test of cognitive function by glucose tolerance subgroup. We used SAS version 8.2 (SAS Institute Inc, Cary, NC) and S-Plus version 6.1 (Insightful Corp, Seattle, Wash) for our analyses.

In adjusted models that included diabetic individuals alone, we added glycohemoglobin and lipid levels, systolic and diastolic blood pressure, and body mass index separately and together to determine whether these factors mediated the effect of diabetes on change in cognitive function. We also evaluated the effect of existing microalbuminuria, retinopathy, stroke, and coronary heart disease on change in cognitive function among those with diabetes. Statistical differences between the regression estimates for diabetes mellitus and IGT that differ by the inclusion of potential mediators were assessed with an adaptation of methods originally developed for logistic regression models.¹³ Statistical significance of the mediation effect was assessed using a Wald test.

Of 999 participants who performed cognitive function tests at both the first diabetes visit and the 4-year follow-up examination, 632 had normal glucose tolerance values, 249 had IGT, and 118 met glucose tolerance criteria for diabetes at baseline. Overall, persons with diabetes were older, less likely to exercise regularly, and less likely to report heavy alcohol use or cigarette smoking. Sex-stratified characteristics are given in Table 1. Compared with men without diabetes, men with diabetes were more depressed and had higher systolic blood pressure and lower high-density cholesterol level. Compared with women without diabetes, women with diabetes had higher body mass index, systolic blood pressure, and total cholesterol, low-density cholesterol, and triglyceride levels and had lower high-density cholesterol levels. They were also less likely to be using estrogen compared with women without diabetes.

We evaluated the age-adjusted partial correlation between glycohemoglobin and each cognitive function test for men and women separately. Among women only, there was a marginally significant correlation with glycohemoglobin and the baseline VF test (r = −0.08; P = .08) and a statistically significant correlation of glycohemoglobin with the follow-up VF test (r = −0.10; P = .04).

There was no significant difference in the mean age-adjusted or multivariate-adjusted scores on the cognitive function tests among the 3 glucose tolerance subgroups for either sex at the first visit (Table 2). Four years later, women with diabetes had a significant decline in the VF test score and had lower scores than those with IGT or NGT (15.2 ± 0.6 for women with diabetes, 16.7 ± 0.4 for women with IGT, and 17.2 ± 0.2 for women with NGT; P for trend = .007) (Figure 1). This finding remained significant after adjusting for multiple comparisons (Bonferroni method). Performance on the 3 cognitive function tests or the change scores did not differ between participants of either sex with or without an APOE ϵ4 allele.

In adjusted logistic regression models in which major cognitive decline (the greatest 10th-percentile decline in cognitive test score vs the lower 90th-percentile decline) was examined for each glucose tolerance subgroup, only women with diabetes had significantly increased odds of major cognitive decline on the VF test score (odds ratio [OR], 4.46; 95% [confidence interval] CI, 1.72-11.79) (Table 3). When we used a more liberal cutoff for defining major cognitive decline, with the greatest 25th-percentile change in cognitive function score compared with the lower 75th-percentile change, we found that there were significantly increased odds for major decline for the VF test score among women with diabetes (OR, 2.34; 95% CI, 1.06-5.17) and women with IGT (OR, 1.84; 95% CI, 1.07-3.18).

We evaluated whether factors associated with microvascular or macrovascular disease or the existence of concurrent microvascular disease (retinopathy in either or both eyes or microalbuminuria) or macrovascular disease (stroke or coronary heart disease) explained the cognitive decline on the VF test score among women with IGT or diabetes (Table 4). The only variable that significantly attenuated the β-coefficient for women with diabetes was glycohemoglobin (P = .003). The existence of concurrent microvascular or macrovascular disease did not change the association between diabetes and decline in VF test score among women. Even after adjusting for all potential mediators and microvascular or macrovascular disease, women with diabetes still had a significantly greater change in VF test score than women without diabetes.

In a cohort of community-dwelling older adults with NGT, IGT, or diabetes who performed similarly on 3 cognitive function tests at baseline, women with diabetes had a 4-fold increased risk of a major cognitive decline on the VF test after 4 years compared with women without diabetes. Glycemic control may partially mediate the apparent association between diabetes and decline in cognitive performance on this test. Glycohemoglobin was negatively correlated with performance on the VF test among women, but the association by glucose tolerance status was not materially changed after adjusting for glycohemoglobin. There was no significant difference in cognitive test performance at baseline or follow-up in men regardless of glucose tolerance status or glycohemoglobin level.

Four prior studies have evaluated the effect of diabetes on change in cognitive performance over time. In the first study, the Baltimore Longitudinal Study of Aging,¹⁴ only men were included and diabetes was defined using the 1979 National Diabetes Data Group criteria.¹⁵ After 12 years, diabetes was unrelated to change in performance on 2 cognitive function tests (the Benton Revised Visual Retention Test and the vocabulary subset of the Wechsler Adult Intelligence Scale).¹⁴ This study was limited by small numbers of subjects with diabetes and loss to follow-up. The second study examined the 3-year change in MMSE score in 353 men enrolled in the Zutphen Study in the Netherlands.¹⁶ In a subgroup analysis, the investigators compared 47 men with diabetes with 282 men without diabetes and found no significant difference in decline in score on the MMSE between the 2 groups (11% vs 14%). However, 14 men with both diabetes and an APOE ϵ4 allele had higher risk of decline than those without the allele (29% vs 3%). The third prospective study of cognitive change included approximately 9700 white women enrolled in the Study of Osteoporotic Fractures who were followed-up for 3 to 6 years.¹² Diabetes, defined by participant self-report of the diagnosis, was associated with poor cognitive performance at baseline on 3 tests of cognitive function and a greater decline in score on 2 tests that examined attention (the Digit Symbol Substitution test and Trails B). This change was independent of several confounders including heart disease and hypertension. The most recent study to prospectively examine cognitive change by glycemic category was the Epidemiology of Vascular Aging (EVA) Study,¹⁷ in which approximately 1000 men and women took the MMSE and 8 domain-specific tests 4 years apart. Diabetes was defined by self-report of the diagnosis or by the 1997 American Diabetes Association criteria.¹⁸ Participants with diabetes had poorer performance at follow-up on 4 domain-specific tests assessing memory, psychomotor speed, and attention compared with those with impaired fasting glucose or normal glucose level. Although adjusting for blood pressure did not alter their results, body mass index explained the association between diabetes and cognitive decline for 3 of the 4 tests: the Auditory Visual Learning test, the Facial Recognition test, and the Digit Symbol Substitution test, but not the Finger Tapping test. Men and women in the EVA Study were on average 7 to 8 years younger and had higher scores on the MMSE at baseline compared with the Rancho Bernardo Study participants. The EVA Study investigators reported sex-specific results but did not evaluate the effect of glycemia without diabetes, lipid levels, or concurrent microvascular or macrovascular disease in their analyses; however, they suggested that blood glucose level may not be a strong risk factor since they did not see any increased risk of cognitive decline among subjects with impaired fasting glucose level.

There are several theories for cognitive impairment in individuals with diabetes.^19,20 Some pathways by which diabetes can affect cognition involve glucose or insulin metabolism or the formation of advanced glycation end products.^21-23 Increases or decreases in glucose concentrations can affect cognitive function. Some studies have found an association between chronically elevated glucose levels and poor performance on cognitive tests.^24,25 Higher fasting insulin levels in older adults with IGT and higher serum insulin concentrations after a 2-hour glucose challenge in women without diabetes were associated with poorer performance on the MMSE.^21,26

Dyslipidemia, especially elevated triglyceride level, has been associated with cognitive impairment in patients with or without diabetes.^27,28 Hypertension, independent of diabetes, has been reported to be a predictor of poor cognitive test performance.²⁹ This effect may in part be due to increased risk of cerebrovascular disease with hypertension.³⁰ Since diabetes coexists with dyslipidemia, hyperinsulinemia, and hypertension, several of these mechanisms could be operating in concert to produce impaired cognition.

We found that elevated levels of glycohemoglobin attenuated the association between diabetes and decline in VF test score among women, while blood pressure and lipid levels had no effect. This finding is compatible with the hypothesis that improved glycemic control lessens cognitive impairment and other data suggesting a possible microvascular mechanism of cognitive impairment. Although adjustment for other microvascular diseases had no effect on the association between diabetes and cognitive dysfunction, small and short (<6 months) clinical trials have reported that improved glycemic control caused improved cognitive test performance mainly in the areas of attention, learning, and complex psychomotor function.^31-33 A large ongoing clinical trial, the Action to Control Cardiovascular Risk in Diabetes (ACCORD), is measuring cognitive function and may be able to more definitively state whether tight glucose, blood pressure, or lipid control affects cognitive decline in individuals with type 2 diabetes mellitus.

We found cognitive decline on only 1 of 3 tests and only in women with diabetes. One explanation for why the VF test was the only cognitive test in which diabetic and prediabetic subjects had greater decline than nondiabetic participants may be that verbal memory is more affected by subcortical vascular disease. We may have seen a differential result by sex in our study because women with diabetes were on average 2 years older, had higher BDI scores, and had lower VF test scores at baseline compared with men with diabetes. However, even after adjustment for age, depression score, and baseline VF test score, women with diabetes had accelerated decline in their adjusted 4-year VF score and lower follow-up scores compared with those observed for men with diabetes (15.2 ± 0.6 in women vs 17.3 ± 0.7 in men). We controlled our analyses for several confounders such as education, depression score, presence of APOE ϵ4 allele, and oral hormone therapy. We did not adjust our main analyses for use of sedative or anxiolytic drugs, which may be disproportionately represented in women with diabetes. The addition of retinopathy or other macrovascular disease outcomes to our models did not change the results. It is possible that the accelerated cognitive impairment on VF testing in women with diabetes only occurred by chance. However, this finding is significant after adjusting for multiple comparisons. The finding of accelerated cognitive decline in women with diabetes is supported by the strength of the association, a dose-graded response with women with IGT having an intermediate decline compared with women with diabetes or NGT, temporality of the association, biological plausibility, and consistency with limited prior literature.¹²

We had a limited number of cognitive function tests performed on a serial basis. However, the 3 tests we used examine different domains of cognitive function. We did not check glucose levels on the days of cognitive testing. There is a potential of acute hypoglycemia or hyperglycemia to cause poor performance on cognitive testing among those with diabetes. However, none of the participants had type 1 diabetes, and the mean ± SD glycohemoglobin level of those with diabetes was 6.8% ± 1.7% at baseline, implying well-controlled or less severe diabetes, in which daily glycemic fluctuations would be less extreme. Moreover, only 2 of the 118 subjects with diabetes were using insulin and at higher risk for hypoglycemia, and excluding them from our analysis did not change our findings. Finally, since the Rancho Bernardo cohort was mostly white, well educated, and relatively affluent, these findings may not be generalizable to nonwhite racial/ethnic groups or to older adults of different socioeconomic or educational backgrounds.

In conclusion, women with diabetes had accelerated cognitive decline in a semantic memory test results in this well-characterized cohort of older, white Americans. Chronic hyperglycemia may mediate this effect. Improved glycemic control may prevent cognitive decline among women with diabetes, but the association also exists at glycemic levels not diagnostic of diabetes.

Corresponding author and reprints: Alka M. Kanaya, MD, c/o Women's Health Clinical Research Center, 1635 Divisadero St, Suite 600, San Francisco, CA 94115 (e-mail: alkak@itsa.ucsf.edu).

Accepted for publication August 15, 2003.

Dr Kanaya is supported by the UCSF-Kaiser Building Interdisciplinary Research Careers in Women's Health scholarship (5 K12 AR47659) and by grant P30-AG15272 under the Resource Centers for Minority Aging Research program by the National Institute on Aging, the National Institute of Nursing Research, and The National Center on Minority Health and Health Disparities, National Institutes of Health. Dr Barrett-Connor is the principal investigator for the Rancho Bernardo study, which is funded by grant NIA 5R01 AG07181 from the National Institute on Aging and by grant 5R01 DK31801 from the National Institute of Diabetes and Digestive and Kidney Diseases. Dr Yaffe is supported by grant NIA AG00888 from the National Institute on Aging and by the Paul Beeson Faculty Scholars Program.

This study was presented as an abstract at the National American Diabetes Association meeting; June 18, 2002; San Francisco, Calif.

We thank Eric Vittinghoff, PhD, for help with statistical questions.