Magnetic Resonance Imaging of the Brain in Diabetes

RESEARCH DESIGN AND METHODS

The study design and 10 cohorts included in CASCADE have been described in detail elsewhere (24–34). Briefly, we included ongoing community-based studies with the following characteristics: 1) data were available on cardiovascular risk factors and disease collected ≥5 years before the MRI assessment; and 2) the cohort had been followed since inception, and data on vital status were available.

Five of the cohorts (27–31) were developed as World Health Organization–Monitoring Trends and Determinants in Cardiovascular Disease (WHO-MONICA) projects (35) or used procedures suggested in the MONICA protocol. Furthermore, all studies originally drew their samples randomly from population registries or a defined working population. Participants for CASCADE were randomly selected from the baseline cohort within strata of those who were aged between 65 and 75 years at the CASCADE visit. All cohorts included subjects in the complete age range, except for the U.K. cohort (31), which included subjects 65–68 years of age.

Men and women were equally represented in the total sample. Excluded from the CASCADE sample before, or after, recruitment were subjects with a known clinical diagnosis of dementia or with Mini Mental Status Examination scores <15 and those with contraindications for the MRI. Data collection for CASCADE took place between 1996 and 1998 and consisted of a clinical interview, blood pressure measurement with a standard random-zero sphygmomanometer, routine laboratory assessment, MRI, and cognitive function testing. Informed consent was obtained at each center in accordance with guidelines from local institutional review boards. All of the individuals who took part in the exam were mobile and competent to understand the nature of their participation.

Study cohort.

The entire CASCADE cohort comprises 1,805 individuals. The current study included the 1,252 CASCADE participants who had a physician’s diagnosis of diabetes status, information on treatment for diabetes, and measured fasting or nonfasting glucose levels. This information was not available in the Augsburg WHO-MONICA (27) and the Epidemiological Prevention study of Zoetermeer (EPOZ) samples (34), comprising 194 and 267 participants, respectively. These samples were not considered for the current CASCADE analysis. The same applied for 13, 6, 52, and 21 attendees of the Rotterdam scan study (25), the Epidemiology of Vascular Aging (EVA) study (26), the POL-MONICA Krakow study (29), and the Whitehall II study (31), in which fasting or nonfasting glucose levels were not available. The diabetes cohort was not different from the entire CASCADE sample in terms of age, distribution of sex, and frequency of CVD risk factors.

Risk factor definitions.

Diagnosis of prevalent diabetes was based on the 1997 recommendations of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus (36). It was defined as a history of diabetes confirmed by the treating physician or as treatment for diabetes, a fasting glucose level >126 mg/dl, or a nonfasting glucose level >200 mg/dl. A total of 114 individuals had diabetes according to this definition. In 92 subjects the diagnosis relied on the presence of two or more of the diagnostic features. Diabetes diagnosis was based solely on a fasting glucose level >126 mg/dl in 21 subjects, and in one study participant, it relied solely on a nonfasting glucose level >200 mg/dl. Treatment was dichotomized into no treatment or diet only, and the use of oral antidiabetic medications or insulin. A separate analysis for insulin-dependent diabetes was precluded by the small size of this group. According to the American Diabetes Association’s clinical practice recommendations (www.diabetes.org/uedocuments/overview.pdf), glycemic control at the time of glucose measurement in the treated diabetic patients of the study was in the normal range (<110 mg/dl fasting or <120 mg/dl nonfasting plasma glucose levels) in 7 (12.5%) patients, in the goal range (90–130 fasting or 110–150 nonfasting plasma glucose) in 13 (23.2%) patients, and in the range where additional action is suggested (<90 or >150 mg/dl fasting or <110 or >180 mg/dl nonfasting glucose) in 31 (55.4%) patients. In five (8.9%) subjects, the glucose values were below those requiring additional action but above the goal range.

In the current study, the group of treated diabetic patients included all patients who had been given a pharmacological treatment for diabetes, irrespective of the achieved level of glycemic control. Hypertension was defined as a history of hypertension with or without treatment or systolic blood pressure >160 mmHg or diastolic blood pressure >95 mmHg. Among the 559 hypertensive study participants, diagnosis of hypertension was based on history alone in 20 (3.6%) subjects, whereas treatment or elevated blood pressure values were the sole diagnostic feature in 45 (8.1%) and 32 (5.7%) subjects, respectively. In all other cases, the diagnosis of hypertension relied on combinations of the diagnostic criteria.

Coronary heart disease was coded to be present if the study participant reported symptoms of angina pectoris, or was treated for angina or had a history of myocardial infarction, PTCA, or coronary bypass surgery before the entry into the study. Smoking status was assessed by questionnaire, and subjects were defined as never, current, or former smokers. Total cholesterol was determined by using commercially available kits. BMI was calculated as body weight in kilograms divided by the square of height in meters (kg/m²).

MRI

MR acquisition.

All scans were made with a 1.0- or 1.5-tesla machine. The core MRI protocol included proton density and T2-weighted as well as T1-weighted sequences with 20 axial slices that were 5- or 6-mm thick with an interslice gap of 1 or 1.2 mm, respectively. The same mobile MRI machine (Siemens 1.0 tesla) was sent to five study sites (Spain, Italy, Poland, Sweden, and the U.K.) and parked centrally for 1 week. Other centers (Germany and the Netherlands) acquired the images on a 1.5-tesla machine using the core MRI protocol. Subsequent to the start of CASCADE, two other centers with already-collected scans were included (France and Austria); those scans were obtained with protocols comparable to the one used in CASCADE.

MRI reading protocol.

Hard copies of the scans were evaluated at the core radiology center (Department of Radiology, Daniel Den Hoed Cancer Center, Rotterdam, the Netherlands). The evaluation protocol was based on semiquantitative scales with known inter- and intrarater reliability (25,37). All MRI scans were read for periventricular and subcortical white matter lesions, infarcts, and cortical as well as subcortical atrophy. White matter lesions were focal signal hyperintensities on proton density and T2-weighted scans, which were not seen or exhibited only faint hypointensity on T1-weighted images. Periventricular white matter lesions were abnormalities directly abutting the lateral ventricle. They were graded on a severity scale (0–3) at the frontal and occipital horns and the body of the lateral ventricle, with the total periventricular white matter lesion score being the sum of these three scores. Subcortical white matter lesions were hyperintense foci separated from the lateral ventricle. They were graded by size (small, medium, and large) and number. The total volume of subcortical white matter lesions was assessed by multiplying each lesion by a size-dependent constant (0.0042 for small lesions, 0.114 for medium lesions, and 0.95 for large lesions) and by subsequent summation of the results (25). Infarcts were focal abnormalities with clearly defined borders, and they were isointense to cerebrospinal fluid on all sequences with a diameter >3 mm. Cortical atrophy was qualitatively assessed by a semiquantitative scale (0–3) applied to each lobe and to the sylvian fissure. The sum of the lobar scores represented the total cortical atrophy score (range 0–15). Subcortical atrophy was defined as the average of ventricular indexes relative to the width of the brain measured at the level of frontal horns, occipital horns, and caudate nuclei. The extent of white matter abnormalities and atrophy was considered to be severe if it belonged to the upper quintile of the distribution.

One neuroradiologist trained three raters, who scored hard copies of the images. One rater (reading A) scored the scans from all of the studies except those from the Netherlands. The reader was blinded to center. The scans from the Netherlands were read by the two other raters (reading B). Intrarater reliability for the atrophy reading ranged from a κ value of 0.95 (frontal lobe) to a κ value of 0.67 (temporal lobe). There was no significant intrarater difference in the quantitative measure of white matter lesion load. Interrater κ values between reading A and B ranged from 0.35 for the occipital lobe to 0.72 for the frontal lobe and sylvian fissure; there were no significant differences in mean white matter lesion load between reading A and B.

Statistical analysis.

We used the Statistical Package of Social Sciences (PC+, version 10.0.0; SPSS, Chicago, IL) for data analysis. Comparisons of categorical variables between nondiabetic patients and diabetic patients were performed using χ² test. Assumptions of a normal distribution for continuous variables were tested by nonparametric Kolmogorov-Smirnov statistics. Normally distributed continuous variables were compared by Student’s t test, and the Mann-Whitney U test was used for comparison of non-normally distributed variables, including the highly skewed total periventricular white matter lesion score and the total volume of subcortical white matter lesions. Before data pooling we assessed the separate studies for heterogeneity by formal testing of the study × diabetes interaction on each outcome variable. The data were pooled because there was no evidence of heterogeneity.

To examine the independent associations of diabetes to the various types of morphologic brain changes, we performed multiple linear regression analysis. We adjusted for sociodemographics and for CVD risk factors, including hypertension, coronary heart disease, smoking status, BMI, and total cholesterol. Logistic regression analyses, with adjustment for the same covariates, were performed for dichotomized MRI variables infarcts, severe periventricular and severe subcortical white matter lesions (WMLs), and severe cortical and subcortical atrophy.

Possible interactions between diabetes and hypertension, diabetes and coronary heart disease, and diabetes and ever smoking were assessed by stratification. If stratification suggested a possible interaction, the respective interaction terms were tested in the multiple linear regression models.

RESULTS

The descriptives and frequencies of CVD risk factors of the CASCADE cohorts are shown in Table 1. Of the 114 study participants with diabetes, 58 (50.9%) subjects had no treatment or only dietary treatment, 47 (42.0%) received oral antidiabetic medications, and 9 (0.7%) received insulin. As can be seen from Table 2, diabetic patients were more commonly men, had a higher frequency of hypertension and coronary heart disease, and had a higher BMI. The univariate comparison of MRI findings demonstrated more extensive cortical atrophy and a higher ventricle-to-brain ratio in diabetic patients than in nondiabetic patients. There were no significant between-group differences for focal brain abnormalities, including periventricular, subcortical WML, and infarcts (Table 3). After adjustment for sex, age, study, education, and the CVD risk factors hypertension, coronary heart disease, smoking status, BMI, and total cholesterol, a nonsignificant trend toward more pronounced cortical atrophy in diabetic patients was seen (β = 0.44, 95% CI −0.04 to 0.92, P = 0.07). For severe cortical atrophy, the relative risk in diabetic patients remained significantly increased when compared with nondiabetic patients (odds ratio [OR] 1.73, 95% CI 1.06–2.81). Diabetes was not significantly associated with any other brain lesions.

Stratification of the data by hypertension, coronary heart disease, and ever smoking revealed a strong interaction between diabetes and hypertension for the extent of cortical brain atrophy. Figure 1 demonstrates the multiple linear regression results on the association between diabetes and cortical atrophy in normotensive and hypertensive subjects for each substudy and for the entire cohort. A direct relationship between diabetes and cortical atrophy was only seen in the patients who were hypertensive. There was no significant association with cortical atrophy in diabetic patients with normal blood pressure. The positive relationship between diabetes and cortical atrophy in hypertensive diabetic patients was consistently observed in all CASCADE cohorts. It reached statistical significance for the entire cohort after adjustment for all covariates (β = 0.99, 95% CI 0.37–1.62, P = 0.002). The OR (95% CI) for severe cortical atrophy in normotensive diabetic patients was 1.25 (0.64–2.45), whereas in hypertensive diabetic patients, it was 2.34 (1.42–3.86). Hypertension in the absence of diabetes was not significantly related to cortical atrophy (β = 0.19, 95% CI −0.11 to 0.49, P = 0.21). Table 4 shows the effect of the interaction between diabetes and hypertension on cortical atrophy. The first model included the interaction term total diabetes × hypertension, in addition to all sociodemographic factors and CVD risk factors. The second model included the interactions no diabetes × hypertension, untreated diabetes × hypertension, and treated diabetes × hypertension. Only the interactions total diabetes × hypertension and untreated diabetes × hypertension were significant.

DISCUSSION

These cross-sectional data from a pooled large community-based sample demonstrated a direct relationship between diabetes and cortical brain atrophy. The association between diabetes and cortical atrophy was independent of potential confounders. We observed a strong interaction between diabetes and hypertension, such that the presence of both diseases resulted in a substantially greater risk for cortical brain atrophy. Neither diabetes nor hypertension were significantly related to cortical atrophy in the absence of the other risk factor, underscoring the importance of the combination of the two risk factors for the occurrence of this MRI finding. The interaction with hypertension was seen in untreated but not in treated diabetic patients. Diabetes was not associated with any type of focal ischemic brain lesions, including abnormalities caused by small vessel disease. Possibly, these cohorts did not include individuals with the most severe stages of diabetes or with severe sequelae of brain infarcts because they are less likely to attend a health survey. Nonetheless, many previous MRI studies focusing on brain changes caused by small vessel disease presented similar results (3,4,6,9,11–15,18,19). A lack of association between diabetes and small vessel disease–related cerebral lesions has also been observed in other previous population-based MRI studies (18,19). These studies included individuals who on average were older than our cohort.

Our investigation has several strengths. With 1,252 participants, including 114 diabetic patients, it is among the largest community-based MRI investigations. It includes samples from several parts of Europe, CVD risk factor assessment was similar in all centers, and all MRI scans were read centrally to homogenize scan interpretation. Also, results were consistent across cohorts.

A weakness of the study is that fluid-attenuated inversion-recovery sequences, which are more sensitive to white matter abnormalities than standard T2-weighted spin-echo pulse sequences, were not performed because they were still in an investigational state when CASCADE was initiated. Other limitations of the study are the lack of data on diabetes-related factors that might be involved in the evolution of brain abnormalities, such as the duration and complications of diabetes or the frequency of hypo- or hyperglycemic events. Although our results suggest that diabetes treatment affects the interaction between diabetes and hypertension, additional data on the duration and quality of diabetes and hypertension therapy are needed to better interpret this finding.

This epidemiological study cannot determine the mechanism(s) by which diabetes leads to cortical atrophy. The association could be causal or could result from shared risk factors, including a common genetic susceptibility for both diabetes (38) and degenerative brain disease. Lunetta et al. (7) suggested that episodes of hypoglycemia, glycometabolic dysequilibrium, or alterations of the blood-brain barrier may be responsible for brain atrophy in young insulin-dependent diabetic patients. It is also possible that atrophy is only a consequence of a status of dehydration of the brain in patients with diabetes. Another presumed mechanism is neurodegeneration with increased production of advanced glycation end products (38–40). Epitopes of these products have been detected in very early stages of Alzheimer’s disease and are thought to promote the formation of plaques and tangles (38–40). Recently, a postmortem analysis of the Honolulu-Asia Aging study demonstrated a 3-fold increased number of hippocampal neuritic plaques and a 3.5-fold higher count of cortical neurofibrillary tangles in participants with type 2 diabetes and an ε4 allele of the apolipoprotein E gene compared with those with neither risk factor (38).

Promotion of Alzheimer pathology, which has also been reported for hypertension (41), could be one pathway in the development of cortical brain atrophy in which diabetes and hypertension exert interactive effects, as seen in the current investigation. Another could be alteration of the blood-brain barrier. Both risk factors are known to cause blood-brain barrier disturbances by reduction in the density of capillaries and by thickening of the basal membrane, which can lead to cerebral hypoxia with subsequent brain atrophy (42–44). Importantly, the Framingham study also described a strong interaction between non–insulin-dependent diabetes and high blood pressure as a risk factor for poor cognitive performance in the elderly (45). Although these data suggest a link between diabetes and hypertension with degenerative processes, it is important to realize that brain atrophy as seen in the current study cannot a priori be considered an equivalent of cerebral degeneration. To evaluate the causative mechanisms of this finding, studies on the clinical consequences and pathological substrates of cortical atrophy in patients with diabetes and coexisting hypertension are needed.