Introduction

Cognitive dysfunction has been found in multiple studies comparing both adolescents and adults with type 1 diabetes with non-diabetic individuals. These effects are most apparent on measures of psychomotor efficiency, intelligence, sustained attention, speed of information processing, cognitive flexibility and visual perception [1]. The pathophysiological basis for these findings remains poorly understood. Although earlier reports attributed such deficits to the effects of recurrent, moderately severe hypoglycaemia, a growing body of research has failed to substantiate such linkages [1–4]. This was particularly apparent in our analyses of the long-term follow-up data from 1,144 participants in the Diabetes Control and Complications Trial (DCCT)/Epidemiology of Diabetes Interventions and Complications (EDIC) study [2]. In those analyses, we found no relationship between decline in cognitive functioning over an 18 year follow-up period and the occurrence of one or more episodes of hypoglycaemia-associated seizure or coma.

On the other hand, results from the same DCCT/EDIC study [2], as well as findings from several recent cross-sectional and longitudinal studies [1, 3, 5–7], identified associations between mild cognitive dysfunction and chronic hyperglycaemia, defined as a long history of poor metabolic control and elevated HbA1c values. Two well-known outcomes of chronic hyperglycaemia, diabetic retinopathy [3, 4, 6] and distal symmetric polyneuropathy [7, 8], have previously been associated with cognitive dysfunction in several small studies of adults with type 1 diabetes, leading to speculation that diabetic cognitive complications may be largely microvascular in origin [9].

Cognitive function can also be influenced by other biomedical factors that are commonly associated with diabetes. Macrovascular risk factors like cigarette smoking [10], elevated blood pressure and hypertension [11], hypercholesterolaemia [12], and subclinical atherosclerosis have been reported to impair cognition in non-diabetic adults [13] and in older adults with type 2 diabetes [14, 15]. Despite those observations, no study has comprehensively assessed the degree to which macro- and microvascular complications individually and/or synergistically influence cognitive performance in large numbers of young and middle-aged adults with type 1 diabetes.

The DCCT and the follow-up EDIC study provide an excellent means of examining those inter-relationships in a very large sample of carefully monitored adults with type 1 diabetes. Participants completed an extensive battery of cognitive tests, which was repeated over the course of the DCCT/EDIC study period. They underwent frequent and systematic biomedical evaluations to ascertain changes in metabolic control over time and were also assessed for occurrence of micro- and macrovascular risk factors and clinically significant complications.

These data, collected prospectively over a mean of 18.5 years, allowed us to test the hypothesis that the cognitive changes occurring in adults with type 1 diabetes are associated with four broad classes of biomedical variables: (1) macrovascular risk factors (hypertension, smoking, abnormal lipid levels, obesity); (2) development of microvascular complications (retinopathy, nephropathy); (3) evidence of sub-clinical macrovascular disease (carotid intima–media thickening [IMT], coronary calcification); and (4) time-weighted HbA1c levels. We examined the relative contribution of these biomedical variables and carried out additional analyses to determine whether effects of these factors on cognitive performance were influenced by sex, age, history of severe hypoglycaemia or presence of the APOE ε4 allele.

Methods

Participants

Between 1983 and 1989, 1,441 patients from 13 to 39 years of age were enrolled in the DCCT and followed until 1993 [16]. At the end of the DCCT, 1,375 (96%) of the 1,428 surviving participants volunteered to continue in the EDIC observational follow-up study. In 2004, 85% of the surviving eligible participants (n = 1,144) (Fig. 1) were evaluated using the same cognitive test battery as that used at baseline in the DCCT. This group constitutes the sample reported on here [2]. DCCT participants who did not participate in this follow-up study were comparable to the 1,144 participants in terms of baseline demographic, biomedical and cognitive measures. This research protocol was approved by each site’s institutional research ethics committee; each participant gave written informed consent prior to the initiation of research activities.

Fig. 1
figure 1

Flow diagram of participants who took part in cognitive testing at EDIC year 12

Full size image

Cognitive assessment

Cognitive testing was conducted at each site by personnel trained and certified by the DCCT/EDIC Central Neurobehavioral Coding Unit [17]. The test protocol, which took 4 to 5 h to complete, included the following widely used, well-validated tests that were administered initially during the DCCT: six subtests (similarities, comprehension, digit span, digit symbol, block design and object assembly) from the Wechsler Adult Intelligence Scale [18]; four subtests (category test, tactual performance test, trail making test and finger tapping test) from the Halstead–Reitan Neuropsychological Battery [19]; the logical memory and visual reproductions subtests from the Wechsler Memory Scale [20]; the Digit Vigilance Test [21]; the Grooved Pegboard Test [22]; the Verbal Fluency (FAS) Test [23]; the Four-Word Short Term Memory Test [24]; the Symbol-Digit Learning Test [25]; and the Embedded Figures Test [25]. Tests were administered in a fixed sequence and capillary blood glucose levels were routinely monitored to verify absence of hypoglycaemia during testing [2].

During the DCCT, 24 test variables were chosen a priori to be of particular diagnostic value when applied to patients with type 1 diabetes. A standardised (z) score was calculated for each variable, with the mean and standard deviation from the baseline assessment of the DCCT cohort used as reference. These procedures have been outlined elsewhere [17]. To reduce the number of comparisons, the 24 standardised scores were grouped into one of eight cognitive domains consistent with standard neuropsychological assessment strategies [26]. For each domain, the simple average of the standardised scores was used to represent the change from baseline.

Biomedical evaluations

During EDIC, participants had an annual history update, physical examination, electrocardiogram and laboratory testing, using the same methods employed during the DCCT [16]. Participants self-reported the presence of sensory symptoms of peripheral neuropathy [27]; best-corrected visual acuity was measured at 4-year intervals during EDIC. The visual acuity examination closest to the cognitive re-evaluation was used in these analyses. Participants were asked quarterly in the DCCT and annually in EDIC to provide information about smoking and alcohol consumption.

HbA1c values were measured in a central laboratory by high-performance liquid chromatography [28], with tests done quarterly during DCCT and annually during EDIC. Time-weighted values across the entire duration of the DCCT/EDIC study were used in our analyses. They were calculated from DCCT inception by weighting each DCCT value by one quarter of a year and weighting each EDIC value by 1 year. Severe hypoglycaemia was defined as any event leading to seizure or coma [2, 29]. Fasting lipid profiles and 4 h urine collections for measurement of albumin excretion rate and creatinine clearance were obtained in alternate years during EDIC. Renal complications were defined as a centrally measured serum creatinine level of at least 177 μmol/l, or treatment with dialysis or transplantation for chronic renal failure [27]. Hypertension was defined as systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg or documented hypertension or the use of anti-hypertensive agents for the treatment of hypertension. Hypercholesterolaemia was defined as LDL-cholesterol ≥ 3.4 mmol/l or the use of lipid-lowering agents. Retinopathy was assessed by seven-field stereoscopic fundus photography according to the DCCT/EDIC protocol [30]. For these analyses, retinopathy was defined by the presence of proliferative diabetic retinopathy (PDR) and/or history of panretinal scatter-photocoagulation (laser) therapy.

The measurement of intima–media thickness was performed during years 1, 6 and 12 of EDIC; for these analyses, the year 12 examination results were used. A single longitudinal lateral view of the distal 10 mm of the right and left common carotid arteries was obtained using B-mode ultrasonography [31]. Coronary artery calcification was measured with computed tomography on one occasion during years 7 and 9 of EDIC and scans were read centrally [32]. Genotyping of the APOE haplotype, rs7412 (ε2/ε3) and rs429358 (ε3/ε4), and of the ACE intron 16 indel (rs4340) was performed as described previously [33].

Statistical analyses

Separate analysis of covariance models were used to assess the relationship between each of the eight cognitive domain scores and HbA1c values, macrovascular risk factors (hypertension, waist circumference, hypercholesterolaemia, smoking history), macro- and microvascular complications, and alcohol use. Adjustments were made for sex, baseline age, duration of follow-up, education level, the number of interval cognitive tests taken, presence of painful neuropathy and visual acuity at EDIC year 12. Additionally, five multivariable models were evaluated using psychomotor efficiency as the dependent variable to assess the independent effects of multiple complications and risk factors. Statistical analyses were performed using SAS version 8.2 statistical analysis software (Cary, NC, USA). Univariate comparisons are indicated as being significant at p < 0.006 after Bonferroni correction. For the final model, where all previously significant variables were entered, the Hochberg correction (p < 0.05) was used to control for the false discovery rate [34].

Results

Demographic and clinical characteristics of sample

When restudied an average of 18.5 years after their DCCT baseline assessment, participants ranged in age from 29 to 62 years (mean 45.7, SD 6.8) and were in relatively good health despite an average duration of 25 years of diabetes (Table 1). As reported previously, participants treated with intensive or conventional therapy during the DCCT had indistinguishable HbA1c levels during follow-up in EDIC, with values declining from 9.0 ± 1.6% (mean ± SD) at DCCT baseline to 7.7 ± 1.2% in the former conventional treatment group and to 7.8 ± 1.2% in the former intensive treatment group. There were no statistically significant (p < 0.01) differences between the two treatment groups for any of the demographic characteristics at DCCT baseline and EDIC year 12.

Table 1 Demographic characteristics at DCCT baseline and EDIC year 12
Full size table

Association of micro- and macrovascular risk factors and complications with change in cognitive function

After adjusting for sex, baseline age, baseline education, presence of painful neuropathy and visual acuity at EDIC year 12, length of follow-up and the number of interval cognitive tests taken, modest associations were found between complications and cognitive performance (Tables 2 and 3). The most consistent associations appeared in the domains of psychomotor efficiency (renal complication, PDR) and motor speed (PDR). Macrovascular risk factors were also found to be associated with cognitive outcomes, including hypertension (with psychomotor efficiency) and smoking (with learning, immediate memory delayed recall, spatial information processing and psychomotor efficiency). HbA1c was also associated with psychomotor efficiency and motor speed.

Table 2 Single predictor models for macrovascular and microvascular risk factors
Full size table
Table 3 Single predictor models for macrovascular and microvascular complications
Full size table

Multivariable modelling of cognitive decline

To evaluate the independent effects of multiple complication and risk factor predictor variables, we selected psychomotor efficiency as the dependent variable because it has been identified in other studies as the cognitive domain most often affected in patients with diabetes [1, 3, 4, 35] and, as expected, showed the strongest associations with vascular complications and risk factors in this study (Table 4). The domain included five test scores: verbal fluency (total correct); digit symbol (total correct in 90 seconds); trail making Part B (total time); and grooved pegboard (completion time for dominant and non-dominant hands). Five multivariate models were evaluated: (1) base model using the demographic covariates evaluated in the separate analysis of covariance models; (2) addition of macrovascular risk factors and HbA1c to the base model; (3) addition of macrovascular disease to the base model; (4) addition of microvascular complications to the base model; and (5) evaluation of all previously significant variables in a single model.

Table 4 Multivariable models for domain seven (psychomotor efficiency)
Full size table

Based on the fifth (final) model described above (Table 4), five variables predicted decline in psychomotor efficiency over the 18.5 year follow-up period: two demographic variables (age, education), two microvascular complications (renal complication, PDR) and HbA1c. One additional variable (common carotid IMT) was significant at p < 0.01, but was marginally beyond the established level of Hochberg-corrected statistical significance (Hochberg-corrected p = 0.08). Taken together, these variables explained 23% of the variance in change in psychomotor speed. Somewhat surprisingly, other well-known candidate variables (e.g. hypertension, hypercholesterolaemia, smoking) were unrelated to cognitive outcomes in these multivariable models. We also found no interactions between any of the predictors and sex, age (older vs younger), number of severe hypoglycaemic events or presence of the APOE ε4 allele (data not shown). The correlation matrix between all possible covariates was examined to assess for potential multicollinearity. Interactions were fit in the final model for correlations exceeding 0.40 and none were found to be statistically significant.

Discussion

This study indicates that over an 18 year period, the development of clinically significant microvascular complications is associated with modest declines in cognitive functioning in relatively healthy young and middle-aged adults with type 1 diabetes. Multivariable modelling demonstrated that the level of glycaemic control, indexed by HbA1c level, the occurrence of serious diabetic retinopathy (defined as presence of PDR and/or history of panretinal laser therapy) and renal complications were each independently associated with declining performance in measures of psychomotor efficiency. These relationships persisted even after controlling for demographic characteristics (age, education) and health factors (symptomatic peripheral neuropathy, reduced visual acuity) known to influence cognitive performance. In contrast, test scores were unrelated to recurrent episodes of severe hypoglycaemia, presence of the APOE ε4 allele or diabetes-related comorbid conditions like hypertension or hypercholesterolaemia. Because participants were treated for hypertension and lipid disorders during the entire DCCT and EDIC follow-up period, the cognitive effects of such variables may have been lessened by the interventions. Previously we had reported that level of depression measured using the SCL-90 was associated with cognitive function [2], but found that it did not interact with biomedical predictors in that set of analyses.

Several earlier, smaller studies reported associations between retinopathy and psychomotor efficiency in adults with [4, 6] and without diabetes [36]. Our study is the first to show that the presence of another common microvascular complication, significant kidney dysfunction, also independently predicts cognitive decline in adults with type 1 diabetes. The relationship between cognitive decline and clinically significant microvascular complications suggests that a ‘microvascular’ pathway may contribute to cerebral dysfunction by reducing the efficient delivery of substrate to neurons as a consequence of microangiopathy [5]. Studies of non-diabetic adults [36] have shown that the presence of retinal microvascular abnormalities is associated with the same pattern of psychomotor slowing as seen in our cohort and is also associated with evidence of cerebral atrophy detected by magnetic resonance imaging [37] and subclinical cerebral infarcts [38]. Given the close anatomical and physiological relationships between retinal and cerebral vascular networks [39], it now appears that retinal microvascular disease may be a marker of damage to cerebral microcirculation [40, 41]. Our work also suggests that diabetic renal disease is a marker of brain dysfunction. The underlying mechanism remains poorly understood, but may reflect the magnitude of microvascular complications in multiple organ systems.

In addition to detecting relationships between microvascular complications and cognitive outcomes, our study also found an independent association between change in cognition and time-weighted HbA1c values. This apparent ‘non-vascular’ relationship may reflect the deleterious effects of high brain glucose levels on neuronal integrity. Recent research using proton magnetic resonance spectroscopy has detected elevated levels of brain glucose and glutamate (an excitatory neurotransmitter that can cause neuronal damage in high concentrations) within prefrontal regions in a large sample of young adults with childhood onset of type 1 diabetes compared with non-diabetic control participants [42]. Among patients with type 1 diabetes, higher HbA1c values were associated with elevated prefrontal glutamate concentration. Elevated glutamate was, in turn, negatively correlated with performance in measures of psychomotor speed. Duration of poorer metabolic control is also inversely correlated with levels of N-acetylaspartate and choline-containing compounds in white matter and in deep grey matter [43]. These and other [44] brain magnetic resonance spectroscopy findings suggest that axonal injury, demyelination and increased membrane proliferation secondary to gliosis have occurred and may be a consequence of glucose neurotoxicity [45]. This non-vascular pathway could explain, at least in part, the recent observation that adults with a long history of type 1 diabetes manifest psychomotor slowing associated with microstructural damage to white matter fibre tracts in the corpus callosum and corona radiata [46].

The magnitude of cognitive decline reported in this study is modest, but comparable in size and scope to that reported in other studies of adults with type 1 diabetes [1]. It now appears that tests of mental and motor slowing, particularly those requiring ‘psychomotor integration’ or eye–hand coordination, are especially sensitive to the effects of chronic hyperglycaemia and are associated with severity of retinal microangiopathy [4, 6]. The neuroimaging literature suggests that optimal performance of these types of cognitive tasks requires the integrity of multiple small and distinct anatomical substrates in the left and right hemispheres, compared with other types of cognitive tasks, which may activate more limited neuroanatomic areas [47]. Thus, if chronic hyperglycaemia leads to significant hypoperfusion in multiple brain regions and also disrupts the integrity of associative axonal networks, one might expect to see greater decline in psychomotor speed than with changes in other cognitive domains. This prediction is consistent with our data, as well as with the cognitive literature on type 1 diabetes [1].

In many ways it is not surprising that, as a group, our participants showed minimal cognitive dysfunction. All are relatively healthy young and middle-aged adults who developed diabetes in adolescence or early adulthood (mean age at diagnosis 21 years) and whose diabetes has been carefully managed and monitored. To date, two other studies have measured cognitive function in patients with type 1 diabetes over an extended period of time [6, 44]. Only one of those studies explicitly examined the relationship between changes in cognitive function and the development of vascular complications [6]. In that study, the authors reported stronger relationships between cognitive change and vascular complications in their relatively small (n = 103) study sample, but their participants differed from the DCCT/EDIC cohort in several potentially important ways, including longer duration of diabetes and poorer metabolic control.

Very few cognitive studies have been conducted on older adults with type 1 diabetes [3], but extensive research on older adults with type 2 diabetes clearly demonstrates that in those over the age of 65, the presence of diabetes accelerates the rate of cognitive decline. Relative to their non-diabetic peers, individuals with type 2 diabetes showed 1.2- to 1.5-fold faster declines in cognitive function over time [48]. Such declines in cognitive performance may be worsened by the presence of diabetes-related comorbid conditions like hypertension [49] or macrovascular disease (e.g. carotid IMT) [50]. The current high level of cognitive function within our study cohort may explain the small amount of variance accounted for by our glycaemic and microvascular predictors. Moreover, the absence of a statistically reliable relationship between cognitive change and early macrovascular complications may be a consequence of the relatively low prevalence of those complications in this carefully monitored study cohort [32]. However, as our patients with type 1 diabetes continue to age, they may show more pronounced evidence of cognitive decline as their exposure to chronic hyperglycaemia increases and their diabetic vascular complications worsen. Those diabetic patients who are successful in maintaining optimal metabolic control and in slowing or preventing the development of complications should also manifest relatively less cognitive decline over time.