Effects of Intensive Diabetes Therapy on Neuropsychological Function in Adults in the Diabetes Control and Complications Trial

Methods

Study Sample

Patients recruited for the DCCT were 13 to 39 years of age, had had IDDM for 1 to 15 years, and were in generally good health [1]. Exclusion criteria were advanced retinopathy, nephropathy, or neuropathy; a history of drug or alcohol abuse; psychotic episodes; eating disorders; epilepsy; recurrent episodes of ketoacidosis; and recurrent episodes of coma or seizure due to hypoglycemia. In the first 278 randomly assigned patients followed for 12 months, a history of severe hypoglycemia was identified as a risk factor for severe hypoglycemia [15]. Potential volunteers were subsequently excluded if they had had more than two episodes of hypoglycemic seizure or coma in the previous 2 years. The demographic and clinical characteristics of the 1441 patients are listed in Table 1.

Extent of Follow-up

The entire cohort of 1441 patients was followed for 3.5 to 9 years (mean, 6.5 years), yielding approximately 9300 patient-years of observation. Two hundred sixty-eight patients (19% of the total cohort) were studied for 9 years, and 1088 (76%) were studied for 5 years. Almost all patients (99.7%) were followed for at least 3 years. Final follow-up study data were collected on 1422 patients (99% of the total cohort). Of the 19 patients who did not complete the study, 8 dropped out and 11 died. More than 95% of expected visits were held over the 9-year study period, and 98% of the expected neuropsychological protocols were completed.

Adherence to Assigned Treatment

We randomly assigned 711 patients to receive intensive treatment, which consisted of insulin administered three or more times per day by injection or by continuous subcutaneous infusion with an external pump. Blood glucose levels were monitored four or more times per day, and the results, coupled with anticipated meal content and exercise, were used to adjust the insulin dose. Treatment goals were preprandial blood glucose levels between 3.89 and 6.66 mmol/L, postprandial blood glucose levels less than 9.99 mmol/L, a weekly 0300 h measurement greater than 3.61 mmol/L, a monthly measured glycosylated hemoglobin (HbA_1c) level within the nondiabetic range (< 6.05%), and avoidance of severe hypoglycemia. Conventional therapy (730 patients) consisted of one or two daily insulin injections; the goal was freedom from symptoms of hyperglycemia and frequent or severe hypoglycemia. The intensive and conventional treatment groups maintained a separation of median HbA_1c level of about 2 percentage points throughout the follow-up period (7.1% compared with 9.0%; P < 0.001) [1]. Patients' adherence to randomly assigned treatment was high; more than 97% of study time was spent receiving the assigned therapy [1].

Definition and Frequency of Severe Hypoglycemia

All episodes of severe hypoglycemia were reported to the Coordinating Center as soon as possible after their occurrence. Severe hypoglycemic episodes were defined as those in which the patient had incapacity sufficient to require the assistance of another person. In addition, the definition of severe hypoglycemia required that 1) the blood glucose level was measured and found to be less than 2.78 mmol/L or 2) the clinical manifestations were reversed by the administration of oral carbohydrate, subcutaneous glucagon, or intravenous glucose. Approximately one third of severe hypoglycemic episodes involved coma, seizure, or suspected seizure. In the intensive treatment group, 61 severe hypoglycemic episodes occurred per 100 patient-years compared with 19 episodes per 100 patient-years in the conventional treatment group (relative risk, 3.28 [95% CI, 2.65 to 4.05]; P < 0.001) [16]. In the intensive treatment group, 16 severe hypoglycemic episodes involving coma, seizure, or suspected seizure occurred per 100 patient-years compared with 5 such episodes in the conventional treatment group (relative risk, 3.02 [CI, 2.36 to 3.86]; P < 0.001) [16].

Neuropsychological Test Protocol

Neuropsychological testing was done by trained personnel at baseline; years 2, 5, and 7; and study end. The Central Neuropsychological Coding Unit trained and certified the personnel at the clinical centers. Results from the final assessment were assigned to the closest year of expected scheduled neuropsychological evaluation, up to year 9. The test protocol, which required 4 to 5 hours to complete, included the Wechsler Adult Intelligence Scale [17] for patients 16 years of age or older; the Wechsler Intelligence Scale for Children, Revised [18], for patients younger than 16 years of age; four subtests (category test, tactual performance test, trail making test, and finger tapping test) from the Halstead-Reitan Neuropsychological Battery [19]; the digit vigilance subtest from the Lafayette Clinic Repeatable Battery [20]; the logical memory and visual reproduction subtests from the Wechsler Memory Scale [21]; the arithmetic subtest from the Wide Range Achievement Test [22]; the Grooved Pegboard Test [23]; the Verbal Fluency Test [24]; and several additional specialized measures (Symbol Digit Learning Test, Four-Word Short-Term Memory Test, Embedded Figures Test) [25]. Tests were administered in a fixed order, with rest breaks scheduled at 45- to 60-minute intervals. The same battery of tests was administered at each evaluation. Capillary blood glucose levels were monitored to rule out the presence of hypoglycemia during testing.

The tests were scored by technicians at the Central Neuropsychological Coding Unit who were masked to treatment assignments. These results were sent to the Coordinating Center, where the data were entered, verified, and edited for out-of-range values and other errors.

Outcome Measurements

Clinically Rated Neuropsychological Worsening

We had initially planned that all neuropsychological test protocols would be rated by a team of expert clinicians to determine the presence of a clinically significant change in functioning between baseline and reassessment as part of a program to monitor safety. However, during the feasibility phase, we found that the neuropsychological tests could not be rated quickly enough to meet safety needs. We therefore developed a computer-based screening algorithm to identify patients with a high probability of neuropsychological worsening [26]. Protocols so identified were reviewed by two expert clinical neuropsychologists who independently rated each protocol for change from baseline. If a protocol was rated as significantly worsening since baseline, the local center was notified so that additional neurologic or psychological studies could be done and the patient could be more closely monitored. This screening process was found to have high sensitivity (100%) and specificity (87.3%) for identifying cases of significant worsening [26]. The gold standard was the expert clinicians' rating of the neuropsychological data. Although this screening process was initially implemented to monitor the safety of the DCCT participants, the results of the expert clinicians' ratings were used to define one of the outcome measures, neuropsychological worsening.

Cognitive Domains

Each neuropsychological assessment yielded more than 70 individual test scores; 25 of these had been chosen a priori to be of particular diagnostic value when applied to patients with IDDM. For each of these 25 test variables, a standardized (Z) score was calculated, with the mean and standard deviation from the baseline assessment of the DCCT cohort used as reference. These standardized scores provided a unit-free measurement of the relative improvement (positive sign) or deterioration (negative sign) in performance compared with the total group at baseline. To further reduce the number of comparisons, the 25 standardized scores were grouped into one of eight cognitive domains. Within each domain, the simple average of the standardized scores was used to represent the change from baseline; that is, equal weights were assigned to each test. The eight domains, their constituent tests, and their means and standard deviations at baseline are given in Table 2.

Statistical Analysis

Previously described quality control procedures resulted in highly reproducible test results (98.6% test-retest concordance for significant worsening [κ, 0.78; P < 0.001] [26]. The Wilcoxon rank-sum test was used to evaluate the differences between the two groups for ordinal and numeric variables [27]. The contingency chi-square test was used for categorical variables; when the sample size was small, the Fisher exact test was used [27]. For recurrent events, the crude event rates were presented as the number of events per 100 patient-years of follow-up. The variance of the crude event rate was based on a distribution-free estimator that included an adjustment for over-dispersion [28]. A spline function was used to fit a model-free smooth line to describe associations among variables [29].

The method of generalized estimating equations, a regression method for longitudinally measured repeated data, was used to assess the effects of covariates on measures of cognitive function [28]. These models included age, sex, and years of education as baseline covariates and a set of time-dependent contrast variables for time effects (year 5 compared with year 2, year 7 compared with year 2, and year 9 compared with year 2). These time-effect contrast variables were used to evaluate practice effects or familiarity effects. After adjustment for these covariates, other variables were entered to address specific study questions.

We used two types of models. For the first, the dependent variable was the standardized score of each cognitive domain, treated as a quantitative variable. The second type used a dichotomous dependent variable indicating whether the patient was clinically rated as having neuropsychological worsening. The resulting variable estimates (β) in the model for a covariate represents the average increase or decrease of the standardized score per unit change in the covariate. In the second type of analysis, the β value is the change in the log-odds of worsening per unit change in a covariate. For example, the β value for the treatment group term (intensive-conventional) represents the average difference between the two treatment groups in the standardized score or in the log-odds of worsening. The two-tailed Wald test (a Z-test using the ratio of the coefficients to its standard error) was used for tests of significance of the β values [28].

We examined the association between domain scores and hypoglycemia at fixed time points, such as year 5. We first did rank transformation of the domain scores and then used the usual general linear regression and analysis of variance method [27] to test the overall effect of hypoglycemia, adjusting for age, sex, and years of education. We also specifically tested the contrast between no event and more than five events. The rank transformation was done to reduce the effect of outliers.

We report all results that were nominally significant (P < 0.05), whether adjusted for covariates or not. However, because of the multiple tests, these nominally significant results should be interpreted with caution.

Results

Comparisons between the Intensive and Conventional Treatment Groups

Clinically Rated Neuropsychological Worsening

The results of the clinical rating of the neuropsychological assessments for each treatment group are summarized in Table 3. Because fewer patients were followed for 5, 7, and 9 years and because no new cases were rated as having clinically worsened after 5 years of follow-up, these comparisons are restricted to assessments completed at year 2 and year 5. At year 2, 9 patients in the conventional treatment group and 8 in the intensive treatment group were clinically rated as having significantly worsened. At year 5, 6 patients in the conventional and 3 in the intensive treatment group were rated as having significantly worsened. Cumulatively, at year 5, 14 patients in the conventional group were clinically rated as having worsened compared with 9 in the intensive treatment group. No significant differences were seen between treatment groups at year 2 or at year 5. Additional analyses done by the generalizing estimating equations method, which included all years of follow-up in the model, also showed no significant treatment group effect (P = 0.24).

Compared with patients who did not have clinically significant worsening, the 23 patients who were rated as having worsened significantly by year 5 were more likely to have significantly lower baseline verbal IQ scores (111.6 ± 10.5 compared with 104.5 ± 11.8; P = 0.0015) and performance IQ scores (113.5 ± 10.8 compared with 106.2 ± 16.0; P = 0.0013) as well as less education (14.1 ± 2.3 years compared with 13.1 ± 2.4 years; P = 0.034). The two groups did not, however, differ with regard to IDDM variables such as duration of diabetes, HbA_1c level at study entry, or history of hypoglycemic coma before randomization or with regard to psychosocial variables such as history of depression or alcohol intake. The lower IQs could have been a result of neurologic damage from mild repeated episodes of hypoglycemia that were not detected by the baseline history.

Review of patients' clinical characteristics, including cumulative HbA_1c level; the percentage of patients with major accidents; and the percentage of patients who developed nephropathy, retinopathy, or depression during follow-up and at the time of the occurrence of the clinically rated neuropsychological worsening, showed no significant differences between patients whose performance did and patients whose performance did not worsen (Table 4). In addition, the most severe adverse events experienced by the entire DCCT study group did not occur among the patients rated as having neuropsychological worsening. These severe events include death, loss of vision, myocardial infarction, renal failure, surgical amputation, and attempted suicide. However, the patients with neuropsychological worsening had a significantly higher frequency of psychiatric hospitalization during the study (6.01 events/100 patient-years) than did patients without neuropsychological worsening (0.71 events/100 patient-years; relative risk, 8.5 [P < 0.001]).

Although 23 patients were rated as having clinical worsening since baseline at one of their neuropsychological assessments, 16 of them had had one or more subsequent assessments, the results of which were not judged to be clinically worse than the results of the baseline assessment. Among the remaining 7 patients, further evaluation at the local clinics suggested that, except in the case of 1 patient, the worsenings could be explained by intercurrent illness, temporary life disturbances, or other factors. This patient (patient A), a 31-year-old man in the intensive therapy group, had a bicycle accident that caused severe head injuries during the third year of the trial. This patient was clinically rated as showing neuropsychological worsening at the fifth and subsequent years.

Cognitive Domains

For all domains, the relative performance of patients in the intensive therapy group was almost identical to that of patients in the conventional treatment group. The method of generalized estimating equations was used to test relative performance differences for each of the eight cognitive domains across the duration of the study for all patients. After adjustment for age, sex, years of education, and three time-dependent contrast variables to control for practice effects, we noted no significant differences between treatment groups (β values, 0.02 to 0.03; all P values more than 0.20) in six of eight cognitive domains. Treatment group had a significant or near-significant influence on two domains: immediate memory (β equals 0.06; P = 0.06) and motor speed (β equals 0.11; P = 0.004); intensive treatment was associated with improved function. However, the relative improvement in performance levels was only slightly higher among the patients receiving intensive treatment than among those receiving conventional treatment.

Figure 1 plots the Z scores for motor speed over the HbA_1c levels measured at the annual visit at the time of the neuropsychological assessment for both treatment groups combined. This figure shows that relative improvement in motor speed was associated with a lower HbA_1c level. In a regression analysis adjusting for baseline HbA_1c level, this association was significant within each year. Because the mean HbA_1c level during the trial differed significantly between the intensive and conventional treatment groups (7.1% compared with 9.0%), the difference in motor speed between groups may be attributed to the difference in HbA_1c levels. Practice effects are evident insofar as the overall level of performance tended to improve from year 2 to year 7. In the conventional treatment group, the Z score decreased by 0.057 per 1% increase in the HbA_1c level; a similar increase of the HbA_1c level in the intensive therapy group was associated with a 0.061 decrease (P < 0.001 for both treatment groups).

Comparisons between Patients with and without Severe Hypoglycemia

Clinically Rated Neuropsychological Worsening

Table 5 shows the cumulative number of severe hypoglycemic episodes that occurred before the neuropsychological assessment. The mean rate (per 100 patient-years) of severe hypoglycemia in patients with clinical worsening was not significantly higher than the rate in patients without worsening at either year 2 or year 5. Although the percentage of patients with at least one episode of severe hypoglycemia before neuropsychological assessment was slightly higher among patients clinically rated as having worsened, this difference was not statistically significant. Regression analyses also showed no significant association between the history of hypoglycemia as a time-dependent covariate and the odds of worsening.

We also examined the relation between high event rates of hypoglycemia (> 1 event/y) and clinically rated neuropsychological worsening. By year 5, only 2 of the 151 patients who had more than one event requiring assistance per year were clinically rated as having worsened; of the 24 patients with more than one episode of coma or seizure per year, 1 patient (patient A) was clinically rated as having worsened. These rates did not significantly differ from those for patients with one or fewer episodes of severe hypoglycemia per year.

Cognitive Domains

We used regression analyses to examine the association between hypoglycemia and standardized scores [Z scores] over time for the eight domains. Covariates included age, sex, years of education, and the three time-dependent contrast terms of visits. After adjustment for these variables, models included two time-dependent hypoglycemic variables: 1) whether the patient had an episode [coma or seizure] and 2) the cumulative number of hypoglycemic episodes (coma or seizure) up to the time of the neuropsychological assessment. Among all patients drawn from both treatment groups, no association was seen between the occurrence of hypoglycemia or the cumulative number of hypoglycemic episodes and performance in seven of the eight cognitive domains. Modeling the problem-solving domain showed a weak but nominally significant effect of the cumulative number of severe hypoglycemic episodes (β equals − 0.015; P = 0.047).

We used simple linear regression to examine the effect of hypoglycemia on domain scores at year 5. Patients who had assessments in year 5 were grouped on the basis of the cumulative number of previous hypoglycemic events (coma or seizure): no event, one to five events, and more than five events. Table 6 shows the domain scores according to the three hypoglycemia groups. In general, the performance level for all domains at 5 years was better than the level at baseline (all positive scores), presumably because of the presence of a learning effect. The overall difference for each domain among the three hypoglycemia groups was not statistically significant. Further, the contrast between patients who had had no event up to year 5 and those who had had more than 5 episodes did not significantly differ. We did the same analyses on data from year 2 and the cumulative number of hypoglycemic events up to year 2. Again, no differences were seen.

To further examine the potential effect of a high frequency of severe hypoglycemic events on performance, we plotted the neuropsychological results for each of the domains at year 5 for all patients who had more than one coma or seizure episode per year (n = 24) Table 6, along with the results from the rest of the patients. This is shown in Figure 2 for two arbitrarily selected domains. The 24 patients who had had at least one episode of seizure or coma per year were randomly distributed among the other patients, showing the lack of differentiation between this group of patients and all others having fewer episodes of hypoglycemia. Patient A; however, fell outside of the coordinates of the graph. We obtained a similar pattern of results (not shown) with the other cognitive domains.

Discussion

Our experience with more than 9300 patient-years of follow-up data leads us to conclude that intensive diabetes therapy, as provided by DCCT clinicians, is not associated with the development of significant cognitive impairment. Results from clinical ratings provide no evidence to show that intensive therapy increases the likelihood of clinically significant neuropsychological worsening. Moreover, our analysis of individual test scores from the extensive battery of well-known neuropsychological measures, subsumed into eight cognitive domains, provides no evidence that the performance of the intensively treated group worsened over time relative to the conventionally treated group. If anything, there was a slight but consistent improvement in performance on at least one domain—motor speed—that appears to be associated with improved metabolic control.

We could not directly address a causal relation between brain damage and hypoglycemia because both neurobehavioral performance and hypoglycemia were outcomes of the trial. Given the study design and the neurobehavioral test intervals, we could only examine the association between the two outcomes. The DCCT data provide no evidence to show that repeated episodes of severe hypoglycemia, as defined, are associated with the development of clinically significant, or subtle, cognitive impairment. Some patients who had severe hypoglycemia, such as patient A, may have developed cognitive deficits either as a direct consequence of central nervous system damage secondary to neuroglycopenia or as an indirect consequence of a hypoglycemia-mediated accident. However, in the aggregate, we saw no excess of cases of cognitive dysfunction in patients who had more than one episode of severe hypoglycemia yearly compared with patients who had fewer episodes of severe hypoglycemia over the same time period.

Our pattern of results confirms the previous report from the SDIS [12, 30]. In that prospective comparison of diabetic adults intensively and conventionally treated for 5 to 7.5 years, the SDIS investigators administered a briefer, computerized test battery and used raw test scores rather than clinical ratings of impairment. Patients in the intensive treatment group had a nearly threefold increase in the rate of severe hypoglycemia (1.1 compared with 0.4 episodes per patient per year), an outcome similar to that in our study. Although the SDIS researchers did not specifically examine the effects of hypoglycemia on cognitive functioning, results from their earlier analysis of 3-year data showed no consistent hypoglycemia-associated cognitive impairment [3]. For example, they found that patients with 2 to 4 hypoglycemic episodes showed greater deterioration on a maze test than did patients with 0 or 1 episodes and patients with 5 or more hypoglycemic episodes. Moreover, patients who had 5 or more severe hypoglycemic episodes showed a marked improvement on an auditory reaction time test compared with patients with fewer than 5 episodes. Interpretation of these results—particularly the latter finding of improved functioning after hypoglycemia—must be tempered by the relatively small sample size: Only 57 patients had 1 or more episodes of severe hypoglycemia during a 3-year period. Nevertheless, the SDIS data do not support the view that repeated episodes of severe hypoglycemia are necessarily associated with cognitive dysfunction [31].

Both our findings and those from SDIS vary from results reported previously from several cross-sectional studies. Wredling and associates [11] evaluated 17 diabetic adults who reported five or more episodes of severe hypoglycemia (defined as loss of consciousness and need for external help) during a 3-year period and compared their performance with that of 17 demographically similar diabetic adults who reported that they had never had severe hypoglycemia. Administering the same computerized test battery that was used in the SDIS, Wredling and colleagues found significant differences on 7 of 18 variables and concluded that repeated episodes of severe hypoglycemia were associated with cognitive impairment.

In their cross-sectional study of 100 diabetic adults, Deary and associates [10] administered several information-processing measures and many traditional clinical neuropsychological tests similar to those used in the DCCT. Small but statistically significant associations (correlation coefficients of 0.04 to 0.12) were shown between the estimated lifetime frequency of severe hypoglycemia and scores on the performance subtests from the revised Wechsler Adult Intelligence Scale and scores on several information-processing tests. No association was found between frequency of hypoglycemia and measures of verbal fluency or learning and memory tests. In a subsequent analysis, Deary and colleagues [13] compared the scores of these diabetic patients with those of 100 adults without diabetes. Although between-group differences were found in both verbal IQ (104.7 for persons without diabetes compared with 100.6 for patients with diabetes) and performance IQ (104.0 compared with 99.4), the difference between performance IQs was abolished when frequency of hypoglycemia was entered as a covariate. Deary and colleagues interpreted these results as evidence that a significant relation exists between the frequency of severe hypoglycemia and cognitive functioning [13]. However, as in the study by Wredling and coworkers [11], the between-group differences were small and were restricted to a few measures. Moreover, cross-sectional data must be interpreted cautiously with regard to causality. Finally, because damage to the hippocampus is commonly seen after severe hypoglycemia in rodents [32], nonhuman primates [33], and humans [34] and because memory disorders are typically associated with such damage [34, 35], the absence of significant learning and memory disturbances in the persons studied by Langan [10] and Wredling [11] and their colleagues is inconsistent with their hypothesis that cognitive impairment is necessarily a consequence of repeated hypoglycemic damage.

In summary, repeated evaluations of neuropsychological functioning of participants in the DCCT provide no evidence of significant cognitive deterioration associated with intensive therapy as practiced in the DCCT or associated with repeated episodes of severe hypoglycemia as occurred in the DCCT. This finding does not preclude the possibility that prolonged or severe hypoglycemia may cause cognitive impairment. Our conclusion must also be tempered by the knowledge that DCCT participants were 13 to 39 years of age at study entry and were followed for 3.5 to 9 years; patient volunteers with a history of frequent hypoglycemic seizure or coma were excluded from entering the trial. In addition, cognitive impairment associated with hypoglycemia may not become manifest until more time has elapsed. Further follow-up of the DCCT population should clarify this possibility. Whether the population of patients with IDDM who are more vulnerable to the occurrence of severe hypoglycemia will also be immune to cognitive deterioration, as were the DCCT volunteers, is unknown. We cannot conclude that similar results would have been obtained had young children or older adults with diabetes been studied. However, because several previous cross-sectional studies in children have shown that cognitive functioning may be adversely affected in diabetic children who had repeated episodes of even mild hypoglycemia within the first 5 or 6 years of life [36, 37], we urge caution in implementing intensive insulin therapy in young children. Although the failure to show any cognitive deficits in adult DCCT participants with frequent hypoglycemia is reassuring, hypoglycemia remains a potentially dangerous side effect of intensive therapy. Intensive therapy should be adjusted to minimize the occurrence of severe hypoglycemia.