Risk Factors for Decline in IQ in Youth With Type 1 Diabetes Over the 12 Years From Diagnosis/Illness Onset

A constant supply of glucose is critical for normal cerebral metabolism (1,2). Thus the brain is one of the major organ systems affected in type 1 diabetes, as glucose homeostasis is frequently disrupted, even in well-controlled diabetes (3). Developing brains have high cerebral energy needs associated with brain growth and neural pruning and may be more sensitive than adults to glucose fluctuations (3). It is important to document specific illness-related risk factors for brain changes in young people with type 1 diabetes. Better understanding of the impact of childhood-onset disease on the central nervous system (CNS) will facilitate evidence-based pediatric management regimens and proactive interventions to minimize academic underachievement and functional impairment.

There is a growing literature documenting pathophysiological CNS changes (4–13) and neurocognitive deficits (14–23) in youth with type 1 diabetes. However, attempts to identify specific diabetes-related risk factors for CNS compromise have produced inconclusive findings. An association between cognitive deficits and an early age of disease onset is a robust finding (14), but associations between glycemic perturbations and CNS effects are less consistent. Given the alteration in conscious state that accompanies severe episodes, the focus traditionally has been on hypoglycemia as the explanation for cognitive deficits in children with type 1 diabetes. However, neuroimaging findings are inconsistent, with some studies finding no association between hypoglycemia history and structural brain changes (4,12) while others report positive correlations, but with little consistency between the brain regions affected (5,7,9–11). Meta-analytic cognitive studies also provide contrasting findings, with one reporting positive association between hypoglycemia history and cognitive deficit (23) and another finding no association (14).

There is now increasing concern that chronic exposure to hyperglycemia may impact on the developing brain, although research findings to date are inconsistent and probably reflect, in part, difficulty in obtaining accurate reports of lifetime metabolic control, particularly when measured retrospectively. Volumetric reductions in both gray and white matter and altered diffusion tensor imaging (DTI) parameters have been associated with greater exposure to hyperglycemia in several studies (4,6,10–13), but not in others (5,7–9). Similarly, an association between long-term poor metabolic control and cognitive deficits was found in some (5,16–18,20,22) but not other (9,14) studies. There is now compelling evidence that type 1 diabetes impacts upon the developing brain, but our understanding of the causal mechanisms that underlie CNS changes is limited. This makes it difficult to weigh the relative neurotoxicity of glycemic extremes and to tailor management regimens accordingly.

Controlled, longitudinal studies are particularly informative in documenting illness-related changes in cognitive function. The only study to date that has examined within-subject change in intelligence quotient (IQ) from childhood to follow-up in early adulthood (24) is limited by the fact that initial assessment of IQ occurred years after diabetes onset. Thus participants had already been exposed to a period of glycemic dysregulation before baseline assessment. Our group has previously reported that youth with type 1 diabetes had lower full-scale IQ (FSIQ) and verbal IQ (VIQ) scores than healthy community control participants (HCs) when assessed cross-sectionally 12 years after disease onset (9). In the current study, we used data from diabetes onset and from follow-up 12 years later to examine change in IQ scores in youth with type 1 diabetes and HCs. Within the diabetes group, we examined associations between changes in IQ and age of diabetes onset, history of serious hypoglycemia, and longer-term metabolic control.

Consecutive admissions to the Royal Children’s Hospital, Melbourne, Australia, with newly diagnosed type 1 diabetes between 1990 and 1992 (n = 133), together with HCs (n = 126), stratified for age and sex, formed the original cohort. FSIQ <70 or history of CNS trauma or disease (e.g., epilepsy) were exclusions. Children with attention deficit hyperactivity disorder or learning difficulties were not excluded. Twelve years later, participants who could be located (125 youth with type 1 diabetes and 93 HCs, 94 and 74%, respectively) were invited for follow-up. A total of 106 participants with type 1 diabetes and 75 HCs consented (rates of 85 and 81%, respectively). Two participants with type 1 diabetes who developed a seizure disorder after diabetes onset were included in the current analyses.

This report is of participants who were assessed on the Wechsler Preschool and Primary Scale of Intelligence-Revised (WPPSI-R; 34 with type 1 diabetes and 21 HCs) (25) or Wechsler Intelligence Scale for Children-Revised (WISC-R; 61 with type 1 diabetes and 49 HCs) (26) at study inception (baseline) and on the Wechsler Abbreviated Scale of Intelligence (WASI) (27) at follow-up 12 years later (95 with type 1 diabetes and 70 HCs). We did not included participants with type 1 diabetes onset before age 3 years who were assessed on the Bayley Scales of Infant Development (28) at baseline (n = 10) or HCs who were recruited later to match this younger cohort and had no baseline data (n = 10). Three HCs without socioeconomic status (SES) information at baseline were excluded from analyses. The sample reported in this article comprised 95 participants with type 1 diabetes and 67 HCs.

All assessments were conducted by psychology-trained research assistants. Blood glucose levels (BGLs) of type 1 diabetes participants were determined prior to assessment by capillary sample to ensure a reading between 4 and 18 mmol/L. There were no significant associations between BGL and IQ score at either time point. This study was approved by the Human Ethics Research Committee of the Victorian Government Department of Human Services.

SES was measured using the Daniel Scale of Occupational Prestige (29). This scale is based on parental occupation and rated on a six-point scale (1 = high, 6 = low SES). The WPPSI-R (25) and the WISC-R (26) are standardized measures of general intelligence used to assess children aged <7 years and 7–16 years, respectively. The WASI is a standardized, brief measure of intelligence for individuals aged 6 to 90 years. These measures provide FSIQ, VIQ, and performance IQ (PIQ) scores (mean = 100, SD = 15). The WPPSI-R, WISC-R, and WASI use similar tasks to assess comparable skills of general intelligence, and summary scale scores are highly correlated (>0.8) (27). The use of a longitudinally assessed HC group should control for variability related to the use of different test instruments at different time points.

The age of onset of diabetes was documented. Episodes of hypoglycemia associated with seizure/coma were recorded and corroborated through medical records. HbA_1c measurements from diagnosis were obtained for each patient (number of measurements, range = 9–55, median = 36) from hospital and clinic databases. HbA_1c levels prior to 1994 were measured using Bio-Rad (Hercules, CA) affinity column chromatography method and after 1994 using a Bayer (Calabria, Barcelona, Spain) DCA 2000 latex immunoagglutination method. Equilibration between the two methods was achieved using the following: HbA_1c (measured by immunoagglutination) = 1.2 + 1.07 × HbA_1c (measured by column chromatography). This equation was derived after regression analyses were performed on 50 patients with paired samples using both measuring techniques. The percentage of total time from diagnosis that HbA_1c was ≥9.0% (74.9 mmol/mol) was calculated to estimate overall metabolic control. Previous episodes of diabetic ketoacidosis (DKA) were collected via self-report.

For the purpose of illustration and reporting mean change in IQ scores, each illness-related risk was dichotomized into two groups: diabetes onset <6 years (early-onset diabetes) and onset ≥6 years (late-onset diabetes); participants with one or more episode of hypoglycemia with seizure/coma (seizure history) versus no history of seizures/coma (no seizure); and participants with HbA_1c readings ≥9.0% (74.9 mmol/mol) for more than one-third of lifetime measurements (poor control) and those with HbA_1c ≥9.0% for less than one-third of lifetime measurements (good control).

For each participant, a change in VIQ, PIQ, and FSIQ score from baseline to 12-year follow-up was calculated. Positive change scores indicated IQ increase over time; negative change scores indicated decline. ANCOVA was used to investigate the difference in changes in VIQ, PIQ, and FSIQ scores between participants with type 1 diabetes and HCs, covarying for IQ and SES at baseline and length of follow-up. Next, within the type 1 diabetes group, three linear regression analyses were conducted with change in VIQ, PIQ, and FSIQ scores, respectively, entered as the dependent variable. In the first block, age of diabetes onset (onset), number of hypoglycemic episodes (seizures), and percentage of time that HbA_1c was ≥9.0% (metabolic control) were entered as predictor variables. IQ and SES at baseline and length of follow-up were entered as covariates. In block two, interaction terms of the diabetes-related variables were entered (onset × metabolic control; onset × seizures; metabolic control × seizures). Here we interpret β-values.

Because participants with an earlier age of onset of diabetes <7 years were assessed by WPPSI-R at baseline and those ≥7 years were assessed by WISC-R, psychometric properties of the tests used may confound analyses. To overcome this, we tested whether change in IQ differed by baseline test (WPPSI-R vs. WISC-R) in HCs, any difference assumed to be a psychometric effect. HCs with WPPSI-R showed significantly greater loss in PIQ than HCs with WISC-R. To account for this, we created a test-standardized change in PIQ score for each participant with diabetes, which was the observed score minus the expected score. The observed score was the change in PIQ over the 12-year follow-up. The expected score was the mean decline in PIQ over time experienced by the HC group for WISC-R and WPPSI-R participants, respectively (essentially the psychometric effect). The test-standardized score for PIQ was used as the dependent variable in regression.

To test potential effects of fewer HbA_1c measurements, we selected ≥30 HbA_1c values over the follow-up period as a reasonable number of measurements (an average of 2.5 per year of diabetes). The sample with type 1 diabetes was dichotomized into those with ≥30 HbA_1c values and <30 values. Correlations were conducted between the percentage of time HbA_1c was ≥9.0%, and change in IQ scores for each group separately. Next, regression analyses were repeated with those participants with ≥30 HbA_1c values only.

To test for effects of concurrent BGLs on cognitive performance, the mean BGL for each illness-related risk group (early or late onset, seizure history or no seizures, poor or good control) were compared using independent t tests. Additionally, regression analyses were repeated covarying for BGLs at baseline and 12-year follow-up assessment.

Episodes of DKA were indexed via self-report but were not verified on medical records. Thus analyses of the effect of DKA on change in IQ are exploratory, and results should be interpreted with caution. We conducted ANCOVAs to explore differences in change in IQ for participants reporting one or more episode of DKA and those who did not, covaried for IQ and SES at baseline and length of follow-up.

Sample characteristics are presented in Table 1. At baseline, participants with type 1 diabetes did not differ significantly from HCs on age, sex, SES, or IQ scores. As previously reported (9), participants with type 1 diabetes had significantly lower VIQ and FSIQ scores than HCs at follow-up.

Mean changes in IQ scores for participants with type 1 diabetes, HCs, and each risk group are presented in Table 2. Those with type 1 diabetes showed significantly greater decline in VIQ and FSIQ over time than HCs (change in VIQ, F[1, 157] = 6.47, P = 0.01; change in FSIQ, F[1, 157] = 5.22, P = 0.02). There was no significant group difference in change in PIQ (F[1, 157] = 1.28; P = 0.3).

Regression analyses within the type 1 diabetes group are presented in Table 3. For any given baseline score, these models predict the IQ score at 12-year follow-up as predicted by each of the illness-related variables and covariates.

For each regression model, the overall model was significant. Baseline IQ and SES were always significantly associated with change in IQ, and length of follow-up was never significantly associated with change in IQ. A younger age at diabetes onset was significantly associated with negative change in PIQ and FSIQ. Based on the model with change in PIQ as the dependent variable, for any given PIQ score at baseline, the predicted follow-up PIQ score would be reduced by 2.36 points for each year earlier that diabetes onset occurred. Similarly, based on the model for change in FSIQ, for any given baseline FSIQ score, the predicted follow-up FSIQ score would be reduced by 0.97 points for each year earlier that diabetes onset occurred. A greater number of hypoglycemic seizures was significantly associated with negative change in VIQ. Thus, according to the model with VIQ as the dependent variable, for any given baseline VIQ score, VIQ at follow-up would be reduced by 1.19 points for each additional seizure. Metabolic control did not significantly predict change in IQ score. None of the interaction terms were significant.

Correlations between percentage of time HbA_1c was ≥9.0% and change in VIQ, PIQ, and FSIQ conducted separately for those participants with ≥30 HbA_1c values (n = 67; 70%) and those with <30 values were low for both groups (r <0.2) and not statistically significant. All regression analyses were repeated including only participants with ≥30 HbA_1c values. Metabolic control still did not contribute significantly to change in VIQ, PIQ, and FSIQ (P > 0.9; P = 0.2; P = 0.4, respectively).

Mean BGL for participants with type 1 diabetes and each illness-related risk group (early or late onset, seizure history or no seizures, poor or good control) are reported in Table 2. The poor control group had significantly higher BGL prior to assessment at follow-up compared with the good control group (t[81.5] = 3.64; P < 0.001). There were no other significant group differences. Regression analyses were repeated covarying for BGL at baseline and then BGL at 12-year follow-up. Results were similar and are not reported here.

Thirty-one participants (32.6%) reported one or more episode of DKA (range 1–8; data missing on two participants). Means, SDs, and comparisons are presented in Table 4. Participants with one or more episode of DKA showed a greater loss in VIQ, PIQ, and FSIQ than participants without DKA episodes, but these were not statistically significant.

To our knowledge, this is the first study to document change in IQ from diagnosis with type 1 diabetes in childhood to follow-up in young adulthood. Some decline in IQ scores was evident in both type 1 diabetes and HC groups. This probably reflects, in part, the use of a more recently normed test at follow-up. However, the decline in VIQ and FSIQ was greater for participants with type 1 diabetes compared with HCs. PIQ did not appear to be negatively impacted by diagnosed diabetes. There was evidence of selective impact of specific illness risk factors on IQ. Early age of diabetes onset was associated with a decline in PIQ and FSIQ scores, while a history of hypoglycemic seizures was associated with a decline in VIQ. These findings are consistent with previous reports documenting adverse effects on cognition in those with diabetes onset very early in life (14,17,20) and in those who experience serious hypoglycemia (5,17,20,22,23). Thus the associations reported are not novel, but our access to IQ scores from diagnosis and follow-up provides particularly compelling evidence that the timing of diabetes onset and exposure to serious hypoglycemia are significant factors in the cognitive sequelae of type 1 diabetes.

To examine the effect of age of onset on change in PIQ, we used a test-standardized change in PIQ score to account for the psychometric effects of different baseline IQ tests. We feel confident that declines in PIQ and FSIQ are real effects of an earlier age of diabetes onset. We also note that the current findings are consistent with our cross-sectional findings at 12-year follow-up reported previously (9).

The early-onset effect identified in the meta-analysis conducted by Gaudieri et al. (14), and replicated by others (17,20), has traditionally been interpreted as a surrogate for the effect of hypoglycemia on an immature brain. Our findings do not support this hypothesis. Very few of our participants (n = 5) experienced hypoglycemic seizure early in life (30), and our regression analyses do not support an interaction between early-onset diabetes and known serious hypoglycemia, although we cannot rule out unrecognized event, particularly at night. While we demonstrated a very clear association between early-onset disease and deterioration in PIQ and FSIQ, as well as deficits in a range of specific cognitive deficits in a previous report (20), our neuroimaging findings provide a different picture. Contrary to our expectations, the early-onset subgroup of the cohort showed less evidence of brain changes than those with later-onset diabetes (31).

There are several possible explanations for these discrepant findings. Early exposure to glycemic perturbations may disrupt a later stage of neurodevelopment, a “sleeper” effect not yet evident in our younger participants. Alternatively, our neuroimaging techniques (volumetric measurement of white and gray matter) were insufficiently sensitive to detect subtle changes in brain architecture and connectivity in the early-onset subgroup. Recent studies using more sophisticated neuroimaging techniques such as DTI have shown altered diffusivity and anisotropy (a measure of the integrity of neural circuitry used in DTI) in young children with type 1 diabetes (4,6,12), suggesting compromised integrity of white matter structures critical for neural connectivity and efficient information processing.

In our sample, a history of hypoglycemic seizure/coma compromised VIQ, consistent with other data documenting an association between poorer language skills and positive seizure history (20,21,23,32). These findings suggest that clinical management regimens designed to avoid such events are warranted well into adolescence, despite the apparent resilience to hypoglycemic events shown by adults (33). We have previously suggested that the specific sensitivity of language skills to hypoglycemia may reflect a semantic memory deficit secondary to hypoglycemia-related hippocampal damage (30). However, our neuroimaging findings 12 years after disease onset (9) showed reduced volumes in the thalamus, a brain structure critical for arousal and vigilance, in those who had experienced a hypoglycemic seizure. Furthermore, the compromised white matter integrity demonstrated in recent DTI studies (4,6,12) is likely to degrade efficient processing of incoming information. Language development is highly sensitive to environmental input, particularly educational experience, thus subtle compromise of attention and information processing skills may lead to cumulative deficits in verbal knowledge stores over childhood.

We did not demonstrate an association between chronic poor metabolic control and changes in IQ. This null finding contrasts recent reports demonstrating the negative impact of poorly controlled diabetes on cognition (5,16–18,20,22), suggesting that it would be premature to discount the neurotoxic impact of hyperglycemia on the developing brain. It is possible that our measure of long-term metabolic control (percentage of time since diagnosis that HbA_1c was ≥9.0%) was insufficiently sensitive to discern changes in IQ. The number of available measurements of HbA_1c also varied widely across participants. Although our analyses suggested that people with fewer values were not driving the null findings, it would seem likely, intuitively, that those with missed clinic appointments and hence missing HbA_1c values, were more poorly controlled than regular attendees. Furthermore, participants with a known history of poor metabolic control were overrepresented in attrition from the sample at 12-year follow-up, undermining the power of our study to detect hyperglycemia-related changes in IQ.

There is now growing evidence on the potential effect of DKA on the developing brain, including cognitive ability (34). Self-reported DKA was not corroborated by scrutiny of medical records. Thus we lack complete confidence that every episode of DKA was reported. Exploratory analyses showed that although participants with and without DKA history did not differ significantly on change in IQ, those with a history of DKA showed greater mean IQ loss in absolute terms. This suggests a possible effect that should be studied in future research using more robust samples and more comprehensive DKA history.

It is necessary to interpret these findings in light of limitations of our study. We did not collect data on academic performance throughout the schooling of this cohort, although fewer participants with type 1 diabetes than HCs completed year 12, the preuniversity year of schooling in Australia (35), providing some external validation of our findings. Moreover, this study does not provide information on the timing or trajectory of changes in IQ, which limits the ability to implement specific interventions during the highest-risk periods.

A comprehensive understanding of the impact of type 1 diabetes on CNS structure and function remains an ongoing challenge. Our findings show that VIQ and FSIQ are negatively impacted over time in young people with type 1 diabetes. We demonstrated declines in IQ associated with the specific diabetes-related risks of early diabetes onset and history of hypoglycemic seizure, but not hyperglycemia. Recent in vitro and in vivo studies suggest that in addition to glycemic extremes, constantly fluctuating glycemic levels and disturbances in insulin homeostasis and counterregulatory hormone responses may exert toxic effects on the developing CNS (2,36). Laboratory studies that manipulate metabolic variables under experimentally controlled conditions may be needed to test explanatory models if we are to fully understand the impact of type 1 diabetes on the developing brain. More frequent assessment of youth with type 1 diabetes will help clarify the timing and trajectory of functional and structural brain changes to optimize intervention.

Acknowledgments. The authors thank the participants and their families for their commitment to this longitudinal study.

Funding. This research was supported by Juvenile Diabetes Research Foundation Internationalhttp://dx.doi.org/10.13039/100000901 (JDRF grant 1-2003-135). A.L. is funded by a National Health and Medical Research Council (Australia) Early Career Fellowship (1072593). E.A.N. is currently receiving research support from a JDRF Clinical Trial Network Pilot and Feasibility Study grant.

Duality of Interest. E.A.N. is currently receiving research support from Pfizer Australia (grant WI179088). G.A.W. reports receiving consulting fees from Ipsen and equity in Antisense Therapeutics Limited (Australia) and Neuren (Australia) and is currently receiving research support from Novo Nordisk and Sandoz. F.J.C. reports receiving lecture fees and/or consultancy fees from Novo Nordisk, Eli Lilly, and Medtronic as well as research grants from Eli Lilly and Medtronic. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.L. and E.A.N. were primarily responsible for data collection and analysis and the drafting of the manuscript. G.A.W. and F.J.C. contributed to the interpretation of the findings and the final editing of the manuscript. E.A.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.