Participants

At the age of 13 to 19 years offspring of women in the register were invited to a follow-up examination (singletons and one child per mother) (n = 746). A control group matched according to gender, age and postal code (as a marker of socioeconomic status) was identified through the Danish Central Civil Registration System. At inclusion all control participants and/or their parents were asked about a prior history of gestational diabetes, and in addition all available obstetric records were scrutinized for information on gestational diabetes.

We excluded 16 diabetes exposed offspring due to: maternal diabetes diagnosis had been reclassified to MODY or type 2 diabetes (n = 12), no contact between mother and child (n = 2), drug abuse (n = 1), pregnancy at the time of recruitment (n = 1).

Reasons for exclusion of control offspring were: adopted child (n = 4), the obstetric record revealed a maternal diagnosis of gestational diabetes (n = 3), birth place outside of Denmark (n = 1).

The EPICOM study examined 278 diabetes exposed offspring (exposed group, 37.1% of invited offspring) and 303 unexposed offspring (control group, 15.7% of invited offspring). Participants were all born in Denmark and mainly of white European ethnicity (98.8%). A total of 269 (36.1% of invited) exposed and 293 (15.3% of invited) control offspring completed the computer based attention test (Fig 1). All non-completions of the test (n = 9 in the diabetes exposed group, n = 10 in the control group) were due to technical difficulties related to computer problems. The inclusion process and the follow-up study have recently been described in detail [12, 25].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of the study.

https://doi.org/10.1371/journal.pone.0169308.g001

The protocol was in accordance with the Helsinki Declaration and approved by the regional ethical committee of Central Denmark (M-20110239). We obtained written informed consent from all participants or their parents if the participant was younger than 18 years.

Principal component analysis

Conners CPT-II provides a large number of individual test measures. To reduce the number of variables and to obtain a clearer picture of components of attention, we followed Egeland & Kovalik-Gran [29] and subjected the correlations among the different test performance indicators to a principal component analysis. Thus we analyzed T scores on the following variables: number of omissions, commissions, perseverations, hit reaction time, hit reaction time standard error (HR consistency), variability of standard errors or change in consistency style), hit reaction time by block (the slope of change in reaction time over six blocks), hit reaction time standard error by block (change in consistency over time), hit reaction time by differing inter-stimulus intervals and hit standard error by inter-stimulus interval (whether longer intervals are associated with more variance in performance) and response style. Like Egeland & Kovalic-Gran [29] we did not include detectability in the principal component analysis, but unlike these authors measures of change in omissions and commissions were not available for our analysis.

Only three factors had eigenvalues larger than 1, but varimax rotation with three, four and five retained components showed a pattern similar to that obtained by Egeland & Kovalic for only five components. Thus, the first five components explained 83% of the variance, and the rotated matrix of loading showed the following attention factors:

Factor 1, Focused attention: defined by the loadings of omissions, hit reaction time standard error, variability and perseverations. Factor 1 explained 24.0% of the variance.

Factor 2, Vigilance: defined by the loadings of hit reaction time by differing inter stimulus intervals and hit reaction time standard error by inter stimulus interval. Factor 2 explained 17.5% of the variance.

Factor 3, Hyperactivity/impulsivity: defined by the loading of commissions and hit reaction time. Factor 3 explained 16.1% of the variance.

Factor 4, Sustained attention: defined by the loadings of hit reaction time by block, hit reaction time standard error by block. Factor 4 explained 15.0% of the variance.

Factor 5, Response style: defined by the loading of response style. Factor 5 explained 10.3% of the variance.

The first four factors clearly corresponded to four of the Egeland & Kovalic factors with the same test scores having high loadings on the same factors except for the response style test score. Egeland & Kovalic called their fifth factor change in control and this fifth factor reflected the measures of change in omissions and commissions which were not included in our analysis. In their analysis Response Style loaded together with commissions and hit reaction time on the hyperactivity/impulsivity factors, but this was not the case in our analyses with rotation of 4 or 5 factors.

Variables

Statistical analysis

Continuous data with normal distribution are presented as mean and standard deviation [31], while continuous data with skewed distribution are presented as median and interquartile range. Comparisons of groups were performed using student`s T-test, Mann-Whitney, Chi-squared and Fisher`s exact tests where appropriate.

We adjusted for confounders and mediators on each principal component factor in multivariate linear regression analyses with maternal diabetes as the independent variable and each of the principal component factors as outcome. The regression coefficient β with 95% confidence interval corresponds to the mean difference between the diabetes exposed and the control group.

Statistical analyses were performed with IBM statistics SPSS version 22, with a significance level of 0.05. Correction for multiple testing with the Bonferroni method was applied in analysis of our five principal components and in these tests the level of significance was 0.05/5 = 0.01.

Attention scores

Differences between exposed and control offspring were not found in any of the individual test measures derived from Conners CPT II (Table 2). Similarly, the multivariate regression analysis of attention factors did not show significant differences between the two groups in either crude or adjusted analyses of attention factors 1 to 4. The exposed offspring had a marginally higher score on the fifth factor (Response style) in crude analysis (β = 0.169, p = 0.05) and in analyses adjusted for potential confounders (β = 0.238, p = 0.02) and mediators (β = 0.267, p = 0.05) (Table 3). Differences in the fifth factor were no longer significant after correction for multiple testing.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. T-scores in individual test measures in Conners Continuous Performance Test II.

Offspring of women with type 1 diabetes (n = 269) compared to an unexposed control group (n = 293).

https://doi.org/10.1371/journal.pone.0169308.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Regression analyses of factor scores from Conners Continuous Performance Test-II.

Offspring of women with type 1 diabetes (n = 269) compared to an unexposed control group (n = 293).

https://doi.org/10.1371/journal.pone.0169308.t003

No interaction between gender and exposure was found. Gender differences were evaluated in a model controlling for exposure group and a higher male score was found for factor 3 (hyperactivity/impulsivity) (β = 0.33, p<0.001), but no significant differences in the remaining factors.

We did not observe any significant interaction between maternal diabetes and low parental education on the risk of attention deficits in the offspring.

We found low correlations between age at follow up and each of the five factors: correlation coefficients ranged from -0.13 to 0.16.

Strengths and limitations

One of the major strength of this study is the use of a large, well-characterized and prospectively followed cohort of adolescent children of women with type 1 diabetes and a control group identified from the background population, both with detailed baseline information available.

In order to describe different aspects of attention we applied a high-signal-to-noise continuous performance test, a widely used test of attention deficits, in particular ADHD, and we identified attention factors in principal component analyses. Unfortunately, other sources of information on attention performance were not available. Impaired executive function and cognitive flexibility are known to be present in individuals with ADHD [28], and performance based tests of these functions as well as subjective ratings (parental and/or self-reported) of attention, impulsivity, executive function and cognitive flexibility would have enabled a more comprehensive assessment of attention deficits and cognition in our cohort. We were not able to specify the underlying diagnosis of the attention deficit (e.g. ADHD) with the applied test, since it would require a more comprehensive examination including diagnostic interviews and rating scales.

We did consider some of the potential problems with the use of Conners CPT II in our analyses, and chose not to include the overall confidence index. The identification of attention factors was done to focus on the different aspects of attention and to have a basis for deriving composite measures presumably more reliable than the individual test scores.

A limitation to our study was the risk of selection bias owing to the limited and different proportions of participation in our two groups. We have previously shown that both groups had intelligence scores above the population mean indicating selection from the better performing part of the population [12], which also might apply for the continuous performance test used to assess attention deficits. A systematic difference of participation between our groups would most likely be associated to positive selection of the control group, implying larger group differences. Thus, elimination of selection bias would result in differences supporting our null-hypothesis, which is equivalent to our conclusion.

Low socioeconomic status is known to have a strong effect on the risk of development of attention deficits in the offspring through different pathways [7]. The effects of maternal diabetes has been suggested to be more pronounced in low SES environments, which might explain the discrepancy between our findings and findings in similar studies due to a possibility of sample bias in our study [8–11]. Participation in a follow-up study with a comprehensive examination program requires personal resources and non-participation is in general associated with poorer health and lower socioeconomic status [34]. In our study the interaction between maternal diabetes and low parental educational level was not significant, and thus there was no evidence of stronger effects on attention in children from families with low parental educational level.

Finally, statistical power should be sufficient with 269 participants in the exposed group and 293 in the control group. Thus, the statistical power would be 84% for detecting a group difference of one quarter of a standard deviation.

Other studies

In a nested case-control study, including individuals from the Norwegian Birth and Prescription Databases, only 3.0% (n = 88) of diabetes exposed offspring had been prescribed ADHD medication in the period 2004–2012 [8]. However, offspring exposed to maternal type 1 diabetes had higher odds of ADHD (OR = 1.7, 95% CI = 1.3–2.1) in models adjusted for parental diagnosis of ADHD and perinatal and socioeconomic confounders. We found a difference between our groups in the use of ADHD medication: 2.2% of the diabetes exposed offspring and none in the control group had self-reported use of ADHD medication at the time of follow-up. This could indicate a difference in attention deficits not revealed by administration of Conners CPT-II. Medication might reduce symptoms in the affected individuals and modify test scores towards the normal range. Sample bias with selection of the better performing part of the population in both groups is supported by these numbers, as the population prevalence of attention deficits has been estimated to 3–4% [1]. The controls represent the background population and we would expect some individuals with attention deficits in this group. Both groups could include individuals with attention deficits without current use of ADHD medication. We did not have permission to corroborate the self-reported information with information from the Danish National Prescription Register or from the National Patient Register on admissions to psychiatric departments.

Increased risk of ADHD symptoms in offspring exposed to maternal gestational diabetes (GDM) has been described previously [9, 10], in particular when GDM was present together with low socioeconomic status. One study combined a battery of neuropsychological tests with parental interviews on 212 preschool children (GDM offspring, n = 21) in New York and found a small increased risk in those exposed to maternal GDM. When GDM was present with low socioeconomic status the risk of ADHD at age six was synergistically increased suggesting an interaction of these risk factors [10]. Similar findings are reported in a German nationwide survey of children and adolescents [9].

In line with this, a register-based study on the EPICOM cohort found that offspring born to mothers with poor glycemic control in pregnancy attained lower grades in primary school compared to matched controls, while those born to mother with good glycemic control obtained better grades than their matched controls in an analysis adjusted for parental educational level [35]. Achievement of good glycemic control and management of a complex chronic disease are related to a high level of personal and social resources, which most likely will be reflected in the parenting role and upbringing of the child. However, in this study of attention deficits, interaction between parental educational level and exposure to maternal diabetes did not affect measures of attention in the offspring.

It is possible that not only intrauterine exposure to diabetes but also a shared genetic susceptibility to both type 1 diabetes and ADHD could influence attention in the offspring. A German register study found an increased frequency of ADHD in children with type 1 diabetes compared to those without, while a Norwegian study found no association between type 1 diabetes and ADHD [36, 37]. Offspring of a mother with type 1 diabetes have a diabetes lifetime risk of 3% [38], and we included three diabetes exposed offspring with type 1 diabetes in our cohort. All offspring without a prior diabetes diagnosis completed an oral glucose tolerance test at the day of cognitive testing, ensuring that all individuals with diabetes should be identified. With only three cases of type 1 diabetes genetic association should only be a minor contributor to our results.