Question:

Association of Type 1 Diabetes With Standardized Test Scores of Danish

Schoolchildren
Introduction
Type 1 diabetes leads to microvascular and macrovascular complications 1 and increased
risk of cardiovascular mortality. The elevated risk of neuroglycopenia and hyperglycemia
associated with diabetes has been suggested to affect cognitive performance, 2 but results
are mixed. Some studies have linked diabetes in children to lower academic performance, 3-
7
 while other studies have found no differences in performance when comparing children
with and without diabetes.8-11 Other studies have focused on the association between
adverse diabetes outcomes such as hypoglycemia, diabetic ketoacidosis, and poor
metabolic control and cognitive skills. A recent study found that hypoglycemia adversely
affects the working memory of the brain.12 Furthermore, research has shown that mild
hypoglycemia has a negative effect on language processing, 13 and poor metabolic control
has been associated with lower academic achievement. 4,8,9,14 However, most of these studies
were conducted on smaller, nonrandom samples of patients with diabetes.
In Denmark, all children in public schools are tested in reading and math using standardized
tests. The primary objective of this study was to compare test scores for children with
diabetes with scores for children without diabetes, with and without adjustment for
socioeconomic status.
Methods
The study was approved by the Danish Data Protection Agency. The Danish Registry of
Childhood and Adolescent Diabetes (DanDiabKids) was approved by the Danish Health
Data Authority (file No. 14/915976). These approvals constitute the necessary legal
requirements, and informed consent is not required.
This study was a population-based retrospective registry study of all children attending
grades 2, 3, 4, 6, and 8 in Danish public schools from January 1, 2011, to December 31,
2015. Because tests were only mandatory for students in public schools, the sample was
restricted to students who attended public schools. Students were tested in reading in
grades 2, 4, 6, and 8 and in math in grades 3 and 6.
Students were initially identified through Statistics Denmark and subsequently linked with
background information from different administrative registries, based on unique personal
identity numbers. These data were then augmented with test scores from the Danish
National Tests. The DanDiabKids database15 was used to identify the population of children
diagnosed with diabetes (including yearly measures of hemoglobin A 1c [HbA1c]). Children
with diabetes who had at least 1 test score and children with diabetes who had at least 1
HbA1c measurement before a test were included. DanDiabKids is a national database
containing information based on patient journals, clinical examination, and blood samples.
Data are uploaded to DanDiabKids by clinicians from all pediatric clinics in Denmark, in
connection to yearly visits scheduled in close proximity to each patient's birthday.
Outcome Measures
The pooled reading and math test score was the primary outcome. Each test evaluates a
student's abilities within 3 profile areas. In reading, these are language comprehension,
decoding, and reading comprehension. In math, they are numbers and algebra, geometry,
and mathematics in use.16 The tests are computer-based, adaptive (students are presented
with questions of varying difficulty drawn from a national question bank based on a
computerized continuous assessment of the student's proficiency level), and objective
(teachers cannot influence scoring or questions asked). Tests are initially scored on a
computer according to a Rasch model17 within each profile area18 and then mapped (by the
Danish Ministry of Education) into a single score in math and a single score in reading, with
scores ranging between 1 and 100.
The tests have been shown to be highly predictive of exit examination grades and have
been used for other research purposes.18-21 The tests are considered low stakes compared
with exit examinations because they are only used to provide feedback on learning progress
to students and not used for sanctioning of teachers or schools.
Because socioeconomic status can influence academic outcomes, the association between
diabetes and test scores was adjusted for this potential confounder. In addition, subgroup
analyses of the primary outcome were performed to assess factors that might affect the
association, including early onset of diabetes, duration of diabetes, and diabetes control
measured by HbA1c level and presence of diabetic ketoacidosis.
Secondary outcomes were investigated only in children with diabetes to assess whether
diabetes control might influence test scores, including the association between HbA 1c level,
diabetic ketoacidosis, and hypoglycemia and the pooled test score.
All outcomes and analyses were repeated for math and reading test scores separately.
Clinical Measures
All clinical measures were obtained from DanDiabKids, including diabetes type, HbA 1c level,
self-care (including self-monitoring of blood glucose levels), treatment (including pen or
pump use; insulin dose), episodes of severe hypoglycemia or diabetic ketoacidosis, and
other clinical measures (eg, blood pressure, body mass index). 15 HbA1c levels were
measured centrally using the standardized method described by the International
Federation of Clinical Chemistry and determined using high-pressure liquid chromatography
(Tosoh Bioscience), with a reference range of 4.3% to 5.8%. Severe hypoglycemia and
diabetic ketoacidosis were defined according to the International Society for Pediatric and
Adolescent Diabetes guidelines.22 Severe hypoglycemia was defined as episodes with loss
of consciousness or seizure with blood glucose level less than 63 mg/dL (3.5 mmol/L) or
regain of consciousness after treatment with glucagon or glucose. Episodes of severe
hypoglycemia were the number of self-reported episodes treated at home and
hospitalizations for hypoglycemia the last year before data upload. Episodes of diabetic
ketoacidosis were the registered number of hospital admissions for diabetic ketoacidosis
the last year before data upload.
The mean value of the 3 latest HbA1c measurements (recorded yearly) before the school test
was used as the measure of HbA1c level (if fewer measurements were available, the mean
of those was used). The treatment target for Danish children with diabetes was HbA 1c level
less than 7.5%. Poor control was defined as HbA 1c level greater than 8.6%, which is the
cutoff used by several pediatric clinics in Denmark as an indicator for the need for extra
visits or psychosocial interventions.
Diabetic ketoacidosis and hypoglycemia were recorded if the child had any episode of either
diabetic ketoacidosis (dichotomous) or severe hypoglycemia (dichotomous) in the same
time window as the HbA1c measures. Dates of birth, diabetes onset, and school test were
used to calculate age at onset and diabetes duration at school test.
Socioeconomic Status
Data on socioeconomic covariates were obtained from administrative registries
with Statistics Denmark, including information on sex and birth order for all children. For
parents, information included age, income, highest completed education, number of children
in the household, marital status, residential zip code, whether the parent had insulin-
dependent diabetes measured by insulin prescription claims, and whether the parent was
an immigrant or descendent. Immigrants are Danish residents not born in Denmark, with
neither of their parents born in Denmark. Descendants are Danish residents born in
Denmark, with neither of their parents born in Denmark. All socioeconomic covariates were
measured when the child was 5 years old.
Statistical Analysis
Means of selected variables were compared between children with and without diabetes.
Linear regression models were estimated to investigate the relationship between diabetes
and test scores. Diabetes was coded with the value 1 for children with a confirmed
diagnosis of diabetes and 0 for all other observations. Children who developed diabetes
between tests were coded as 0 before onset and 1 after onset. The regression-based
comparison between children with and without diabetes was performed considering test
scores in math and reading as distinct observations, which were then pooled and analyzed
jointly in the regression analyses. Cluster-robust standard errors were used to adjust for
within-individual correlation attributable to multiple observations per child (a child could
possibly be observed with several test scores over time and test topic). For test scores to be
comparable across grades and topics, all regression models were adjusted for grade-,
topic-, and year-specific effects. Additionally, models were adjusted for socioeconomic
status.
To assess the robustness of these results, a case-sibling analysis was performed. Children
with diabetes were compared with their siblings without diabetes using linear regression
models and adding maternal fixed effects. Additionally, a grade- and topic-specific
comparison between children with and without diabetes was performed by repeating the
linear regression analysis separately by grade and topic.
Linear regression was also used to compare subgroups of children with diabetes to children
without diabetes. To analyze aspects of associations with early disease onset and test
scores, children diagnosed before school start (before age 6 years) were considered in 1
subgroup. Associations of diabetes duration and test scores were investigated in the
subgroup of children with disease duration longer than 4 years (cutoff identified post hoc
from mean value in data). To identify associations of diabetes control and test scores,
subgroups of children with mean HbA1c levels greater than 7.5% and greater than 8.6%
were considered. The last subgroup consisted of children with diabetic ketoacidosis at
onset. Only test score observations for which the subgroup criteria were met at the point of
the test were included in each subgroup. These models were estimated with and without
adjusting for socioeconomic status. Differences between subgroups were tested by
estimating the linear regressions jointly (Seemingly Unrelated Regression Equations) and
then testing for equality of coefficients of interest across regressions.
For secondary outcomes among children with diabetes, we assessed the relationship
between HbA1c levels and test scores by estimating kernel-weighted local means ("lpoly"
command in Stata), with and without adjustment for socioeconomic status. Furthermore, we
estimated linear regression models to assess the linear relationship between test scores
and HbA1c levels, diabetic ketoacidosis, and hypoglycemia within the group of children with
diabetes (with and without adjustment for socioeconomic status).
To assess the importance of missing data on test scores, all children with missing test
scores were assigned the lowest possible score, and the main analysis re-estimated to
determine if conclusions changed. Among children with diabetes, HbA 1c level was compared
by missing or nonmissing test score. Among children with diabetes, test scores were
compared by missing or nonmissing values of HbA 1c.
The primary analysis was replicated, discarding observations in which diabetes onset was
within 6 months from the day of the test. All main analyses and the kernel-weighted local
mean analysis of test scores and HbA1c level among children with diabetes were
prespecified. The subgroup analyses were specified post hoc.
All statistical analyses were performed using Stata 15 (StataCorp). All statistical testing was
2-sided, and P < .05 was considered statistically significant. Because of the potential for
type I error due to multiple comparisons, findings for analyses of secondary end points
should be interpreted as exploratory.
Results
Of 744 516 children attending grades 2 through 8 overall, 2473 had diabetes. Of these
children, 631 620 (85%) attended public schools, and of these, 2031 had diabetes. Children
with diabetes were not more likely to attend public schools than children without diabetes
(odds ratio [OR], 0.97 [95% CI, 0.88 to 1.07]). A total of 608 655 students took at least 1
test, and 1 561 770 test scores were identified (n = 1 037 006 in reading and n = 524 764 in
math). Of the children with diabetes, 1878 had at least 1 test score (n = 4234 test scores)
and 1729 had HbA1c levels measured before a test (n = 3684 test scores). Children with
diabetes were more likely to miss a test than children without diabetes (6.3% vs 3.8%;
difference, 2.5 percentage points [95% CI, 1.8 to 3.2]).
Mean age of the study population was 10.31 (SD, 2.42) years (at first test), and 51% were
male. The comparison of background characteristics between children with and without
diabetes (Table 1) showed that fathers of children with diabetes were less likely to have a
master's degree (8.5% vs 10.6%; difference, −2.1 percentage points [95% CI, −3.3 to −0.8])
or be immigrants or descendants (12.5% vs 14.9%; difference, −2.5 percentage points [95%
CI, −4.1 to −0.9]) and were more likely to have insulin-dependent diabetes (6.5% vs 0.8%;
difference, 5.8 percentage points [95% CI, 4.7 to 6.8]). Mothers of children with diabetes
were less likely to be immigrants or descendants (11.4% vs 14.7%; difference, −3.2
percentage points [95% CI, −4.6 to −1.8]) and more likely to have insulin-dependent
diabetes (2.5% vs 0.5%; difference, 2.1 percentage points [95% CI, 1.4 to 2.7]). Among
children with diabetes, mean duration of diabetes was 4.5 (SD, 3.3) years (first percentile,
0.1 year; 99th percentile, 13 years), 18.3% had diabetic ketoacidosis at onset, and 64.0%
used insulin pumps.
Primary Outcome
The overall mean pooled test score was 56.11 (SD, 24.93). There was no statistically
significant difference between test scores from children with vs without diabetes (mean,
56.56 vs 56.11; difference, 0.45 [95% CI, −0.31 to 1.22]). The estimated difference from the
linear model adjusted for grade-, topic-, and year-specific effects was 0.24 (95% CI, −0.90
to 1.39]). Adjusting for socioeconomic status, the difference was 0.45 (95% CI, −0.58 to
1.49) (Table 2 and Figure 1).
Sensitivity and Subgroup Analyses of the Primary Outcome
In the case-sibling analysis (Figure 1), 7344 test scores were identified, of which 2966 were
from children with diabetes (n = 1257 children) and 4378 were from siblings without diabetes
(n = 1606 children). Children with diabetes had slightly higher raw mean test scores than
their siblings (56.43 vs 55.03; difference, 1.40 [95% CI, 0.20 to 2.60]), but this difference
was insignificant in both linear models (adjusted for grade-, topic-, and year-specific effects,
0.32 [95% CI, −0.92 to 1.57]; additionally adjusted for socioeconomic status, −0.03 [95% CI,
−1.30 to 1.23]).
In the analysis of differences in test scores between children with and without diabetes
separately by topic and grade (Figure 1), only reading in fourth grade with adjustment for
socioeconomic status showed a significant difference (1.89 [95% CI, 0.09 to 3.68]; higher
scores for children with diabetes).
When comparing test scores in subgroups of children with diabetes with test scores of
children without diabetes (Figure 2), having an HbA1c level greater than 7.5% (n = 1155
children, n = 2354 test scores) was associated with lower test scores (mean, 53.84 vs
56.11; difference, −2.26 [95% CI, −3.27 to −1.26]), but the association was not statistically
significant in the linear model adjusted for socioeconomic status (difference, −1.06 [95% CI,
−2.37 to 0.25]). Having an HbA1c level greater than 8.6% (n = 472 children, n = 867 test
scores) was associated with lower scores (mean, 49.95 vs 56.11; difference, −6.15 [95%
CI, −7.80 to −4.50]), and the difference was statistically significant in the linear model with
adjustment for socioeconomic status (difference, −2.47 [95% CI, −4.50 to −0.45]). In the
statistical tests for interaction between subgroups of children with diabetes (separately
compared with children without diabetes), there was a statistically significant difference
between subgroups with HbA1c levels greater than 7.5% vs 7.5% or lower and HbA 1c levels
greater than 8.6% vs 8.6% or lower (P < .001 for both subgroups, both with and without
adjustment).
Diabetic ketoacidosis at onset (n = 336 children, n = 755 test scores), diabetes onset at age
younger than 6 years (n = 564 children, n = 1391 test scores), and diabetes duration longer
than 4 years (n = 799 children, n = 1659 test scores) were not associated with significantly
different test scores compared with the children without diabetes (Figure 2). In the statistical
tests for interaction between subgroups with diabetic ketoacidosis at onset vs no diabetic
ketoacidosis at onset (separately compared with children without diabetes), there was a
statistically significant difference without adjustment (P = .01) but not with adjustment (P 
= .13). The tests for interaction between children with diabetes onset at age younger than 6
years vs 6 years or older (P = .77 without adjustment; P = .81 with adjustment) and children
with diabetes duration longer than 4 years vs 4 years or less (P = .89 without adjustment; P 
= .72 with adjustment) did not reach statistical significance.
Secondary Outcomes
Among children with diabetes, levels of HbA 1c and indicators for severe hypoglycemia and
hospitalizations for diabetic ketoacidosis could be assigned to 87% of the test score
observations (n = 3684 from n = 1729 children with diabetes). Mean HbA 1c level was 7.9%
(SD, 1.1%). HbA1c level was significantly higher when the test score was missing (8.3% vs
7.9%; difference, 0.4% [95% CI, 0.3 to 0.6]).
The unadjusted kernel-weighted test scores for children with diabetes and HbA 1c levels less
than 8% were higher than the population mean (Figure 3A). Conversely, test scores for
children with HbA1c levels in the range 8% to 11% were lower than the population mean.
Adjusting for socioeconomic status attenuated the association between HbA 1c level and test
scores. Test scores from children with mean HbA 1c levels less than 7.5% were slightly above
the population mean. In contrast, test scores were lower than the population mean among
children with HbA1c levels ranging from 8.5% to 9.5% (Figure 3B).
The unadjusted linear model (Figure 3A) showed that each 1% increase in HbA1c level was
associated with a 3.48-point lower test score (95% CI, −4.32 to −2.64). The corresponding
coefficient from the adjusted model (Figure 3B) was −1.59 (95% CI, −2.53 to −0.66).
Experiencing at least 1 episode of diabetic ketoacidosis within the past 3 years was
associated with lower test scores (50.80 vs 56.91; difference, −6.11 [95% CI, −10.15 to
−2.07]). This association was not statistically significant when controlling for HbA 1c level,
socioeconomic status, or both (eTable 1 in the Supplement). No association was found
between having experienced at least 1 episode of severe hypoglycemia and test scores
(eTable 1 in the Supplement).
All outcomes and analyses were repeated for math and reading test scores separately
(eTables 1 and 2 and eFigures 1-5 in the Supplement). The results were similar to those for
the pooled test score.
Missing Values
To address the potential of the increased propensity to miss a test among children with
diabetes, all missing test scores were replaced with 1 (assuming that children who missed a
test performed worst of all). Children with diabetes scored slightly lower (53.08 vs 54.02;
difference, −0.94 [95% CI, −1.76 to −0.13]). The difference was not statistically significant in
adjusted models (linear model, −1.05 [95% CI, −2.21 to 0.12]; with adjustment for
socioeconomic status, −0.88 [95% CI, −1.94 to 0.17]).
To investigate the importance of missing HbA 1c values among children with diabetes, pooled
test score means were compared. Mean test score was 55.38 when HbA 1c values were
missing and 56.74 when values were available (difference, −1.35 [95% CI, −3.71 to 1.00]).
In models in which test scores from children with diabetes onset near the date of the test
(within 6 months) were discarded, results were unchanged (n = 558 test scores; difference
adjusted for grade-, topic-, and year-specific effects, 0.43 [95% CI, −0.75 to 1.61]; with
additional adjustment for socioeconomic status, 0.62 [95% CI, −0.44 to 1.69]).
Discussion
In this population-based cohort study, there was no statistically significant difference in
pooled reading and math test scores between children with and without diabetes. The
estimated difference was small and statistically insignificant. Among children with diabetes,
a statistically significant (linear) association between test scores and diabetes control
(assessed by HbA1c level) was found.
A key question is whether the tests measure school performance. Previous research 18 has
demonstrated that test scores predict 48% to 51% of the variation in ninth-grade math and
Danish examination grades.
The main finding of no statisticially significant difference in test scores contrasts with the
findings of some previous studies.3-7 However, most existing studies are based on small,
nonrandom samples of individuals with diabetes matched to random controls. Closest to
this study is the study by Persson et al, 6 which found small negative effects of diabetes on
school grades for the population of Swedish children with diabetes born in the 1970s.
Besides school grades (potentially) measuring more dimensions than test performance,
most of the children in the current study were born after the turn of the millennium. It is
possible that advances in treatment modalities over recent decades (64% of children with
diabetes in this study used an insulin pump) have improved not only gaps in mortality and
morbidity1,24 between individuals with diabetes and the overall population but also have
improved gaps in school performance. This study also documented that it is important to
adjust for confounders such as socioeconomic status when assessing the association
between diabetes control and test scores or school performance.
While demonstrating a negative association between poor glycemic control and test scores,
the results showed that children reaching the treatment target (HbA1c level <7.5%) scored
higher than the population mean. While no conclusive evidence can be provided from an
observational study, one interpretation could be that the parents of children with diabetes
direct more of their resources toward the affected child, giving rise to both better school
performance and better metabolic control. This might happen to a lesser extent among
children with poor diabetes control. A second interpretation could be that being able to
maintain good glycemic control and school performance is confounded by ability (cognitive
and noncognitive skills), which is correlated with socioeconomic status.
Limitations
The study has several limitations. First, there was a higher proportion of missing test scores
for children with diabetes. However, the sensitivity analysis showed that setting all missing
test scores to 1 did not change the results. Second, Denmark has considerable diabetes
awareness and good medical and social care by international standards, and hence the
findings may not apply to other countries. It is possible that the gaps in test performance are
larger in countries where access to health care is associated with socioeconomic status.
Third, in this study tests were performed after short disease duration (mean, 4.5 years), and
diabetic complications usually manifest after a longer duration. However, no association
was found after limiting to children with longer disease duration. Further, even though
diabetic ketoacidosis at onset (or later) may directly affect cognitive skills, no association
with test scores was found.
 
Article Link: https://drive.google.com/file/d/1CVrpPsNPD5ix0YOGQPsu5p_iaYm9zSiy/view?
usp=sharing
Required
Writea brief (but more than one sentence) introduction to provide the background and the
purpose for the study.   Why was the study conducted? What was it about? What is the
objective of the study? Include a brief overview of the literature review if it is included in the
article. (As you write this, assume the reader has no knowledge about this study)
Hypotheses: What was the research hypothesis/hypotheses that was/were tested?
Method:
(Writeas a paragraph)

 Describe the type of study.  Is it an Experimental or Observational study?

o If Experimental, was it Control, Randomization, or Replication?
o If Observational, was it a Prospective or a Retrospective Study?
 Where was the study done?
 What was the sample size?
 What was the target population?
 If the subjects were grouped, how were they grouped?
 If the treatment was applied to the subjects in the group, describe the specifics of the
treatment
 What were the variables measured? (name of the variables, name the dependent
and the independent variables, if applicable)
 When the study was conducted (time frame)? How long did the study last?
 What statistical package, SPSS, SAS etc. was used for analysis?
 What Statistical Analysis Tools were used for analyzing the data (For example,
Descriptive Statistics, Confidence Intervals, Paired t-test, Correlation, ANOVA etc.)