Numeracy Skills and Glycemic Control in an Observational, Multicenter, Cross-Sectional, and International Study of Children with Type 1 Diabetes

Abstract

Aims

This study examined the possible association between numeracy skills and glycemic outcomes in children with type 1 diabetes.

Methods

The study used a cross-sectional design and collected data from 7 centers of the Hvidoere Study Group. HbA1c was measured centrally. Numeracy was assessed using the specific 5-item Diabetes Numeracy Test (DNT-5) and the international, general Wordless Mathematical Test (WMT). The HbA1c predictive multivariate generalized linear model was constructed using the adjusted R-squared index for model selection. Pearson’s correlation coefficient was calculated between observed and predicted HbA1c levels in the training and testing datasets.

Results

Overall,306 adolescents aged 12–18 (mean age 14.96 ± 1.68) years and diabetes duration of 6.57 (±3.75) participated in this study. Numeracy skills, as assessed by the WMT but not DNT-5, predicted the HbA1c levels after adjustment for sociodemographic and clinical factors. The correlation between observed and predicted HbA1c levels was consistent in both datasets and was 0.34 (N = 155) and 0.37 (N = 61) for the training and test datasets, respectively (p = 0.412). The effect size for the WMT-based predictive model of HbA1c adjusted for clinical and socioeconomic factors was significantly higher (p < 0.05) than the single-parameter-based model.

Conclusions

Numeracy, as assessed by an international general math test, is a good predictor of HbA1c in children and adolescents with type 1 diabetes. The basic and short WMT is a potentially effective tool in personalized clinical pediatric diabetes practice. Therapy planning should consider adjusting therapy to compensate for lower numeracy skills and/or training to improve the patient’s numerical proficiency.

Introduction

Type 1 diabetes (T1D) is one of the most common chronic diseases of childhood. The incidence continues to increase, leading to increased disease years characterized by greater medical and psychosocial complexity. Despite significant therapeutic options and technological progress, glycemic targets are still not universally met, especially in the pediatric population [1–3]. This suggests that other important parameters play a role in determining diabetes care outcomes, e.g., numeracy skills.

Self-management of diabetes requires applying numeracy skills to several key decisions and tasks. Numeracy is recognized as a relevant aspect of literacy, but there is no internationally agreed definition. It has been broadly defined as the ability to understand and use numbers in daily life tasks. It is especially important in diabetes management [4–7]. Numeracy skills may limit the achievement of optimal blood glucose outcomes when interpreting food labels and calculating carbohydrates, which remains important even in the era of automatic insulin delivery systems [8]. Furthermore, they are crucial for ensuring insulin dose adjustment to glycemia or comprehending when to refill medications. Low numeracy skills have been associated with higher HbA1c, particularly in adults [5, 9, 10]. The inadequate numeracy skills are common in the general pediatric population. According to the PISA report [11], 31% students performed below level 2, which is considered the baseline level of proficiency students need to participate fully in society. Low numeracy skills should not to be confused with dyscalculia (a learning disability); numeracy is more complex and may be related to socioeconomic status, parental numeracy skills, age, gender, and country. The commonness makes it likely to occur, therefore likely the case in children with T1D as T1D reflects the background population regarding socioeconomic status [12–14].

To assess the numeracy skills of children and their families, Huizinga et al. [15] developed and validated a Diabetes-specific Numeracy Test (DNT) to assess diabetes-related numeracy skills. A shortened version of DNT, called DNT-5, was developed for children and released in 2013 [16]. The alternative is the generic, rapid, and reliable test to assess numeracy skills: the Wordless Mathematical Test (WMT) [17].

The Hvidoere International Study Group on Childhood Diabetes is a longstanding scientific and clinical collaboration of 26 pediatric diabetes centers from 24 countries across 4 continents. This group has verified the effect on glycemic outcome of insulin treatment regime, goal setting, diabetes team dynamics, family dynamics, and quality of life [18]. However, the contribution of numeracy skills to the observed variability in glycemic outcomes in a multinational context has not been assessed previously. The aim of the present study was to explore numeracy skills and their potential impact on glycemic outcomes in children with T1D.

Methods

Study Design

We conducted an observational, cross-sectional, descriptive, multicenter, international study.

Study participants: study participants were randomly recruited during their regular outpatient visits in 7 centers of the Hvidoere study group in 7 different countries. Participants were aged 12–18 years, with T1D of at least 12 months and no co-existent conditions (e.g., coeliac, thyroid disease, cognitive or visual impairment).

Questionnaires

Demographic data, including sex, age, age at diagnosis, ethnicity, caregiver’s occupation (either part time, full time, or at home) and their educational level (first or second level vs. third or more), and child’s school performance (either appropriate class for age or requiring assistance), were collected through questionnaires. Data on diabetes duration, diabetes management (severe adverse events over the previous 3 months, HbA1c values of the last year), and current insulin treatment (i.e., treatment modality and use of CGM) were filled in by the attending physician based on the child’s health record.

Numeracy skills were assessed by the general Wordless Mathematical Test (WMT) and the diabetes-specific 5-item Diabetes Numeracy Test (DNT-5), both scored from 0% to 100%. The WMT has been adopted by the Organization for Economic Cooperation and Development (OECD) as part of the Program for International Student Assessment [11]. The WMT aims to assess general mathematical reasoning and problem-solving abilities and is stripped of linguistic or verbal components to evaluate pure mathematical skills without interference from language comprehension. Whilst the WMT principally concentrates on general cognitive math reasoning, the DNT 5, that is the shortened version of the original DNT questionnaire, focuses on functional numeracy and consists of 5 questions in 4 domains: nutrition, exercise, blood glucose monitoring, and medication, all of which are essential components of daily diabetes self-management.

In addition, the 5-item Psychological Well-being Questionnaire (WHO-5) and the Diabetes Family Responsibility Questionnaire (DFRQ) were administered. The original DFRQ had 17 items, but a revised version only contains 10 of the 17 items. Each task is rated by the respondent as belonging primarily to the parent (score 1), being shared equally by the parent and child (score 2), or belonging primarily to the child (score 3), with the total score ranging from 10 to 30, with higher scores indicating that the child assumes greater responsibility; for the parent, the score was reversed [19]. All questionnaires were translated from English into the official language of the participating center and back into English by native English speakers to ensure their readability and the correct translation of the items.

ΗbA1C

Two capillary blood samples from a single finger pricking were collected during the child’s visit for local and centralized HbA1c analysis. The analysis was centralized at Copenhagen University Hospital at Herlev and reported in International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) units (mmol/mol). HbA1c was determined by high-performance liquid chromatography.

Data analysis: all data were analyzed anonymously at the Silesian University of Technology, Department of Data Science and Engineering, Gliwice, Poland. Descriptive statistics and a 95% confidence interval were calculated for continuous and ranked variables. Frequency tables were used to describe categorical variables. Missing data were not imputed to avoid confounding the complex relationship between variables. The Shapiro-Wilk test was used to test the hypothesis of normality of the distribution. The F test and Bartlett’s tests were used to verify the homogeneity of variances. Two-sample tests or ANOVA with appropriate post hoc tests were used to test hypotheses of equality of means between patient subpopulations. p values were supported by ANOVA-based η² effect size (ES) estimates and post hoc tests adjusted for Hedge’s d or rank-biserial correlation coefficient ES measures. Pearson’s and Spearman’s correlation coefficients (r) measured the association between variables. The Benjamini-Hochberg false discovery rate (FDR) was calculated to correct for multiple testing. FDRs less than or equal to 0.05 were treated as significant. ES measures were quantified as no effect (absolute value below 0.1 for r family and below 0.2 for d family), small effect (ranged 0.1−0.3 and 0.2−0.5 for r and d family, respectively), medium effect (ranged 0.3−0.5 and 0.5−0.8, respectively), and large effect (above 0.5 and above 0.8, respectively). In the case of ANOVA ES measure, the thresholding values equal to 0.01, 0.06, and 0.14 were used to distinguish among small, medium, and large ES [20]. For the contingency tables of size 2 by p, the limits for Cramer’s V ES measure set to 0.1, 0.3, and 0.5 were used to qualify ES levels. If the minimal dimension of the contingency table was higher than 2, the corrected threshold values were used, following the guidance in Reference [21].

Model building and validation: the regression prediction generalized linear model was built by optimizing the adjusted R-squared index. The stepwise forward feature selection was used to identify the subset of the best predictors. Only predictors with at least a small ES from the univariate analysis were included in the initial subset of potential descriptors. The dataset was split into training and testing, where the training consisted of cases with complete data for the chosen initial subset of predictors (N = 155 children). The remaining 151 children with incomplete data were used to build the testing datasets for the final model validation. The Pearson’s correlation coefficient between observed and predicted HbA1C levels and F statistics were used for model diagnosis. p values less than or equal to 0.05 were treated as significant.

Results

Overall, 306 children with T1D were included in the study. Of those, 11 were from Canada, 37 were from Greece, 56 were from Luxembourg, 82 were from Poland, 38 were from Portugal, 42 were from Spain, and 40 were from Sweden (Table 1).

Table 1.

Demographic and clinical characteristics of the cohort

Variable	Summary statistics
Gender, N = 299
Boys	161 (53.85%)
Girls	138 (46.15%)
Age, N = 306, years	14.96 (±1.68)
Age at diagnosis, N = 303, years	8.47 (±3.58)
Diabetes duration, N = 306, years	6.57 (±3.75)
Ethnicity, N = 299
Caucasian	289 (96.66%)
Other	10 (3.34%)
Type of insulin treatment, N = 303
CSII	170 (56.11%)
MDI	133 (43.89%)
CGM, N = 254
Yes (any type)	158 (62.20%)
No	96 (37.80%)
HbA1c, N = 275, mmol/mol	62.07 (±14.32)
DKA (number of episodes over the previous 3 months), N = 297
0	270 (90.91%)
≥1	27 (9.09%)
Severe hypoglycemias, according to the ISPAD 2022 definition (number of episodes over the previous 3 months), N = 304
0	283 (93.09%)
≥1	21 (6.91%)
Responsibility for diabetes management, N = 274
Child or parent	148 (54.01%)
Both	126 (45.99%)
Mother’s work, N = 300
Full time	187 (62.33%)
Part time	56 (18.67%)
At home	57 (19.00%)
Father’s work, N = 279
Full time	253 (90.68%)
Part time	6 (2.15%)
At home	20 (7.17%)
Maternal education, N = 300
First or second level	183 (61.00%)
Third or post-graduate level	117 (39.00%)
Paternal education, N = 285
First or second level	193 (67.72%)
Third or post-graduate level	92 (32.28%)
Child’s school performance, N = 298
Age-appropriate class	288 (96.64%)
Requiring assistance	10 (3.36%)

Numerical variables are represented by mean ± standard deviation; for categorical variables, the frequency distribution is provided.

A comprehensive analysis encompassing all the factors under scrutiny revealed that the majority of centers exhibited no substantial disparities in the prevalence of missing data. However, a significant disparity in the completeness of the ethnicity of the father was identified, ranging from 0% to 19.6% and yielding a p value of less than 0.001 and an ES V of 0.31. This considerable variation precluded the incorporation of this characteristic into subsequent analyses. With regard to test scores, we observed nonuniformity across centers in the occurrence of missing data for the DNT-5 test scores of the parents. Consequently, we opted against incorporating this trait as a potential predictor.

Numeracy Skills: Performance on DNT-5 and WMT

The DNT-5 score quartiles were 40% for the lower and 80% for the upper quartile (with a median of 60%). The scores were correlated with age (r_s = 0.15, small ES, p = 0.02, FDR <0.10), WHO scores (r_s = 0.16, small ES, p = 0.01, FDR <0.10), and WMT (r_s = 0.3, small ES, p < 0.01, FDR <0.01). Scores were better with higher paternal education (p = 0.01, small ES, FDR <0.05) and shared responsibility for diabetes management (p = 0.02, small ES, FDR <0.10).

The median WMT score was 86% (with a lower quartile of 64% and an upper quartile of 93%). WMT scores were positively correlated with age (r_s = 0.21, small ES, p < 0.01, FDR <0.01) and DNT-5 (r_s = 0.29, small ES, p < 0.01, FDR <0.01). In addition, scores were better with higher maternal and paternal education (p < 0.01, small ES, FDR <0.01), age-appropriate school attendance (p = 0.03, small ES, FDR <0.05), and shared responsibility for diabetes management (p < 0.01, small ES, FDR <0.01). They were also better in pump users (p = 0.03, small ES, FDR <0.10).

HbA1c

As shown in Table 1, the mean HbA1c was 62.1 (±14.3) mmol/L. There were significant differences in HbA1c between centers (p < 0.001, large ES, FDR <0.01). HbA1c was lower in children on pump treatment (Table 2, p = 0.04, small ES, FDR <0.10). It was lower in families where the responsibility for the care of diabetes is shared rather than assumed by either the parents or the child (p = 0.04, small ES, FDR <0.10). HbA1c was also decreased with a higher level of paternal education (p < 0.01 small ES, FDR <0.05) and when the child was following a class appropriate for his age (p = 0.039, small ES, FDR <0.10).

Table 2.

Differences between the two treatment modality groups (CSII vs. MDI)

Variable	NCSII	NMDI	CSII	MDI	p value/FDR (ES)
Age	170	133	14.95 (±1.69)	14.97 (±1.65)	0.902/0.910 (none)
Age at diagnosis	169	131	7.79 (±3.50)	9.33 (±3.42)	<0.001/<0.001 (small)
Diabetes duration	170	133	7.21 (±3.61)	5.78 (±3.79)	<0.001/0.001 (small)
HbA1c	152	120	60.43 (±12.97)	64.06 (±15.20)	0.038/0.096 (small)
Mathematical wordless	161	129	81.0% (±17.6%)	76.1% (±20.3%)	0.034/0.093 (small)
DNT-5	155	90	56.5% (±28.9%)	59.1% (±27.8%)	0.514/0.593 (none)
WHO	160	128	66.7% (±17.4%)	67.6% (±16.5%)	0.615/0.683 (none)
Responsibility for diabetes management	151	121			0.231/0.365 (none)
Child or parent			87 (57.62%)	60 (49.59%)
Both			64 (42.38%)	61 (50.41%)

HbA1c was negatively correlated with the WMT score (r_s = −0.18, p < 0.01, small ES, FDR <0.05), but there was no correlation with the DNT-5 (r_s = −0.02, p = 0.83, FDR >0.10) test. A weak, negative correlation was found with the WHO (r = −0.13, p = 0.04, small ES, FDR <0.10). Moreover, HbA1c was positively correlated with diabetes duration (r_s = 0.22, p < 0.01, small ES, FDR <0.01). The Circos diagrams in Figure 1 visualize the observed relations between all clinical, educational, and socioeconomic descriptors analyzed and the level of HbA1c (panel A) and WMT scores (panel B).

Fig. 1.

Diagrams representing the relation between the potential explanatory variables and the level of HbA1c (a) and results of the child’s WMT (b). Gray lines demonstrate the association of at least a small ES between the analyzed indicator and the chosen explanatory variables. The darker the line, the stronger the association observed. Green marks the continuous variables, while yellow denotes the categorical variable. The prefix C- means the child’s test results, while P- denotes the test results of the parent. The detailed results of the comparative analysis are presented in online supplementary Table S1 (for all online suppl. material, see https://doi.org/10.1159/000545419).

WHO Questionnaire

Regarding the WHO questionnaire results, the median value was 68% (quartiles equal to 60% for the lower and 80% for the upper one). The WHO scores correlated negatively with HbA1c (r_s = −0.12, small ES, p = 0.04, FDR >0.10) and only positively with the DNT-5 (r_s = 0.16, small ES, p = 0.01, FDR <0.10) score. Scores were higher in males (small ES, p < 0.01, FDR <0.01) and lower with a high number of DKA hospitalizations (p = 0.01, small ES, FDR <0.05).

Multivariate HbA1c Predictive Model

The multivariate stepwise feature-forward model construction used a subset of 11 potential predictors pre-selected based on the univariate analysis: WMT score, DNT-5 score, WHO test results, age at diagnosis and visit, duration of DM, maternal and paternal education, type of insulin therapy, use of CGM, and responsibility for diabetes care. The training dataset comprised 155 children and included complete information on the above-preselected descriptors. Four predictors: WMT test results, duration of DM, type of insulin injection (MDI vs. CSII), and use of CGM (no/yes) remained in the final model. Of those, insulin injection type and diabetes duration are the leading statistically significant factors (p < 0.05) with the remaining WMT and use of CGM as the important correcting factors (Table 3). According to the obtained model, the level of HbA1C decreases with better WMT results, increases with longer duration of DM, and decreases with CSII, with additional correction toward lower HbA1C when CGM is used.

Table 3.

Parameters of the final multivariate ΗbA1c prediction model (F stat = 4.98, p = 0.000852, adjusted R² = 0.094)

Explanatory variable	Beta estimate	Standard error of beta estimate	±95% CI	t statistics	p value
Intercept	67.8930	4.8284	58.3546−77.4314	14.061	<0.0001
WMT	−9.7648	5.3442	−20.3222 to 0.7926	−1.827	0.0697
DM duration	0.9257	0.2954	0.3421−1.5093	3.134	0.0021
Insulin injection type (CSII)	−5.2853	2.3327	−9.8935 to −0.6771	−2.266	0.0249
CGM (yes)	−3.1963	2.0252	−7.1971 to 0.8045	−1.578	0.1166

The obtained correlation coefficient between observed and predicted HbA1c for the training dataset is 0.3423 (N = 155), which is consistent with the results obtained for the test data (r = 0.3673, N = 61 of 151, as only these patients have complete data on the required four predictors). It is a significantly stronger (p < 0.05) correlation than for any of the univariate models (0.19 was the best coefficient achieved).

Discussion

We confirm that numeracy skills matter independently of diabetes regimen and this suggests that assessment of numeracy skills is an important factor to personalize treatment and choose the best approach. Previous studies have mainly assessed caregivers’ numeracy skills, which have been found to be inversely correlated with their children’s HbA1c [22, 23]. Studies directly assessing children’s numeracy and its relationship to their HbA1c are lacking [24], but the importance of correct carbohydrate counting skills in users of automated insulin delivery systems has been documented [8]. Extensive studies in adults with diabetes have shown that improving numeracy skills can modify diabetes outcomes [25–27]. Promising early results have been published on improving numeracy in children and caregivers with diabetes, such as interactive video games for young people with diabetes (e.g., Power Defense, an avatar-based video game developed by a team in Toronto) [28]. Therefore, it is essential to assess the numeracy skills of children and adolescents and their families at the outset and provide additional support as needed [29]. The WMT result, but not the diabetes-specific DNT-5 result, is associated with HbA1c level in children with diabetes, independently of the applied technologies. This finding is in contrast to the results of other research groups, in which the DNT-5 and other numerical literacy measures were associated with HbA1c level. However, these studies were conducted in mixed cohorts of adult T1D/T2D subjects [9, 30–32]. The correlation between the scores from the DNT-5 and the WMT in children is only weak, suggesting that these tests assess different aspects of numerical literacy. One hypothesis is that the DNT-5 reflects diabetes education in clinics, while the WMT measures general maths skills. The WMT is a good and quick tool, indicative and helpful in assessing a child’s ability to make logical processes and predictions, a skill that is particularly useful in self-care of T1D, i.e., adjusting insulin doses, predicting and avoiding hypoglycemia, etc.

WMT scores improved with shared responsibility for diabetes management between the child and caregivers. Better-educated families are more likely to share responsibility [33]. The complex Web of correlations among important factors such as shared responsibility, parental education, mental thriving (WHO-5), ΗbA1c, injection type, and math scores makes it difficult to differentiate the importance of each factor, but the cross-sectional design does not allow for distinguishing the effects of each factor. The multivariate model, which allowed us to identify the most informative and least redundant predictors, showed that the best predictors of HbA1c were WMT performance, DM duration, and CSII use, further improved by CGM use. Diabetes duration and technology are well-known factors associated with HbA1c [34–39] and may attribute to both physiological and behavioral factors (i.e., gradual decline in beta-cell reserve and insulin secretion, as well as decreasing adherence to treatment [40, 41]), whereas the association to WMT performance is new. The other factors like shared responsibility and mental thriving were not significant in the multivariate model. They may still be relevant as areas for intervention given the associations between shared responsibility and WMT as well as HbA1c and mental thriving. The final model had a moderate effect level, indicating that variables other than those included in our study are still important for HbA1c.

In a recent study of 944 people with T1D aged 9–25 years, lower scores on the WHO-5 item questionnaire have been associated with higher ΗbA1c; however, differences were statistically significant only between people with an ΗbA1c >9% and those with an HbA1c <7.5% [42]. Indeed, results on the relation of psychological well-being and HbA1c are not conclusive. Interventions to improve health-related quality of life in adolescents failed to improve ΗbA1c, despite the improvement in measures of well-being [43]. However, psychological issues are prevalent among individuals with T1D [44], which may elucidate the absence of the WHO test result from the final predictive model, despite its association with HbA1c level.

The main limitation of our study is the absence of an external testing dataset. The internal hold-out data of similar distribution of HbA1c and explanatory variables were employed solely for the purpose of validating the obtained prediction model, which may introduce a degree of bias into the estimation of the prediction quality.

Conclusions

Children’s mathematical skills are important in optimizing the most effective and safest insulin treatment for T1D, both MDI and CSII. The international, short WMT is a potentially effective tool for assessing basic math skills and may be used to better tailor education and diabetes management taking numeracy skills into account. It can also be employed as a predictor of diabetes outcomes in children. Customizing diabetes education to address poor mathematic skills, as assessed by WMT, could potentially improve diabetes management and optimize glycemic control and diabetes outcomes in adolescents with type 1 diabetes mellitus.

Acknowledgments

Authors would like to thank Jette Høgsmose and Nour Houssein Jaqub for assisting with the HbA1c analysis; Russell Rothman, Hilary Hoey, and Andreas Neu for helpful discussions during the early stages of research design; Bogna Szotowski for sharing her experience with WMT usage; Taha Arshad and Marise Abdou for their assistance with the study conducted in Toronto, Prof. Christina Kanaka-Gantenbein for her assistance with the study conducted in Athens, and Zuzanna Gosławska and Halla Kamińska for their assistance in the study conducted in Katowice; and all children with diabetes and their caregivers for their volunteer participation in the study.

Statement of Ethics

This study protocol was reviewed and approved by Ethics Committee of Medical University of Silesia, Katowice, Poland (approval No. KNW/0022/KB1/17/18). Written informed consent/assent was obtained from the children and one of their parents. The study protocol was reviewed and approved by each of the participating sites.

Conflict of Interest Statement

All authors declare no conflict of interest related to this project.

Funding Sources

The study was partially financially supported by Eli Lilly Polska, grant H60-PL-0013.

Author Contributions

J.P. and I.K. analyzed data, contributed to the discussion, and wrote the first draft of the manuscript. J.S., C.B., T.G.B., T.C.S., F.C., and P.J.-C. were involved in project planning and execution and critically revised the analysis and manuscript. I.K., P.J.-C., M.B., L.A.C., C.C.-C., and S.S. collected the blood samples and supervised the numerical skill testing. M.P. participated in research design, funding application, and data acquisition. P.J.-C. is the guarantor of this work and, as such, has 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. All authors reviewed the manuscript before submission.

Funding Statement

The study was partially financially supported by Eli Lilly Polska, grant H60-PL-0013.

Data Availability Statement

The data that support the findings of this study are not publicly available due to GDPR rules and local authorities. Data may be available in anonymized format upon reasonable request to the corresponding author.