Socioeconomic Inequality in Metabolic Control Among Children With Type 1 Diabetes: A Nationwide Longitudinal Study of 4,079 Danish Children Free

Aspects of family background such as socioeconomic status (1,2), household composition (3–5), and ethnicity (1) have been linked to metabolic control—a key determinant of diabetes-related complications (6–9)—in children with type 1 diabetes. Specifically, children with well-educated or affluent parents (1,2,10,11) and those who grow up in two-parent households (3–5) tend to have better glycemic control than children of less educated parents and children in single-parent households. Despite a growing number of studies examining the relationship between family background and outcomes among children with type 1 diabetes, the underlying mechanisms are less well understood. In this regard, a key difficulty is to disentangle the interaction between various aspects of family background and treatment outcomes because several mechanisms may be simultaneously at play. For instance, access to health care and adherence to medical advice are two channels that may both be tied to maternal educational level and diabetes outcomes.

Denmark offers a unique opportunity for studying the relationship between diabetes outcomes and individual socioeconomic factors. The population of children diagnosed with type 1 diabetes is registered (12) with detailed information on diabetes-related outcomes at onset and yearly follow-ups that can be linked through national registries to third-party reported information on (maternal) socioeconomic characteristics including education, income, and employment status. Further, with universal health care offering equal access to care, an important confounder is eliminated.

The aim of this study was to examine the degree of inequality among Danish children with type 1 diabetes from onset and during follow-up across maternal education and to explore the relative importance of key predictors/mediators in explaining the gap between children of mothers with different educational backgrounds.

All Danish children diagnosed with type 1 diabetes between years 2000 and 2013 (n = 4,911) were identified in the Danish Registry of Childhood and Adolescent Diabetes (DanDiabKids) (12). DanDiabKids contains demographic and clinical information measured annually for all Danish children diagnosed with diabetes since 1996. The sample was restricted to individuals born in 1987 or later (n = 4,376) to allow for a linkage between children and maternal education in the year before birth (access to maternal educational information was available from 1986). Of the 4,376 children, maternal educational information was identified for n = 4,079 children. Due to the longitudinal structure of DanDiabKids, each child was observed six times on average (range 1–16), giving a total of n = 24,432 observations for the 4,079 children with type 1 diabetes and maternal educational information.

Annual demographic and clinical information from DanDiabKids was augmented with background information on mother’s highest completed education, mother’s age at birth, mother’s employment status, mother’s wage percentile, child’s number of siblings, and child’s birth order by merging DanDiabKids with the Danish administrative registries through unique personal identity numbers. Clinical information was registered in DanDiabKids by clinicians from all pediatric clinics in Denmark, and blood samples taken at the appointment closest to each patient’s birthday were sent to a central laboratory. Maternal education was grouped into four categories: ≤high school (n = 1,643), vocational or 2-year college (n = 1,548), bachelor’s degree (n = 695), and ≥master’s degree (n = 193). Maternal education was assessed in the year before birth to avoid any feedback of children’s diagnosis of type 1 diabetes on the measure of maternal educational attainment.

The main outcome was HbA_1c, the main exposure was maternal education before birth, and the main characteristic of interest (apart from maternal education) was the number of daily blood glucose tests. HbA_1c levels were measured centrally using the high-pressure liquid chromatography method (Tosoh Bioscience, South San Francisco, CA) with a normal range of 23.5–40 mmol/mol (4.3–5.8%). The number of daily blood glucose tests was registered as the mean number of daily tests over the past weeks before the visit, based on blood glucose downloads in the clinic.

Secondary outcomes were a dichotomous variable indicating at least one episode of diabetic ketoacidosis (DKA) within the last year and a dichotomous variable indicating at least one severe hypoglycemia episode within the last year, both defined according to International Society for Pediatric and Adolescent Diabetes guidelines (13). Specifically, episodes of DKA during follow-up were defined as hospitalizations with DKA and episodes of severe hypoglycemia were defined as blood glucose <63 mg/dL (<3.5 mmol/L) and unconsciousness. Severe hypoglycemia included both self-reported episodes and episodes treated at a hospital. At onset, DKA was classified into mild (7.20 ≤ pH < 7.30 or bicarbonate <15 mmol/L) or moderate/severe (pH <7.20 or bicarbonate <10 mmol/L).

For statistical tests, two-sided P values were calculated where relevant and P values <0.05 were considered statistically significant.

Means, SDs, and proportions of selected variables were compared for children across the four groups of maternal education. For handling of repeated measurements in the comparison of means and proportions, one observation per child was obtained, following the procedure described below, before averaging across maternal education group. Means and proportions of baseline characteristics were calculated from the first observation for each child. Means and proportions of onset characteristics were calculated for children where onset information was available. Follow-up measures of HbA_1c, duration of diabetes, and number of daily glucose tests were calculated as averages during follow-up for each child before calculation of the average within each group of maternal education. For episodes of DKA or severe hypoglycemia, the share of children who experienced at least one episode during follow-up was calculated. Finally, we calculated the treatment mode regimen during follow-up for each child and used this to present the proportion of children in each regimen across groups. In case of ties, the first mode was chosen following this hierarchy: pump, basal/bolus, premixed/other. In the comparison of means and proportions, existence of group differences was tested by ANOVA F tests and P values were presented.

For assessment of the relationship between maternal education and metabolic control, mean HbA_1c levels were compared repeatedly from onset until 5 years after onset across maternal education group and depicted graphically with 95% CIs. Similarly, the mean numbers of daily glucose tests were compared over time across groups to examine the relationship between maternal education and treatment adherence. In a further investigation of the joint relationship between maternal education, metabolic control, and adherence, the mean level of HbA_1c was plotted for each number of daily glucose tests across maternal educational group and presented with 95% CIs.

Linear regression models were estimated with the primary goal being to assess the association between maternal education group and HbA_1c levels for children with type 1 diabetes with and without adjustment for baseline and treatment characteristics. Results from the linear regressions were presented as regression coefficients with 95% CIs and P values. Logistic regression models were estimated to examine the association between maternal education group and risk of DKA and severe hypoglycemia. For the logistic regressions, results were presented as odds ratios (OR) with 95% CIs and P values. In both linear and logistic regressions, cluster-robust SEs were used to allow for arbitrary within-individual correlation due to repeated measurements (14).

For evaluation of the relative importance of various characteristics in explaining the raw gap in HbA_1c levels across maternal education, a linear decomposition (15,16) was performed between children of mothers with at most high school education and children of mothers with at least a master’s degree (details on the Oaxaca-Blinder decomposition can be found in Supplementary Table 2).

For investigation of differences between children with and without maternal educational information, HbA_1c at follow-up and risk of DKA at onset across children with and without maternal educational information were tested.

For testing of the sensitivity of measuring maternal education before birth, maternal education was measured at each visit and the linear regressions of HbA_1c onto maternal educational attainment were reestimated.

All statistical analyses were conducted in Stata 15 (17).

The comparison of means and proportions for children with type 1 diabetes across maternal education group (Table 1) showed that factors such as maternal employment, number of siblings, and living with both parents varied significantly between groups (P < 0.001 in all cases). In contrast, as shown in Table 1, no significant differences were shown between sex or share of parents with type 1 diabetes. HbA_1c at onset did not differ according to maternal education level, but children of mothers with master’s degrees were more likely than children of less educated mothers to be diagnosed before age 6 years and to be diagnosed with moderate-to-severe DKA at onset (test for group difference P < 0.001 and P = 0.02, respectively).

All variables measured during follow-up varied significantly across groups of maternal education (P < 0.001 for all variables except for diabetes duration, with P = 0.01). On average, children of the most highly educated mothers had better metabolic control, lower incidence of DKA, and fewer events of severe hypoglycemia than children of less educated mothers. These children also measured blood glucose more often and were more likely to use pumps than children of less educated mothers.

Figure 1A shows that children of the most highly educated mothers had the lowest average levels of HbA_1c already after 6 months and that group differences were fully manifested (in the sense that there was no overlap of CIs) after a diabetes duration of 2 years. This group difference remained constant from 2 to 5 years after onset, with an almost monotonic relationship between mothers’ educational levels and children’s average HbA_1c levels. Figure 1B shows that children of the most highly educated mothers had the highest mean number of daily glucose measurements from 6 months after onset and beyond, with an expansion in inequality during the first years after onset.

Figure 2 reveals a negative association between the number of daily blood glucose tests and HbA_1c across all levels of maternal education and that children of the most highly educated mothers had the lowest levels of HbA_1c over the entire span of daily glucose measurements. Further, Fig. 2 displays that the differences in HbA_1c across maternal education were largest among children who measured their blood glucose relative rarely (≤3 times per day), while the differences were smallest among those who measured their blood glucose often (≥6 times per day).

The linear regression models with HbA_1c as the outcome and maternal educational attainment as the main explanatory variables (Table 2) showed that maternal education group was significantly related to HbA_1c levels both with and without adjustment. Without adjustment (Table 2) (model 1), children of mothers with vocational or 2-year college had lower levels of HbA_1c than children of the least educated mothers (−2.41 mmol/mol [−0.22%], P < 0.001). The same was true for children of mothers with bachelor’s degrees (−5.42 mmol/mol [−0.50%], P < 0.001) and children of mothers with at least a master’s degree (−8.78 mmol/mol [−0.80%], P < 0.001). These associations were only altered slightly after baseline and onset characteristics were controlled for (Table 2) (model 2) but decreased substantially once treatment characteristics were controlled for (Table 2) (model 3). After addition of the full set of controls, the gap in HbA_1c between children of the least educated mothers and the other groups of children was −0.86 mmol/mol (−0.08%), P = 0.01; −2.48 mmol/mol (−0.23%), P < 0.001; and −4.27 mmol/mol (−0.39%), P < 0.001 for children of mothers with vocational or 2-year college, bachelor’s degrees, or master’s degrees, respectively. Maternal income was not associated with different HbA_1c levels in the full model. Furthermore, higher numbers of daily glucose tests were associated with lower HbA_1c (Table 2). Specifically, each extra daily glucose test was associated with −2.16 mmol/mol (−0.20%) HbA_1c (P < 0.001).

Maternal education was significantly associated with episodes of DKA and severe hypoglycemia (Supplementary Table 1). With no adjustments (model 1), there was a monotonic decrease in the OR of both DKA and severe hypoglycemia as maternal education increased. Compared with the reference category (mothers with high school or less), the OR for DKA was 0.22 (95% CI 0.07, 0.71; P = 0.011) if the mother had a master’s degree or more. For severe hypoglycemia, the corresponding OR was 0.50 (95% CI 0.33, 0.78; P = 0.002). These associations remained significant after adjustment for baseline and onset characteristics (model 2). In the fully adjusted model (model 3), most of the association between maternal education and DKA and severe hypoglycemia disappeared. The number of daily glucose tests was associated with lower risk of DKA (OR 0.83 [95% CI 0.77, 0.89]) but was not associated with the risk of hypoglycemia.

The Oaxaca-Blinder decomposition (Supplementary Tables 3 and 4) showed that 41.2% of the gap in HbA_1c between children of the least educated mothers and children of the most highly educated mothers could be explained by differences in observable characteristics. Differences in the daily number of glucose tests explained 22.5% of the entire gap between the two groups. Living with both parents explained 6.0% of the gap, while employment status of the mother, diabetes duration, and number of siblings all explained a significant, but minor, part of the remaining gap (3.3%, 3.4%, and 1.7%, respectively).

Children with maternal educational information had lower HbA_1c at follow- up than those without background information (n = 3,914 vs. n = 287, HbA_1c [mmol/mol] 66.2 vs. 71.1, HbA_1c [%] 8.21 vs. 8.65; P < 0.001), but there were no significant differences in the rate with DKA at onset (proportion with no DKA at onset 84% vs. 80%, proportion with mild DKA at onset 8% vs. 9%, and proportion with moderate or severe DKA at onset 8% vs. 10%; all nonsignificant).

In the sensitivity regression of HbA_1c and maternal education measured at each visit instead of before birth (Supplementary Table 5), the regression coefficients on the contemporaneously measured maternal education were always contained in the 95% CIs of the main estimates from Table 2, and conclusions were unchanged.

In this nationwide longitudinal study, we found strong evidence that metabolic control (measured by levels of HbA_1c) varied significantly by maternal educational background. A considerable fraction (22.5%) of this difference was explained by differences in the number of daily blood glucose tests, indicating that the lower level of HbA_1c among children of highly educated mothers may be partially caused by more intensive self-monitoring.

While the association between patient education or socioeconomic status and treatment compliance is generally well established in adults (18), it is less documented in the pediatric population. In children with type 1 diabetes, associations have been found between parental income or education and better self-management (2) as well as health-related quality of life (19). Other research on children with type 1 diabetes (20,21) has shown that blood glucose monitoring is an important predictor of glycemic control. Taken together, evidence from the literature thus supports the idea that parental education and socioeconomic status may affect patients’ glycemic control through the adherence channel. However, the existing evidence is scarce and we are not aware of any other large-scale studies exploring these relationships simultaneously.

Even though differences in the number of daily glucose tests may explain parts of the observed inequality in metabolic control by maternal education, this is not the full story. In the joint graphical representation of HbA_1c, the number of daily blood glucose tests, and maternal education (Fig. 2), considerable differences across groups remained even when the number of daily blood glucose tests was held fixed. Similarly, the regression analysis showed that the associations between maternal education and HbA_1c remained significant even after adjustments for the full set of covariates (model 3) (Table 2). Potential explanations for the remaining differences are that highly educated mothers may be more capable of helping with diabetes management, be more able to educate their children in how to control their diabetes, and have increased focus on (to us) unobserved aspects of treatment and adherence. For example, previous research has shown associations between low levels of education and low levels of diabetes knowledge among mothers of children with type 1 diabetes (22), linked maternal education to maternal health investments for children in utero (23), and found higher levels of maternal education to be associated with lower BMI and better fitness in children (24,25). Relatedly, higher socioeconomic status of parents has been linked to better executive function and cognitive control in children (26,27), which has furthermore been connected to health behavior among children with type 1 diabetes (28).

The longitudinal analysis of metabolic control (Fig. 1A) showed that levels of HbA_1c were similar across maternal educational groups at onset and that the difference between groups evolved during the first 2 years after onset. This supports an idea that inequality in metabolic control develops during the first years after onset rather than being present at onset. A similar pattern of expanding disparities during the first 24 months after onset was found across ethnic groups in New Zealand (29). The longitudinal analysis of treatment adherence (Fig. 1B) showed that the inequality in the number of daily glucose measurements was more prevalent ≥1 year after onset than 6 months after onset. This once again highlights how the number of daily glucose measurement may contribute to the inequality in HbA_1c across maternal education.

The number of daily blood glucose tests was simultaneously an important predictor of HbA_1c (model 3) (Table 2 and Fig. 2) and the main factor in explaining the gap in HbA_1c by maternal educational level (Supplementary Tables 3 and 4). The number of daily glucose tests has previously been shown to be associated with HbA_1c (30,31). Naturally, measuring glucose levels more frequently does not improve glycemic control per se—unless followed up by a correction bolus or other type of action—but the number of glucose measurements is an important part of the daily disease monitoring and may also be related to other aspects of treatment adherence. Consequently, we consider the number of daily glucose measurements as an indicator of treatment adherence and interpret the associations found between the number of daily glucose measurements and HbA_1c as capturing the link between treatment adherence and outcomes.

In addition to the main findings, a few interesting patterns emerged after full adjustment.

First, we found premixed insulin to be associated with lower HbA_1c, which is probably an artifact of the lack of change in treatment regimen if treatment target was reached on premixed insulin. A somewhat similar selection-driven pattern was found in a previous study (30) where children with two injections/day were doing slightly better than those with multiple injections/day. Second, family structure (number of siblings and living with both parents) seemed to be important factors in relation to levels of HbA_1c. Family structure has previously been found to be associated with HbA_1c (1,5). A potential explanation is that the number of siblings is directly related to the amount of parental resources available to each child and that living in a two-parent household increases the amount of parental resources available. A second interpretation is related to socioeconomic inequality in the risk of divorce, in particular that resourceful parents are less likely to choose to divorce (32), thereby driving the association between living in a two-parent household and metabolic control.

The magnitudes of the unadjusted inequalities found across maternal educational groups were large and clinically relevant. Previous studies from the DCCT/EDIC (Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications) have shown each 10% relative decrease in HbA_1c to be associated with a risk reduction of 44% for retinopathy progression (33), a 28% lower risk of cardiovascular events (hazard ratio 0.72) (7) (30-year follow-up), and a 39% lower mortality risk (hazard ratio 0.61) (9) (27-year follow-up). Therefore, the relative gap in HbA_1c of ∼10% between children of the least and most educated mothers found in this study may potentially translate into considerable differences in risk of complications related to type 1 diabetes.

The findings from our analyses were in line with previous research demonstrating associations between socioeconomic background and glycemic control (1,2,5). The study was strengthened by the population design and the longitudinal nature of the data. Further, the detailed clinical data allowed us to shed light on some of the potential mechanisms for the differences by socioeconomic status, specifically the role of daily glucose tests. Due to the Danish universal tax–financed health care system—including free glucose testing equipment and strips—disparities in access to care were essentially eliminated as a confounding factor. Accordingly, it was shown that maternal income was not associated with any of the outcomes. However, in spite of equal access, a significant gap in metabolic control between children of mothers with varying levels of education remained.

The study has several limitations. First, it was not possible to obtain information on all the mothers of children with type 1 diabetes in the registries. We were not able to assess why this information was missing, but evidence suggested that it was nonrandom, as HbA_1c varied systematically according to whether the mother was identified. Still, it was only for 6.8% of the children that the information was missing, so the impact on the interpretation of the results is likely modest. Data on HbA_1c at onset were also missing for a large proportion of the children (68%), but coverage was better during follow-up, with only 4% missing HbA_1c measures.

For clinicians and policymakers, our results suggest that it may be beneficial to provide extra support to the least privileged children during the first few years of diabetes. Recent research found both high levels of HbA_1c and low socioeconomic status to be negatively associated with standardized test scores in reading and math among children with type 1 diabetes (34). This suggests that efforts addressing numeracy and literacy to both parents and children may be warranted. For example, diabetes educators could apply alternative pedagogical approaches based on education research, e.g., by using more visual learning tools such as graphs and pictures. Other research has shown the importance of social and emotional support for health outcomes (35), which points toward the significance of increasing decentralized support. For instance, health care professionals could help facilitate the formation of patient groups through which families could help support each other outside the clinic.

Ultimately, more research is needed to understand why differences in behavior (such as differences in the number of glucose measurements) develop.

Funding. This study has been funded by the Independent Research Fund, Denmark, grant 8019-00055B. N.F.N. is affiliated with the Center for Economic Behavior and Inequality (CEBI). The activities of CEBI are financed by a grant from the Danish National Research Foundation.

Duality of Interest. J.S. serves as an adviser to Medtronic, Janssen, and Novo Nordisk; owns shares in Novo Nordisk; and has received fees for speaking on behalf of Medtronic, Sanofi, Novo Nordisk, and Bayer AG. N.F.N. is employed by Coloplast. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. N.F.N. and N.S. wrote the first draft of the manuscript. N.F.N., A.G., T.M.E., J.S., and N.S. contributed to the interpretation of data, critically revised the manuscript, and approved the final version for submission. N.F.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.