Pediatric Diabetes Consortium Type 1 Diabetes (T1D) New Onset (NeOn) Study: Factors Associated with HbA1c Levels One Year after Diagnosis

Abstract

Objective

To identify determinants of HbA1c levels one year after the diagnosis of type 1 diabetes (T1D) in participants in the Pediatric Diabetes Consortium (PDC) T1D New Onset (NeOn) Study.

Research Design and Methods

Diabetes-specific as well as socioeconomic factors during the first year following diagnosis were analyzed in 857 participants (mean age 9.1 years, 51% female, 66% non-Hispanic White) not participating in an intervention study who had an HbA1c value at 12 months.

Results

Mean ± SD HbA1c at one year was 62 ± 16 mmol/mol (7.8% ± 1.5). In univariate and multivariate analyses, clinical center, non-Hispanic White race, private health insurance, living with both parents, higher frequency of self-monitoring of blood glucose (SMBG), and lower insulin requirements were associated with lower HbA1c concentrations at one year (p<0.01). No association was found with gender, age, Tanner stage, BMI, DKA at onset, number of positive autoantibodies or HbA1c at onset, or number of visits to diabetes physician during the first year.

Conclusions

White race, higher socioeconomic status, two-parent household, more frequent SMBG and low insulin requirements are associated with lower HbA1c concentration one year after the onset of T1D in children.

Introduction

Glucose control, as reflected by HbA1c, is one of the strongest predictors of chronic complications in adults (1) and children (2) with type 1 diabetes (T1D) including proliferative retinopathy (3), diabetic nephropathy (4), cardiovascular disease and age-adjusted mortality (5, 6) rates. Poor glucose control during the first few years of T1D in children predicts development of microalbuminuria (7) and retinopathy (8) in adulthood.

Although there are reports on factors associated with HbA1c in patients with established T1D (9–19), there are no published studies on the association of diabetes-specific and socioeconomic factors with HbA1c during the early stages of the disease in a representative sample of children who receive their care at pediatric diabetes centers in the United States.

The Pediatric Diabetes Consortium (PDC) T1D New Onset (NeOn) study aims to improve the care of children with diabetes through sharing of best practices (20). This large and geographically and ethnically diverse cohort of youth with T1D offered us the unique opportunity to study possible factors that may be associated with metabolic control one year after the diagnosis of T1D.

Study Design and Methods

Patients

The PDC T1D NeOn Study enrolled 1,052 patients with T1D between July 2009 and April 2011. The protocol was approved by the Institutional Review Board (IRB) at each of the 7 participating centers. Informed consent was obtained from participants 18 years of age and older and from parents of those less than 18 years of age. Assent was obtained from participants <18 years of age as required by local IRB regulations. To be eligible for enrollment in the study, patients had to be <19 years of age and managed at one of the 7 participating PDC centers within 3 months of diagnosis. A detailed description of PDC and the design of the study have been published previously (20). The analyses reported herein included data from 857 participants; 163 were excluded from this analysis due to no HbA1c measurement available at one year (319–455 days from diagnosis), and 32 were excluded due to participation in an intervention study.

Data Collection

Demographic, socioeconomic status (SES) and clinical characteristic data were collected from medical records and from interviews with the patient and/or parent. Follow-up visits were completed per usual care and all visits during the first year post-diagnosis were entered in standardized electronic case report forms for the study.

Body mass index (BMI) at diagnosis was computed from the closest height and weight within ± 14 days of diagnosis. BMI percentiles and Z scores adjusted for age and gender were calculated using 2000 CDC population growth chart data (21). If height and/or weight was missing within ± 14 days of diagnosis or if the participant was <2 years of age at the time of diagnosis, BMI percentile was not calculated.

Diabetes ketoacidosis (DKA) was defined according to DCCT criteria of pH <7.3 or HCO3 <15mEq/L and treatment in a healthcare facility (22). Tests for diabetes autoantibodies were included any time prior to diagnosis or within 91 days after diagnosis for insulinoma-associated autoantibody (IA-2) or glutamic acid decarboxylase (GADA) and within 14 days after diagnosis for insulin autoantibody (IAA).

Statistical Analysis

Least squares regression was used to assess the association of baseline (demographic, SES, and clinical) and one-year (total daily insulin dose, SMBG frequency, and number of visits to diabetes management providers [excluding visits during the first 3 weeks following diagnosis, which correspond to initial education in centers where patients are not routinely admitted at diagnosis]) factors with HbA1c at one year from T1D diagnosis. Initial models were constructed for each baseline factor individually. A multivariate model of baseline factors was then constructed using stepwise selection with p-values <0.10 required to be included in the model. Due to multiple comparisons, only factors with p-values <0.01 were considered statistically significant although factors with p-values <0.10 were included in the model to adjust for potential confounding. Interaction terms were tested for all variables included in the final multivariate model with a p-value <0.01 required to be included. A second multivariate model added the one-year factors to the significant baseline factors. Continuous variables were tested for linearity by adding a quadratic term to the model. If significant non-linearity was detected the variable was divided into categories and analyzed as discrete variables. Missing covariates were treated as a separate category for discrete variables and a missing value indicator was added to the model for continuous variables. All reported p-values are two-sided. All analyses were conducted using SAS version 9.3 (SAS Institute, Cary, NC).

RESULTS

At the time of diagnosis, the mean ± SD age of the 857 participants was 9.1 ± 4.1 years, 51% were female, 66% were Non-Hispanic White, and 68% had private health insurance. Mean ± SD HbA1c was 102 mmol/mol ± 25 (11.4% ± 2.3), 19% of participants were overweight or obese (BMI ≥85^th percentile for age and gender) and 33% had DKA. Of the 510 participants who were tested for all three autoantibodies, 95% were positive for at least one (Table 1).

Table 1.

Participant Characteristics at Diagnosis (N=857^a)

	%
Gender: Female	51%
Age (years)
<5	19%
5–<12	56%
12–<19	25%
Mean ± SD	9.1 ± 4.1
Range	0.7 − 18.8
Tanner Stageb
I	67%
II	11%
III	7%
IV	7%
V	7%
BMI Percentilec
<85th	81%
85th–<95th (Overweight)	10%
≥ 95th (Obese)	9%
Race/Ethnicity
White	66%
Hispanic or Latino	21%
Black/African American	7%
Other/More than one race	6%
Health Insurance
Private	68%
Military/CHP/Medicaid/Medicare	30%
None	2%
Parent Education
High school or less	32%
AA/BS/BA	44%
MS/MA or professional degree	23%
Family Income
<$25,000	13%
$25,000−$49,999	19%
$50,000−$74,999	17%
$75,000−$99,999	14%
≥ $100,000	37%
Family Structure
Lives with both parents	72%
HbA1c mmol/mol (%)
<86 (<10%)	28%
86−<119 (10–<13%)	45%
≥ 119 (≥ 13%)	27%
Mean ± SD	102 ± 25 (11.4 ± 2.3%)
Range	37−180 (5.5−18.6%)
Number of Positive Autoantibodiesd
0	5%
1	20%
2	39%
3	35%
DKA	33%

^a.Number of participants with missing data: Tanner stage (412), BMI percentile (361), race/ethnicity (20), health insurance (15), parent education (163), family income (280), family structure (1), HbA1c (39), DKA (27).

^b.Imputed as stage 1 for girls < 8 years of age and boys < 10 years of age.

^c.BMI percentile not calculated for n=36 participants less than 2 years of age.

^d.Limited to those tested for all three auto antibodies (510).

One year after diagnosis of T1D, mean ± SD HbA1c was 62 mmol/mol ± 16 (7.8% ± 1.5) with a median insulin dose of 0.6 units/kg/day (interquartile range [IQR] 0.5 to 0.8). The median number of self-reported SMBG tests per day was 5 (IQR 4 to 6) and the median number of visits to diabetes providers during the first year of T1D was 4 (IQR 4 to 5). In cases where a blood glucose meter download was available (n=469), the median (and IQR) number of SMBG tests per day was similar to self-report. Insulin pump therapy was being used by 34% of participants at one year and 59% of all participants were taking three or more injections of insulin per day (Table 2).

Table 2.

Participant Characteristics at One Year (N=857^a)

	%
HbA1c mmol/mol (%)
<53 (<7%)	29%
53−<64 (7–<8%)	34%
64−<75 (8–<9%)	20%
≥ 75 (≥ 9%)	17%
Mean ± SD	62 ± 16 (7.8 ± 1.5%)
Range	31−140 (5.0−15.0%)
Insulin Delivery
Pump	34%
1−2 injections/day	7%
3–4 injections/day	45%
5 or more injections/day	14%
Median (IQR)b	4 (3 to 4)
Rangeb	1−12
Insulin Dose (units/kg/day)
<0.5	32%
0.5–<0.8	44%
≥ 0.8	25%
Median (IQR)	0.6 (0.5 to 0.8)
Range	0.02−4.6
SMBG (# tests/day)
1−3	13%
4–6	66%
≥ 7	22%
Median (IQR)	5 (4 to 6)
Range	2−18
Visits to Diabetes Team Providers through 1 Year
1−3	12%
4–5	76%
≥ 6	11%
Median (IQR)	4 (4 to 5)
Range	1−9

^a.Number of participants with missing data: number of injections (305), insulin dose (124), blood glucose monitoring (360).

^b.Limited to those not on an insulin pump (552).

Several baseline (i.e., at diagnosis) characteristics were significantly associated with lower HbA1c at one year in univariate analyses: clinical site, white race, private health insurance, higher household income, higher parent education level, living with both parents, and absence of DKA at diagnosis (p-value <0.01 for each, Table 3). In the multivariate analysis of the baseline characteristics, clinical site, race, health insurance status and family structure remained significant. The effects of family income and parent education were confounded with the other SES factors so that possible independent effects of these factors cannot be confirmed nor ruled out. In the multivariate analysis, there was no detectable association with gender, age, Tanner stage, BMI, number of positive anti-islet autoantibodies, DKA at onset or HbA1c at onset. Positivity for each of the autoantibodies, i.e., GADA, IAA or IA-2, was not associated with one-year HbA1c (p-value >0.10; data not shown).

Table 3.

Factors at Diagnosis Associated with HbA1c at One Year (N±857^a)

^a.Number of participants with missing data: race/ethnicity (20), health insurance (15), family structure (1), household income (280), parent education (163), Tanner stage (412), BMI percentile (361), DKA (27) and HbA1c (39).

^a.“Other” could be lives with mother, lives with father, splits time with mother and father, lives with legal guardian who is not parent, lives away at school or other.

^c.Analyzed as an ordinal variable.

^d.Analyzed as a continuous variable. Categories are for display purposes in this table.

^e.Missing values imputed as stage I for girls < 8 years of age and boys < 10 years of age.

^f.Limited to those tested for all three auto antibodies (510).

^g.The multivariate model contains all factors with an adjusted p-value <0.10 to account for potential confounding, but only p-values <0.01 are considered statistically significant in this analysis. Factors with blank entries in the multivariate columns were excluded from the model because p ≥0.10.

In the multivariate model for one-year factors, adjusting for the significant baseline factors indicated above, lower one-year HbA1c was associated with higher frequency of SMBG testing and a lower total daily insulin dose per kg of body weight (both p-value <0.001; Table 4, Figure 1).

Table 4.

Factors at One Year Associated with HbA1c at One Year (N±857^a)

^a.Number of participants with missing data: insulin dose (124) and blood glucose monitoring (360).

^b.Also adjusted for all factors (p<0.10) from diagnosis model in Table 3.

^c.Analyzed as continuous variables. Categories are for display purposes in this table.

^d.Visits during the first 3 weeks following diagnosis were excluded due to differences among centers in initial diabetes education.

Figure 1. Factors Associated with HbA1c at One Year (N=857a).

Bottom and top of each box denote the 25^th and 75^th percentiles. Horizontal line inside each box denotes the median and the dot denotes the mean.

a. Number of participants with missing data: race/ethnicity (20), health insurance (15), family structure (1), insulin dose (124), # daily BG tests (360).

DISCUSSION

This study describes a large cohort of racially diverse children from seven pediatric diabetes treatment centers across the United States and reports characteristics associated with glycemic control one year following diagnosis of T1D. We observed that race/ethnicity, socioeconomic factors, frequency of SMBG and lower dose of insulin were independently associated with HbA1c one year after diagnosis of T1D in children.

There have been few previous studies analyzing factors associated with HbA1c at the early stages of T1D in children. In a study of 275 European children, Mortensen et al. identified ethnic minority, higher HbA1c at diagnosis and positivity for GAD65 autoantibodies as predictors of higher HbA1c one year after diagnosis (23). Our study found that the association between race/ethnicity and one-year HbA1c was independent of SES, frequency of SMBG, pump use and insulin dose. Higher levels of HbA1c have been previously reported in African-American adults with and without diabetes (24) and minority youth with diabetes (9) compared with their white counterparts. Whether this difference reflects higher glycemia, or higher rates of glycosylation, is currently unknown (25, 26). It is also possible that the distribution of socioeconomic factors is narrower for minorities thus not allowing studies to completely rule out confounding. It is concerning, however, that African-Americans also have higher mortality rates and risk of diabetic complications than whites (27).

Several socioeconomic factors were significant in univariate analysis, but because of colinearity, were not included in the multivariate model. Health insurance, the SES factor that was included in the multivariate analysis, was independently associated with HbA1c at one year. Our analysis does not allow us to completely rule out a possible independent effect of family income, structure or education. Socioeconomic factors were not analyzed in Mortensen’s study (23) but analyses in established, long-duration T1D have demonstrated a strong impact of health insurance type (28, 29), family structure (12–14, 18), education (18), family income (19, 30) and other social circumstances (e.g. immigration) (17, 31, 32) on glucose control. Although health insurance type may impact the choice of insulin delivery modality (28), in our analysis both factors were analyzed as independent variables associated with HbA1c.

Higher frequency of SMBG was associated with lower one-year HbA1c after adjustment for the other variables in the model. The association between frequency of SMBG and glucose control has been reported in patients with established T1D in several (15, 16, 18, 19, 33–35) but not all (14, 36) studies. The disagreement among studies might be explained by the strong and complex relationships between SMBG and other factors, such as use of insulin pumps or family structure and support (14), which also affect HbA1c. Our and others’ studies (18) suggest that frequency of SMBG is associated with HbA1c independently of those factors, and thus frequent BG monitoring should be encouraged.

We found that lower dose of insulin was independently associated with HbA1c at one year. In patients who retain residual endogenous insulin secretion, the dose of exogenous insulin adjusted for body weight is a measure of both beta-cell function and insulin resistance. In this fashion, insulin dose has been used to define the partial remission period (37, 38). The long known association between greater beta-cell function and lower HbA1c (39) provides the rationale for trials that aim to preserve beta-cell function and prolong the partial remission period in an effort to prevent complications of diabetes.

Clinical center was independently associated with HbA1c concentrations after adjustment for other variables. Differences in the therapeutic approaches by different centers may explain this finding, highlighting the importance of networks that promote sharing clinical practices among centers.

In our study, age was not associated with HbA1c. This is contrary to studies in established T1D that have found that age, and in particular late adolescence, is associated with higher HbA1c (11, 14, 16, 31). A likely explanation for this discrepancy is that we studied children and adolescents only one year after onset and therefore before the deterioration in compliance with diabetes tasks and glucose control that are typical during adolescence (40, 41). Alternatively, the challenges of managing diabetes in teenagers may have been offset by the accelerated rate of autoimmune destruction of beta-cells in younger patients, compared with adolescents. Mortensen et al. (23) also did not find an association with age.

Several studies have demonstrated that the influence of glucose control on risk of complications persists for many years (5). HbA1c in the first five years of childhood-onset T1D predicts microalbuminuria (7) and retinopathy (8) on follow-up. HbA1c at onset of diabetes correlates with glucose control in subsequent years (36, 42) and predicted microalbuminuria 11 years later (6). Our study sought to understand the factors associated with metabolic control one year after T1D diagnosis in children. By addressing characteristics that are associated with HbA1c at the early stages of disease, long-term chronic complications could be prevented in patients with childhood-onset T1D.

Although the list of predictors that our study was able to assess was comprehensive, one limitation is that we did not explore psychological factors. Previous literature indicates that diabetes-specific conflict (15, 34), self-care behavior, eating disturbances, depression (11), and family behavior factors (12, 43) influence glucose control in established T1D likely through their effect on adherence to treatment (40). It is likely that these factors also affect HbA1c at the early stages of disease. Missing data for some of the factors studied at one year of onset is another limitation of the study.

In summary, we studied a large sample of children with T1D and analyzed a comprehensive series of variables at diagnosis and during the first year of T1D. We found that sociodemographic factors, SMBG frequency and low insulin requirements are associated with HbA1c at one year of childhood-onset T1D. Further investigation is needed to identify potentially modifiable factors that mediate these associations.

Abbreviations

HbA1c

Hemoglobin A1c

T1D

type 1 diabetes

PDC

Pediatric Diabetes Consortium

NeOn

New Onset Study

SMBG

self-monitoring of blood glucose

BMI

body mass index

DKA

diabetic ketoacidosis

IRB

Institutional Review Board

SES

socioeconomic status

CDC

Centers for Disease Control and Prevention

DCCT

The Diabetes Control and Complications Trial

IA-2

insulinoma-associated autoantibody

GADA

glutamic acid decarboxylase

IAA

insulin autoantibody

SD

standard deviation

BG

blood glucose

The Pediatric Diabetes Consortium Study Group

Clinical Centers: (Listed clinical center name, city, and state. Personnel are listed as (PI) for Principal Investigator, (I) for co-Investigator and (C) for Coordinators.) (1) Baylor College of Medicine, Houston, TX: Morey Haymond, MD (PI); Maria J. Redondo, MD, PhD (I); Krishna Hassan, MD (C); Kathy Shippy, RN, CCRP (C); Chris George (C); Mariam Pontifes (C); (2) Children’s Hospital of Los Angeles, Los Angeles, CA: Jamie Wood, MD (PI); Brian Ichihara, BA (C); Megan Lipton, MA, CCRP (C); Marisa Cohen, MPH (C); (3) Stanford University, Stanford, CA: Bruce Buckingham, MD (PI); Breanne Harris, BS (C); Satya Shanmugham, BS (C); (4) Barbara Davis Center for Childhood Diabetes, University of Colorado School of Medicine, Aurora, CO: Georgeanna J. Klingensmith, MD (PI); Eric Cruz, BA (C); Heidi Haro, BA, BS (C); Maria King, BA (C); Katherine Manseau (C); (5) University of Florida, Gainesville, FL: Desmond Schatz, MD (PI); Janet Silverstein, MD (I); Michael J. Haller, MD (I); Erica Dougherty, BS (C); (6) Yale University, New Haven, CT: William V. Tamborlane, MD (I); Eda Cengiz, MD (PI); Melody Martin, CCRP (C); Amy Steffen, BA (C); Lori Carria, MS (C); Darryll Cappiello (C); (7) University of Michigan, Ann Arbor, MI: Joyce M. Lee, MD, MPH (PI); Surair Bashir (C); Ashley Eason (C); Coordinating Center: Jaeb Center for Health Research, Tampa, FL: Roy W. Beck, MD, PhD; Katrina J. Ruedy, MSPH; Craig Kollman, PhD; Crystal G. Connor, MS, MPH; Beth Stevens.