Longitudinal Continuous Glucose Monitoring Study in Young Children With Presymptomatic Type 1 Diabetes Followed in the Environmental Determinants of Islet Autoimmunity (ENDIA) At-Risk Cohort Study

Abstract

OBJECTIVE

To characterize longitudinal continuous glucose monitoring (CGM) data in young children with presymptomatic type 1 diabetes.

RESEARCH DESIGN AND METHODS

Between 2021 and 2024, children in the Australian ENDIA study with persistent multiple islet autoimmunity underwent blinded CGM assessments every 3–6 months. CGM-derived metrics (SD sensor glucose, coefficient of variation, mean sensor glucose, and percent CGM time >7.8 mmol/L [140 mg/dL]) were determined for each child by time from islet autoantibody detection.

RESULTS

A total of 178 CGM assessments were analyzed for 36 children (median [Q1, Q3] age at first assessment 4.5 [3.5, 6.0] years) who underwent a median of 5.5 (2.0, 7.0) assessments of 11 (9, 15) days duration each. High within-person variability was observed in serial CGM metrics, including percent CGM time >7.8 mmol/L (140 mg/dL) (intraclass correlation coefficient 0.30).

CONCLUSIONS

Further research is needed to inform interpretation of CGM-derived metrics in young children with presymptomatic type 1 diabetes.

Graphical Abstract

Introduction

Current consensus guidance on monitoring individuals identified with presymptomatic type 1 diabetes highlights the need for additional research on the role of continuous glucose monitoring (CGM) for disease staging and prediction of clinical onset in this population (1). CGM can identify early dysglycemia in children and adults with presymptomatic type 1 diabetes (2–6), including those with normal HbA_1c and oral glucose tolerance test (OGTT) results (7). Several studies have reported percent CGM time >7.8 mmol/L (140 mg/dL), derived from a single CGM assessment, being predictive of risk of progression to clinical onset with varying sensitivity and specificity (2,3,7,8). Aside from a recent analysis of longitudinally measured CGM, HbA_1c, and OGTTs in children and adults with presymptomatic type 1 diabetes, which found that intermittent CGM combined with HbA_1c was almost as effective as OGTT for predicting progression to stage 3 type 1 diabetes (6), longitudinal CGM data collected during presymptomatic type 1 diabetes are limited.

Since 2021, children followed in the Australian Environmental Determinants of Islet Autoimmunity (ENDIA) study (9,10) have been invited to participate in a substudy that involves 3–6 monthly CGM assessments from the time of islet autoantibody detection (clinical trial reg. no. ACTRN12613000794707). The current study aimed to describe CGM data from longitudinal assessments conducted between 1 January 2021 and 31 December 2024 to provide novel insights into individual patterns in this young cohort.

Research Design and Methods

Study Population

Commencing in 2013, the ENDIA study (9,10) is an Australia-wide at-risk cohort study longitudinally following 1,473 children from pregnancy/early-life to 10 years of age or clinical onset of (stage 3) type 1 diabetes. Children are tested for islet autoantibodies (insulin, IA-2, GAD, zinc transporter 8 antibody) every 3 months to age 2 years and every 6 months thereafter. Persistence is defined as detected on two or more consecutive visits at least 3 months apart (9,10).

Study Design

The ENDIA CGM substudy has been previously described (5). Briefly, since 2021, all ENDIA children with persistent islet autoimmunity have been invited to undergo serial blinded CGM using the Dexcom G6 system. Participants wear two consecutive sensors at 3–6 monthly intervals according to their age and stage. Following CGM assessment, CGM data are reviewed by pediatric endocrinologists, who then provide feedback to participant families.

CGM Data Validation

CGM data were downloaded from the Dexcom Clarity database as comma-separated value files. Sensor glucose readings for 12 h following each sensor insertion were excluded from analysis due to potential discrepancies in measurements during this period (7). Following this, CGM assessments with <4 days with >70% sensor glucose readings were excluded from further analysis. In addition, CGM daily profiles were manually and independently reviewed by three pediatric endocrinologists experienced in using CGM in children with type 1 diabetes. Reviewers were blinded to participant identifier, islet autoantibody profile, and time since autoantibody detection. CGM data deemed erroneous by two or more reviewers were excluded from analysis.

Variables

The study assessed the following variables:

• CGM-derived metrics: Standard CGM metrics (11) were derived for each CGM assessment completed for each child.

  • Flag for >10% CGM time >7.8 mmol/L (140 mg/dL): A binary indicator variable was created and set to 1 for CGM assessments with >10% time >7.8 mmol/L (140 mg/dL).
  • Proxy for fasting glucose: Mean sensor glucose between 3:00 and 5:00 a.m., rather than 6:00 a.m., calculated as a proxy for fasting glucose for each CGM assessment, as children may wake early and eat breakfast before 6:00 a.m.

• HbA_1c: Following islet autoantibody detection, HbA_1c was measured at each ENDIA visit in accredited clinical laboratories. HbA_1c measurements within 3 months of the CGM assessment were obtained.

• Time from last CGM assessment to date of clinical diagnosis: The diagnosis of stage 3 type 1 diabetes was measured according to standard criteria (12), with the commencement of insulin therapy being an ENDIA end point.

Statistical Analysis

Spaghetti plots of SD sensor glucose (SDSGL), coefficient of variation (CV), mean SGL, and percent CGM time >7.8 mmol/L (140 mg/dL) were generated to present these metrics for each child by time from islet autoantibody detection. Intraclass correlation coefficients (ICCs) for each CGM metric were estimated using linear mixed models to estimate within- and between-person variability in serial measures. Measures were log-transformed where residuals and/or random effects showed substantial deviation from a normal distribution. An exploratory analysis of children with one or two consecutive CGM assessments flagged as having >10% time >7.8 mmol/L (140 mg/dL) was conducted to estimate the sensitivity, specificity, and positive predictive value (PPV) (along with 95% Wilson CIs) of a single/double consecutive positive flag status in predicting progression to clinical diagnosis of (stage 3) type 1 diabetes in the following 12 months.

Ethics Statement

Ethics approval was obtained nationally for the ENDIA CGM substudy from the Women and Children’s Hospital in Adelaide (2020/HRE01400) and Child and Adolescent Health Service in Western Australia (HREC RGS 0000002402). Written informed consent was provided by each child’s parent/caregiver.

Data and Resource Availability

Deidentified participant data will be made available following completion of the ENDIA CGM substudy to investigators whose proposed use of the data has been approved by an independent review committee (learned intermediary) identified for this purpose. Requests for data can be made by e-mail to the ENDIA Study Chief Operating Officer at endia@adelaide.edu.au.

Results

Clinical Characteristics

Between 1 January 2021 and 31 December 2024, 36 of 45 (80.0%) ENDIA children with persistent multiple islet autoimmunity eligible to enrol in the ENDIA CGM substudy were consented. Median (Q1, Q3) age of participants at the time of their first CGM was 4.5 (3.5, 6.0) years, and 21 (58.3%) were boys. Eight of 36 (22.2%) participants progressed to clinical onset of type 1 diabetes during the study period, with time from last CGM assessment to diagnosis ranging between 0.2 and 13.8 months.

CGM Data

A total of 185 CGM assessments were conducted over the study period, ranging from 1 to 12 assessments per participant over a median 2.0 (0.9, 2.7) years of follow-up. After excluding CGM assessments determined to be erroneous (saw-tooth appearance/expert reviewer consensus), 12 h after each sensor insertion, and CGM assessments with <4 days with >70% CGM data (n = 7), 178 CGM assessments for 36 participants were available for data analysis (median 5.5 [2, 7] CGM assessments per participant of a median 11 [9, 15] days duration) (Supplementary Table 1). The median number of CGM assessments for the eight participants who progressed to clinical diagnosis was 2.5 (1.75, 7), ranging from 1 (n = 2) to 8 (n = 2).

Spaghetti plots of SDSGL, CV, mean SGL, percent CGM time >7.8 mmol/L (140 mg/dL), mean 3:00–5:00 a.m. sensor glucose, and HbA_1c are presented for each child by time from islet autoantibody detection and progressor status, illustrating heterogeneity in longitudinal serial measurements (Figs. 1A–D and 2A and B). Examination of estimated ICCs for each measure found that 70% of the variability observed in percent CGM time >7.8 mmol/L (140 mg/dL) was explained by within-person, rather than between-person, sources (Table 1). Visual inspection of serial CGM assessments available for six of the eight participants who progressed to clinical type 1 diabetes suggested progressively increased hyperglycemia (Fig. 1A and Supplementary Appendix 1), with all having consecutive CGM assessments with >10% time >7.8 mmol/L (140 mg/dL) prior to diagnosis (Fig. 1A).

Figure 1.

Percent CGM time >7.8 mmol/L (140 mg/dL) (A), SDSGL (B), CV SGL (C), and mean SGL (D) by time since islet autoantibody detection and progressor status at end of the study period. Each line represents an individual participant. Time from last CGM assessment to clinical diagnosis and insulin commencement for progressors was 0.2 months (pink), 0.6 months (orange), 1.9 months (light purple), 3.3 months (dark green), 3.6 months (brown and blue), 7.9 months (dark purple), and 13.8 months (light green).

Figure 2.

HbA_1c percent (A) and mean 3:00–5:00 a.m. sensor glucose (B) by time since islet autoantibody detection and progression status at the end of the study period. Each line represents an individual participant. Time from last CGM assessment to clinical diagnosis and insulin commencement for progressors was 0.2 months (pink), 0.6 months (orange), 1.9 months (light purple), 3.3 months (dark green), 3.6 months (brown and blue), 7.9 months (dark purple), and 13.8 months (light green).

Table 1.

Within- and between-person metrics

Metric	Overall mean (SD)	Within-person SD	Between-person SD	ICC
Log(percent time >7.8 mmol/L [140 mg/dL])	2.1 (0.9)	0.81	0.53	0.30
Log(SDSGL)	0.2 (0.3)	0.18	0.21	0.58
Mean SGL	6.3 (0.6)	0.53	0.44	0.40
Log(CV)	19.1 (4.5)	0.16	0.17	0.54
Mean SGL (between 3:00 and 5:00 a.m.)	5.8 (0.7)	0.55	0.39	0.34
HbA1c	5.2 (0.3)	0.25	0.15	0.27

Exploratory analysis of participants with CGM assessments flagged as >10% time >7.8 mmol/L (140 mg/dL) with ≥12 months follow-up after the flag, or progression to clinical diagnosis of (stage 3) type 1 diabetes within this period, identified 23 participants with a single flagged CGM assessment and 11 with two consecutive flagged CGM assessments. Three of the 23 single flagged participants progressed to clinical diagnosis within 12 months of the flag (sensitivity 100% [95% CI 43.8%, 100%], specificity 25.9% [95% CI 13.2%, 44.7%], PPV 13.0% [95% CI 4.5%, 32.1%]). Of the 11 participants identified with two consecutive flagged CGM assessments, 4 progressed to clinical diagnosis within 12 months of the second flag (sensitivity 100% [95% CI 51.0%, 100%], specificity 69.6% [95% CI 49.1%, 84.4%], PPV 36.3% [95% CI 15.2%, 64.6%]).

Conclusions

The detailed longitudinal CGM-derived metrics in young children with presymptomatic type 1 diabetes presented in this study provide new insights into individual trajectories of CGM metrics over time. A key finding is significant within-person heterogeneity observed in serial CGM metrics over a median duration of 2 years’ follow-up, including percent CGM time >7.8 mmol/L (140 mg/dL). High within-person variability in CGM metrics has also been reported recently in a study of longitudinal CGM conducted in Belgian children and adults with presymptomatic type 1 diabetes (6), and discordant CGM metrics from assessments conducted 6 months apart were observed in German youth (8).

Potential sources of within-person variability in serial CGM assessment could include concurrent fever/illness, medication use (e.g., steroids), physical activity, measurement error, and dietary intake. For example, it is possible that parents/caregivers may have made sustained, self-directed changes to the dietary (e.g., carbohydrate) intake of their children after receiving feedback on their child’s glycemic status. Within-person variability in longitudinally measured CGM metrics may also reflect variable β-cell function, variable α-cell function, and/or insulin resistance over the course of progression to type 1 diabetes (13). For instance, an increasing trend in percent CGM time >7.8 mmol/L (140 mg/dL) might be expected in a cohort of children with presymptomatic type 1 diabetes if all are progressing toward clinical onset at, albeit, a variable rate, which would result in high within-person variability in serial measurements.

The low number of participants with serial CGM assessments who progressed to clinical (stage 3) type 1 diabetes minimized the feasibility of investigating the utility of CGM-derived metrics for predicting progression to clinical diagnosis. Previous studies have reported percent CGM time >7.8 mmol/L (140 mg/dL) predicting progression to clinical diagnosis with varying specificity and sensitivity (2,3,7). In this study, exploratory analyses conducted for spending >10% CGM time >7.8 mmol/L (140 mg/dL) at baseline suggest that this flag has high sensitivity and moderate specificity for predicting progression in this population.

Strengths of this study include availability of 178 validated CGM assessments with a median duration of 11 days conducted over a median 2-year follow-up period in participants undergoing a median of five CGM assessments each, all using the same CGM system. These data are globally unique and provide clinically relevant insights regarding longitudinal CGM data in young children with presymptomatic type 1 diabetes who may progress more rapidly to clinical (stage 3) type 1 diabetes. As more invasive measures of glycemia (e.g., random/fasting blood glucose and OGTT) have lower feasibility and acceptability in this age-group (14,15), it is important to address the evidence gap regarding optimal use of CGM for monitoring progression in this population (1).

Further research is needed to extend the understanding of whether heterogeneity in serial CGM metrics observed in our study reflect a dynamic bidirectional nature of glycemic progression in presymptomatic type 1 diabetes or a modifiable within-person source of variability (e.g., dietary intake) while undergoing CGM assessments. Meanwhile, while providing additional clinical insights helpful for counseling families with children identified as having presymptomatic type 1 diabetes, its role in accurately staging individuals to determine their eligibility for disease-modifying therapies or predicting time to clinical onset remains unresolved.

This article contains supplementary material online at https://doi.org/10.2337/figshare.29640536.

Funding Statement

This study was supported by grants from Diabetes Research Western Australia, the Women and Children’s Hospital Research Foundation, the Australasian Paediatric Endocrinology Group, and The Leona M. and Harry B. Helmsley Charitable Trust (grant 2205-05241). A.H. was supported to lead this study by a JDRF postdoctoral fellowship (grant 3-PDF-2020-939-N) and Raine Medical Research Foundation priming grant. The ENDIA study is supported by Breakthrough T1D, the recipient of the Commonwealth of Australia grant for Accelerated Research under the Medical Research Future Fund (grants 3-SRA-2023-1374-M-N, 3-SRA-2020-966-M-N, 1-SRA-2019-871-M-B, and 4-SRA-2015-127-M-B). Dexcom, Inc. provided CGM products via an investigator-initiated research agreement (Steeprock ID OUS-2019-029).