The differential impact of automated insulin delivery systems on body mass index in children with type 1 diabetes

Introduction

Automated insulin delivery (AID) systems represent a transformative advancement in the management of type 1 diabetes (T1D) in children and young people (CYP). Given the dynamic and unpredictable nature of childhood activity levels and dietary patterns, continuous insulin adjustments without repeated parental or caregiver intervention are essential for maintaining optimal glycaemic control [1]. AID systems integrate real-time continuous glucose monitoring (CGM) with insulin pumps, using algorithm-driven adjustments to automate insulin delivery [2]. Multiple studies and systematic reviews have demonstrated that AID systems significantly enhance glycaemic outcomes, increasing time in range (TIR, 3.9–10.0 mmol/L) whilst reducing hyperglycaemia and hypoglycaemia, and improving the quality of life for both people living with T1D and their caregivers [3], [4], [5]. These benefits have led to their widespread recommendation as the best therapeutic option for CYP with T1D [6], 7].

While the glycaemic benefits of AID therapy are well established, its impact on body weight remains unclear. Real-world data from over five hundred adult users suggest that body mass index (BMI) increases during the first five months of AID therapy [8]. In contrast, for CYP transitioning to AID systems, BMI z-score appears to remain stable in both clinical [9] and real-world settings [10].However, previous analyses have largely relied on univariate assessments, without considering key underlying factors such as baseline BMI status, gender, ethnicity, socioeconomic status (SES), and diabetes duration. Furthermore, historical data from the Diabetes Control and Complications Trial suggests that improvements in glycaemic control are strongly linked to weight gain over time [11]. This raises concerns about a potential increase in obesity-related complications, particularly among individuals at higher risk. Large international paediatric datasets consistently report that advanced diabetes technologies and intensified insulin regimens are associated with increased overweight and obesity prevalence [12], [13], [14]. While prior research suggests that, on average, BMI z-score does not significantly change upon transition to AID therapy [9], unstratified assessments may obscure patterns of BMI redistribution across different BMI categories [15]. Given the widespread adoption of AID systems, it is essential to investigate their real-world impact on BMI with a more comprehensive and nuanced approach. One study examining meal distribution and BMI in children using AID found that while AID improved TIR and increased both meal frequency and insulin use, it had no significant impact on BMI [16]. However, further studies are needed to determine whether AID has differential effects based on individual characteristics.

Following the implementation of a clinically validated hybrid virtual onboarding programme for AID systems in our tertiary centre [17], we set out to evaluate longitudinal changes in BMI z-score over 12 months. This study aims to address the existing knowledge gap by assessing the weight trajectory of CYP using AID, considering key demographic and clinical variables to better understand the real-world effects of these systems on BMI z-score changes.

The primary aim was to assess changes in BMI z-score from baseline to 6 and 12 months after transitioning to AID therapy in children and young people with type 1 diabetes. The secondary aim was to identify predictors of BMI z-score change over six months using a mixed-effects model.

Materials and methods

Data collection

Data was collected at AID initiation (baseline), 6 and 12-months post AID-initiation. Information was extracted from the electronic patient record system (TWINKLE) and included:

1. Demographics: Age, gender, and ethnicity, based on self-reported data and classified according to the UK Census categories (White, Asian, Black, Mixed, and Other) [23].

2. Socio-economic status (SES): Determined using the index of multiple deprivation (IMD), which is based on 37 indicators across seven domains. IMD scores range from 1 (most deprived) to 32,844 (least deprived) [24].

3. Main carer education: Categorised as: No qualifications, 1–4 GCSEs (grades A–C), ≥5 GCSEs (grades A–C), ≥2 A-levels or equivalent, or university degree [23].

4. Anthropometric data: Height (cm), weight (kg), BMI z-score, calculated using 1990 UK data [25].

5. BMI z-score groups: Participants were classified into five groups according to their baseline BMI z-score to assess changes in BMI z-score over time. Using the World Health Organisation (WHO) BMI classifications [15] with the healthy weight category split by below and above 0 (50th centile), the predefined thresholds were obese (BMI z-score ≥2), overweight (BMI z-score ≥1 to≤2), healthy weight above average (BMI z-score >0 to<1), healthy weight below average (BMI z-score >−2 to≤0) and underweight (BMI z-score ≤−2) [15].

6. Disease and treatment specifics: duration of diabetes was captured in years and months. Insulin delivery methods were categorised as multiple daily injections (MDI) or continuous subcutaneous insulin infusion (CSII). The AID systems in use included MiniMed 780G using Guardian 3 and 4 Sensors, CamAPS FX (CAMS) using Dexcom G6 sensors, t:slim X2 with Control IQ (CIQ) using Dexcom G6 sensors, and Omnipod 5 (OP5) using Dexcom G6 sensors.

HbA_1c measurements

HbA_1c (mmol/mol) readings were obtained at three time points: at initiation of AID (baseline), followed by assessments at 6 and 12 months. These measurements were conducted using the Siemens DCA Vantage point-of-care device, which is aligned with International Federation of Clinical Chemistry (IFCC) reference standards. The device has been shown to closely correspond with laboratory-based HbA_1c testing (r=0.99) [26] and has a maximum measurable value of 130 mmol/mol.

Glycaemic metrics

CGM metrics were collected for the 90-days preceding HbA_1c measurements at baseline, 6-months and 12-months. Also, for the first 90-days (3-months) post-initiation. Internationally agreed CGM metrics [27] were collected from Dexcom Clarity and Glooko.

1. Mean blood glucose (MBG, mmol/L).

2. Time below range level 2 (TBR2): Percentage of time glucose was<3.0 mmol/L.

3. Time below range (TBR): Percentage of time glucose was<3.9 mmol/L.

4. Time in tight range (TITR): Percentage of time glucose was 3.9–7.8 mmol/L.

5. Time in range (TIR): Percentage of time glucose was 3.9–10.0 mmol/L.

6. Time above range (TAR): Percentage of time glucose was>10.0 mmol/L.

7. Time above range level 2 (TAR2): Percentage of time glucose was>13.9 mmol/L.

8. Coefficient of variation (COV %).

9. Percent sensor use (PSU %).

Daily insulin units

At baseline, 90-day average daily insulin usage for CSII patients were collected from Glooko and Carelink. Similarly, AID system insulin usage at 6 and 12-months was collected from the same platforms. MDI patients self-reported average daily insulin usage for the 90-days prior to starting AID therapy. Insulin units per kilogram (u/kg/day) was calculated for each patient using synchronised weight for all time points.

Weighted averages of 6-months data

To minimise bias in BMI z-score group comparisons, both between and within groups, a weighted average approach was applied to time-dependent glycaemic variables [28]. This method ensured that periods with higher sensor utilisation contributed proportionally more to the overall metric and used the following formula:

The 3-month value and 6-month value refer to the recorded 90-day average CGM at each time point (e.g., MBG, TBR2, TBR, TITR, TIR, TAR, TAR2, COV, PSU). PSU at 3 months and PSU at 6 months represent the percentage of time CGM data were available at each respective time point. This weighting method ensured that periods with greater CGM adherence contributed proportionally more to the final estimate, preventing over-representation of time points with lower data capture. The baseline 90-day CGM metrics were then deducted from 6-month 180-day weighted averages to get the Δ6 month CGM metrics.

Results

Of the 285 CYP in the service at the time of the study, 180 on AID systems were screened for inclusion. Of those screened, eight were excluded due to a diagnosis under three months and two were excluded for less than 50 % CGM data at baseline.

Sub-group analysis

Baseline BMI z-score groups

Participants were classified into five BMI z-score groups at baseline: obese (n=34), overweight (n=45), healthy weight above average (n=45), healthy weight below average (n=38), and underweight (n=8). At baseline and six months, significant differences were observed in the underweight and healthy weight below average as explained in the subsequent text. (Table 2).

Table 2:

Anthropometric and glycaemic metrics by BMI z-score group before and six months after automated insulin delivery system initiation.

	Obese BMI z-score ≥2 n=34			Overweight BMI z-score ≥1 to ≤−2 n=45			Healthy weight above average BMI z-score >0 to<1 n=45			Healthy weight below average BMI z-score >−2 to≤0 n=38			Underweight BMI z-score ≤−2 n=8
Characteristics	Before AID Mean (±SD)	6 months post-AID Mean (±SD)	p-Value	Before AID mean (±SD)	6 months post-AID Mean (±SD)	p-Value	Before AID mean (±SD)	6 months post-AID Mean (±SD)	p-Value	Before AID mean (±SD)	6 months post-AID Mean (±SD)	p-Value	Before AID mean (±SD)	6 months post-AID Mean (±SD)	p-Value
Anthropometrics
BMI Z score	2.44 (±0.38)	2.45 (±0.52)	0.859	1.40 (±0.27)	1.45 (±0.43)	0.814	0.46 (±0.27)	0.52 (±0.47)	0.321	−0.59 (±0.36)	−0.21 (±0.68)	<0.001	−2.31 (±0.35)	1.19 (±1.33)	0.032
Daily insulin, U/kg/day	0.96 (±0.28)	0.87 (±0.27)	0.028	0.90 (±0.35)	0.87 (±0.38)	0.353	0.83 (±0.20)	0.78 (±0.19)	0.113	0.87 (±0.16)	0.85 (±0.12)	0.473	1.08 (±0.35)	0.93 (±0.27)	0.088
HbA1c, mmol/mol	62.7 (±10.8)	55.9 (±8.7)	<0.001	62.0 (±9.7)	55.6 (±10.4)	<0.001	61.4 (±10.0)	54.4 (±7.19)	<0.001	60.7 (±10.3)	52.7 (±6.9)	<0.001	64.7 (±9.7)	55.0 (±7.6)	0.001
CGM metrics
MBG, mmol/L	11.0 (±1.73)	9.5 (±1.1)	<0.001	10.7 (±1.6)	9.3 (±1.1)	<0.001	10.4 (±1.5)	9.0 (±0.9)	<0.001	10.5 (±2.0)	9.1 (±1.0)	<0.001	11.1 (±1.6)	9.3 (±0.5)	<0.001
TBR2%: <3.0 mmol/L	0.3 (±0.6)	0.3 (±0.6)	0.385	0.5 (±0.8)	0.3 (±0.4)	0.083	0.6 (±0.9)	0.6 (±0.8)	0.802	0.5 (±1.0)	0.4 (±0.9)	0.486	0.3 (±0.5)	0.4 (±0.7)	0.297
TBR%: <3.9 mmol/L	1.8 (±1.8)	1.8 (±1.2)	0.985	2.5 (±2.3)	1.9 (±1.1)	0.079	2.4 (±1.9)	2.4 (±1.8)	0.852	2.0 (±2.1)	1.8 (±1.5)	0.466	1.4 (±0.9)	1.4 (±0.9)	0.952
TITR%: 3.9–7.8 mmol/L	27.5 (±10.5)	40.9 (±9.7)	<0.001	28.2 (±10.5)	42.1 (±8.9)	<0.001	30.7 (±10.3)	45.2 (±9.9)	<0.001	30.6 (±11.4)	44.3 (±9.3)	<0.001	26.0 (±7.9)	41.5 (±4.6)	<0.001
TIR%: 3.9–10.0 mmol/L	46.0 (±14.1)	62.0 (±10.7)	<0.001	47.3 (±13.3)	63.6 (±9.7)	<0.001	50.3 (±12.4)	65.9 (±9.3)	<0.001	49.9 (±14.5)	65.7 (±9.7)	<0.001	45.6 (±13.0)	62.5 (±5.3)	<0.001
TAR%: >10.0 mmol/L	52.2 (±15.2)	37.4 (±11.8)	<0.001	50.2 (±14.4)	35.7 (±11.1)	<0.001	47.2 (±12.8)	32.0 (±9.8)	<0.001	48.1 (±15.0)	34.1 (±11.6)	<0.001	53.0 (±13.4)	37.0 (±8.0)	0.005
TAR2%: >13.9 mmol/L	25.4 (±14.2)	14.2 (±8.3)	<0.001	23.3 (±12.3)	12.6 (±8.4)	<0.001	32.0 (±9.8)	21.7 (±12.2)	<0.001	23.0 (±14.1)	11.1 (±7.1)	<0.001	23.9 (±12.4)	12.3 (±3.0)	0.016
COV, %	39.7 (±5.0)	40.2 (±5.1)	0.604	40.9(±5.0)	40.0 (±9.3)	0.542	39.9 (±5.3)	39.3 (±4.8)	0.452	39.2 (±4.7)	39.2 (±5.3)	0.962	37.8 (±5.1)	38.6 (±2.9)	0.659
PSU, %	83.6 (±15.7)	92.7 (±6.9)	0.005	79.6 (±16.9)	91.7 (±10.4)	<0.001	84.7 (±13.2)	94.0 (±4.4)	<0.001	85.9 (±15.1)	94.1 (±3.0)	0.002	85.8 (±17.1)	95.5 (±5.4)	0.132

1. MDI, multiple daily injections; CSII, continuous sub-cutaneous insulin infusion; GCSE, general certificate of secondary education; AID, automated insulin delivery; AutoAID, automated cloud-integrated AID; ManualAID, automated cloud-integrated and algorithm updated AID; 780G, MiniMed 780G system; CAMS, CamAPS Fx; CIQ, t:slim X2 with control IQ (CIQ); OP5, Omnipod 5 system; COV, co-efficient of variation; GMI, glucose management indicator; MBG, mean blood glucose; TBR2, time below range level 2; TBR, time below range; TITR, time in tight range; TIR, time in range; TAR, time above range; TAR2, time above range level 2; PSU, percent sensor use. Bold denotes statistical significance at p < 0.001.

BMI z-score: baseline to 6 months post-AID initiation

BMI z-score changes were minimal among those in the obese group (2.44–2.45, ΔBMI z-score: 0.01, 95 % CI: −0.09 to 0.11, p=0.859), overweight group (1.40–1.45, ΔBMI z-score: 0.05, 95 % CI: −0.05 to 0.16, p=0.814) and in the healthy weight above average group (0.46–0.52, ΔBMI z-score: 0.06, 95 % CI: −0.04 to 0.17, p=0.321). However, participants in the healthy weight below average category showed a significant increase in BMI z-score from −0.59 to −0.21 (ΔBMI z-score: 0.38, 95 % CI: 0.20 to 0.57, p<0.001) whilst the underweight group exhibited the most substantial increase in BMI z-score, rising from −2.31 to −1.19 (ΔBMI z-score: 1.13, 95 % CI: 0.13 to 2.12, p<0.001 Table 2 and Figure 1A).

Figure 1:

(A) Changes in BMI z-score, (B) change in time in range (3.9–10.0 mmol/L), (C) change in insulin dose (u/kg/day,) over six months of AID therapy by baseline BMI z-score category.

A one-way ANOVA confirmed a significant overall between group effect of baseline BMI z-score on ΔBMI z-score (p<0.001), with post-hoc Bonferroni tests revealing that the underweight group had a significantly greater increase compared to all other groups (p<0.001, Table 3 and Figure 1A). However, no significant differences were observed between the healthy weight above average, overweight, and obese groups (Table 3 and Figure 1A). Furthermore, the healthy weight below average group had a significantly greater increase in ΔBMI z-score compared to all other groups other than the underweight group (p<0.05, Table 4 and Figure 1A).

Table 3:

Comparison of changes in anthropometric and glycaemic metrics between BMI z-score groups from baseline (CGM) to six months post-AID initiation.

Characteristics	Obese BMI z-score ≥2 n=34 Δ mean (95 % CI)	Overweight BMI z-score ≥1 to ≤−2 n=45 Δ mean (95 % CI)	Healthy weight above average BMI z score>0 to<1 n=45 Δ mean (95 % CI)	Healthy weight below average BMI z-score >−2 to≤0 n=38 Δ mean (95 % CI)	Underweight BMI z-score ≤−2 n=8 Δ mean (95 % CI)	p-Value
Anthropometrics
ΔBMI z-score	0.01 (−0.09 to 0.11)	0.05 (−0.05 to 0.16)	0.06 (−0.04 to 0.17)	0.38 (0.21–0.57)b	1.13 (0.13–2.12)a	<0.001
ΔDaily insulin:, U/kg/day	0.09 (−0.17 to −0.01)	−0.03 (−0.09 to 0.03)	−0.05 (−0.11 to 0.01)	−0.02 (−0.8 to 0.4)	−0.15 (−3.3 to 0.03)	0.373
ΔHbA1c, mmol/mol	−6.8 (−9.8 to −3.7)	−6.4 (−9.3 to −3.4)	−7.1 (−9.7 to −4.4)	−8.0 (5.0–11.0)	−9.8 (−14.3 to −5.2)	0.845
CGM metrics
ΔMBG, mmol/L	−1.5 (−1.9 to −1.1)	−1.4 (−1.8 to −1.0)	−1.4 (−1.8 to −1.0)	−1.4 (−1.9 to −0.9)	−1.7 (−2.6 to −0.8)	0.976
ΔTBR2%: <3.0 mmol/L	0.1 (−0.1 to 0.3)	−0.2 (−0.5 to 0.3)	0.0 (−0.3 to 0.4)	0.1 (−0.3 to 0.2)	0.1 (−0.2 to 0.6)	0.420
ΔTBR%: <3.9 mmol/L	0.0 (−0.6 to 0.6)	−0.6 (−1.4 to 0.1)	0.0 (−0.7 to 0.6)	−0.2 (−0.7 to 0.3)	0.0 (−0.8 to 0.9)	0.582
ΔTITR%: 3.9–7.8 mmol/L	13.4 (10.7–15.9)	13.9 (10.8–17.0)	14.5 (10.8–18.2)	13.7 (10.6–16.9)	15.5 (11.9–19.0)	0.978
ΔTIR%: 3.9–10.0 mmol/L	16.1 (12.5–19.6)	16.3 (12.8–19.9)	15.6 (12.0–19.1)	15.8 (12.0–19.6)	16.9 (9.9–24.0)	0.997
ΔTAR%: >10.0 mmol/L	−14.9 (−18.7 to −10.9)	−14.6 (−18.5 to −10.6)	−15.3 (−19.3 to −11.2)	−14.0 (−17.9 to −10.0)	−16.0 (−25.5 to −6.5)	0.988
ΔTAR2%: >13.9 mmol/L	−11.2 (−14.7 to −7.8)	−10.6 (−13.8 to −7.5)	−10.4 (−13.6 to −7.3)	−11.9 (−15.6 to −8.1)	−11.6 (−20.3 to −2.9)	0.978
ΔCOV, %	0.5 (−1.5 to −2.5)	−0.9 (−4.1 to 2.1)	−0.7 (−2.4 to 1.1)	0.0 (−1.6 to 1.5)	0.8 (−3.2 to 4.8)	0.881
ΔPSU, %	9.1 (3.0–15.2)	12.3 (7.1–17.6)	9.4 (−5.6 to 13.2)	8.2 (3.3–13.2)	8.9 (−3.4 to 21.3)	0.794

1. MDI, multiple daily injections; CSII, continuous sub-cutaneous insulin infusion; GCSE, general certificate of secondary education; AID, automated insulin delivery; AutoAID, automated cloud-integrated AID; ManualAID, automated cloud-integrated and algorithm updated AID; 780G, MiniMed 780G system; CAMS, CamAPS Fx; CIQ, t:slim X2 with Control IQ (CIQ); OP5, Omnipod 5 system; COV, co-efficient of variation; GMI, glucose management indicator; MBG, mean blood glucose; TBR2, time below range level 2; TBR, time below range; TITR, time in tight range; TIR, time in range; TAR, time above range; TAR2, time above range level 2; PSU, percent sensor use. ^aUnderweight group is significantly different from all other groups (p < 0.001). ^bHealthy weight below normal group is significantly different from Healthy weight above normal, overweight, and obese, (p < 0.05). Bold denotes statistical significance at p < 0.001

Table 4:

Final mixed-effects model for BMI z-score change from baseline to six months.

Predictor	Estimate (B)	Std. Error	95 % CI lower	95 % CI upper	df	t-value	p-Value
Intercept	−0.47	0.22	−0.90	−0.04	166	−2.17	0.032
Baseline MBG	0.07	0.02	0.03	0.11	166	3.43	<0.001
Baseline BMI z-score	−0.14	0.20	−0.19	−0.09	166	−5.22	<0.001
BMI z-score × MBG	−0.04	0.02	−0.07	0	166	−2.27	0.025

Insulin usage: baseline to 6 months post-AID initiation

Daily insulin requirements did not change across BMI z-score groups, except in the obese category, where there was a small but statistically significant decrease from 0.96 to 0.87 u/kg/day (Δu/kg/day: −0.09, 95 % CI: −0.17 to −0.01, p=0.028, Table 3). No significant differences were observed in ΔBMI z-score between groups (Table 3 and Figure 1C).

Glucose metrics: baseline to 6 months post-AID initiation

HbA_1c decreased significantly across all BMI z-score groups with ΔHbA_1c ranging from −9.8 to −6.8 to (p<0.001) with no between group differences. MBG, TAR, and TAR2 followed a similar trend, with changes between −1.7 and −1.4 mmol/L (p<0.001) for MBG, −16.0 % to −14.0 % (p<0.001) for TAR, and −11.9 % to −10.4 % (p<0.001) for TAR2, all with no differences between groups. In contrast, TITR, TIR and PSU improved across all groups, with TITR increasing by 13.4–15.5 % (p<0.001), TIR by 15.6–16.9 % (p<0.001), and PSU by 8.2–12.3 % (p<0.001 for most groups), again, all with no between group differences. Neither TBR2, TBR nor COV showed any significant changes over six months or differences between groups. Full analysis is detailed in Tables 2 and 4 and illustrated in Figure 1.

Discussion

This study provides new insights into the impact of AID systems on BMI z-score trajectories in children and young people with type 1 diabetes. Consistent with earlier reports [8], 30], we observed stability in BMI for most of the cohort, with increases primarily among those underweight or below the average range at baseline. Importantly, our mixed-effects modelling demonstrated that baseline BMI and mean blood glucose (MBG) were the independent predictors of BMI change, while other demographic and clinical factors, including age, sex, ethnicity, socioeconomic status, carer education, therapy type, and sensor wear, were not significant. This modelling provides a robust explanation for the redistribution effect we observed and addresses the potential confounders raised by reviewers.

These BMI changes occurred in parallel with clinically meaningful improvements in glycaemic outcomes. Across all BMI groups, time in range increased by 15–17 % and HbA_1c improved by approximately 6 mmol/mol, sustained at 12 months. Time above range fell substantially, and importantly, there was no increase in hypoglycaemia. These findings are consistent with and extend prior real-world studies of AID confirming that substantial glycaemic benefit is achieved without an accompanying rise in obesity risk [8], 10], 30].

Concerns about weight gain with improved glucose control are longstanding. Historical data from the DCCT demonstrated a uniform link between tighter glucose control and increased weight [11]. More recently, data suggests rising BMI trajectories in young people using intensive insulin regimens and advanced technologies [31]. This reinforces concerns that technology-driven improvements in glycaemia may exacerbate obesity risk. Large international paediatric datasets have further demonstrated that intensified insulin regimens and advanced technologies are associated with increased prevalence of overweight and obesity [12], [13], [14]. Our results refine this narrative by showing that AID acts as a metabolic modulator: underweight or leaner children gain weight appropriately, while overweight or obese children improve glycaemic control without further weight gain.

The interaction between baseline BMI and MBG further clarified the observed pattern: higher MBG predicted greater increases in BMI among those starting with lower BMI z-scores, but this association was attenuated in participants already overweight or obese. This supports a mechanistic explanation in which improved glycaemic control through AID reduces caloric loss via glycosuria in leaner individuals, facilitating weight restoration, while those with higher BMI benefit metabolically without additional weight gain. This nuance contrasts with the uniform effect described in the DCCT [11] and underscores the weight-redistributive rather than weight-promoting nature of AID.

A further strength of this analysis was the use of weighted CGM metrics across the 3- and 6-month follow-up periods. This approach ensured that periods of higher sensor wear contributed proportionally more to the calculation of glycaemic variables, minimising bias from incomplete data. To our knowledge, this method has rarely been applied in paediatric AID studies and represents a novel way of safeguarding against artefacts from variable device adherence. By addressing missingness explicitly, we provide greater confidence that the observed associations reflect true metabolic relationships rather than technical artefacts.

Notably, HbA_1c was not retained as a predictor of BMI change in the mixed-effects model. This is consistent with our recent work showing that children and young people of Black heritage have HbA_1c values 4–5 mmol/mol higher than White or South Asian peers despite equivalent MBG and time in range, confirming HbA_1c is not a reliable surrogate for direct glucose exposure in diverse groups. Direct CGM-derived measures such as MBG are therefore more physiologically relevant [32].

Our findings also speak to concerns regarding subgroup differences. Age and sex were tested and found not to be significant predictors in either univariate or multivariable analyses, addressing the possibility that puberty or sex-specific factors might explain the results. Second, pubertal status was not directly captured, and while age was tested in the model, pubertal staging would have provided more granular insight. Similarly, prior therapy type (multiple daily injections vs. pump therapy) and sensor duration were not associated with BMI change in the model. These null findings provide reassurance that the observed redistribution was not driven by baseline therapy characteristics or CGM exposure but reflects the combined effect of BMI status and glycaemia.

Previous studies have sometimes reported no overall change in BMI following transition to AID [10], 30] but unstratified reporting may have masked important subgroup effects. Supporting this, a study examining meal distribution in children using AID found that while TIR improved and meal frequency increased, BMI did not change overall [16]. By applying stratified analyses and modelling, we highlight redistribution that aligns with physiological need.

Limitations must be acknowledged. First, the underweight subgroup was small (n=8), which reduces the precision of estimates for this group. While the direction and magnitude of effect were consistent, these results should be interpreted cautiously. BMI is a proxy for adiposity and does not distinguish fat from lean mass; future work should include body composition. Finally, diet, physical activity, and eating behaviours were not measured.

Despite these limitations, strengths include longitudinal weighted CGM approach to minimise bias, and inclusion of a large, ethnically diverse, socioeconomically deprived population. A further strength of this analysis was the use of mixed-effects modelling to maximise statistical power in a single-centre cohort. Together, these factors maximise relevance to real-world practice.

From a clinical standpoint, these results provide reassurance. Overweight and obese children can expect improved glucose control without further weight gain, while those underweight or below average can achieve catch-up growth. Taken together, the findings suggest that AID is a weight-redistributive rather than weight-promoting intervention, aligning weight trajectories more closely with physiological need.

These results also have policy implications. The recent NICE technology appraisal mandating national rollout of AID [7] is further supported by our finding that AID improves glucose control and supports healthy weight redistribution without promoting excess gain. This strengthens the case for universal provision, particularly in diverse and socioeconomically deprived cohorts.

Key messages

1. AID therapy increases BMI z-score in underweight children while maintaining stability in those overweight or obese.

2. Mixed-effects modelling confirms baseline BMI and mean glucose, not age, sex, HbA1c, or therapy type, predict BMI change.

3. Weight redistribution occurs alongside a 15–17 % increase in time in range and sustained HbA1c reduction without added hypoglycaemia.

Conclusions

Automated insulin delivery supports healthy weight redistribution in children and young people with type 1 diabetes. Underweight or leaner individuals experience appropriate weight restoration, while those already overweight or obese maintain weight stability. These outcomes occur alongside significant improvements in time in range and HbA_1c without increasing hypoglycaemia. Baseline BMI and mean blood glucose, rather than age, sex, or therapy type, predict weight change. Taken together, these findings show that AID systems act as metabolic modulators, improving glycaemic control while aligning weight trajectories with physiological need.