O-04: Evidence of clinically meaningful subclusters of pediatric patients with T1D identified using unsupervised clustering analysis: preliminary findings from the everyone study
A Clarke, J Curtis, Y Elia, D Wherrett, A D Paterson, D Korczak, F Babalola, M Aitken, C Ma, F MahmudAbstract
Introduction
Pediatric populations with Type 1 Diabetes (T1D) are heterogeneous, with distinct phenotypes emerging in youth with T1D.
Objectives
To identify clinically meaningful subclusters of patients with T1D using unsupervised clustering analysis based on a limited numbers of clinical features related to endogenous insulin production and insulin resistance.
Methods
Clusters were generated using K-means clustering and pilot data from the EveryONE study, which were compared using ANOVA with a pairwise comparison of means.
Results
Complete data from 91 participants with T1D aged 5 to 18 years clustered into 3 distinct phenotypes. Cluster 1 (n=32) was characterized by younger participants (12.2 ± 3.2 years), shorter diabetes duration (3.2 ± 2.0 years), higher residual insulin production and lower insulin resistance. Cluster 2 (n=33) was characterized by older participants (16.0 ± 2.0 years), longer diabetes duration (10.5 ± 3.5 years), little to no autoimmunity or residual insulin production, and preserved insulin sensitivity. Cluster 3 (n=26) was characterized by higher weight Z (1.6 ± 1.1), BMI Z (1.8 ± 0.8) and waist-to-hip ratio (0.85 ± 0.07), and showed evidence of insulin resistance according to several markers including higher total daily insulin dose (1.1 ± 0.3U/kg), lower estimated insulin sensitivity (SEARCH; 6.7 ± 2.1) & higher TG-to-HDL ratio (1.0 ± 0.5). Significant differences between clusters were observed based on HbA1c (C1: 7.5 ± 1.1% vs. C2: 7.9 ± 1.6% vs. C3: 8.4 ± 1.3%; P=0.02) and insulin dose-adjusted A1c (C1: 10.3 ± 1.4 vs. C2: 11.5 ± 2.4 vs. C3: 12.6 ± 1.6; P<0.0001).
Conclusions
Unsupervised clustering using a limited set of clinical factors identified clinically meaningful subtypes of youth with T1D that were associated with differences in glycemic management. The resulting profiles provide a foundation for precision medicine approaches to Type 1 Diabetes, including risk stratification and tailored treatment approaches.