Machine Learning–Based Clustering to Predict Functional Beta-Cell Depletion in Type 1 Diabetes

Introduction and Objective: No studies to date have predicted functional pancreatic beta cell depletion in type 1 diabetes (T1D) via machine learning. The aim of this study was to develop a prediction model using machine learning to stratify patients by depletion risk. Methods: Records for 293 patients with T1D (180 with acute-onset T1D [AT1D] and 113 with slowly-progressive T1D [SPT1D]) were obtained from the TIDE-J multicenter longitudinal database of Japanese T1D patients. We identified clinical factors associated with functional beta cell depletion, which was defined as C-peptide [CPR] level below detection limit, using a Cox proportional hazards model. We also applied Random Survival Forest (RSF)-based survival analysis to predict functional beta cell depletion. The RSF-derived case-wise similarity matrix was visualized using Uniform Manifold Approximation and Projection (UMAP), followed by k-means clustering based on depletion risk. Results: In Cox analysis, ZnT8 antibodies (FDR-adjusted P = 0.001), IA-2 antibodies (P = 0.020), and BMI (P = 0.027) were associated with functional beta cell depletion. The RSF model also identified urinary ketone bodies, body weight, weight loss, and GAD antibodies as significant predictors. Patients were clustered into four groups (Clusters 1-4). In confirmation of clustering using a decision tree, body weight, weight loss, and ZnT8 antibodies (cutoff: 205 U/mL) were identified as the determinants of cluster assignment. All cluster 1 individuals retained measurable CPR for more than 6 years, while 50% of cluster 4 patients progressed to functional beta cell depletion within 4 years. Both, AT1D and SPT1D were distributed across all clusters. Conclusion: Machine learning approaches enable prediction of functional beta cell exhaustion across T1D subtypes using limited clinical indicators. Disclosure: A. Shirozu: None. E. Kawakami: Consultant; Current; THE PHAGE. T. Ishikawa: None. A. Imagawa: Consultant; Current; Sanofi. T. Hanafusa: None. R. Suzuki: Consultant; Current; Novo Nordisk. Advisory Panel; Ended; Eli Lilly and Company, Sanofi K.K. H. Kajio: None. N. Kodani: None. D. Chujo: Speaker's Bureau; Current; Sanofi K.K.