2396-P: Comparing Cluster-Based and Extreme-Value of Polygenic Risk Score Approaches to Stratify Patients with Type 2 Diabetes

Introduction and Objective: Pathway-specific polygenic risk scores (pPRS) provide a biologically informed framework for characterizing genetic heterogeneity in type 2 diabetes (T2D). However, how pPRS-based stratification strategies inform therapeutic selection and comorbidity risk remains unclear. Methods: In an East Asian cohort of patients with T2D (n=19,734), we compared 2 pPRS-based classification approaches: (1) a cluster-based strategy using K-means clustering across several pathway-specific pPRS to derive 7 patient subgroups, and (2) an extreme-value strategy defining high-risk groups as the top 20% of each pPRS. Genetic architectures and clinical phenotypes relevant to treatment decisions were systematically evaluated. Results: The cluster-based approach identified 7 patient subgroups with multidimensional genetic profiles, shaped by both dominant and secondary pPRS contributions. These subgroups demonstrated clear separation in clinical traits, including age, BMI, HbA1c, triglyceride, HDL-C, and GPT levels (all P<.0001). By comparison, the top 20% pPRS approach largely reflected a single predominant pathway and provided limited separation of clinical characteristics among patients. Cluster-based subgroups also demonstrated stronger associations with medications use/comorbidities: several K-means subgroups, such as subgroup-obesity (S-obesity)(OR 1.60), S-body fat (OR 1.54), S-metabolic syndrome (OR 1.28) and S-residual glycaemic (OR 1.98) (all P<.0001), were significantly associated with injectable T2D medications. Furthermore, S-obesity (OR 1.34), S-metabolic syndrome (OR 1.14), and S-beta cell-PI (OR 1.17) subgroups showed significant associations with hypertension. Conclusion: Compared with an extreme-value pPRS cutoff, K-means clustering across multiple pPRS showed better distinguishing genetically and clinically T2D patient subgroups, with potential to inform individualized treatment decisions and targeted interventions for precision diabetes care. Disclosure W.H. Sheu: None. J.I. Rotter: None. Funding This work was supported by the grants MG-112-PP-18, MG-112-SP-09 and MG-113-GP-03 from the National Health Research Institutes in Taiwan and National Science and Technology Council, Taiwan (NSTC 114-2314-B-400-014).