Introduction and Objective: We applied AI-based SHAP (Shapley Additive Explanations) clustering to identify high-risk individuals for diabetes in a population-based Korean cohort.Methods: KoGES data were analyzed, comparing individuals who developed type 2 diabetes (T2DM) within five years to those who remained diabetes-free. A predictive model utilizing transfer learning and intermediate fusion was used. Among 27,745 individuals, 12 SHAP-derived variables (Age, BMI, FBS, HbA1c, T2DM family history, Exercise, Smoking, Drinking, SBP, DBP, History of hypertension, History of dyslipidemia) underwent PCA and k-means clustering. Baseline characteristics and genetic profiles were compared across clusters.Results: Three clusters were identified based on future diabetes risk. Cluster 1 (C1) had the lowest risk, while C2 and C3 had higher risks (cumulative incidence: C1 1.6%, C2 20.1%, C3 23.6%). Compared to C3, C2 individuals were older with higher BMI and HbA1c, but had lower fasting glucose and HOMA-IR. C3 had more males, higher fasting glucose, and greater insulin resistance. SNP analysis showed IGF2BP2 (rs4402960, rs1470579, rs676951) risk alleles were enriched in C2, while CDKAL1 (rs9368222) was associated with C3.Conclusion: AI-based SHAP clustering identified two high-risk diabetes subgroups: one older with higher HbA1c, the other younger with elevated insulin resistance. Identifying these subgroups may facilitate targeted prevention.
I. Jung: None. S. Han: None. Y. Jung: None. E. Kang: None. M. Kim: None. S. Park: None. D. Lee: None. J. Yu: None. J. Seo: None. T. Ahn: None. N. Kim: None.
This work was supported by the National IT industry Promotion Agency (NIPA) grant funded by MSIT (No. S0252-21-1001), Development of AI Precision Medical Solution (Doctor Answer 2.0).
Source link
Leave a Reply