1327-P: Novel Subgroups of the Three-Year Risk of Incident Diabetes and Complications in Chinese Diabetes-Free Adults

Aims: The aim of this study was to analyze common clinical indicators in diabetes-free cohorts population using innovative clustering methods to derive characteristic clusters and to assess the utility for stratified prediction of diabetes risk and complications.

Methods: We analyzed the data of 13,829 diabetes-free adults from a physical examination cohort of 51,400 people follow-up from 2014 to present in Kunshan, China, and used the Ensemble clustering based on ending indicators and Weighted Naive Bayesian Classification method to select the most important variables from the candidate variables and assessed the 3-year risk of diabetes and the complications.

Result: We screened 13 clinically important variables. In the Kunshan cohort we identified 3 clusters. Cluster 2 (n=4,622) was characterized by the worst control of glucose and lipid metabolism, highest risk of diabetes and complications of FLD. Cluster 3 (n=2,456) was characterized by the oldest age, the highest SBP, BMI, waist circumference, intermediate risk of diabetes and the highest risk of CVD and stroke. Cluster 1 (n=6,751) population had good indicators and a low risk of diabetes and complications such as CVD, FLD and stroke. In the post-correction cox survival analysis for FLD, taking cluster 1 as the reference, HR of cluster 2 was 2.357(95% CI: 2.161-2.571, p<0.001) and HR of cluster 3 was 1.903(95% CI: 1.718-2.108, p<0.001). In the post-correction cox survival analysis of CVD, HR of cluster 2 was 1.193 (95% CI: 0.975-1.459, p=0.087) and HR of cluster 3 was 1.295 (95% CI: 1.041-1.611, p=0.02). The risk of diabetes and CVD predicted by PRS was consistent with that predicted by clinical clustering methods.

Conclusions: Phenotypes derived from the analysis of clinical characteristics using the new clustering method help to identify and stratify the risk of diabetes and related complications.