1484-P: Six-Class Clustering and Its Complication Prediction in Chinese Aged Diabetes Cohorts Based on Deep Contrast Clustering Analysis

Aims: This work aims to establish an innovative clustering method predicting variable categories of diabetic complications in aged Chinese diabetes cohort.

Methods: We selected and extracted data of elderly diabetic patients (n=4980) from a medical examination group of 51,400 people followed-up yearly from 2014 to now in Kunshan, China.

Result: The patients were clustered into 6 categories by analyzing 20 indicators. Cluster 1 (n=592) was characterized by heavy smoking and high risk of complications of CVD (Cumulative 10-year incidence rate was 6%). Cluster 2 (n=1130) was characterized by high alcohol consumption, high aminotransferases, the highest risk of complications of stroke (Cumulative 10-year incidence rate was 12.3%) and high risk of FLD (60.5%). Cluster 3 (n=837) was characterized by high blood lipids, high risk of complications of FLD (69.6%) and stroke (11.8%). Cluster 4 (n=886) was characterized by good healthy indicators and a low risk of complications such as FLD (Cumulative 10-year incidence rate was 27.3%), stroke (Cumulative 10-year incidence rate was 6.9%) and CVD. Cluster 5 (n=749) was characterized by older age, higher uric acid, higher creatinine, and the highest risk of complications of CVD. Cluster 6 (n=786) was characterized by high waist circumference, high BMI, high blood pressure and the highest risk of complications of FLD. All three complications, stroke, CVD, and FLD, were analyzed by the Kruskal-Wallis test (p<0.05). The gene of nonalcoholic fatty liver disease (NAFLD) in the cluster 2 was the highest risk coefficient: 1.277 (95% CI:1.522-1.165, P<0.05). It was consistent with the cluster 2 having a higher prevalence of FLD.

Conclusions: A new clustering method has been developed in two large Chinese cohorts of aged diabetes, which may effectively predict complications by clustered in different categories.