The Metabolic-Driven High Risk phenotype had 3.9-fold higher complication odds than the Age-Driven Moderate Risk phenotype (OR 3.94; 95% CI 1.89-8.21) despite being 7 years younger.
Observational (n=217)
Yes
Does unsupervised machine learning identify distinct clinical phenotypes with differential complication burden in patients with Type 2 Diabetes?
Unsupervised machine learning identified three distinct T2DM phenotypes driven primarily by triglycerides and BMI rather than HbA1c, revealing significantly different complication burdens that could guide precision treatment.
Odds Ratio: 3.94 (95% CI 1.89–8.21)
Absolute Event Rate: 71% vs 48%
Introduction and Objective: Traditional diabetes classification relies on disease duration and glycemic control, yet patients with similar HbA1c exhibit vastly different complication profiles. Unsupervised machine learning can help to identify distinct type 2 diabetes phenotypes with differential complication burden and guide precision treatment strategies. Methods: K-means clustering was performed on 217 T2DM patients with complete data from approximately 400 patients across multiple diabetes screening camps including age, gender, BMI, HbA1c, lipid panel (total cholesterol, triglycerides, LDL, HDL, VLDL), diabetes duration, and metabolic markers. Each phenotypes metabolic profile and complication prevalence (bone disease, neuropathy, dyslipidemia, hepatic steatosis) were characterized. Results: Three distinct phenotypes emerged: Phenotype 1;Metabolic-Driven High Risk; (32% of cohort): mean age 58 years, HbA1c 7.8%, triglycerides 210 mg/dL, BMI 30.5, multi-complication rate 71%—represents metabolically unhealthy diabetes with severe dyslipidemia. Phenotype 2 “Age-Driven Moderate Risk”; (44%): mean age 65 years, HbA1c 7.5%, triglycerides 145 mg/dL, BMI 26.2, multi-complication rate 48%—typical older diabetics with moderate control. Phenotype 3; “Young Well-Controlled”; (24%): mean age 45 years, HbA1c 6.8%, triglycerides 115 mg/dL, BMI 24.8, multi-complication rate 18%—successfully managed diabetes. Remarkably, Phenotype 1 had 3.9-fold higher complication odds than Phenotype 2 (OR 3.94, 95% CI 1.89-8.21) despitebeing 7 years younger. High triglycerides (200 mg/dL) and BMI (30) were strongest phenotype discriminators, while HbA1c showed minimal discriminatory power. Conclusion: Unsupervised machine learning gives data-driven phenotypes and guide precision treatment: aggressive triglyceride/weight management for Phenotype 1, standard care for Phenotype 2, maintenance for Phenotype 3. Disclosure S. Patil: None.
SHUBHASHREE PATIL (Fri,) conducted a observational in Type 2 Diabetes (n=217). Metabolic-Driven High Risk phenotype (Phenotype 1) vs. Age-Driven Moderate Risk phenotype (Phenotype 2) was evaluated on Multi-complication rate (bone disease, neuropathy, dyslipidemia, hepatic steatosis) (OR 3.94, 95% CI 1.89-8.21). The Metabolic-Driven High Risk phenotype had 3.9-fold higher complication odds than the Age-Driven Moderate Risk phenotype (OR 3.94; 95% CI 1.89-8.21) despite being 7 years younger.