Data-driven cardiometabolic phenogroups reveal distinct subclinical cardiac and proteomic profiles

Background Cardiometabolic disease is heterogeneous and incompletely resolved by conventional classification. We aimed to use expanded variables to identify data-driven phenogroups and characterise their echocardiographic and proteomic features. Methods Latent class analysis was applied to a discovery cohort (RESET; n=1,034) using fourteen cardiometabolic variables. A decision tree model was constructed using the most important variables to enable practical phenogroup classification and facilitate external validation. External validation used three cohorts: PICMAN (n = 120, proteomics), the UK Biobank (n = 344,817, proteomics and outcomes) and CHARLS (n = 12,145, outcomes). Results Five latent phenogroups were identified: Metabolically Preserved (lowest burden of adiposity, insulin resistance and hyperglycemia), with and without hypertension (each n=244; 23.6%); Lean-Insulin Resistant (IR) (n=140; 13.5%); Obese-Insulin Sensitive (IS) (n=211; 20.4%); and Obese-IR (n=195; 18.9%). Although Lean-IR and Obese-IS showed discordant adiposity and insulin/glycemic status, both exhibited greater subclinical diastolic dysfunction (36.7% and 38.8%, respectively) than the Metabolically Preserved group (16.5%; P<0.001), with impaired global longitudinal strain more prominent in Obese-IS (22.5% vs 13.6%; P<0.001). A decision tree model nominated four variables as most discriminative for the phenogroups (visceral adiposity, IR, elevated SBP, and HbA1c), classifying individuals with an AUC of 0.84 (0.81-0.86), and was used for external validation. Validation in the PICMAN cohort using plasma proteomes showed signatures for the two discordant phenotypes: an inflammatory emphasis in Lean-IR (IL-6, PLCB2) and a stronger hepatic metabolic/injury signature in Obese-IS (ADH4, KRT8/KRT18), upon a shared insulin-endocrine core in both (LEP, IGFBP1, FGF21). In the UK Biobank, phenogroups assigned using surrogates for visceral adiposity and IR (waist circumference and TG:HDL-C ratio) replicated the proteomic signature, and were associated with incident myocardial infarction and stroke. Hazard ratios for the composite outcome after adjusting for age and sex were 1.86 (1.75-1.98) for Lean-IR, 1.85 (1.75-1.97) for Obese-IS, and 2.75 (2.56-2.95) for Obese-IR, compared with the Metabolically Preserved group. Conclusion Data-driven classification with expanded variables reveals underlying heterogeneity. Phenogroups with discordant adiposity and insulin/glycemic status are associated with subclinical cardiac dysfunction, but differ in their proteomic pathways. The inferred divergent biology implies that individuals of different phenogroups may benefit from differing targeted prevention.