Unsupervised cardiometabolic phenotyping unmasks residual MACCE risk beyond LDL-C in acute myocardial infarction after revascularization.
Post-percutaneous coronary intervention (PCI) outcomes in acute myocardial infarction (AMI) remain heterogeneous despite guideline-directed therapy and lower low-density lipoprotein cholesterol (LDL-C) targets. We employed unsupervised clustering of cardiometabolic biomarkers to uncover novel clinical phenotypes that conventional stratifications overlook.
K-means clustering was applied to 13 standardized cardiometabolic variables in 668 AMI patients who underwent successful PCI. Cluster stability was assessed by bootstrap resampling. Clinical outcomes were evaluated by Kaplan-Meier analysis and multivariable Cox regression. Receiver operating characteristic curves and DeLong testing compared discriminative performance between phenotype classification and LDL-C target attainment.
Three phenotypes were identified: Ph0 Metabolically Balanced (n = 391, 58.5%), Ph1 Hyperglycemic-Dyslipidemic (n = 107, 16.0%), and Ph2 Decompensated Inflammatory-Catabolic (n = 170, 25.4%). Over a median follow-up of 31.9 months, major adverse cardiovascular and cerebrovascular events (MACCE) occurred in 19.4%, 31.8%, and 49.4%, respectively (P < 0.001). After full covariate adjustment, Ph2 remained independently associated with MACCE (hazard ratio 2.97, 95% confidence interval 2.00-4.41, P < 0.001), while Ph1 was attenuated to non-significance (P = 0.136). Despite having the lowest LDL-C, Ph2 carried the highest event rate. LDL-C target attainment (<1.8 mmol/L) did not discriminate MACCE risk [area under the curve (AUC) 0.503], whereas phenotype classification yielded an AUC of 0.626 (DeLong P < 0.001).
Unsupervised cardiometabolic phenotyping identified a decompensated inflammatory-catabolic phenotype that carried the highest MACCE risk despite having the lowest LDL-C, representing a high-risk subgroup unrecognizable by conventional lipid-centric stratification. These findings suggest that multi-dimensional metabolic profiling may complement LDL-C targets for residual risk identification in post-PCI AMI patients.
K-means clustering was applied to 13 standardized cardiometabolic variables in 668 AMI patients who underwent successful PCI. Cluster stability was assessed by bootstrap resampling. Clinical outcomes were evaluated by Kaplan-Meier analysis and multivariable Cox regression. Receiver operating characteristic curves and DeLong testing compared discriminative performance between phenotype classification and LDL-C target attainment.
Three phenotypes were identified: Ph0 Metabolically Balanced (n = 391, 58.5%), Ph1 Hyperglycemic-Dyslipidemic (n = 107, 16.0%), and Ph2 Decompensated Inflammatory-Catabolic (n = 170, 25.4%). Over a median follow-up of 31.9 months, major adverse cardiovascular and cerebrovascular events (MACCE) occurred in 19.4%, 31.8%, and 49.4%, respectively (P < 0.001). After full covariate adjustment, Ph2 remained independently associated with MACCE (hazard ratio 2.97, 95% confidence interval 2.00-4.41, P < 0.001), while Ph1 was attenuated to non-significance (P = 0.136). Despite having the lowest LDL-C, Ph2 carried the highest event rate. LDL-C target attainment (<1.8 mmol/L) did not discriminate MACCE risk [area under the curve (AUC) 0.503], whereas phenotype classification yielded an AUC of 0.626 (DeLong P < 0.001).
Unsupervised cardiometabolic phenotyping identified a decompensated inflammatory-catabolic phenotype that carried the highest MACCE risk despite having the lowest LDL-C, representing a high-risk subgroup unrecognizable by conventional lipid-centric stratification. These findings suggest that multi-dimensional metabolic profiling may complement LDL-C targets for residual risk identification in post-PCI AMI patients.
Authors
Xu Xu, Zhang Zhang, Er Er, Gu Gu, Zhang Zhang, Li Li, Dong Dong, Cao Cao, Wang Wang
View on Pubmed