Unsupervised cardiometabolic phenotyping unmasks residual MACCE risk beyond LDL-C in acute myocardial infarction after revascularization

Figure 1. Patient selection and study flowchart. AMI, acute myocardial infarction; ECMO, extracorporeal membrane oxygenation; MCS, mechanical circulatory support; PCI, percutaneous coronary intervention; TIMI, Thrombolysis in Myocardial Infarction.

Figure 2. Determination of optimal cluster number using four validity indices. Selection of optimal cluster number (K = 2–6) using four internal validity indices. (A) Silhouette score; (B) Calinski-Harabasz index; (C) Davies-Bouldin index; (D) Gap statistic. Dashed vertical lines and red markers indicate the selected solution of K = 3.

Figure 3. Standardized variable profiles across three cardiometabolic phenotypes. Heatmap of standardized variable means (Z-scores). Red indicates values above the cohort mean; blue indicates values below. CRP, C-reactive protein; HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro-B-type natriuretic peptide.

Figure 4. UMAP projection of cardiometabolic clustering variables. Each point represents one patient (n = 668). Colours and marker shapes indicate phenotype assignment. UMAP parameters: n_neighbours = 30, min_dist = 0.3, Euclidean metric. UMAP, uniform manifold approximation and projection.

Figure 5. Cumulative MACCE incidence by cardiometabolic phenotype. Kaplan–Meier curves with 95% confidence intervals. Numbers at risk are shown below the x-axis. P values from log-rank test with Bonferroni correction for pairwise comparisons. MACCE, major adverse cardiovascular and cerebrovascular events.

Figure 6. Cumulative all-cause mortality by cardiometabolic phenotype. Kaplan–Meier curves with 95% confidence intervals. Numbers at risk are shown below the x-axis. P values from log-rank test with Bonferroni correction for pairwise comparisons.

Figure 7. Multivariable Cox regression for MACCE and all-cause death. Hazard ratios (95% CI) for Ph1 and Ph2 relative to Ph0 (reference). Model 1: age, sex, BMI. Model 2: Model 1 plus smoking, hypertension, diabetes, dyslipidemia, prior MI, prior revascularization, prior stroke, and CKD. Model 3: Model 2 plus STEMI status, multivessel disease, implantation of ≥2 stents, admission GRACE score, beta-blockers, and PCSK9 inhibitors. All-cause death was modeled using Models 1 and 2 only owing to limited events (n = 61). BMI, body mass index; CI, confidence interval; CKD, chronic kidney disease; GRACE, global registry of acute coronary events; HR, hazard ratio; MI, myocardial infarction; PCSK9, proprotein convertase subtilisin/kexin type 9; STEMI, ST-elevation myocardial infarction.

Figure 8. Cumulative MACCE incidence stratified by LDL-C target attainment. Kaplan–Meier curves comparing patients with LDL-C < 1.8 mmol/L vs. ≥1.8 mmol/L, corresponding to the ESC 2019 guideline target for very-high-risk patients. Numbers at risk are shown below the x-axis. P value from log-rank test. LDL-C, low-density lipoprotein cholesterol; MACCE, major adverse cardiovascular and cerebrovascular events.

Figure 9. ROC curves comparing phenotype classification and LDL-C stratification for MACCE prediction. Cardiometabolic phenotype classification (Ph2 vs. Ph0 + Ph1) vs. LDL-C stratification (≥1.8 vs. < 1.8 mmol/L). AUC values with 95% confidence intervals are shown. P value from DeLong test. AUC, area under the curve; LDL-C, low-density lipoprotein cholesterol; MACCE, major adverse cardiovascular and cerebrovascular events; ROC, receiver operating characteristic.

Figure 10. Subgroup Cox regression analyses for MACCE by phenotype. Hazard ratios (95% CI) for Ph1 and Ph2 relative to Ph0 across subgroups defined by diabetes status and AMI type. Model definitions as in Figure 7. AMI, acute myocardial infarction; CI, confidence interval; HR, hazard ratio; NSTEMI, non-ST-elevation myocardial infarction; STEMI, ST-elevation myocardial infarction.