Prediction of Major Adverse Cardiovascular Events in Peripheral Artery Disease: Integrating Metabolomics and Proteomics for Risk Stratification

Figures

• 

  Fig. 1.. Study design and multiomics workflow for risk prediction in PAD. This figure illustrates the framework of a prospective cohort study enrolling 164 patients with PAD. At hospitalization, clinical parameters including demographics, routine blood tests, biochemistry, and cardiovascular characteristics, as well as blood samples for plasma proteomic and metabolomic profiling, were collected. All patients were prospectively followed for MACE. Clinical, proteomic, and metabolomic data were analyzed to explore disease mechanisms and build risk prediction models. CP, clinical parameters; PAD, peripheral artery disease; MACE, major adverse cardiovascular events.

• 

  Fig. 2.. Proteomics differential analysis reveals proteomic alterations and pathway enrichment underlying major adverse cardiovascular events (MACE) in PAD patients. (A) Bar plot of significantly regulated proteins (adjusted P < 0.05) identified by proteomics differential analysis. A total of 144 proteins were up-regulated and 2 were down-regulated in MACE versus non-MACE patients. (B) Volcano plot of proteomic changes between MACE and non-MACE, highlighting especially up-regulated proteins with |log2 fold change| > 0.58 and adjusted P value < 0.05. Key inflammatory and matrix-remodeling factors are annotated. (C) Radar plot underscores the magnitude of dysregulation in these effector proteins, supporting their roles as potential biomarkers and therapeutic targets. (D) Functional enrichment analysis of differentially expressed proteins, with GO/KEGG terms covering humoral immunity, extracellular matrix, and metabolic processes. (E) Chord diagram mapping enriched KEGG pathways (P value < 0.05) to their contributing proteins, showing areas of immune activation and remodeling. (F) Protein interaction network of KEGG-enriched proteins, presenting functional connectivity and hubs of disease…

• 

  Fig. 3.. Differential metabolomic profiles and pathway enrichment analysis between PAD patients with and without MACE. (A) Volcano plot of differential metabolites: Shows significantly up-regulated and down-regulated metabolites between MACE and non-MACE groups (adjusted P < 0.05, |log2FC| > 1). (B) PLS-DA score plot: Depicts clear separation of MACE and non-MACE groups based on their global metabolic profiles. (C) Top 20 VIP metabolites in PLS-DA: Lists the most discriminatory metabolites driving class separation. (D) Pie chart of metabolite classes: Displays the proportion of major compound classes among significant metabolites, with carboxylic acids and derivatives most prevalent. (E) Dot plot of KEGG pathway enrichment, displaying enrichment ratios and statistical significance for major metabolic pathways. (F) Bubble plot of pathway enrichment results, visualizing enrichment significance and metabolite involvement across pathways. (G) Pathway interaction network: Illustrates connectivity among enriched metabolic pathways, with tryptophan metabolism at the core of multiple metabolic disturbances. (H) Metabolite–metabolite interaction network of significant metabolites. Nodes represent…

• 

  Fig. 4.. Integrative analysis of proteomic and metabolomic data in PAD with and without MACE. (A) Joint pathway enrichment analysis of differential proteins and metabolites. (B) Protein–metabolite interaction network. (C) Correlation analysis between differential proteins and metabolites. AGE-RAGE, advanced glycation end products and receptor for advanced glycation end products; PPAR, peroxisome proliferator-activated receptor; pathway impact, measure of the importance of the pathway in topology analysis.

• 

  Fig. 5.. Identification and prognostic analysis of key proteomic predicting MACE occurrence. (A) Variable selection frequency of the top proteins from nested cross-validated LASSO regression, with mean test AUC indicated for each protein. (B) Top 30 variable importance plot from RF analysis on the training data, highlighting the most influential proteins. Color intensity represents variable importance. (C) Venn diagram showing the overlap of stable features identified by LASSO and RF approaches. (D to H) Kaplan–Meier survival curves for MACE-free survival, stratified by high vs. low expression of the 5 selected proteins (MMP3, ADAM15, CHGB, TNFRSF6B, and TFPI2). Log-rank P values and hazard ratios (HR [95% CI]) are reported. (I) Forest plot summarizing univariate Cox proportional hazards regression results for the 5 candidate proteins, with HRs and 95% confidence intervals (averaged across 5-fold cross-validation). (J) ROC curves illustrating the predictive performance of each selected protein for MACE, with area under the curve (AUC [95% CI]) for MMP3, ADAM15, CHGB, TNFRSF6B, and TFPI2. RF, random forest; LASSO, least absolute shrinkage and selection operator; AUC, area under the ROC…

• 

  Fig. 6.. Construction of a clinical prognostic model for MACE in PAD patients. (A) Partial likelihood deviance plot for LASSO-Cox regression, illustrating the selection of optimal regularization parameter (λ) and identification of informative clinical predictors. (B) Multivariate Cox proportional hazards regression forest plot showing hazard ratios (HR) and 95% confidence intervals (CI) for each selected variable. (C) Kaplan–Meier survival curves for high- versus low-risk groups classified by the clinical risk score, demonstrating a significantly higher incidence of MACE in the high-risk group. (D) Clinical nomogram integrating hypertension, Cr, and BNP for individualized prediction of 1-year and 3-year survival probability in PAD patients. MACE, major adverse cardiovascular events; PAD, peripheral artery disease; LASSO, least absolute shrinkage and selection operator; HR, hazard ratio; CI, confidence interval; Cr, creatinine; BNP, B-type natriuretic peptide.

• 

  Fig. 7.. Comparative performance evaluation of the protein-based, clinical, and combined prognostic models for MACE prediction in PAD patients. (A) Time-dependent ROC curves at the 12-month landmark for the clinical (red), protein-based (pro, blue), and combined (orange) models, with AUC and 95% confidence intervals indicated. (B) Bar plot summarizing performance metrics including AUC, accuracy, F1 score, Gmean, precision, recall, and specificity for each model at 12 months. (C) Calibration curves for the clinical, protein, and combined models; Brier scores and 95% CI are shown, demonstrating good concordance between predicted and observed risk for the combined model. (D) Decision curve analysis showing the net clinical benefit of each model across varying risk thresholds for MACE. (E to G) Kaplan–Meier survival curves stratified by predicted risk groups (high vs. low) for the clinical model (E), protein model (F), and combined model (G), with cutoff values, hazard ratios (HR), 95% CI, and log-rank P values indicated. (H) Incremental model performance comparison by ΔAUC, IDI, and continuous NRI versus the combined model, with P values shown for each comparison. AUC, area under the curve;…