Plasma proteomics improves risk prediction in heart failure and reveals unique biology in chronic chagas cardiomyopathy

Fig 1. Kaplan–Meier survival by heart failure etiology. Kaplan-Meier survival by heart failure etiology. (A) Overall two-year survival curves for all etiologic subgroups (idiopathic, hypertensive, ischemic, Chagas cardiomyopathy, alcoholic, and the whole set of patients), with proportion of deaths shown to the right of each curve; significance (“*”) assessed by log-rank test. (B) Survival curve comparing Chagas versus non-Chagas etiology, demonstrating a significantly steeper decline in Chagas patients (p = 1.05e-05). (C–F) Separate survival comparisons for each non-Chagas etiology versus the remaining cohort: ischemic, hypertensive, alcoholic, and idiopathic HF curves.

Fig 2. Distribution of predictive performance across biomarker models. Distribution of predictive performance across biomarker models. (A) Boxplots show the F1-macro scores obtained from 1,000 random train-validation permutations (70/30 split, employing only the original train set) using logistic regression classifiers built with three feature sets: NT-proBNP (“BNP”), the 9-protein panel (“P9”), and the full proteomic panel with 734 proteins (“All Prots.”). (B) Predictive performance for the endpoint of all-cause mortality (“Death”) and for the composite outcome of death or heart transplantation between marker sets. Boxplots depict the distribution of F1-macro scores across 1,000 test set bootstraps using logistic regression models built with NT-proBNP (BNP) and the 9-protein panel (P9). A two one-sided tests (TOST) procedure was applied to formally assess equivalence between models for the same outcome (distributions were not significantly equivalent). (C) Time-dependent discrimination of biomarker models P9, P9 + MAGGIC, BNP, and MAGGIC for right-censored Cox proportional hazards survival prediction in the test set, under a cumulative/dynamic framework. Stepwise curves show AUC(t)…

Fig 3. Subgroup analysis of predictive performance differences between the P9 panel and NT-proBNP. Subgroup analysis of predictive performance differences between the P9 panel and NT-proBNP models. Points represent the absolute difference in F1-macro (P9 minus BNP) across the overall test set and clinically relevant subgroups, including etiology, LVEF strata, and NYHA class. Vertical error bars denote 95% confidence intervals based on 1,000 bootstraps on test set. Interrogation mark denotes small power due to sample size (alcoholic on test set: N = 31, deaths = 3).

Fig 4. UK Biobank (UKBB) compared to GENIUS-HF. UK Biobank (UKBB) compared to GENIUS-HF, according to the MAGGIC score per sample; left ventricular ejection fraction (LVEF) was excluded from the analysis as a covariate because data were missing for more than 90% of the UKBB participants. (A) Distribution of UKBB HFrEF cohort (left) against GENIUS-HF (right). A total of 431 HFrEF samples (total deaths = 28) were found for UKBB; for the test set of GENIUS-HF the total is 364 samples (57 deaths). (B) Comparative classification performance of P9 vs. BNP models in the GENIUS-HF and UKBB cohorts, analyzed overall and within the highest-risk (3rd) MAGGIC tercile. Metrics include F1-macro, iAUC, sensitivity, specificity, positive predictive value (PPV), and net reclassification improvement (NRI total, by event, and by non-event). Δ values represent the change in performance achieved by P9 relative to BNP. Asterisks denote statistical significance based on 1,000 bootstrap-based 95% confidence intervals excluding zero.

Fig 5. Enriched pathways from GSEA analysis of proteins. Enriched pathways from GSEA analysis of proteins. (A) Enriched pathways (“Term”) in the train set, that were further validated in the test set, for each etiology (hypertense and idiopathic groups had no validated pathways); main higher-level metabolic processes to the left (“Process”). NES values (normalized enrichment score, degree to which a term is over- or under-represented) to the right. (B) Network of genes present in more than one significant enriched pathway (“Term”) for Chagas and ischemic patients.