Harnessing the Plasma Proteome to Predict Mortality in Heart Failure Subpopulations

BACKGROUND: We derived and validated proteomic risk scores (PRSs) for heart failure (HF) prognosis that provide absolute risk estimates for all-cause mortality within 1 year. METHODS: Plasma samples from individuals with HF with reduced ejection fraction (HFrEF; ejection fraction <40%; training/validation n=1247/762) and preserved ejection fraction (HFpEF; ejection fraction ≥50%; training/validation n=725/785) from 3 independent studies were run on the SomaScan Assay measuring ≈5000 proteins. Machine learning techniques resulted in unique 17- and 14-protein models for HFrEF and HFpEF that predict 1-year mortality. Discrimination was assessed via C-index and 1-year area under the curve (AUC), and survival curves were visualized. PRSs were also compared with Meta-Analysis Global Group in Chronic HF (MAGGIC) score and NT-proBNP (N-terminal pro-B-type natriuretic peptide) measurements and further assessed for sensitivity to disease progression in longitudinal samples (HFrEF: n=396; 1107 samples; HFpEF: n=175; 350 samples). RESULTS: In validation, the HFpEF PRS performed significantly better (P≤0.1) for mortality prediction (C-index, 0.79; AUC, 0.82) than MAGGIC (C-index, 0.71; AUC, 0.74) and NT-proBNP (PRS C-index, 0.76 and AUC, 0.81 versus NT-proBNP C-index, 0.72 and AUC, 0.76). The HFrEF PRS performed comparably to MAGGIC (PRS C-index, 0.76 and AUC, 0.83 versus MAGGIC C-index, 0.75 and AUC, 0.84) but had a significantly better C-Index (P=0.026) than NT-proBNP (PRS C-index, 0.75 and AUC, 0.78 versus NT-proBNP C-index, 0.73 and AUC, 0.77). PRS included known HF pathophysiology biomarkers (93%) and novel proteins (7%). Longitudinal assessment revealed that HFrEF and HFpEF PRSs were higher and increased more over time in individuals who experienced a fatal event during follow-up. CONCLUSIONS: PRSs can provide valid, accurate, and dynamic prognostic estimates for patients with HF. This approach has the potential to improve longitudinal monitoring of patients and facilitate personalized care.