Proteomics-based clustering outperforms clinical clustering in identifying people with heart failure with distinct outcomes

Abstract

BACKGROUND: Heart failure (HF) clustering typically relies on clinical characteristics which may not reflect underlying pathophysiology relevant for personalized medicine. We aimed to identify plasma protein profiles of HF patients with reduced ejection fraction (HFrEF). METHODS: Using latent class analysis, we derived clusters based on 1) clinical characteristics, and 2) proteomics (SomaScan) from 379 HFrEF patients (median age 64 years [Q1 56; Q3 72], 73% male). Survival analysis assessed associations with major cardiovascular (CV) events (HF hospitalization, CV death, or advanced therapy), HF hospitalization, CV death, and all-cause mortality. Associations were validated in 511 external patients (median age 72 years [Q1 63; Q3 79], 70% male). We identified differentially expressed proteins and explored whether proteins are targets of developmental or approved drugs. RESULTS: We show that clinical clustering identifies three patient clusters without distinct disease progression. Contrary to this, clustering based on plasma proteomics identifies three patient clusters with clear differences in disease, which are validated in the external cohort. The slowly progressing cluster 1 includes younger patients with fewer comorbidities, while the rapidly progressing cluster 3 consists of older patients with more atrial fibrillation and renal failure, and the hospitalization cluster 2 is intermediate in many characteristics. Medication use is similar across clusters. Relative to cluster 1, patients in cluster 2 have an increased risk of major CV events (HR 2.31, 95%CI 1.23; 4.36) and HF hospitalization (HR 2.30, 95%CI 1.10; 4.78). Patients in cluster 3 experienced increased event rates of major CV events (HR 5.84), HF hospitalization (6.50), CV death (8.58), and all-cause mortality (5.07). Twelve proteins are differentially expressed across the identified clusters, including druggable CD2, GDF-15, ABO, IGFBP-1, IGFBP-2, and RNase1. CONCLUSIONS: Proteomics-based clustering identifies three HFrEF clusters associated with distinct outcomes that remain undetected using only clinical characteristics.