Multi-Protein Signatures Beat Single Biomarkers
Why a proteomic panel predicts cardiovascular death in heart failure far better than NT-proBNP alone.
The Limit of a Single Marker
NT-proBNP is the canonical readout of cardiac congestion, and it is genuinely prognostic: stratifying heart failure patients by NT-proBNP tertile gives a hazard ratio of 1.77 per SD (95% CI 1.62–1.93, log-rank p = 2.7e-26). But one analyte measures one process. At 15 years, the high-NT-proBNP tertile still has roughly 65% survival versus about 87% in the low tertile, a separation of around 22 percentage points. Useful, but it leaves most of the prognostic picture untold.
A Panel Tells the Fuller Story
Kaplan-Meier survival by risk tertile over 15 years, from the analysis. Left: NT-proBNP alone. Right: multi-protein Cox score. Shaded bands are 95% confidence intervals. UK Biobank heart failure cohort (n=804 low, 805 high).
A multi-protein Cox signature roughly doubles the hazard ratio to 3.50 (95% CI 3.18–3.85, log-rank p = 1.3e-86) and widens 15-year survival separation between high- and low-risk tertiles to about 46 percentage points (roughly 48% vs 94% survival). The panel resolves risk that NT-proBNP alone cannot.
| Stratifier | NT-proBNP alone | Multi-protein model |
|---|---|---|
| Hazard ratio (per SD) | 1.77 | 3.50 |
| 95% CI | 1.62–1.93 | 3.18–3.85 |
| 15-yr survival, low vs high tertile | ~87% vs ~65% | ~94% vs ~48% |
| Log-rank p | 2.7e-26 | 1.3e-86 |
Proteins Carry the Prognostic Signal
C-index for 15-year CV-death prognosis, by data modality. UK Biobank heart failure cohort.
Proteomics alone takes discrimination from a C-index of 0.60 (baseline demographics) to 0.74, almost matching the best multimodal model at 0.76. Adding clinical labs and polygenic risk scores contributes only a little more. For a prognostic test, a proteomic panel is close to sufficient on its own.
What the Panel Detects
The discovered proteins span the distinct biological processes that drive heart failure progression, which is why they add up to more than NT-proBNP:
| Process | Representative panel proteins |
|---|---|
| Congestion / wall stress | NT-proBNP, ANGPT2 |
| RAAS / renal perfusion | Renin, EDN1 |
| Fibrosis / tissue remodelling | SPON1, HGF, SCGB3A1 |
| Metabolic / systemic stress | ANGPTL4, GDF-15, REG4 |
Why It Matters for Trials and Diagnostics
A panel-based prognostic test enriches trials and stratifies patients far more sharply than a single analyte, concentrating events and improving statistical power. Aurora discovers and validates these signatures on population-scale data, then ports them to other endpoints by changing the cohort definition. The result is a faster route from biomarker hypothesis to a defensible, multi-protein prognostic readout.
CEO, Hurdle
He/Him. Tom is CEO at Hurdle, a diagnostic-as-a-service company. Tom is a specialist in Epigenetics, Machine Learning, and Computational Biology.
LinkedIn →Working on something similar? We can talk through how Hurdle would approach it.
Talk to us →