165-Protein Blood Score Predicts Atrial Fibrillation Better Than Genetics Alone
A new plasma protein risk score from 51,680 participants significantly outperforms existing AF prediction models, including polygenic risk scores.
Summary
Researchers developed a protein risk score (PRS) using 165 plasma proteins measured in over 51,000 UK Biobank participants. Added to existing clinical tools—CHARGE-AF, NTproBNP, and polygenic risk score—the protein score raised the C-index from 0.771 to 0.816 for predicting incident atrial fibrillation. Each standard deviation increase in the score corresponded to a 2.20-fold higher AF hazard. The score also reclassified 5.4% of patients into more accurate risk categories and increased net clinical benefit from 3.8 to 5.4 per 1,000 people. External validation in the ARIC study confirmed consistent improvement in risk stratification, suggesting proteomics could meaningfully enhance AF screening from a single blood draw.
Detailed Summary
Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and a leading cause of stroke and heart failure. Existing prediction tools like CHARGE-AF incorporate clinical risk factors, but leave substantial room for improvement. This study asked whether large-scale plasma proteomics—measuring hundreds of proteins simultaneously from a single blood sample—could close that gap.
The research team drew on the UK Biobank Pharma Proteomics Project (UKB-PPP), which profiled 1,459 unique plasma proteins in 51,680 adults free of AF at baseline. Using LASSO-penalized Cox regression on a 70% derivation set (36,176 individuals, 2,155 AF events), they built a protein risk score. The remaining 30% (15,504 individuals, 910 events) served as the internal test set, and the Atherosclerosis Risk in Communities (ARIC) study (11,012 individuals, 1,260 events) provided external validation.
The final protein risk score incorporated 165 plasma proteins, 15 of which were mechanistically linked to atrial remodeling—structural and electrical changes in the atrium that predispose to AF. Per standard deviation increase in the score, the hazard ratio for incident AF was 2.20 (95% CI 2.05–2.41). When added to a baseline model comprising CHARGE-AF, NTproBNP, and polygenic risk score, the C-index improved from 0.771 to 0.816—a clinically meaningful gain of 0.044. Net reclassification improved by 5.4% using a 5-year risk threshold of 5%, and decision curve analysis showed net benefit rose from 3.8 to 5.4 per 1,000 people at the same threshold. The ARIC replication cohort confirmed these findings with consistent directional improvements in risk stratification.
The biological relevance of the 165-protein panel is notable: the inclusion of proteins tied to atrial remodeling suggests the score is capturing genuine disease biology rather than spurious statistical associations. This distinguishes it from purely data-driven scores and supports potential mechanistic insights into AF pathogenesis.
Clinically, a validated protein risk score derived from a routine blood draw could enable earlier identification of high-risk individuals who might benefit from enhanced monitoring, lifestyle interventions, or prophylactic anticoagulation. The integration with polygenic risk scores also points toward a multi-omic stratification framework. Caveats include the observational design, the need to assess cost-effectiveness of mass proteomic profiling, and questions about generalizability beyond predominantly European-ancestry cohorts.
Key Findings
- 165-protein plasma score raised AF prediction C-index from 0.771 to 0.816 when added to CHARGE-AF, NTproBNP, and polygenic risk score.
- Each SD increase in protein risk score conferred a 2.20-fold higher hazard for incident AF (95% CI 2.05–2.41).
- Score reclassified 5.4% of individuals into more accurate risk categories at a 5-year, 5% risk threshold.
- Net clinical benefit increased from 3.8 to 5.4 per 1,000 people after adding the protein score.
- External validation in ARIC (11,012 participants) confirmed consistent improvement in AF risk stratification.
Methodology
LASSO-penalized Cox regression was applied to 1,459 plasma proteins in 36,176 UKB-PPP participants (70% split) to derive the protein risk score, tested internally in 15,504 participants and externally in 11,012 ARIC participants. Discrimination was assessed by C-index; clinical utility was evaluated by net reclassification improvement and decision curve analysis at a 5-year, 5% AF risk threshold.
Study Limitations
The study populations are predominantly of European ancestry, limiting generalizability to other ethnic groups. Large-scale proteomic profiling remains expensive and not yet routine in clinical settings, raising cost-effectiveness concerns. Observational design precludes causal inference, and the incremental clinical benefit of proteomic screening at population scale requires prospective evaluation.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.