17-Protein Blood Panel Predicts Heart Failure Risk in Diabetics Far Better Than Standard Tests
A proteomics screen of 2,920 plasma proteins identifies a 17-protein score that lifts heart failure prediction accuracy to 0.833 C-index in type 2 diabetes patients.
Summary
Researchers analyzed blood proteins in over 2,000 people with type 2 diabetes from the UK Biobank, tracking them for more than 13 years. They found 455 proteins linked to heart failure risk (447 positively, 8 inversely). Using machine learning to narrow the list, a panel of just 17 proteins predicted who would develop heart failure better than standard clinical tests, genetic risk scores, and even NT-proBNP — the current gold-standard cardiac biomarker. The top risk-raising protein was WAP 4-disulfide core domain protein 2 (WFDC2), while apolipoprotein C-I was protective. The combined 17-protein score achieved a C-index of 0.833, an increment of 0.091 over the baseline model. These findings point toward biological pathways — including cytokine signaling, cell adhesion, and extracellular space interactions — relevant to heart failure in diabetics.
Detailed Summary
Heart failure is one of the most dangerous and common complications of type 2 diabetes, yet predicting who will develop it remains difficult with current clinical tools. This study set out to determine whether measuring hundreds of blood proteins simultaneously could sharpen that prediction.
Researchers used data from 2,198 UK Biobank participants with type 2 diabetes, measuring 2,920 plasma proteins at enrollment and following participants for a median of 13.1 years. During that time, 298 individuals developed heart failure. Cox proportional hazards models were used to test each protein individually, and a LASSO machine-learning method then distilled the most predictive proteins into a compact panel.
The results were striking. A total of 455 proteins showed statistically significant associations with heart failure risk. The strongest risk-increasing protein was WFDC2 (WAP 4-disulfide core domain protein 2), with an 90% higher risk per standard deviation increase. Apolipoprotein C-I was the most protective, associated with a 25% lower risk per SD. Enriched biological pathways included cell adhesion, cytokine signaling, and extracellular space interactions — pointing to inflammation and structural remodeling as key drivers.
The final 17-protein risk score achieved a C-index of 0.833 when added to clinical variables, polygenic risk scores, and NT-proBNP — an improvement of 0.091 over the baseline model. Net reclassification and integrated discrimination metrics confirmed meaningful gains in risk stratification.
For clinicians managing diabetic patients, these findings suggest that proteomic profiling could one day enable far more precise identification of those at highest heart failure risk, enabling earlier intervention. Caveats include the study's reliance on a largely European cohort, the abstract-only availability of full methods, and the need for external validation before clinical translation.
Key Findings
- 455 of 2,920 plasma proteins were significantly associated with heart failure risk in type 2 diabetes patients (447 positively, 8 inversely).
- WAP 4-disulfide core domain protein 2 (WFDC2) conferred a 90% higher heart failure risk per SD increase (HR 1.90, 95% CI 1.65-2.19) — the strongest single predictor found.
- A 17-protein score reached a C-index of 0.833 with an increment of 0.091, outperforming clinical variables, polygenic risk, and NT-proBNP alone.
- Apolipoprotein C-I was the top protective protein (HR 0.75 per SD, 95% CI 0.66-0.85).
- Enriched pathways included cell adhesion, extracellular space, signaling receptor activity, and cytokine-cytokine receptor interaction.
Methodology
Prospective cohort study of 2,198 UK Biobank participants with type 2 diabetes; 2,920 plasma proteins measured at baseline using the Olink or SomaScan platform (not specified in abstract). Cox proportional hazards models assessed protein-HF associations; LASSO with 10-fold cross-validation selected the 17-protein predictive panel. Model performance evaluated via Harrell's C-index, calibration slope, NRI, IDI, and decision curve analysis.
Study Limitations
This summary is based on the abstract only; full methods, supplementary analyses, and protein platform details are unavailable. The cohort is drawn from the UK Biobank, which skews toward European-ancestry participants, limiting generalizability. External validation in independent, diverse cohorts is needed before clinical deployment of the 17-protein score.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.