Large-scale plasma proteomic profiling enhances prediction of heart failure among hypertensive individuals: prospective insights from the UK Biobank.

Over a median 15.1 years of follow-up, 1,074 of 16,128 hypertensive UK Biobank participants without baseline HF developed incident heart failure.

• Participants were from the UK Biobank Pharma Proteomics Project.
• Plasma levels of 2,911 proteins were measured using the Olink Explore platform.
• Participants were randomly split 1:1 into training and validation sets.

In the training set, 534 proteins were associated with incident HF after adjustment for demographic and clinical risk factors.

• Associations were identified with Cox models.
• The proteins were enriched in cytokine signalling, extracellular matrix organisation, and immune-cell migration pathways.

A 101-protein protein risk score (PRS) built by LASSO regression was independently associated with incident HF.

• LASSO regression was applied to the HF-associated proteins from the training set.
• The PRS was evaluated in the validation set.

Adding the PRS to age and sex raised the validation-set C-index from 0.670 to 0.786.

• Performance was assessed with Harrell's C-index, time-dependent ROC, NRI, IDI, calibration, and bootstrap validation.
• The paper reports improved reclassification and discrimination with the PRS.

Adding the PRS to the PCP-HF model raised the validation-set C-index from 0.727 to 0.787.

• PCP-HF is the Pooled Cohort Equations to Prevent Heart Failure model.
• Reclassification and discrimination metrics (NRI, IDI) were reported as improved.

PRS levels were elevated years before HF diagnosis.

• This suggests the proteomic signature captures early molecular changes preceding overt disease.

A parsimonious six-protein panel (NT-proBNP, GDF15, BCAN, WFDC2, CTHRC1, CNTN5) achieved an AUC of 0.775, outperforming NT-proBNP alone.

• The panel was derived using LightGBM and sequential forward feature selection.
• The AUC of 0.775 compares with the lower AUC of NT-proBNP alone.