Targeted proteomics improves CV risk prediction

A plasma protein-based risk model demonstrated superior performance compared to the PREVENT risk model in predicting cardiovascular outcomes among UK Biobank participants without CVD.

This summary is based on the publication of Ho FK, Mark PB, Lees JS et al. - A Proteomics-Based Approach for Prediction of Different Cardiovascular Diseases and Dementia - Circulation. 2025 Feb 4;151(5):277-287. doi: 10.1161/CIRCULATIONAHA.124.070454.

Introduction and methods

Background

The addition of individual circulating cardiometabolic biomarkers such as NT-proBNP, high sensitivity cardiac troponins and hs-CRP to conventional risk scores leads to improvements in discrimination [1-4]. It is unknown whether a protein-based risk model, using a proteomic panel, improves CVD risk prediction.

Aim of the study

The authors evaluated the performance of a plasma proteomics-based risk model versus the PREVENT (Predicting Risk of CVD Events) risk score for CVD risk assessment.

Methods

In this prospective cohort study, 51,859 adults without CVD and with proteomic measurements from the UK Biobank were included. Plasma samples were processed and in the proteomic analysis a panel of 3072 circulation protein biomarkers were measured, which included targets from cardiometabolic, inflammation, neurology and oncology pathways [5].

An exposome-wide association approach was used to estimate the association between individual proteins with clinical outcomes after adjusting for cofounders.

To assess the performance of the proteomics-based risk model on incident clinical outcomes, 3 prediction models were created: (1) the PREVENT model, (2) the protein model which included age, sex, and proteins and (3) the PREVENT plus proteins model. The PREVENT model included PREVENT risk factors age, sex, current smoking, eGFR, BMI, systolic BP, total and HDL cholesterol, antihypertensive medications, and lipid-lowering medications. The machine-learning least absolute shrinkage and selection operator (LASSO) method was used to identify a subset of predictor proteins, and selected proteins in each model varied depending on the outcome and baseline model. Data was randomly divided into two dataset. 80% of the data was used in a derivation set for model training, and the remaining 20% of the data was used for validation.

Outcomes

The primary outcome was a composite of fatal and nonfatal coronary heart disease, stroke, or HF (major adverse cardiovascular events, MACE). Secondary outcomes were ASCVD, MI/revascularization, ischemic stroke, aortic stenosis, HF, abdominal aortic aneurysm (AAA) and dementia.

Main results

• Over a median follow-up period of 13.6 years (IQR: 12.8-14.4), 4857 participants developed MACE.
• After adjusting for demographic, lifestyle, and clinical factors, NT-proBNP had the strongest positive association with MACE (HR: 1.68 per SD increase), followed by proADM (HR: 1.60) and GDF-15 (HR: 1.47) in the exposome-wide association analysis.
• To predict all outcomes of interest, 222 proteins were selected in the protein model and 177 proteins in the PREVENT plus proteins model in the derivation cohort.
• To predict the risk of MACE, 86 proteins were selected in the protein model.
• Compared with the PREVENT model, the protein model improved risk discrimination (c statistic: +0.051) and reclassification performance (net reclassification index: +0.09) for MACE in the validation cohort.
• The protein model also improved the c statistic for all secondary outcomes:
  • ASCVD (+0.035)
  • MI/revascularization (+0.023)
  • stroke (+0.024)
  • aortic stenosis (+0.015)
  • HF (+0.060)
  • AAA (+0.0244)
  • dementia (+0.068)
• Compared with the PREVENT model, the protein model improved the net reclassification index for most secondary outcomes with the exception of aortic stenosis. 
• There were modest improvements in the c statistic with the PREVENT plus proteins models compared with the protein models alone.

Conclusion

This study demonstrated that targeted proteomics with approximately 200 biomarkers has superior predictive performance for MACE and other specific CVD events compared with the PREVENT model.

“This methodology, when less expensive and more automated, offers the potential for advancing precision medicine to improve cardiovascular screening and the improved tailoring of preventive strategies,” concluded the authors.

Find this article online at Circulation

References

1. Welsh P, Doolin O, Willeit P, Packard C, Macfarlane P, Cobbe S, Gudnason V, Di Angelantonio E, Ford I, Sattar N. N-terminal pro-B-type natriuretic peptide and the prediction of primary cardiovascular events: results from 15-year follow-up of WOSCOPS. Eur Heart J. 2013;34:443–450. doi: 10.1093/eurheartj/ehs239
2. Willeit P, Welsh P, Evans JDW, Tschiderer L, Boachie C, Jukema JW, Ford I, Trompet S, Stott DJ, Kearney PM, et al. High-sensitivity cardiac troponin concentration and risk of first-ever cardiovascular outcomes in 154,052 participants. J Am Coll Cardiol. 2017;70:558–568. doi: 10.1016/j.jacc.2017.05.062
3. Willeit P, Kaptoge S, Welsh P, Butterworth AS, Chowdhury R, Spackman SA, Pennells L, Gao P, Burgess S, Freitag DF, et al. Natriuretic peptides and integrated risk assessment for cardiovascular disease: an individual-partici pant-data meta-analysis. Lancet Diabetes Endocrinol. 2016;4:840–849. doi: 10.1016/S2213-8587(16)30196-6
4. Welsh P, Kimenai DM, Shah ASV, Gadd DA, Marioni RE, Woodward M, Sudlow CLM, Campbell A, Cleland JGF, Pellicori P, et al. Multiple car diac biomarkers to improve prediction of cardiovascular events: findings from the Generation Scotland Scottish Family Health Study. Clin Chem. 2024;70:403–413. doi: 10.1093/clinchem/hvad205
5. UK Biobank. Olink proteomics data. Version 1.0. March 2023. Accessed September 28, 2024. https://biobank.ndph.ox.ac.uk/showcase/ukb/docs/ Olink_proteomics_data.pdf