A Proteomics Model for Predicting Recurrent ASCVD Risk Outperforms Clinical Risk Model

ascvd risk
ascvd risk
Researchers evaluated the ability to improve risk stratification of patients with ASCVD via a novel machine-learning model.

A proteomics-based risk model was found to be superior for predicting recurrent atherosclerotic cardiovascular disease (ASCVD) compared with a clinical risk model, according to results of a study published in the European Heart Journal.

Data for this study were sourced from the Second Manifestations of ARTerial disease (SMART) cohort which is an ongoing, prospective, single-center cohort that started in 1996, from University Medical Center Utrecht in the Netherlands. Data from individuals (n=870) who had a 10-year SMART risk score of greater than 15% and had available blood samples were used as the derivation cohort. Data for the validation cohort were sourced from the Athero-Express study, which observed 700 patients who received carotid and femoral endarterectomy in 2002 for 3 years. Using these data, a proteomic-based risk model was developed and tested for its ability to predict risk for recurrent ASCVD.

The derivation and validation cohorts were aged median 65 (IQR, 9) and 70 (IQR, 9) years; 75.5% and 68.4% were men; mean BMI was 26.9±3.9 and 26.2±3.8; and 30.2% and 18.6% of patients had recurrent ASCVD, respectively.

The relative importance of 6 of the 50 assessed proteins (N-terminal pro-brain natriuretic peptide [NT-proBNP], kidney injury molecule-1 [KIM1], matrix metallopeptidase 7 [MMP7], growth/differentiation factor-15 [GDF-15], hydroxyacid oxidase 1 [HAOX1], transforming growth factor b induced [TGFBI]) were found to be greater than 0.4.

Using the importance of these proteins, the developed proteomics model had a receiver operating characteristic (ROC) curve area under the curve (AUC) of 0.810 (95% CI, 0.797-0.823) in the derivation cohort and 0.801 (95% CI, 0.785-0.817) in the validation cohort. These values were superior to the clinical model for both the derivation (AUC, 0.750; 95% CI, 0.734-0.765) and validation (AUC, 0.765; 95% CI, 0.743-0.784) cohorts.

Combining the protein and clinical models improved the prediction among the derivation cohort (AUC, 0.824; 95% CI, 0.812-0.835) but not the validation cohort (AUC, 0.792; 95% CI, 0.771-0.811).

After recalibrating all models, the protein model outperformed the clinical model (D AUC, 0.036; 95% CI, 0.020-0.051; P <.001) and the combined model was not superior (DAUC, -0.007; 95% CI, -0.023 to 0.004; P =.996).

Stratified by low (£2 mg/L) and high (>2 mg/L) C-reactive protein (CRP) status, the top 10 most predictive proteins differed between cohorts, in which a-1-microglobulin/bikunin precursor (AMBP), nidogen 1 (NID1), vasorin (VASN), and transferrin (TF) were found to be important among patients with low CRP but not among those with high CRP.

This study may have been limited by the proteins included in the proteomics panel used.

“We show that a panel of 50 proteins is superior to a clinical risk model in predicting recurrent ASCVD events,” the study authors wrote. “…large prospective studies will have to confirm the value of proteome-based risk scores in secondary prevention before routine clinical implementation can be advocated.”

Disclosure: Multiple authors declared affiliations with industry. Please refer to the original article for a full list of disclosures.


Nurmohamed NS, Periera JPB, Hoogeveen RM, et al. Targeted proteomics improves cardiovascular risk prediction in secondary prevention. Eur Heart J. Published online February 9, 2022. doi:10.1093/eurheartj/ehac055