Plasma proteomics for prediction of subclinical coronary artery calcifications in primary prevention

• Published in: The American heart journal Vol. 271; pp. 55 - 67
• Main Authors: Royer, Patrick, Björnson, Elias, Adiels, Martin, Álvez, María Bueno, Fagerberg, Linn, Bäckhed, Fredrik, Uhlén, Mathias, Gummesson, Anders, Bergström, Göran
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.05.2024; Elsevier Limited
• Subjects: Algorithms; Arteriosclerosis; Atherosclerosis; Biomarkers; Biomarkers - blood; Blood Proteins - analysis; Cardiac and Cardiovascular Systems; Cardiovascular disease; Cardiovascular diseases; Computed tomography; Coronary artery; Coronary Artery Disease - blood; Coronary Artery Disease - diagnosis; Coronary Artery Disease - diagnostic imaging; Coronary Artery Disease - epidemiology; Coronary vessels; Diabetes; Disease prevention; Feature selection; Female; Glucose; Glucose tolerance; Heart diseases; Humans; Ischemia; Kardiologi; Machine Learning; Male; Medicin och hälsovetenskap; Microbiota; Middle Aged; Plasma; Plasma proteins; Predictions; Predictive Value of Tests; Primary Prevention - methods; Proteins; Proteomics; Proteomics - methods; Questionnaires; Risk Factors; Sweden - epidemiology; Tomography, X-Ray Computed - methods; Vascular Calcification - blood; Vascular Calcification - diagnostic imaging; Sweden; SBP; OR; CAC; IHD; ASCVD; CI; ROC; T2D; MRMR; AUC; CACS; CT; HDL; QC; XGB; IGTM study; FINDRISC; LASSO; NPX; SCAPIS; OGTT
• ISSN: 0002-8703; 1097-6744; 1097-6744
• DOI: 10.1016/j.ahj.2024.01.011

Recent developments in high-throughput proteomic technologies enable the discovery of novel biomarkers of coronary atherosclerosis. The aims of this study were to test if plasma protein subsets could detect coronary artery calcifications (CAC) in asymptomatic individuals and if they add predictive value beyond traditional risk factors. Using proximity extension assays, 1,342 plasma proteins were measured in 1,827 individuals from the Impaired Glucose Tolerance and Microbiota (IGTM) study and 883 individuals from the Swedish Cardiopulmonary BioImage Study (SCAPIS) aged 50-64 years without history of ischaemic heart disease and with CAC assessed by computed tomography. After data-driven feature selection, extreme gradient boosting machine learning models were trained on the IGTM cohort to predict the presence of CAC using combinations of proteins and traditional risk factors. The trained models were validated in SCAPIS. The best plasma protein subset (44 proteins) predicted CAC with an area under the curve (AUC) of 0.691 in the validation cohort. However, this was not better than prediction by traditional risk factors alone (AUC = 0.710, P = .17). Adding proteins to traditional risk factors did not improve the predictions (AUC = 0.705, P = .6). Most of these 44 proteins were highly correlated with traditional risk factors. A plasma protein subset that could predict the presence of subclinical CAC was identified but it did not outperform nor improve a model based on traditional risk factors. Thus, support for this targeted proteomics platform to predict subclinical CAC beyond traditional risk factors was not found.