Mass Spectrometry Based Proteomic Profiling Improves Long-Term Mortality Risk Prediction in Chronic Coronary Artery Disease

Moritz von Scheidt (München)1, S. Steigerwald (Martinsried)2, R. Lermer (München)1, T. Keßler (München)1, P. Sen (München)1, D. Bongiovanni (Augsburg)3, M. Heinig (Neuherberg)4, A. Kastrati (München)1, M. Mann (Martinsried)2, H. Schunkert (München)1

1Deutsches Herzzentrum München Klinik für Herz- und Kreislauferkrankungen München, Deutschland; 2Max Planck Institute of Biochemistry Department of Proteomics and Signal Transduction Martinsried, Deutschland; 3Universitätsklinikum Augsburg I. Medizinische Klinik Augsburg, Deutschland; 4Institute of Computational Biology, Computational Health Center, Helmholtz Zentrum München Deutsches Forschungszentrum für Gesundheit und Umwelt (GmbH) Neuherberg, Deutschland


Aims: Patients with stable but severe coronary artery disease (CAD) suffer from a high event rate. Given a limited performance of clinical prediction models in such patients we assessed the utility of proteomic profiling to improve prediction of long-term mortality risk.

Methods: 1,200 consecutive patients with 2- or 3-vessel disease diagnosed by coronary angiography in the years 2005 and 2006 were analyzed. Deep mass spectrometry differentiated 735 unique serum proteins. Coxnet was used for clinical and proteomic variable selection. Internal validation was performed by randomly splitting CAD cases into training (n=885) and validation (n=315) cohorts.

Results: During a median follow-up of 16.5 years, 650 (54.2%) patients died. 66 (out of 317) proteins differed significantly between living and deceased individuals. Low serum levels of A2M, B2M, CFD, FGB and CRP showed strongest correlations with improved survival. Predicting long-term mortality risk, a proteomic supported model (13 clinical and 31 proteomic variables) outperformed the clinical model (13 variables) in both training (ROC-AUC 0.77 vs. 0.73, P(C-statistics)<0.001) and validation sets (AUC 0.75  vs. 0.72, P<0.001). Moreover, inclusion of proteomics data improved integrated discrimination index (IDI) 0.09 (P<0.001) and net reclassification index (NRI) 0.28 (P<0.001). Most relevant for model performance were age, LV-EF, smoking, CAD complexity, and the proteins APOC3, C1RL, CD14, CLEC3B, FCGBP, IGHG2, IGHV1-2, IGHV3-38, IGKV3D-20, IGLV2-14, PIGR and SERPINA4. Enrichment analysis revealed coagulation, stress response, inflammation, and immune system related pathways were strongest associated with mortality risk.

Conclusion: Proteomics is a powerful tool for improving predictive accuracy in models of long-term mortality risk in chronic CAD.
Diese Seite teilen