Proteomic Biomarker Discovery for Acute Pulmonary Embolism

Background and Objectives: Pulmonary embolism (PE) is common yet challenging to diagnose. The current framework for acute PE diagnosis includes a combination of pre-test probability estimation, D-dimer testing, and computed tomographic pulmonary angiography (CTPA). D-dimer testing is sensitive but not specific which has contributed to a 450% increase in CTPA imaging since 2004. Novel biomarkers may improve the accuracy of PE diagnostics. We conducted the largest proteomic analysis to date of patients being evaluated for acute PE to identify new candidate biomarkers for acute PE.

Methods: We leveraged our biorepository of emergency department patients evaluated for possible acute PE. We collected blood samples within 24 hours of CTPA. PE+ and PE- were based on CTPA results. Using stratified random sampling based on PE diagnosis and enriched to include a larger portion of PE+ samples, we selected 86 samples (56 PE+, 30 PE-). We excluded hemolyzed samples and those with processing/freezing times >90 minutes. Relative expression of 5,400+ proteins was measured using the OlinkTM Explore HT platform (a technique using DNA-tagged matched antibody pairs that allows DNA hybridization and qPCR). These results were evaluated using a validated proteomics machine learning pipeline that applies and aggregates multiple supervised learning models, including random forests, support vector machines, elastic nets, and extreme gradient boosting, to identify candidate biomarkers that best differentiate PE+ from PE- patients.

Results: We identified 52 proteins that were statistically significantly differentially expressed in PE+ compared to PE- samples with XGBoost-based classifier accuracy of 90.6% in 5-fold cross validation. Sixteen proteins remained significant after false discovery rate correction; 5 related to angiogenesis or tumorigenesis, 3 to clotting, 2 to the endothelium, 1 to sex hormones, and 5 that were physiologically novel.

Conclusion: This study represents the first of its kind proteomic analysis for acute PE. These results suggest that there is a proteomic signature associated with acute PE that may facilitate a new biomarker-based diagnostic or risk-stratification tool. These results warrant further investigation with an increased sample size.

Presenter:

  • Drew A. Birrenkott, MD, DPhil
Authors
  • Birrenkott Drew 2025

    Drew Birrenkott, MD, DPhil

    Emergency Physician

    Mass General Brigham

    Drew Birrenkott, MD, DPhil, is an emergency physician and biomedical engineer. He is currently a fellow in clinical innovation and research translation in vascular emergencies at Mass General Brigham. He completed his doctorate in engineering science at Oxford University using signal processing and AI to estimate respiratory rate from cardiac waveforms. His doctoral work received international patents and has been licensed by start-up companies for use in medical devices. He completed medical school at Stanford University and residency at the Harvard Affiliated Emergency Medicine Residency (HAEMR) at Mass General Brigham. As a member of the Massachusetts General Hospital Center for Vascular Emergencies, he is currently working on novel applications using both AI and -omics to improve the acute diagnosis of pulmonary embolism (PE) including systems to improve PE triage risk and expedite diagnosis. He is a member of the Automated Registry of CardioVascular Emergencies (ARCVE), a multi-center consortium of hospital systems and venous thromboembolism researchers which has created one of the largest registries of patients evaluated for VTE in the emergency department.

Related Media