Predictors of ischemic stroke and major bleeding among patients with atrial fibrillation in clinical practice
American Heart Journal
Pishoy Gouda MBBCh, BAO, MSc, Josephine Harrington MD, Kelly Arps MD, Gretchen Sanders RN, Anqi Chen Msc, Karen Chiswell PhD, Paul Hofmann BS, Keith Marsolo PhD, Kirubel Asfaw MSc, Rosa Coppolecchia DO, John H. Alexander MD, MHS, Jonathan Piccini MD, Christopher B. Granger MD, Heidi T. May MD, Elizabeth Chrischilles MD, Benjamin A. Steinberg MD, Alanna M. Chamberlain MD, Jeffrey VanWormer MD, William Schuyler Jones MD, Manesh R. Patel MD
Summary
Background
Numerous risk scores exist to quantify the risk of ischemic stroke risk in patients with atrial fibrillation or flutter (AF/AFL). However, how these tools perform in cohorts or those already on oral anticoagulation (OAC) therapy is unclear.
Methods
Using two electronic health records datasets, individuals with AF/AFL were identified [Duke University Health System (DUHS) – 2009-2018; 7 health systems participating in the Patient-Centered Clinical Research Network (PCORnet®) – 2015-2019]. Individuals were stratified based on their OAC status at the time of entry into the cohort. Multivariable competing risk models for ischemic stroke and major bleeding were developed in the DUHS prevalent AF/AFL population and independently replicated in an incident AF/AFL within the PCORnet® population. Discrimination of the final model for ischemic stroke was assessed against the CHA2DS2-VASc model alone.
Results
During the study period, 214,749 unique patients with AF/AFL were identified (DUHS – 43,896 with median CHA₂DS₂-VASc score 3; PCORnet – 170,853 with median CHA₂DS₂-VASc 3), of which 41.1% were not on OAC, 32.1% were on a direct oral anticoagulant (DOAC), and 26.8% were on warfarin. The CHA₂DS₂-VASc model alone demonstrated modest discrimination for ischemic stroke across OAC strategies (C-index range 64%–69%). Incorporation of routinely available clinical variables substantially improved discrimination for ischemic stroke (C-index 74%–76% for no OAC, 74%–75% for DOAC, and 70%–72% for warfarin) and for major bleeding (C-index 75%–76% for no OAC, 75%–76% for DOAC, and 72%–73% for warfarin). Several factors not captured by CHA₂DS₂-VASc, including race and prior major bleeding, were independently associated with ischemic stroke risk.
Conclusions
The CHA₂DS₂-VASc score provides modest discrimination for ischemic stroke and major bleeding across AF/AFL populations, including patients already treated with OAC. Incorporation of routinely available clinical factors substantially improves risk discrimination and highlights important sources of residual risk not captured by existing models.
Citation
Gouda, Pishoy, et al. “Predictors of ischemic stroke and major bleeding among patients with atrial fibrillation in clinical practice.” American Heart Journal (2026): 107490.