Patients with atrial fibrillation, post-acute coronary syndrome and venous thromboembolism are prescribed antithrombotic drugs (antiplatelet agents and oral anticoagulants) in dual and triple combinations. These drugs independently cause gastrointestinal bleeding (GIB) from mucosal defects or vascular abnormalities of the gastrointestinal tract. Cardiac patients are the fastest growing at-risk group due to the rapidly aging US population, their high-burden of co-morbidity, frequent antithrombotic polypharmacy and prescription of new antithrombotic drugs with higher incidence of GI adverse events. The real-world GIB risk of antithrombotic agents used in combination in a diverse cardiac population has yet to be characterized. Furthermore, we have previously demonstrated poor performance of existing risk scores (HAS-BLED, CHA2DS2-VASC etc.) for the prediction of non- warfarin antithrombotic drug bleeding; scores frequently used in the clinical setting despite their inaccuracy. Absence of knowledge regarding the real-world risk of antithrombotic-GIB, and the inability to accurately predict which patients will bleed, hampers patient counselling regarding a frequently occurring adverse event which is known to cause morbidity and mortality among cardiac patients. We propose to fill this knowledge gap by quantifying GIB risk in a large, geographically diverse cohort of elderly and non-elderly cardiac patients with atrial fibrillation, venous thromboembolism or post-acute coronary syndrome. GIB associated with antithrombotic drug combinations will be stratified by underlying cardiac conditions; incidence rates (events/100 patient-years) and propensity-matched Cox proportional models (with 95% confidence intervals) will be used to estimate outcome. We will examine heterogeneity of safety effects related to age, chronic co-morbidity, and hepatic or renal dysfunction. Machine learning techniques will then be used to derive and validate a highly sensitive algorithm for the prediction of antithrombotic-related GIB. Discovery of such an algorithm is the first step in the future application of a predictive model in any evidence-based clinical delivery platform, such as a decision rule or risk calculator.

Public Health Relevance

Gastrointestinal bleeding (GIB) in cardiac patients is common, deadly and on the rise due to an older population, antiplatelet and anticoagulant drugs used in combination, and availability of new drugs with higher GIB risk than their predecessors. Lack of data regarding the magnitude of GIB and limited knowledge of which patients are most at risk, prevents accurate counselling regarding cardiac drug safety. We will quantify this risk, study risk factors; and use machine learning techniques to derive and validate a clinical algorithm to better predict at-risk patients.

National Institute of Health (NIH)
Agency for Healthcare Research and Quality (AHRQ)
Research Project (R01)
Project #
Application #
Study Section
Healthcare Patient Safety and Quality Improvement Research (HSQR)
Program Officer
Bartman, Barbara
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Mayo Clinic, Arizona
United States
Zip Code