A New Predictive Model for Acute Gastrointestinal Bleeding in Patients Taking Oral Anticoagulants: A Cohort Study

Background We developed a predictive model of long-term gastrointestinal (GI) bleeding risk in patients receiving oral anticoagulants and compared it with the HAS-BLED score. Methods We periodically followed a cohort of 508 patients taking oral anticoagulants (66 direct oral anticoagulants users and 442 warfarin users). Absence of GI bleeding at an initial examination, and any subsequent GI bleeding, were confirmed endoscopically. The bleeding model was developed by multivariate survival analysis and evaluated by Harrell's c-index. Results During a median follow-up of 31.4 months, 42 GI bleeds (8.3%) occurred; 42.8% in the upper GI tract, 50.0% in the lower GI tract, and 7.1% in the middle GI tract. The cumulative 5- and 10-year probability of GI bleeding was 12.6% and 18.5%, respectively. Patients who bled had a significantly higher cumulative incidence of all-cause mortality (hazard ratio 2.9, p <0.001). Multivariate analysis revealed that absence of proton pump inhibitor (PPI) therapy, chronic kidney disease, chronic obstructive pulmonary disease, history of peptic ulcer disease and liver cirrhosis predicted GI bleeding. The c-statistic for the new predictive model using these five factors was 0.65 (p <0.001), higher than the HAS-BLED score of 0.57 (p = 0.145). Conclusions Gastrointestinal bleeding increased the risk of subsequent mortality during follow-up of anticoagulated patients, highlighting the importance of prevention. We developed a new scoring model for acute GI bleeding risk based on five factors (no-PPI use, chronic kidney disease, chronic obstructive pulmonary disease, history of peptic ulcer disease and liver cirrhosis), which was superior to the HAS-BLED score.