Abstract P097: Predictive Model of Bleeding Events in Atrial Fibrillation Patients Using Dabigatran
Objective: Develop a model to predict incident bleeding events in atrial fibrillation (AF) patients on dabigatran, using clinically available variables.
Hypothesis: A predictive model of bleeding in dabigatran users developed from claims data will be clinically useful.
Methods: We studied AF patients initiating dabigatran, a new oral anticoagulant, in the MarketScan dataset, a large healthcare utilization database, in the period 2010-2012. Two thirds of the sample was randomly selected and used to derive the predictive model (training dataset), which was then validated in the remaining third (testing dataset). Predictors were selected from diagnosis, procedural, and medication codes potentially associated with bleeding using health claims. The outcome of interest were intracranial bleeding or gastrointestinal hemorrhage, defined by validated algorithms. A Cox model with backwards elimination of variables (p<0.05 threshold) was used to select variables for the predictive model. Discrimination was determined with a C-statistic for survival analysis and calibration with a chi square test.
Results: The training data set included 26,283 individuals and 404 bleeding events. The testing dataset included 13,224 individuals and 205 events. Median follow-up time was 436 days (interquartile range 256-591 days). The final model included the following variables: age, hematologic disorders, heart failure, kidney disease, prior vascular procedure, and prior use of warfarin (table). The internal training c-statistic was 0.75 (0.72-0.77) with and adjusted calibration chi-square p-value=0.0827. The testing validation c-statistic was 0.77 (0.74-0.80) with an adjusted calibration chi-square p-value=0.0215.
Conclusion: A simple model using clinical variables was able to identify AF patients at higher risk of bleeding when using dabigatran. This model could assist clinical decisions about anticoagulant use. However, the model may need recalibration before being used in an external population.
Author Disclosures: I.R. Rapson: None. L.Y. Chen: None. R.F. Maclehose: None. P.L. Lutsey: None. A. Alonso: None.
- © 2015 by American Heart Association, Inc.