Abstract 9387: A Better Personalized Prediction for International Normalized Ratio Levels in Inpatients Receiving Warfarin
Objectives: The objectives of this study were: 1) to test the effectiveness of a novel prediction method for International Normalized Ratio (INR) levels in inpatients receiving warfarin therapy; 2) to compare the prediction accuracy between this method and existing methods.
Methods: We developed a novel personalized pharmaco-kinetic and pharmaco-dynamic (PKPD) model to predict the trajectory of INRs over time. We tested this model on 60 inpatient cases in New York Presbyterian Hospital from 2012 to 2013 using two different methods. The first method (fixed horizon) predicts future INR levels by the initial five observed INRs; the second method (rolling horizon) makes predictions based on the latest five observed INRs on a rolling basis. Prediction accuracy is evaluated by the Absolute Percentage Prediction Error (PAPE), which is |Predicted INR - Actual INR|/Actual INR.
Results: A personalized INR prediction example:
Summary statistics for prediction errors:
Conclusions: The rolling horizon method continuously updates the PKPD model based on the individuals’ latest INR responses, and thus significantly improves the prediction accuracy as compared to the fixed horizon method (p<.001). The rolling horizon method provides accurate predictions for INR trajectory five days into the future, and can help clinicians improve warfarin dosing to reduce adverse outcomes due to INRs that are not within a therapeutic range.
Author Disclosures: Y. Zhao: None. N. Liu: None. Y. Wang: None. K. Hickey: None.
- © 2014 by American Heart Association, Inc.