An Expert System to Predict Warfarin Dosage in Turkish Patients Depending on Genetic and Non-Genetic Factors

Abstract

Warfarin which is a vitamin K antagonist is one of the most widely used oral anticoagulants worldwide. Genetic factors affecting warfarin (CYP2C9, CYP4F2 and VKORC1) have been shown in different studies. Apart from genetic factors, the effects of age, height, weight and bleeding condition also have been proven. The use of prescribed warfarin drug in the wrong doses leads to irreparable disasters for the patients. The amount of warfarin the patients have to take is determined by the INR machine and this takes a lot of time. Since dose estimation takes a long time with conventional methods, use of data mining algorithms has been proposed for prediction of warfarin dose. In this paper, unlike previous studies, it was shown that the amount of warfarin was calculated not by numeric prediction but by classification, and better accuracy rates than previous success accuracy rates were obtained. Using the data obtained from the Turkish patients in the study, the dose range required for daily use of the patient's warfarin drug dose was classified by Bayesian and K-Nearest Neighbor (KNN) algorithms. The result of this study using Bayesian algorithm calculated as %59.01 and using KNN algorithm calculated as %50.52.