PREDICTION OF OPTIMAL WARFARIN MAINTENANCE DOSE USING ADVANCED ARTIFICIAL NEURAL NETWORKS

Introduction. The individual response to vitamin K antagonists (VKA) is highly variable, being influenced by clinical factors and genetic variants of enzymes that are involved in the metabolism of VKA (CYP2C)) and vitamin K (VKORC1). Currently, the dose of VKA is adjusted based on measurements of the prothrombin time. In the last years, mathematical algorithms were developed for estimating the appropriate VKA dose, based on different mathematical approaches working on clinical and genetic data. Artificial Neural Networks (ANN) are computerized algorithms resembling interactive processes of the human brain, which allow to study very complex non-linear phenomena like biological systems. Aim. To evaluate the performance of new generation ANN on a large data base of patients on chronic VKA treatment. Methods. Clinical and genetic data from 377 patients (186 m; 191 f) treated with a VKA (warfarin) average weekly maintenance dose (WMD) of 23.7 mg (11.5 SD) were used to create a dose algorithm. Forty-eight variables, including demographic, clinical and genetic data (5 CYP2C9 and 3 VKORC1 genetic variants) were entered into Twist® system, which can select fundamental variables during their evolution in search for the best predictive model. The final model, based on 23 variables expressed a functional approximation of the actual dose within a validation protocol based on a tripartite division of the data set (training, testing, validation). Results. In the validation cohort, the pharmacogenetic algorithm reached high accuracy, with an average absolute error of 5.7 mg WMD. In the subset of patients requiring ≤21 mg (45 % of the cohort) and 21-49 mg (51 % of the cohort) the absolute error was 3.86 mg and 5.45 with a high percentage of subjects being correctly identified (72%, 74% respectively). Conclusion. ANN can be applied successfully for VKA maintenance dose prediction and represent a robust basis for a prospective multicentre clinical trial of the efficacy of genetically informed dose estimation for patients who require VKA.