Dosage individualization of warfarin using artificial neural networks

Area of Science:

• Pharmacogenomics
• Computational biology
• Clinical pharmacology

Background:

• Warfarin dosage individualization is complex.
• Achieving a stable international normalized ratio (INR) of 2-3 is crucial for effective anticoagulation.
• Existing dosing algorithms have limitations.

Purpose of the Study:

• To explore artificial neural networks (ANNs) for warfarin dosage individualization.
• To develop and validate an ANN-based warfarin dosing algorithm.
• To assess the clinical significance of the ANN algorithm compared to existing methods.

Main Methods:

• A cohort of 3,415 patients was used to develop the ANN algorithm.
• A separate cohort of 856 patients was used for validation.
• The algorithm's accuracy was measured by the percentage of patients with predicted dosages within 20% of their actual stable dose.

Main Results:

• The ANN model successfully predicted ideal warfarin dosage in 48% of patients.
• The model explained 48% and 43% of dosage variability in derivation and validation cohorts, respectively.
• Key predictors identified include race, age, weight, CYP2C9 and VKORC1 genotypes, and concurrent medications.

Conclusions:

• An ANN was successfully applied to create a warfarin dosing algorithm.
• The developed algorithm shows potential for recommending more appropriate warfarin dosages.
• ANNs offer a promising tool for personalized warfarin therapy.