Improved accuracy of anticoagulant dose prediction using a pharmacogenetic and artificial neural network-based method
- First Online:
- Cite this article as:
- Isma’eel, H.A., Sakr, G.E., Habib, R.H. et al. Eur J Clin Pharmacol (2014) 70: 265. doi:10.1007/s00228-013-1617-2
The unpredictability of acenocoumarol dose needed to achieve target blood thinning level remains a challenge. We aimed to apply and compare a pharmacogenetic least-squares model (LSM) and artificial neural network (ANN) models for predictions of acenocoumarol dosing.
LSM and ANN models were used to analyze previously collected data on 174 participants (mean age: 67.45 SD 13.49 years) on acenocoumarol maintenance therapy. The models were based on demographics, lifestyle habits, concomitant diseases, medication intake, target INR, and genotyping results for CYP2C9 and VKORC1. LSM versus ANN performance comparisons were done by two methods: by randomly splitting the data as 50 % derivation and 50 % validation cohort followed by a bootstrap of 200 iterations, and by a 10-fold leave-one-out cross-validation technique.
The ANN-based pharmacogenetic model provided higher accuracy and larger R value than all other LSM-based models. The accuracy percentage improvement ranged between 5 % and 24 % for the derivation cohort and between 12 % and 25 % for the validation cohort. The increase in R value ranged between 6 % and 31 % for the derivation cohort and between 2 % and 31 % for the validation cohort. ANN increased the percentage of accurately dosed subjects (mean absolute error ≤1 mg/week) by 14.1 %, reduced the percentage of mis-dosed subjects (mean absolute error 2-3 mg/week) by 7.04 %, and reduced the percentage of grossly mis-dosed subjects (mean absolute error ≥4 mg/week) by 24 %.
ANN-based pharmacogenetic guidance of acenocoumarol dosing reduces the error in dosing to achieve target INR. These results need to be ascertained in a prospective study.