Bayesian Feedback Method for Optimum Phenytoin Dosing

Abstract

The accuracy of Bayesian feedback method for predicting phenytoin dosage was assessed and compared to conventional predictive methods such as the Richens & Dunlop method and Ludden method from 57 patients who took phenytoin for the treatment of grandmal epilepsy, For the comparison of absolute predictability of each method, correlation coefficient between predicted and actual dosage, mean error (ME) and root mean squared error (RMSE) were calculated. For a single data pair, the Bayesian method showed a higher correlation coefficient (0.915) and' less ME & RMSE values( -2.0'1"0 & 8.8'X)) than the Richens & Dunlop methood, and there was a staitistically significant difference between these two methods (p < 0,05). For multiple dat pairs, the Bayesian method also showed a higher correlation coefficient (0,93) and less ME & RMSE values (0.69% & 7.2<1'0), though there was no significant difference. It is suggested that the Bayesian method has better predictability and can be used conveniently even with single Cpss-dosc data pair for optimum phenytoin dosage prediction.