Objective: To compare the predictability of the two methods, Bayesian method (non steady-state assumption) and massbalance algorithm (steady-state assumption), used for predicting serum phenytoin concentrations.
Material and Methods: A prospective study was conducted on adult epileptic out-patients who were treated with phenytoin at Maharaj Nakhon Sri Thammarat Hospital from April 2005 to August 2005. The patients were aged 18-60 years, had no liver dysfunction (serum albumin 3.6-5.0 g/d/L, AST < 42 Units/L, ALT < 50 U/L), had no renal dysfunction (creatinine clearance (Clcr) > 30 ml/min), had not been given drugs that could interact with phenytoin, and had taken phenytoin regularly for at least 30 days. The study exclusion criteria were pregnancy, and co-administration of drugs potentially interacting with phenytoin. Three blood samples were drawn from each patient for phenytoin blood level measurement. Two blood samples were collected on the first visit; the first one was collected at least 8 hours after the last phenytoin administration and the second one was collected 4-6 hours apart. Two predicted steady-state serum phenytoin concentrations obtained from using Bayesian method and mass-balance algorithm were compared with the actual blood levels collected 6 weeks thereafter. The proportions of error prediction (the difference between the predicted and the actual values exceeding ±5 mg/L) of the two methods were compared using chi-square test. Mean phenytoin concentration predicted by each method was compared with the mean actual value using Student's t-test.
Results: Of 80 patients included in the study, 21 patients dropped-out. The reasons were: liver dysfunction (3 patients); phenytoin blood level > 40 mg/L (1 patient), phenytoin blood levels < 2.5 mg/L (3 patients) and drug non-compliance (14 patients). The proportion of predicted error in serum phenytoin using Bayesian method and mass-balance algorithm were 0.22 and 0.24, respectively. The mean (S.D.) serum phenytoin concentration predicted from Bayesian method, mass-balance algorithm and the actual serum phenytoin concentrations were 8.69 (6.58) mg/L, 13.40 (7.20) mg/L and 12.07 (6.63) mg/L, respectively. There was no statistically significant difference in proportions of predicted error in serum phenytoin between Bayesian method and mass-balance algorithm (p = 0.157). There were statistically significant differences between mean actual serum phenytoin concentration and predicted value obtained from Bayesian method (p < 0.005) and that obtained from mass-balance algorithm (p = 0.048). We observed that Bayesian method seemed to underestimate while the other seemed to overestimate the phenytoin concentrations. When the predicted values obtained from both prediction methods were averaged, i.e., linearly combined, they were not significantly difference from the third actual serum phenytoin concentrations (p = 0.384) and had a low percentage of predicted error (11.86%)
Conclusion: Bayesian method and mass-balance algorithm could not provide the good prediction of steady state serum phenytoin concentrations. The mass-balance algorithm was biased towards overestimation, while the Bayesian method was biased towards underestimation of the actual values. The results suggest that averaging, i.e., a linear combination, of prediction from both methods might improve the prediction. This study implies that the pharmacokinetic parameters used in each method may not be at their optimal values, thus limiting the predictability of these methods. It is suggested that further study is needed to clarify the proper values of the pharmacokinetic parameters.