Clinical utility of a Bayesian dosing program for phenytoin

We performed two studies to assess the clinical utility of a Bayesian regression analysis computer program for phenytoin (PHT). In a randomized prospective study of 40 epileptic patients, the dosing program was significantly more accurate (p = 0.002) and less biased (p = 0.02) than a group of physicians at hitting a target PHT serum-concentration. Initial serum PHT concentrations that were not steady state were associated with the largest dosing errors by physicians but did not affect the accuracy of the dosing program. In a second study, we used the dosing program to predict 91 serum concentrations in 31 patients with PHT toxicity after the drug was stopped (initial concentration 26-69 micrograms/ml). The program predicted serum concentrations with a mean error of 3.49 +/- 0.29 micrograms/ml without significant tendency to over- or underpredict. We conclude that this dosing program may aid clinicians by improving dosing accuracy and predicting serum concentrations in patients with PHT toxicity.