Predicting phenytoin dosages using Bayesian feedback: a comparison with other methods.

A Bayesian feedback technique for predicting phenytoin dosage was compared to other dosing methods. Sixty-nine cases were selected on the basis of apparent reliability from 103 medical charts of epileptic patients with multiple phenytoin levels on different dosage regimens. Two published nomograms and a graphical, or computational, technique were compared to the Bayesian technique. Each method was assessed for absolute predictability using measures of bias and precision, i.e., mean percent error and root mean squared percent error, respectively. For a single previous data pair, the Bayesian method was similar to a published nomogram with regard to bias and precision. For multiple data pairs, the graphical or simultaneous equation technique tended to be less biased, but the Bayesian method had better precision. However, none of these differences was statistically significant (p greater than 0.05). The Bayesian method yielded the lowest percentage of predicted doses that exceeded 110% of the actual dose. The Bayesian method conveniently provides a single method applicable to the use of either single or multiple concentration-dosage data pairs and results in fewer extreme dosing errors.