Comparing a mass-balance algorithm with a Bayesian regression analysis computer program for predicting serum phenytoin concentrations.

Abstract

The ability of a mass-balance algorithm to predict non-steady-state phenytoin concentrations in neurosurgery patients was compared with that of Phenda, a computerized Bayesian regression analysis program. Fifty neurosurgery patients who had had two or more initial phenytoin serum concentrations measured at least 60 hours apart and at least 1 hour after any i.v. doses, with the second concentration being not more than twice and not less than half of the first, and who had had a third or final phenytoin measurement (for use in a prediction analysis) were evaluated. The patients' maximum rates of metabolism were calculated by using the two initial phenytoin concentrations and a mass-balance algorithm, and the third phenytoin concentration was predicted. The patients' demographics and phenytoin dosages and concentrations were entered into Phenda, which was used to predict the third phenytoin concentration. The ability of the two methods to predict the third concentration was evaluated by the method of Sheiner and Beal. Fifty observations from 48 patients were evaluated. The mass-balance algorithm had a positive prediction bias of 2.52 mg/L and a precision error of 5.08 mg/L, compared with 2.30 and 5.30, respectively, for Phenda. The difference in the results between the two methods was not significant. There was no significant difference between the mass-balance algorithm and Phenda in the ability to predict phenytoin concentrations.