Predictive performance of two phenytoin pharmacokinetic dosing programs from nonsteady state data

Area of Science:

• Pharmacokinetics and Pharmacodynamics
• Clinical Pharmacology
• Computational Drug Dosing

Background:

• Accurate prediction of drug concentrations is crucial for therapeutic drug monitoring.
• Phenytoin dosing requires careful management due to its narrow therapeutic index.
• Computer programs aid in predicting drug serum levels, but their accuracy varies.

Purpose of the Study:

• To evaluate the predictive performance of two computer programs, Drugcalc and PKS, for phenytoin serum concentrations.
• To compare Bayesian (method 1, method 3) and non-Bayesian (method 2) approaches in predicting drug levels.
• To assess the impact of steady-state and non-steady-state data on prediction accuracy.

Main Methods:

• Performance evaluation of Drugcalc (Bayesian) and PKS (non-Bayesian and Bayesian) programs.
• Prediction of 771 phenytoin concentrations using steady-state, non-steady-state, and combined data.
• Analysis of prediction errors, particularly focusing on dose differences (dD).
• Inclusion of clinical data from 15 patients under routine conditions.

Main Results:

• Excluding dose differences (dD) of 100 mg/day or more significantly increased prediction precision by 87% (method 1), 64% (method 2), and 66% (method 3).
• Non-steady-state data alone yielded clinically acceptable predictions only when combined with steady-state data and using the Bayesian approach (method 3).
• Non-steady-state data effectively detected potential toxicity in 71.4-84.6% of cases across methods.

Conclusions:

• Dose difference (dD) is a major contributor to prediction errors in phenytoin serum concentration monitoring.
• The Bayesian approach, particularly when using non-steady-state data with steady-state values, improves prediction accuracy.
• Non-steady-state data are valuable for identifying potential toxicity and refining dosage adjustments, especially with method 3.