Serum phenobarbital concentration predictions by a personal computer software system

The ability of a personal computer software system to predict actual serum phenobarbital concentrations (SPC) in outpatients taking phenobarbital chronically was assessed by comparing actual with predicted SPC for accuracy, bias, and precision. Data for a four-year period were collected on patients at an outpatient clinic's pharmacokinetic consultation service. The study group included 50 adults and children with at least one SPC taken at a known time after dose administration. Input variables were weight, sex, height, age, concomitant drugs and diseases, phenobarbital dosage regimen, and the time and reported value of all SPC. Initially, SIMKIN (SIMulated KINetics) simulated dosing regimens on the basis of literature estimates of pharmacokinetic parameters; SPC were then estimated for these regimens and compared with actual values. One or two additional SPC were added to the input data and analyzed, and the predicted SPC compared with actual values. Although SIMKIN's accuracy and bias as measured by regression analysis and mean prediction error, respectively, were within clinically acceptable limits, the precision was not. However, these results are limited by the population studied. Patient compliance, concomitant phenytoin therapy, changes in phenobarbital pharmacokinetic parameters with chronic dosing, and disease interactions may significantly affect predictive ability. The clinical effects of these factors need to be evaluated to further improve predictions.