A pharmacodynamic approach to the estimate of carbamazepine autoinduction

Population-based pharmacokinetic prediction algorithms have been developed for several medications. A fundamental assumption has been that the kinetics remain constant over time. Carbamazepine (CBZ), however, induces its own metabolism in a concentration- and time-dependent manner. A Bayesian estimation program is presented that models the changing catabolic enzyme activity, linearly related to hepatic microsomal enzyme concentration, along with the serum drug concentration. An E_max model is used for enzyme formation with respect to drug concentration: elimination of enzyme activity is modeled as a first-order process. This program was tested in 22 drug-naive outpatients begun on CBZ monotherapy. The 1 week concentrations were used to prospectively predict concentrations at 1 month of therapy and were very close to actual measurements: prediction bias (mean error of prediction) = −0.1 μg/mL and precision (median absolute error of prediction) = 1.2 μg/mL. Comparison estimates, made by assuming a constant concentration/dose ratio, had bias = 2.6 μg/mL (p < 0.001) and precision = 2.2 μg/mL (p = 0.01). We conclude that (1) CBZ autoinduction is not complete after 1 week of therapy and (2) the methodology permits accurate estimation of CBZ pharmacokinetics.