EVALUATION OF EQUATIONS FOR UNBOUND SERUM CONCENTRATION PREDICTION OFCARBAMAZEPINE AND CARBAMAZEPINE-10,11-EPOXIDE IN POLYTHERAPY PEDIATRIC-PATIENTS WITH EPILEPSY

We retrospectively evaluated the ability of equations with in vivo pop ulation binding parameters of our previous study (Method 1) or an aver age unbound fraction of 0.25 of Pynnonen (Method 2) to predict the unb ound serum carbamazepine (CBZ) concentration in 50 serum samples from 28 polytherapy pediatric patients with epilepsy. In 12 serum samples f rom 10 patients, the ability of equations for unbound serum carbamazep ine-10, 11-epoxide (CBZ-E) concentration prediction was also determine d in predictive performance with in vivo population binding parameters of our previous study (Method A) or an average unbound fraction of 0. 5 of Pynnonen (Method B). Mean prediction error, mean absolute predict ion error (MAE), and root mean squared error (RMSE) were calculated fo r each method, and these values served as a measure of prediction bias and precision. Method 1 shows a bias to underpredict unbound serum CB Z. The MAE and RMSE were lower in Method 2 (MAE = 0.696 mu M, RMSE = 0 .912 mu M) than in Method 1 (MAE = 0.946 mu M, RMSE = 1.138 mu M). Met hod 2 is superior to Method 1 in accuracy and precision. The effects o f antiepileptic co-medications on predictive performance of Method 1 a re relatively larger in a co-medicated group of serum samples with val proic acid (n = 33, MAE = 0.994 mu M, RMSE = 1.211 mu M) than in a gro up of serum samples without valproic acid cc-medication (n = 17, MAE = 0.853 mu M, RMSE = 0.979 mu M). Furthermore, the effects of the numbe r of antiepileptic co-medications on predictive performance of Method 1 are relatively smaller in a group of serum samples of CBZ with one c o-medication (n = 17, MAE = 0.600 mu M, RMSE = 0.682 mu M) than in eac h group of those of CBZ with two (n = 16, MAE = 1.219 mu M, RMSE = 1.3 78 mu M) and with three to six co-medications (n = 17, MAE = 1.035 mu M, RMSE = 1.246 mu M). For unbound CBZ-E prediction, Method B has a bi as to overprediction. The MAE and RMSE were lower in Method A (MAE = 0 .267 mu M, RMSE = 0.354 mu M) than in Method B (MAE = 0.400 mu M, RMSE = 0.517 mu M). Method A is superior to Method B in accuracy and preci sion.