Comparison of two binding equations for prediction of the concentration of unbound valproic acid in the serum of adult epileptic polytherapy patients.

Because binding of valproic acid to plasma proteins affects the efficacy of the drug in the treatment of epilepsy (only the unbound fraction of the drug is effective) we have compared two methods which use different binding parameters to predict the in-vivo concentration of unbound valproic acid in serum. The study was performed on 46 serum samples from 29 polytherapy adult patients with epilepsy. Mean prediction error, mean absolute prediction error and root mean squared error were calculated for each method; these values served as a measure of prediction bias and precision. The mean absolute prediction errors and root mean squared errors for the two methods were similar in magnitude (Method 1, mean absolute prediction error = 10.0 microM, root mean squared error = 15.0 microM; Method 2, mean absolute prediction error = 10.3 microM, root mean squared error = 13.5 microM). Method 2 had a general tendency to over-predict unbound valproic acid; both methods had a tendency to over-prediction for total concentrations above 500 microM. Method 1 had a tendency to under-prediction at total concentrations below 250 microM. Within the total concentration range of valproic acid investigated, Method 1 was superior to Method 2 for prediction of unbound serum valproic acid. Our approach using Method 1 may be useful for prediction of unbound serum valproic acid concentration in patients with total valproic acid concentrations ranging from 250 to 500 microM; Method 2 may be useful for patients with total valproic acid below 500 microM. Our results suggest that there is wide and unpredictable variability in valproic acid binding to serum proteins among study populations.