Sources of prediction error when using a Bayesian method to evaluate nortriptyline serum concentrations

A Bayesian method was used to evaluate nortriptyline (NTP) serum concentrations (Cps) and predict future Cps in two populations: five simulated groups (n = 20 each) with known clearance (CL) and volume of distribution (Vd), and an actual inpatient group (n = 20). The effects of weight, CL, Vd, and magnitude of Cps on absolute prediction error (APE) were evaluated. In simulated groups, Cps after two doses of NTP and for steady-state were calculated for normal, increased, and decreased Vd and CL. In the actual patient group, Cps were measured in the first few days after starting NTP administration and again during maintenance therapy. The first Cps were used in the Bayesian program to estimate CL and Vd to predict the second Cps. In the simulated group, PE and APE differed significantly between normal and decreased values of CL. A large Vd resulted in less of a change in PE or APE in these subjects, but when combined with low CL led to the largest errors. In the actual patient group, PE was -5.9 +/- 19.2 ng/mL and APE was 15.4 +/- 12.6 ng/mL. In these patients, only body weight was correlated with the percent APE (r = 0.607, P = 0.005). The Bayesian method performs well clinically, but increased Vd and decreased CL can lead to higher PE. Clinically, the only factor that predicted higher APE was obesity. This may reflect an effect on Vd, and in these patients, a high APE may occur.