Implementation and evaluation of a stochastic control strategy for individualizing teicoplanin dosage regimen

A stochastic control strategy for individualizing teicoplanin dosing schedule in neutropenic patients is proposed and compared to the usual Bayesian approach based on the mode of the posterior density of the model parameters. Teicoplanin disposition is described by a bicompartmental model. Age, body weight, serum creatinine, white blood cell count, and sex can be included as covariates. Posterior density of model parameters is obtained by Bayes theorem under a discrete form from which the posterior density of teicoplanin trough concentrations are computed for any dosing schedule. Optimal maintenance dose is determined by minimizing the cost associated, through a logarithmic risk function, to the concentrations being outside the therapeutic range. In Monte Carlo simulation studies on 300 individuals, stochastic control was more accurate than, and equally precise as the usual Bayesian approach. Two-sample based predictions were not better than one-sample based ones. Inclusion of covariates in the model improved dramatically the performances of both strategies. A small retrospective study based on real data (n = 16 patients) shows that reasonable accuracy (bias of 0.7 mg/L) and precision (3 mg/L) in teicoplanin trough concentration prediction is obtained with both strategies provided that covariates are taken into account.