Comparison of some control strategies for three-compartment PK/PD models.

In drug therapy, effective dosage strategies are needed to maintain target drug effects. The relationship between drug dose and drug effect is often described by pharmacokinetic/pharmacodynamic (PK/PD) models where typically the PK model has a multicompartment form and the PD model is the sigmoidal Emax model. The parameters in the PK/PD model are generally unknown in the individual patient, although prior knowledge may be available and can be updated after measurements of drug effect are taken during the therapy. This fact, together with the complexity of the PK/PD model, makes the control problem complex. This paper investigates several control strategies in the framework of a three-compartment PK model plus an effect site with a PD model. Using computer simulations under different assumptions, we show that a MAP (maximum a posteriori) Bayesian type of strategy is effective, nevertheless in high-risk situations a stochastic control strategy hedging against estimation errors provides better performance at computational cost.