A simulation study of the optimal experimental designs for a two-compartment kinetic model

Abstract

The optimal design of nonlinear experiments is based on an optimality criterion that maximizes the Fisher information of an assumed model. In practice, however, we can be less certain about the actual performance or statistical efficiencies of the obtained designs that are jointly affected by the nonlinear model structure, the prior assumption, and the parameter estimation, all of which can make an optimal design underperform. For a realistic numerical evaluation of the designs, we implement a Monte Carlo simulation and parametric bootstrapping approach. In particular, we are interested in a classical second-order kinetic model, the structure of which is however unstable and vulnerable to random noises, as shown in our demonstration. As a result, the corresponding optimal designs are found to be inefficient, even so when the parameter prior is uninformative. Reasonable and robust model and prior assumptions are crucial for careful planning of future optimal design of experiments.