Generalized sensitivities and optimal experimental design

Abstract

We consider the problem of estimating amodeling parameter θ using a weighted least squares criterion for given data y by introducing an abstract framework involving generalized measurement procedures characterized by probability measures. We take an optimal design perspective, the general premise (illustrated via examples) being that in any data collected, the information content with respect to estimating θ may vary considerably from one time measurement to another, and in this regard some measurements may be much more informative than others. We propose mathematical tools which can be used to collect data in an almost optimal way, by specifying the duration and distribution of time sampling in the measurements to be taken, consequently improving the accuracy (i.e., reducing the uncertainty in estimates) of the parameters to be estimated. We recall the concepts of traditional and generalized sensitivity functions and use these to develop a strategy to determine the “optimal” final time T for an experiment; this is based on the time evolution of the sensitivity functions and of the condition number of the Fisher information matrix. We illustrate the role of the sensitivity functions as tools in optimal design of experiments, in particular in finding “best” sampling distributions. Numerical examples are presented throughout to motivate and illustrate the ideas.

Effects of Food and Pharmaceutical Formulation on Desmopressin Pharmacokinetics in Children

“…Sensitivity function optima are marked with green stars. These optima are considered good sampling time points, as the output is the most sensitive to a certain parameter at those points, enabling optimal estimation of this parameter [22], [34]. The most important parameters (i.e.…”

Reprint of “Pharmacokinetic modelling of the anti-malarial drug artesunate and its active metabolite dihydroartemisinin”

“…The sensitivity of a dependent variable x to a change in the parameter p j (considered about parameter vectorp) is the local quantity given by [31,32] …”

“…Note that this formulation normalises only by the presence of the p j , and does not divide by x itself (unlike as in [31,32]) to facilitate interpretation in an identifiability and correlation sense. Note that if the variable x is a function of an independent variable-such as time as here-then so are its sensitivities (S x,p j ).…”

Modeling the Dynamics of the Cardiovascular-respiratory System (CVRS) in Humans, a Survey

“…The design criteria are continuous functions of the Fisher information matrix F(θ 0 ), as for instance 1/ det F(θ 0 ) (D-optimal design) or λ max F −1 (θ 0 ) (E-optimal design) (compare [16,31,81]). In [7] the design criterion was introduced in order to define SE-optimal designs in the context of very general sampling strategies characterized by probability measures on the sampling interval. See [8] for a comparison of D-optimal, E-optimal and SE-optimal designs and [9] for a Monte Carlo based analysis.…”