Experiment design for nonparametric models based on minimizing Bayes Risk: application to voriconazole¹.

Area of Science:

• Pharmacometrics
• Computational Biology
• Experimental Design

Background:

• Individualized therapy often uses nonparametric (NP) population models, which are discrete probability distributions.
• Existing experimental design methods, often based on parametric models, may not fully capture the discrete nature of NP models.
• Designing experiments for NP models is challenging, especially for identifying individuals within these models.

Purpose of the Study:

• To develop an experimental design approach tailored for nonparametric (NP) population models in individualized therapy.
• To address the limitations of existing methods by considering the discrete nature of NP models from a classification perspective.
• To introduce and evaluate novel methods for experiment design in this context.

Main Methods:

• Framing the NP experiment design problem as a classification task.
• Utilizing Bayes Risk as a key information measure for experimental design.
• Developing and detailing the multiple-model optimal (MMopt) experimental design method.
• Simulating examples and conducting a case study in pharmacokinetics.

Main Results:

• The study provides new insights into NP experiment design by viewing it through a classification lens.
• Bayes Risk is proposed as an effective information measure for this discrete context.
• The MMopt method demonstrates minimal computational requirements and significant advantages over existing approaches.
• Simulations and a case study confirm the utility of MMopt in pharmacokinetics.

Conclusions:

• The MMopt method offers an efficient and effective approach for experimental design in individualized therapy with NP models.
• This classification-based perspective provides a more natural fit for discrete NP models.
• The findings have direct applications in pharmacokinetics and personalized medicine, as shown in the voriconazole case study.