The history and future of population pharmacokinetic analysis in drug development.

Abstract

The analysis of pharmacokinetic data has been in a constant state of evolution since the introduction of the term pharmacokinetics. Early work focused on mechanistic understanding of the absorption, distribution, metabolism and excretion of drug products.The introduction of non-linear mixed effects models to perform population pharmacokinetic analysis initiated a paradigm shift. The application of these models represented a major shift in evaluating variability in pharmacokinetic parameters across a population of subjects.While technological advancements in computing power have fueled the growth of population pharmacokinetics in drug development efforts, there remain many challenges in reducing the time required to incorporate these learnings into a model-informed development process. These challenges exist because of expanding datasets, increased number of diagnostics, and more complex mathematical models.New machine learning tools may be potential solutions for these challenges. These new methodologies include genetic algorithms for model selection, machine learning algorithms for covariate selection, and deep learning models for pharmacokinetic and pharmacodynamic data. These new methods promise the potential for less bias, faster analysis times, and the ability to integrate more data.While questions remain regarding the ability of these models to extrapolate accurately, continued research in this area is expected to address these questions.