Abstract: The field of pharmacometrics, in particular pharmacokinetics and pharmacodynamics modeling, uses a step-wise approach to building mathematical models where typically a base structural model is established followed by inter-individual and inter-occasion variability terms and finally evaluation of covariate relationships that affect the structural and variance parameters. This approach evolved primarily as an adaptation to limitations in computational resources (much of the software used for population pharmacokinetics and pharmacodynamics was initially developed in the 1970s). It has been demonstrated that there are significant interactions between the selection of the structural model elements (i.e., fixed effects in a nonlinear mixed effects model) and variance elements (i.e., hierarchies of variance including between subject and residual unexplained variability) that can result in very different structural and variance models (and therefore also covaraites) potentially being selected selected as the “best” description of data. I will present work that uses a hybrid-genetic algorithm approach to evaluate the structural form of population pharmacokinetic models and attempts to explore these interactions more fully by evaluating multiple changes in structural, variance and covariate models simultaneously. This approach also increases the number of potential candidate models assessed dramatically. Thus far, evaluation of population pharmacokinetic models using the hybrid genetic-algorithm approach has identified models that are at least “as good” if not substantially better based on fitness measures such as objective function value, parsimony and uncertainty of parameter estimates. Our conclusion is that this hybrid genetic algorithm approach may facilitate the pharmacometric modeler in assessing the scope of potential model structures and guide further assessment of effects on drug disposition.
Biography: Dr. Bies is currently Associate Professor of Pharmaceutical Sciences at the School of Pharmacy and Pharmaceutical Sciences as well as a member of the Center for Data Sciences and Engineering at the State University of New York at Buffalo. Prior to this, Dr. Bies was Associate Professor of Medicine and Medical and Molecular Genetics at the Indiana University School of Medicine and Director of the Disease and Therapeutic Response Modeling program for the Indiana Clinical Translational Sciences Institute. He serves as: collaborating scientist at CAMH, University of Toronto; North American and executive editor for the British Journal of Clinical Pharmacology; and on the editorial boards of the Journal of Pharmacokinetics and Pharmacodynamics, Clinical Pharmacology and Therapeutics:Pharmacometrics and Systems Pharmacology, Journal of Clinical Pharmacology and Biopharmaceutics and Drug Disposition. Dr. Bies is a member of AAPS, ISoP, ACCP and ASCPT. He is a board member of ISOP and vice-chair for the AAPS CPTR section. Dr. Bies received: a BSc degree in Pharmacy from the University of Toronto (1991); a Pharm.D. from the UTHSCSA and the University of Texas at Austin (1994); and a Ph.D. Pharmacology from Georgetown University in 1998. This was followed by postdoctoral training at the Center for Drug Development Sciences until 2000. His research focuses on the application of pharmacometric approaches in psychiatry, oncology, neurology and cardiovascular disease and novel methods development including machine learning approaches to model selection and optimization methods for parameter optimization in dynamic systems.