New Practices in Computational Modeling

W. R. Greco
2007 Clinical Cancer Research  
In this issue of Clinical Cancer Research, Hu et al. (1) present a specific physiologically based pharmacokinetic/pharmacodynamic study of doxorubicin in the dog, in which the investigators have improved a few critical elements of the traditional physiologically based pharmacokinetic/pharmacodynamic approach, such that their work may serve as a model study, a new paradigm, to be adopted or adapted by other research groups, to increase the overall utility of animal physiologically based
more » ... ally based pharmacokinetic/pharmacodynamic studies in predicting human pharmacokinetic/pharmacodynamics. Physiologically based pharmacokinetic/pharmacodynamic modeling is a mature subfield of pharmacokinetic/pharmacodynamic modeling that has resulted in a large body of published articles. In a Medline search of the 50,000+ articles found with pharmacokinetics as a search term, about 500 articles were also found with physiologically based pharmacokinetics as a search term. Many of the pioneering articles in the field have been highly cited. For example, references 2 -5, authored by Kenneth Bischoff, Robert Dedrick, and coworkers, from the late 1960s and early 1970s, have been cited 152, 90, 399, and 70 times, respectively, up to November of 2006. The general physiologically based pharmacokinetic/pharmacodynamic approach borrows heavily from the field of chemical engineering and models the distribution, metabolism, and effects of agents with real anatomic/physiologic organs, tissues, and compartments. This approach has been often considered a marked improvement over (or at least an important complement to) traditional compartmental pharmacokinetics/pharmacodynamics, which uses concepts of central and peripheral compartments, which only roughly map to real anatomic/ physiologic fluids, organs, and tissues. Because of ethical and practical concerns, traditional compartmental pharmacokinetic/pharmacodynamic studies in humans usually sample only plasma, urine, and/or other body fluids for drug/ metabolite levels, and then make indirect inferences about hypothetical body compartmental parameters such as volumes of drug distribution and drug clearances. In contrast, because of less ethical and practical constraints in animal model systems, physiologically based pharmacokinetic/pharmacodynamic studies commonly directly sample not only body fluids but also actual organs/tissues for drug/metabolites, make direct inferences about real anatomic/physiologic parameters, and then extrapolate the findings to real human organ and tissue compartments. Physiologically based pharmacokinetic/pharmacodynamic studies commonly include some or all of the following steps. An agent (older therapeutic drug already in common clinical use, newer therapeutic drug undergoing development, or important environmentally/occupationally toxic agent) is administered in a controlled fashion to an animal model. At specific times, the animals are sacrificed and the levels of the compound and relevant metabolites are assayed in plasma, urine, other excreta, and a subset of tissues thought to be relevant to the pharmacokinetics or pharmacodynamics of the specific agent. A compartmental diagram with real organ systems/ tissues, plasma flows, metabolism, binding, excretion, and effects of the agent is then constructed. The inflow and outflow of the agent and metabolites in each compartment of the model are then modeled with one or more equations, usually mass balance differential equations. The whole group of individual equations is then made into the minimum logical set of equations that accounts for the complete system of important inputs, bindings, changes, flows, and outputs of the agent. Some parameters of the grand model, such as the distribution ratios of the agent between plasma and specific organs, are calculated from the observed data. Other parameters being used, such as plasma flows and organ sizes for the specific model species, are often taken from the literature (e.g., ref. 6) and then scaled for the specific size and age of the animals in the study. The comprehensive grand model is then simulated with numerical differential equation software, and the simulated time courses of agent concentration and possible effect are compared with the observed data. If there is a mismatch between the simulated and observed data, the model structure and/or model parameters are changed (tweaked) in a logical fashion until there is general concordance. After the mathematical model in the model animal species is finalized, extrapolations are often made to humans, and concordance with human pharmacokinetic/pharmacodynamic data is explored. Many of the past publications in the field (in my view) have lacked the statistical rigor and predictive power to make materially important contributions to the understanding of the pharmacokinetics/pharmacodynamics of agents in humans. Reasons for these deficiencies include an unbalanced emphasis on the engineer's toolbox (especially continuous simulation), as opposed to the statistician's toolbox (especially curvefitting), and the ignoring of fundamental modeling concepts, such as parameter identifiability (the ability to estimate a parameter from actual data). Hu et al. (1) may change these perceptions of statistical rigor and predictive power, however, from this time onwards. The investigators have made a few tweaks in the overall physiologically based pharmacokinetic/
doi:10.1158/1078-0432.ccr-06-2811 pmid:17317813 fatcat:ymzz4munczb3fpkxprxnjxunzm