The process of knowledge discovery from large pharmacokinetic data sets.

The advent of statistical software with powerful graphical and modeling capabilities has revolutionized the manner in which pharmacokinetic and pharmacodynamic analyses are performed. Knowledge discovery from a large (population) pharmacokinetic data set incorporates all steps taken from data assembly to the development of a population pharmacokinetic model and the communication of the results thereof. The process can be formalized into a number of steps: (1) creation of a data set for pharmacokinetic knowledge discovery, (2) data quality analysis, (3) data structure analysis (exploratory examination of raw data), (4) determination of the basic pharmacokinetic model that best describes the data and generating post hoc empiric individual Bayesian parameter estimates, (5) the search for patterns and relationships between parameters and parameters and covariates by visualization, (6) the use of modern statistical modeling techniques for data structure revelation and covariate selection, (7) consolidation of the discovered knowledge into irreducible form (i.e., developing a population pharmacokinetic model), (8) the determination of model robustness (determination of the reliability of model parameter estimates), and (9) the communication and integration of the discovered pharmacokinetic knowledge. This process is discussed, and a motivating example is presented. The use of modern graphical, modeling, and statistical techniques for knowledge discovery from large pharmacokinetic data sets has given the data analyst the freedom to choose statistical methodology appropriate to the problem at hand with the maximization of information extraction, rather than on the basis of mathematical/statistical tractability.