Pharmacometric benchmarking: quantitative methods to assess the predictive performance of population pharmacokinetic modeling programs

Throughout much of history, safe and effective drug doses have been discovered through trial-and-error and validated via anecdote. Such approaches are limited in their ability to define how a drug's safety and effectiveness are influenced by the addition of other co-administered medications and the presence of other acute and/or chronic diseases. Consideration of all these pharmacological and pathophysiological factors is impractical given the complexity of the many interactions that may occur. To further advance clinical pharmacology, it has become necessary to leverage the increasing speed and storage capacity of computers. Developments in mathematics, statistics, and computer science have revolutionized the field of clinical pharmacology by making computers far more than glorified calculators. Today, sophisticated algorithms can be used to interrogate and learn from pharmacological datasets and make informed predictions about the safety and effectiveness of drug dosing regimens. The goal of these population pharmacokinetic analyses is to yield accurate predictions of clinically-relevant pharmacokinetic parameters and improve our understanding of the biological processes that mediate drug disposition. In this dissertation, we present the results of three pharmacokinetic studies that demonstrate the clinical utility of population pharmacokinetic modelling, along the way challenging conventional dosing strategies for vancomycin in preterm neonates and zolpidem among severely burned children. Additionally, we developed a simulation-based iv parameter estimation algorithms. This work lays the foundation for a transparent dialogue regarding the relative strengths and weaknesses of individual algorithms, which heretofore has not been possible. We conclude with a discussion of the additional unanswered questions that may now be investigated using the benchmarking framework developed here. The results of the studies described in this dissertation underscore the importance of enhancing the clinical adoption of population pharmacokinetic models. However, these models must be rigorously evaluated to ensure that they are unbiased and precise. In simulations, three of the most commonly used pharmacokinetic parameter estimation algorithms differentiated themselves when they were applied in different clinical scenarios. This finding highlights an intriguing practical fact that algorithm selection should be guided by the clinical question at hand.