Shaundee Jameson, Jessica Tapia, Robert Trujillo, and Dr. Alona Kryshchenko
The selection of optimal drug dosages is crucial to every phase of the drug development process. Central to understanding these dosages is the study of pharmacokinetics, or, how a drug is affected by the human body after consumption. Pharmacokinetic modeling is used to understand the relationship between a drug’s dosage and its consumer’s blood plasma concentration levels, but it can be challenging due to complex data and the inherent variability within and between human subjects. Pharmaceutical companies traditionally manage these challenges through
the use of parametric statistical modeling to identify the covariates that could influence the dose-concentration relationship. Although this approach suits small sample sizes and sparse data sets well, it requires researchers to specify a structural pharamacokinetic model a priori, which can be cumbersome to adjust when dealing with highly variable and massive data sets. The field of machine learning provides promising alternative that could estimate plasma concentration without relying on complex statistical and pharmacokinetic models. In this research, we compare
the effectiveness of parametric, non-parametric, and machine learning models at predicting plasma concentration in patients using a simulated clinical data set.