| A drug-drug interaction (DDI) occurs between a drug and the other concomitantly administered drug that affects the pharmacokinetic (PK) profile and/or therapeutic or adverse effects of the drug. DDI emerges as a crucial issue in drug development and its multi-fold impact on pharmaceutical industry, public health and society has now been recognized. In this dissertation, we consider three specific problems: (i) How can we predict in vivo DDI from available individual drug's in vivo data and early-stage in vitro DDI data in aid of screening compounds that pose potential DDI from entering later-phase drug development and market? (ii) How can we clinically evaluate the model built in (i) with an observed clinical pharmacokinetic DDI study data? (iii) Non-compartmental model has been a popular PK method; however, it is less useful than compartment model in the predicting problems such as DDI prediction. Therefore, recovering the compartmental model parameters from the published non-compartmental model results has become a crucial step before compartmental modeling can be taken, and this has now been an obstacle facing the pharmacokinetic modelers. In problem (i), we develop an integrated, stochastic, mechanism-based DDI-predication methodology, in which we build a system of PK models connected via the well-stirred enzyme inhibition-based liver model, adopt Bayesian hierarchical nonlinear framework to incorporate the multi-level prior information learned for the PK parameters, and implement a Monte Carlo simulation-based approach to render the DDI prediction evaluation with stochastic properties. The system of differential equations is solved numerically. Next in problem (ii), we propose Monte Carlo based equivalence tests to perform clinical evaluation of the built model at both the mean level and variability level, respectively. Some asymptotic properties of the testing approaches related to the sample size and power are analytically derived. For illustration, we use the ketoconazole-midazalom pair as an example. Lastly, in problem (iii), we propose a Bayesian meta-analytic approach. We develop a three-level Bayesian hierarchical model based on summarized data across studies, where we derive the (nonlinear) functional relationship of the observed non-compartmental model PK parameters to the unknown compartmental model parameters. First-order Taylor approximation is used. The use of hybrid MCMC sampling of Gibbs sampler with Metropolis-Hastings algorithm for posterior inference makes the computation feasible. We evaluate the performance of the proposed model by analyzing both simulated and real data. For illustration, we use midazolam as an example. |
