(494a) Robust Parameter Estimation for Physiologically Based Pharmacokinetics Model with Drug Dissolution Dynamics

Physiologically based pharmacokinetics (PBPK) model can predict absorption, degradation, execration and metabolism in a drug delivery system. Thus, it can be useful for regulating dose and estimating drug concentration at a particular time during the clinical demonstration. While PBPK model is generally expressed as a set of ordinary differential equations with a large number of parameters, in-vivo experimental data are often noisy and sparse. This makes it difficult to estimate parameters with conventional least squares approaches. Therefore, maximum a posteriori (MAP) method based on Bayes’ rule can be used for parameter estimation of PBPK model. Conventional MAP methods require constructing probability distribution function (PDF) using sampling techniques such as Markov Chain Monte Carlo (MCMC) sampling scheme. However, this sampling approach is computationally prohibitive to use because PBPK models often involve a large number of unknown parameters. This work proposes a MAP scheme based on the objective function from Bayes’ rule. The proposed method estimates parameters using the product of prior distribution and likelihood, not from the posterior distribution. Therefore, even if there are a large number of parameters, the sampling approach is not necessary and the estimation problem is converted to a multi-variable numerical optimization problem.

 The prior distribution was specified either uniform or normal distribution according to the prior knowledge on the uncertain parameters. Since the likelihood function is related to the difference between predictions and observations, normal distribution was employed. This implies that there is no mismatch in the model structure. Hence, this study also proposes drug dissolution dynamics based on Noyes-Whitney equation for orally administrated drugs and incorporates them into the PBPK model. The proposed scheme is illustrated on Tegafur, an orally-administrated anti-cancer drug, with rat in-vivo data.

 The robustness of parameter estimation by the proposed scheme was compared against least squares estimation and conventional MAP scheme. The variances of parameters of the suggested method showed the best performance regardless of magnitude of measurement noise and the number of observations. The prediction accuracy was also improved by 73% compared with the PBPK model without the drug dissolution model.