Study and application of nonparametric and parametric population modeling for automatic subpopulation classification to CYP2D6 phenotype compounds and pediatric age groups
Kairui Feng1, Robert H. Leary1, Michael Dunlavey1, Amin Rostami-Hodjegan1,2
1. Certara, USA, 2. University of Manchester, UK
Objectives: Subpopulation classification is an important way to improve the decision making in drug development [1]. Existing population pharmacokinetic methods for subpopulation identification predominantly rely on heterogeneous expression or mixture modeling. If sub-populations distributions overlap, the identification of individual subpopulation based on existing methods becomes difficult and tricky. Taking CYP2D6 phenotype as examples, Dextromethorphan, Bufuralol or Imipramine has many metabolic pathways via other enzymes while Metoprolol or Desipramine or Tolterodine has less metabolic pathways [2]. Finding dosage for pediatric age sub-groups via this polymorphism is tricky. We propose to use nonparametric and parametric population methods combined for automatic subpopulation classification to overcome the problems suffered from current methods [1, 3].
Methods: A selected proportion of CYP2D6 phonotype subpopulation is virtually sampled using physiologically based pharmacokinetic (PBPK) modeling, i.e. to create combinations of heterogeneous virtual patients. CYP2D6 metabolic compounds as above are used to generate the drug exposure with subpopulations distributions overlap. A simplified population PK model is built using parametric method for identifying the initial population distribution. Nonparametric method [4] is added on top of parametric method which reduce the number of support points and automatically determine the number of components of the mixture models to capture the subpopulation. Simulation such as visual predictive check (VPC) is used to confirm the subpopulation.
Results: PBPK modeling successfully creates the overlay mixture distribution samples. The nonparametric support points automatically reduce from a few hundred points to a few points which avoids the current fixed number of components in mixture modeling problems [3]. The converted mixture models then use for simulation nicely in VPC and the subpopulations classification matched the predefined PBPK subpopulation.
Conclusions: The nonparametric methods successfully find differences in exposure in genetic subpopulations and pediatric age/weight groups. The method can be expanded to manage the dose-titration or individual treatment in all patients based on safety and/or efficacy markers, or on Therapeutic Drug Monitoring (TDM), or gene based dosing.
[1] Carlsson et al. Modeling Subpopulations with … a Subpopulation for the Use in Model Analysis and Improved Decision Making, the AAPS Journal, Vol. 11, No. 1, p148-154, 2009
[2] Kaila et al. Mixture models and subpopulation classification: … to metoprolol CYP2D6 phenotype. J PK PD. 2007.
[3] NMusers, VPC with Mixture Model, http://www.cognigencorp.com/nonmem/current/2009-April/0676.html
[4] Wang Y, Maximum likelihood computation for fitting semiparametric mixture models, Stat Comput., 20: 75–8, 2010
