Callies Sophie1, Suriyapperuma Subha1, Barrington Philip1, Yuan Zheng1, Lau Yiu-Keung 1, Tuttle Jay 1, Waters David1, Cocke Patrick.J1, Soon Danny1.
Eli Lilly and Company
Objectives: To assess the variance parameters of the PKPD model using a sensitivity analysis. To describe, using the model, the change of the biomarker input rate following administration of the mAb.
Methods: Approximately 10 PK and 9 PD data per individual were available over a time frame of 2 weeks (or more). Multi-compartments PK models with non-linear TMDD clearance were fit to the mAb serum concentration versus time data (Gibiansky et al, 2008, 2009). An indirect response model, with a sigmoid Emax function describing the effect of the mAb on the input rate of the biomarker, was fit to the pharmacodynamic data. In addition, to account for the biomarker circadian rhythmic change, this model also included a cosine function to describe the input rate of the biomarker. NONMEM version VII.2 was used to model the data with the First Order Conditional Estimation with Interaction (FOCEI) method implemented. Visual predictive checks, standard error on the estimates, objective function values and diagnostic plots were used to drive the model development.
Results: A three-compartment PK model with non-linear target mediated drug disposition clearance adequately fit the mAb PK data. This model was found to be better than a two compartment model. The indirect response PKPD model, mentioned in the method section, adequately describe the mean profiles of the biomarker. All fixed effects – mean parameters (e.g clearances, volumes, EC50, Emax) were reliably estimated with standard error on the estimates (SEE) less than 25 % (except for EC50, SEE 34%). The random effects – variance parameters could only be estimated on a few of the fixed effects parameters and were less precisely estimated than the fixed effects. This issue in the estimation of variances led to inflated distributions in the visual predictive check plots. Therefore a sensitivity analysis was carried out to determine the best estimates for the random variance parameters.
Conclusions: Reliable estimation of variance parameters can be challenging when the structural model is complex (such as the model presented in this abstract with non-linear TMDD model and circadian change of the biomarker). The sensitivity analysis helps better estimate the variance parameters in order to get more reliable predictions of the variability in the biomarker response. This model will be used to help predict biomarker response through simulation and will be applied to make dosing decisions.
References:
[1] Gibiansky L, Gibiansky E. Target-mediated drug disposition model: relationships with indirect response models and application to population PK-PD analysis. 2009; J Pharmacokinet Pharmacodyn; 36:341-351.
[2] Gibiansky L, Gibiansky E, Kakkar T, Ma P. Approximations of the target-mediated drug disposition model and identifiability of model parameters. 2008; J Pharmacokinet Pharmacodyn; 35:573-591.
Reference: PAGE 21 () Abstr 2349 [www.page-meeting.org/?abstract=2349]
Poster: Model evaluation