Amir S. Youssef(1), Joel S. Owen(1), Juan Jose Pérez-Ruixo(2) and Sameer Doshi(2)
(1)Department of Pharmaceutical Sciences, Union University, Jackson, Tennessee, USA; (2)Department of Pharmacokinetics and Drug Metabolism, Amgen Inc., Thousand Oaks, California, USA
Objectives: Analyzing pharmacokinetic (PK) data from small numbers of subjects can be supported by the use of Bayesian priors in NONMEM. Bayesian priors may be considered instead of combining the original and new datasets either for speed of the analysis or when the original data are not available to be combined. This simulation study assessed the influence of the LTBS approach on the type 1 and 2 error rates of clearance estimation when using Bayesian priors.
Methods: An open 2-compartment PK model with linear elimination from the central compartment for a hypothetical biological compound was employed to describe the time course of drug concentrations after intravenous bolus, zero order subcutaneous input was employed to characterize drug absorption. Clearance (CL) and volumes were allometrically scaled. Interindividual variability on parameters was set to 25% and the residual error was defined as 20% CCV. Body weights were randomly drawn from the NHANES dataset (http://www.cdc.gov/nchs/nhanes.htm). Plasma drug concentrations were simulated (500 replicates) for two scenarios: 1) subjects with 1x the prior CL and 2) subjects with 1.4x the prior CL. Each replicate consisted of 15 subjects with 8 optimal samples per subject. Profiles were fit with NONMEM 7 using PK priors and with or without the LTBS approach. The probability of observing Type 1 and 2 errors, bias/precision of parameter estimates, and diagnostic plots were evaluated across the scenarios.
Results: Using an exponential error model (non-LTBS), the type 1 error rate was greater than 70% (1xCL). In the same scenario using the LTBS approach, the type 1 error dropped to less than 5%. Type 2 error (1.4xCL) was greater than 20% with non-LTBS and less than 1% with LTBS. With the LTBS approach the estimation of clearance was accurate and precise, while without the LTBS approach the estimation of clearance was inaccurate and imprecise. Overall, the LTBS approach dramatically improved the model fit and the normality of weighted residuals provided an indication of the benefit of the LTBS approach.
Conclusions: The type 1 and 2 error rates as well as the accuracy and precision of clearance estimation using Bayesian priors and LTBS approach is a suitable method for PK data analysis of compounds that follow a two compartment linear PK model with zero order absorption. Further research is ongoing in order to generalize these findings to other common PK models.
Reference: PAGE 23 () Abstr 3091 [www.page-meeting.org/?abstract=3091]
Poster: Methodology - Other topics