The Influence Of Covariate Distribution On The Prediction And Extrapolation Of Pharmacokinetic Data In Children.
C. Piana(1), M. Danhof(1), O.E. Della Pasqua(1,2)
(1) Division of Pharmacology, Leiden/Amsterdam Center for Drug Research, The Netherlands (2) Clinical Pharmacology Modelling & Simulation, GlaxoSmithKline, Stockley Park, UK
Objectives: The predictive performance of a model in which covariates have been identified should imply the ability not only to accurately describe the data used during model building, but also to make extrapolations to a population in which the covariate values differ from the original population. Extrapolation beyond the covariate range explored is particularly important for the purposes of pharmacokinetic bridging in paediatric indications. The aim of this simulation exercise is to assess model performance when covariate distribution and range of values differ from the original population used during model building.
Methods: A hypothetical drug with a one compartment model was used, in which weight has been linearly and exponentially correlated to clearance. Allometric scaling concepts were taken into account, but the exponents were explored with values higher and lower than 0.75. Plasma concentrations were simulated for a group of 43 children with a weight range between 7.43 and 61.3 Kg. A total of 8 observations per subject were simulated. In addition to full population, two subgroups were created. The first group comprised 20 children with weight between 10 and 15 kg, whilst the second group included 14 children with weight between 30 and 45 kg. The simulated plasma concentration data sets were subsequently fitted to a pharmacokinetic model according to standard model building criteria. Model performance was assessed by comparing the accuracy of AUC and Cmax estimates obtained for each subgroup, based 1) on a model derived from the overall population (n=43) and 2) by extrapolations to subgroup 2 (n=14), based on a model derived from subgroup 1 (n=20). NONMEM VI was used to perform model building and simulations and R for data manipulation and graphical summary of the results.
Results: PK parameters of interest could not be accurately predicted from a pharmacokinetic model obtained from a population with a different covariate range. Factoring the differences in the covariate distribution did not improve model performance. Predictions were accurate only when a model is used for the purposes of interpolation within the population. regarding results.
Conclusions: The use of pharmacokinetic modelling in paediatric research should be limited to interpolations within the range of values observed during model building. The covariate point estimate (from the overall population) must be kept in the model even when extrapolations refer to a subset of the original population.
