Chenel M (1, 2), Ogungbenro K (1), Duval V (2), Laveille C (2), Jochemsen R (2), Aarons L (1)
1. School of Pharmacy, University of Manchester, United Kingdom. 2. Institut de Recherches Internationales Servier, 6 place des Pléiades, 92415 Courbevoie Cedex, France.
Introduction: A phase II dose-ranging study is planned for a drug in clinical development. The pharmacokinetics (PK) of the compound will be part of an ancillary study and will be considered as a secondary objective of the trial. The two aims of the PK analysis will be firstly to estimate PK parameters at steady state using a population approach in the concerned population, and secondly to investigate any potential PK/PD relationship between the exposure (Area Under the Curve) and the activity. As characterising the PK is at best a secondary objective in the clinical study, few samples can be drawn per patient for PK analysis and the sampling schedules must be as flexible as possible. Sampling time windows will be specified in order to get the maximum information from the samples without disturbing the trial too much and a population approach will be used for the PK analysis.
Aim: To determine the optimal blood sampling time windows for the estimation of PK parameters by a population approach under clinical constraints.
Methods: The clinical study will involve 3 dose levels (10, 30 and 90mg) and 150 patients per dose group. The drug will be administered orally once a day and the PK samples will be taken at steady state. All patients will be sampled just before drug administration (trough) and between 2 and 4h after dose. One third of patients in each dose group could as well be sampled between 4 and 10h after dose with a maximum number of samples equal to 6 per patient. Because the trial will be double blind, sampling time windows must be the same in each dose group. Based on previous data, the following model was shown to properly describe the PK of the compound: a 2-compartment model with a first order absorption constant (KA); inter-individual variability (IIV) on the elimination and inter-compartmental clearances (CL and Q, respectively) and on the KA. The residual variability was a combined error model. Based on this model two approaches were developed: M1 method where all parameters were estimated and M2 method where KA and IIV on KA were fixed. Optimal sampling times were determined by optimizing the population Fisher information matrix (PFIM) using PFIM 1.2 under MATLAB for the two approaches. The criterion used was D-optimality and the algorithm was a modified Fedorov exchange algorithm. Optimal sampling time windows were determined by allowing the D-optimal windows design to result in a specified level of efficiency when compared to the fixed-times D-optimal design. Finally, the D-optimal sampling time window design was evaluated, after MATLAB simulations and NONMEM estimation with the FOCE interaction method, by computing the relative error of estimations.
Results: According to the coefficients of variation of the standard error (CVSE) given by PFIM for each parameter, the best results were obtained when KA and IIV on KA were fixed (M2 method). Windows were determined with the M2 approach and 4 optimal sampling time windows : at trough, between 2 and 4h after dose for all patients and only 2 sampling time windows between 4 and 10h after dose for 1/3 of patients; equal to [4h – 5h05’] and [9h10’-10h]. These sampling time windows obtained with a 90% level of efficiency and a uniform sampling distribution are wide enough to be useful in a clinical trial. The trends of CVSE values given by the PFIM are in full agreement with the simulations of 100 datasets with the selected design. Mean population parameters, such as the CL, were quite well estimated but the relative error was high for the IIV on Q, and for the additive random error.
Conclusion: An optimal sampling time windows strategy under clinical constraints was implemented for a Phase II study. Sampling time windows were designed and the PK sampling schedule was evaluated by simulation. As previously described with fixed effect models, the number of samples per patient is equal to the number of fixed parameters, and consequently in the present case only 4 sampling times windows were necessary to estimate parameters as there were 4 mean population PK parameters in the model. The weakness of the design was the lack of information about the absorption phase which might be overcome with a Bayesian approach. However, characterising the absorption phase is not a major concern for the proposed trial as the clinical study will mainly focus on exposure. Therefore, the sampling time windows will then be suggested to define the sampling schedule in the phase II study to come. Without this approach, the PK sampling schedule would be made empirically and results would be at best similar to those described here.
Reference: PAGE 14 () Abstr 699 [www.page-meeting.org/?abstract=699]
Poster: poster