Efficiency criteria generated by optimal design tools should be evaluated in the light of study objectives
I Poggesi (1), D Huntjens (1), H Smit (2), H Kimko (1), A Vermeulen (1)
(1) Advanced PK/PD Modeling & Simulation; Clinical Pharmacology and (2) TA Oncology; Clinical Pharmacology. Janssen Research & Development, a Division of Janssen Pharmaceutica NV, Belgium.
Objectives: The definition of an optimal sampling schedule is of particular importance in applying sparse sampling strategies in clinical studies [1]. The objectives of this study were to assess the efficiency of different pharmacokinetic sampling strategies - logistically constrained or optimal - as estimated using WinPOPT [2] and to further assess these designs using population analyses of simulated datasets.
Methods: An hypothetical compound (ka 1 h-1, CL 650 L/h, Vc 4500 L, Q 785 L/h, Vp 12800 L, log-normal inter-individual variability (IIV) and proportional residual variance) was assumed to be given every 48 h. Two designs for the collection of plasma samples were considered: (i) {0.5, 2, 4 h after the first dose and predose, 0.5, 2, 4 h at steady state} and (ii) {1, 6, 12 h after the first dose and predose, 1, 6, 12 h at steady state}. The designs were evaluated using the WinPOPT software [2]. The same program was used for selecting the optimal design. Simulations were performed using NONMEM [3], and the plasma concentrations were extracted at the relevant times and used to re-estimate the population and individual parameters, which were compared with the ‘true' ones.
Results: Based on the WinPOPT-generated efficiency value and considering all parameters, schedule (ii) was approximately 70% more efficient than schedule (i). The optimal design {3, 9, 22 h after the first dose and 0.236, 2.97 (twice), 48 h after the 11th dose} was 236% more efficient than schedule (i). When the optimization was focused on CL and Vc only, schedule (ii) and the optimal design were 30% and 345% more efficient than schedule (i), respectively. Although less efficient in terms of uncertainty of the parameter estimates, the non-optimal designs provided population and individual parameters in reasonable agreement with the true values. Population clearance in particular was estimated with low bias (-6%) also with the least efficient schedule. Bias for Vc was generally higher, but still within ±20%; slightly larger biases were observed for IIV.
Conclusions: The minimization of the uncertainty around parameters can be an aim of the design of a study (e.g., in pediatric PK studies). However, when accurate individual PK parameters have to be used in a sequential PKPD approach, bias should be also considered. The available optimality sampling design tools are useful in exploring the precision given a sampling schedule and proposing schedules to be assessed using simulations.
References:
[1] Mentré F, Duffull S, Gueorguieva I, Hooker A, Leonov S, Ogungbenro K, Retout S. Software for optimal design in population PKPD: a comparison. PAGE 2007
[2] Duffull S, Denman N, Eccleston J, Kimko H. WinPOPT User Guide version 1.2, 2008
[3] Beal S, Sheiner LB, Boeckmann A, Bauer RJ. NONMEM User's Guides. (1989-2009), Icon Development Solutions, Ellicott City, MD, USA, 2009.
