Optimising PK sampling under the constraint imposed in later phase clinical trials

Patrick Johnson (1), Byron Jones (1), Barbara Bogacka (2), Oleg Volkov (2)

(1) Pfizer Ltd, Sandwich, U.K.; (2) School of Mathematical Sciences, Queen Mary, University of London, U.K

Objectives: When collecting pharmacokinetic (PK) data in a clinical trial an important design factor is the number and times when blood samples are taken. Given prior information on the expected drug concentration-time relationship, theory for computing D-optimal designs is available and gives sampling times that maximise the collected PK information. The usual assumption is that subjects have blood samples taken at the same fixed times post dose. However, fixed sampling times for each subject rarely happen in later phase clinical trials (Phases 2 & 3) so designs obtained using mathematical theory may not be optimal given such a constraint. The purpose of this presentation is to describe a simulation method that aims to maximise the PK information collected given the practical constraints of later phase clinical trials.

Typical characteristics of later phase clinical trials are that the collection of PK is not the primary objective, blood sampling is sparse (e.g., two sampling times during two clinic visits per subject), there are restraints on when sampling can be done and there is a lack of patient compliance. Such characteristics introduce uncertainty in the sampling times. For example, within a clinic visit the first sampling time may occur randomly within a protocol-specified visit widow (e.g. dose taken between 1 and 2 hours before clinic visit) and the second sampling time is conditional on the first plus some additional clinic assessment time. Consequently, all subjects potentially have different sampling times, but hopefully constrained within specified visit windows. Progress is being made but standard optimal designs obtained via theory fail to incorporate this random component in sampling times and may be sub-optimal from a practical viewpoint.

Methods: We describe the use of simulation to account for the randomness in sampling times. As an illustrative example, a first-order absorption one-compartment PK model was assumed. A grid of sampling windows was created, derived from possible visit windows that could be realistically specified in a study protocol. For each possible sampling window combination, random sampling times for n subjects were simulated (2 samples x 2 visits). The data derived were used to fit the PK model and the efficiency of each sampling window combination estimated. The highest efficiency, having lowest standard error, was used to determine the optimum sampling window combination. This optimum sampling window was compared to a D-optimal design (PFIM [1]) and the loss of efficiency due to the randomness in sampling times was estimated.

Conclusion: Advantages of the simulation method over theoretical designs will be discussed but the former does represent a cost in terms of time and resource. It is recommended that a combination of both optimal design theory and simulation is the best compromise as it will lead to both a faster and more applicable design solution.

References:
[1] Retout & Mentré (2003). Optimisation of individual and population designs using Splus. J. Pharmacokinet. Pharmacodyn., 30(6): 417-443.

Reference: PAGE 14 () Abstr 736 [www.page-meeting.org/?abstract=736]

Poster: poster