Relevance of the use of population design evaluation and optimisation methods in the context of drug development projects in Roche

Context: Population designs evaluation and optimisation methods have been widely developed in the last few years. Based on the population Fisher information matrix approach [1, 2], for which relevant results have been shown [3, 4], PFIM and PFIMOPT Splus functions have been proposed for population designs evaluation and optimisation respectively [5]. However, the use of those functions in a daily routine for design determination is still very few and demonstration of their usefulness in the context of drug development projects is still required.

Objective: The objective was to explore the relevance of using PFIM and PFIMOPT in the context of drug development projects in Roche. The idea was to appreciate, for a given good level of parameter estimates precision, both the number of samples that could be saved and then the diminution of design cost that Roche could expect by using those tools.

Method: Four projects with completed studies (including population PK analyses) have been selected for a retrospective analysis of their population designs. Those studies occurred at different stages of the drug development (phases II and III) and had different ways of administration, IV or PO. Their population designs were given in two different forms, either by fixed sampling times or by sampling windows; moreover they involved full PK profile samples (most with around 10 samples per patients) sometimes combined with a peak and trough strategy at different occasions.

For each study, the same strategy was used. First a priori knowledge on the PK of the drug was collected from previous studies (phase I) in order to mimic the knowledge of the pharmacologist at the time of the choice of the design. Based on that knowledge, the efficiency of the population design, in term of precision of the parameter estimates, was then evaluated with PFIM. In order to try to improve the efficiency of the design and / or to reduce the number of samples per subject, the optimal sampling times were computed using PFIMOPT. Based on those a “compromised” design taking into account the clinical constraints was derived. The gain of efficiency and the study cost reduction using the compromised design were then appreciated.

Results: Results given by PFIM on the efficiency of each population design were in accordance with the different level of difficulties encountered during the modelling of the data with NONMEM (2 studies over the 4 with SE% > 100% on most of the parameters). Optimisation of those designs has allowed derivation of compromised designs, with reasonable expected precision of parameter estimates, but also less costly, with a reduction of the number of samples per subject up to near 45%.

References:
1. Mentré, F., A. Mallet, and D. Baccar, Biometrika, 1997. 84(2): p. 429-442.
2. Retout, S. and F. Mentré,  Journal of Biopharmaceutical Statistics, 2003. 13(2): p. 209-27.
3.  Retout, S., F. Mentré, and R. Bruno, Statistics in Medicine, 2002. 21(18): p. 2623-39.
4. Green, B. and S.B. Duffull, Journal of Pharmacokinetics and Pharmacodynamics, 2003. 30(2): p. 145-61.
5. Retout, S. and F. Mentré, Journal of Pharmacokinetics and Pharmacodynamics, 2003. 30(6): p. 417-443.