Optimisation of experimental design for drug screening in behavioural models of pain.
A. Taneja(1) , J Nyberg(2) , M Danhof(3) , O Della Pasqua(4) , Pain Project Members of the TI Pharma mechanism-based PKPD modelling platform(5)
(1,3,4)Division of Pharmacology, LACDR, Leiden University, The Netherlands;(2)Department of Pharmaceutical Biosciences,Uppsala University,Uppsala,Sweden;(4) Clinical Pharmacology & Discovery Medicine, GlaxoSmithKline, Stockley Park, UK;(5) Dutch Top Institute Pharma, Netherlands
Objectives: Recently, an optimal design technique was developed for the analysis of discrete variables[1]. We aimed to evaluate the feasibility of applying ED-optimality to screening of compounds taking model and parameter uncertainty into account. We illustrate these concepts using gabapentin as a paradigm compound .
Methods: The analysis consisted of two sequential steps: 1) model building & validation and 2) evaluation of a hypothetical screening experiment. Binary response in the logit space was used for optimisation of sampling times and dose levels under the assumption of known pharmacokinetic profile and expected potency range. Baseline/placebo effect, maximum effect and residual variability were assumed to be independent of treatment type and derived from historical data (n=45). We assumed drug potency (EC50) to be the parameter of interest. Optimisation scenarios were based on feasibility, with limits for sample size, dose levels and sampling times.Gabapentin concentrations were simulated for the selected range of doses and sampling times using a two-compartment pharmacokinetic model (V2=0.18 L, V3=3.8L, Cl=0.03 L/h, Ka=0.6 h-1, Q=78 L/h, F=0.83). The estimated EC50 was 198 ng/ml. Validation of the optimised design included simulation of response and refitting of the data to the same logistic model. For the prospective use of the method, optimisation factors were reassessed by testing a range of EC50 values for a hypothetical compound with similar pharmacokinetic profile. Response profiles based on candidate designs were then simulated and data analysed using a logistic model. In addition to parameter estimates, dose response curves are also presented. A Monte Carlo (MC) integration technique was used to integrate the FIM with Latin hypercube (LH) sampling. POPED 2.10/ MATLAB 7.9 were used for the optimal design. Simulations were performed in NONMEM 6.
Results: Preliminary results show that dose selection is essential for accurate parameter estimation (i.e., low relative standard error) . Doses of 100 mg/kg or higher were identified for gabapentin, with variable sampling times for each dose level. Sampling windows ranged from 0 to 8.8 hrs post-dose. Relative standard errors in parameter estimates remain sensitive to group size, despite repeated sampling.
Conclusions: We show how optimality concepts can be used to assess drug potency in screening experiments using historical priors to support model parameter estimation.
References:
[1] Nyberg J, Karlsson MO, Hooker AC. Proceedings of the 18th meeting of the Population Approach Group in Europe , 23-26th June 2009.St Petersburgh, Russia.
