CS Ernest II (1,2), J Nyberg(1), MO Karlsson (1), AC Hooker(1)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Sweden; (2) Eli Lilly and Company, Indianapolis, IN, USA.
Objectives: Optimal design (OD) for discrete-type responses have been derived using generalized linear mixed models [1] and simulation based methods for computing the Fisher information matrix (FIM) of nonlinear mixed effect models (NLMEM) [2]. In this work, OD using a closed form approximation of the FIM for dichotomous, non-homogeneous, Markov-chain sleep NLMEM was investigated.
Methods: NLMEM OD was performed by determining the FIM for each Markov component (previously awake, FIM1, or previously asleep, FIM0) and weighted to the average probability of response being awake, p(1), or asleep, (1-p(1)), over the night to derive the total FIM (FIMtotal) using PopED [3]. FIMtotal,i for a design group i was computed as the sum of the two FIMs: FIMtotal,i=pi(1)*FIM1,i+(1-pi(1))*FIM0,i. The NLMEM consisted of transition probabilities (TP) of dichotomous sleep data estimated as logistic functions. Dose effects were added to the TP to modify the probability of being in either sleep stage. The reference designs (RD) were placebo, 1-, 6-, and 10-mg dosing for a 1- to 4-way crossover study in 4 dosing groups. Optimized design variables (ODV) were dosage and number of subjects in each dosage group and a D-optimal design criterion was used. The designs were validated using stochastic simulation/re-estimation (SSE).
Results: The predicted parameter estimates obtained via the FIM were less precise than parameter estimates computed by SSE; likely due to the weighting scheme, small contribution of the awake Markov component, or lack of correlation between population means and variances in the FIM as opposed to SSE. The ODV improved the precision of parameter estimates leading to more efficient designs compared to the RD. The increased efficiency was more pronounced for SSE than predicted via FIM optimization.
Conclusion: Using an approximate analytic solution of parts of the FIM (FIM1, FIM0), the FIMtotal could be calculate without extensive Monte Carlo simulations. The optimized designs provided increased efficiency for the crossover study designs examined and provided more robust parameter estimation.
References:
[1] Ogungbenro K, et al. Population Fisher information matrix and optimal design of discrete data responses in population pharmacodynamic experiments. J Pharmacokinet Pharmacodyn. 2011 Aug;38(4):449-69.
[2] Nyberg J, et al. Population optimal design with correlation using Markov Models. PAGE 20 (2011) Abstr 2233
[3] PopED, version 2.13 (2012) http://poped.sf.net/.
Reference: PAGE 22 () Abstr 2790 [www.page-meeting.org/?abstract=2790]
Poster: Study Design