Stochastic Simulations Assist to Improve a Poorly Designed Clinical Study for the CSF Pharmacokinetics of Doripenem
Nalda-Molina, Ricardo (1), Dokoumetzidis, Aristides (2), Valsami, Georgia (2)
(1) Pharmacy and Pharmaceutical Technology Department, Faculty of Pharmacy, University of Miguel Hernandez de Elche (Spain) (2) Laboratory of Biopharmaceutics & Pharmacokinetics, Faculty of Pharmacy, National & Kapodistrian University of Athens, Greece
Objectives: To assess the impact of the addition of new subjects, with improved sampling design to the results of a previously performed NONMEM analysis of doripenem in the CSF with sampling design problems, by performing stochastic simulations including uncertainty.
Methods: A previous study (1) investigated the CSF pharmacokinetics of doripenem after IV administration to patients, using NONMEM 7.2. Briefly, the model consists of a standard two compartment model (the parameters of which were taken and fixed from the literature), plus a third distribution compartment corresponding to CSF, parameterized in terms of distribution rate constant (Kcsf) and partition coefficient (PC) which were estimated from CSF data. However, a limited sampling design including mostly one sample per subject and the lack of later samples (after 6 hours) led to a high uncertainty in the parameter estimation. Stochastic simulations of three scenarios were performed, corresponding to the 25th, 50th and 75th percentiles of the uncertainty parameter distribution obtained from the bootstrap analysis of the previous NONMEM results. Simulations and estimation of the dataset (36 patients and 47 CSF samples) with 5, 15, 30 and 60 extra patients, with one sample at 12 hours, for each of the three scenarios, were performed with NONMEM 7.2. Guided by the results of the simulations, two new patients were included in the study, with late sampling times, and a final population pharmacokinetic analysis with the full dataset was performed.
Results: The simulations showed a significant reduction in the standard error of the model parameters estimates, even with only five extra-subjects. Also, it is worth to note that the standard error reduction was more pronounced in the 75th percentile scenario, rather than the 50th percentile as would have expected. Based on the simulations we decided to include new patients in the study. However, the number of new patients we were able to include at the end was only two because of a limited time frame. The final analysis of the full dataset (2) showed a reduction of the estimates standard error (by 75% approximately). Moreover, the new parameter estimates were close to the 75th percentile scenario rather than the 50th, as the results of the simulations suggested.
Conclusions: Stochastic simulations are useful to improve the clinical study design, and inform about the impact of new subjects, and new sampling times, even when the original estimates are biased.
References:
[1] Charkoftaki G. et al. Pharmacokinetics of doripenem in cerebrospinal fluid. AAPS Annual Meeting 2011, Washington DC.
[2] Nalda-Molina R., et al. Pharmacokinetics of Doripenem in Cerebrospinal Fluid of patients with non-inflamed meninges. J. Antimicrob Chemother (Accepted)
