2007 - KÝbenhavn - Denmark

PAGE 2007: Methodology- Design
Marylore Chenel

Comparison of uniresponse and multiresponse approaches of PopDes to optimize sampling times for drug-drug interaction studies: application to a Servier compound.

Marylore Chenel1, Kayode Ogungbenro2, FranÁois Bouzom1

1Servier Research Group, France; 2School of Pharmacy, CAPkR, University of Manchester, Manchester, UK

Objectives: In addition to efficacy information, the collection of pharmacokinetic (PK) data in patients during phase II/III trials presents the interest to evaluate any potential risk of drug-drug interaction (DDI) in linking drug exposures to adverse event occurrences. As the number of blood samples is most often limited in patients, optimal design with joint sampling times for at least 2 compounds could be very useful in the collection of PK information. Since multiresponse approach as implemented in PopDes allows optimization of sampling times for two responses, uniresponse and multiresponse approaches were compared to optimize sampling times for SX drug, a potential weak inhibitor of CYP3A4, and midazolam (MDZ), as reference substrate of CYP3A4.

Methods: The anticipated study design included 12 subjects receiving SX dose twice a day over 5 days, and in the morning of the 5th day, a single dose of MDZ was given 2 hours after the first daily dose of SX. This study design was simulated for 100 subjects using a PBPK (Physiologically Based PK) model for both compounds built with in vitro data. The PBPK model allows prediction of plasma concentration-time profiles for both compounds taking into account the potential interaction. Using the PBPK simulated data, SX and MDZ population PK models were built in NONMEM with the FOCEI method. Then, sampling times were optimized by uniresponse and multiresponse approaches in PopDes. Under the uniresponse approach, the design domain was over 12 hours (one dose interval), and over 22 hours for SX and MDZ, respectively. Under the multiresponse approach, the design domain was over 22 hours for both compounds. In all cases, a single group of 12 subjects was considered. Both approaches were compared after simulation and re-estimation of 1000 datasets in terms of estimation accuracy, relative errors (bias) and RMSE (Root Mean Squared Error) of apparent clearance (CL/F) for the two compounds.

Results: SX simulated data were fitted with a 2-compartment model with a fixed first-order absorption rate constant. Inter-individual variability was added on CL/F and on the apparent volume of distribution of the central compartment (Vc/F) with a correlation between these two parameters. The residual error was a combination of additive and multiplicative errors. The MDZ population PK model was similar except that a zero-order constant was used to describe absorption process and inter-individual variability was added on all fixed effect parameters. Under the uniresponse approach, best designs were obtained with 4 sampling times over 1-dose interval and with 5 sampling times over 22 hours for SX and MDZ, respectively. Thus, there was a total of 9 different sampling times with this approach. Under the multiresponse approach, the best design with joint optimal sampling times for both drugs was obtained with 5 sampling times over 22 hours. Whatever the approach used, in addition of the SX PK information collected, MDZ CL/F values were well estimated as empirical RSEs (Relative Standard Error), median RSEs given by NONMEM as well as RSEs given by the population Fisher information matrix were less than 20%. Estimated MDZ CL/F values were in average equal to the reference value and relative RMSEs were less than 15% in all cases.

Conclusions: Under the clinical constraints for these two population PK models, the multiresponse approach with joint optimal sampling times allowed MDZ CL/F values to be well estimated in addition of PK information collection for the SX drug, and allowed to save 4 sampling times compared to the uniresponse approach. Thus, for the clinical trial, the optimal sampling times estimated by both approaches were slided into the full anticipated sampling time design.

When the clinical trial is completed, MDZ CL/F estimated by population PK modelling using the full anticipated sampling time design will be compared to those obtained with the optimal sampling time designs estimated by uniresponse and multiresponse approaches. The interaction evaluation will be also evaluated by population PK modelling using full and optimal designs.




Reference: PAGE 16 (2007) Abstr 1157 [www.page-meeting.org/?abstract=1157]
Poster: Methodology- Design
Click to open PDF poster/presentation (click to open)
Top