Exposure-response analysis of certolizumab pegol in Crohn’s disease population
Maria C. Kjellsson(1), Kristin E. Karlsson(1), Andrew Hooker(1), Brigitte Lacroix(2), Maria Laura Sargentini-Maier(2), Armel Stockis(2), Mats O. Karlsson(1)
(1)Div. of Pharmacokinetics & Drug Therapy, Dept. of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden; (2)Pharmacometrics, Global Exploratory Development, UCB Pharma S.A., Braine-l’Alleud, Belgium
Objectives: The objectives of this population analysis were to: 1) characterize the placebo and drug effect of certolizumab pegol (CZP) in the Crohn's disease (CD) population using data pooled from 3 clinical trials, 2) investigate the presence of exposure-response in different subsets of the data, 3) simulate new study designs using the population model.
Methods: The CDAI (Crohn's Disease Activity Index) data from 1597 subjects were used for non-linear mixed effects modeling using NONMEM (FOCE INTER estimation method). Modeling was based on data from a dose-ranging trial (study 1) and two confirmatory trials (studies 2 and 3). Placebo or CZP was administered subcutaneously at doses of 100 mg (only study 1), 200 mg (only study 1), and 400 mg (all studies) every 4 weeks, with a 400 mg induction dose at week 0, 2, 4 (only studies 2 and 3), for a duration of 8 weeks (Study 1) or 24 weeks (Studies 2 and 3).
Results: An infusion-like model varying with time and an Emax model varying with individual predicted concentrations from the PK model of certolizumab were used for describing the placebo effect and drug effect of CZP, respectively. Two populations were identified: a drug-sensitive population and a non-drug sensitive population. The maximum effect of CZP was fixed to zero for the non-drug sensitive population and the proportion of drug-sensitive subjects was estimated. Differences between the studies were found in baseline CDAI, maximum placebo effect and proportion of drug-sensitive subjects. These parameters were all estimated to lower values for study 2, intermediate for study 1 and higher for study 3. Exposure-response was shown for a number of different subsets of the data, models and measurements of exposure. Individual predicted concentrations from the PK model of certolizumab were shown to be the best measurement of exposure in the majority of the subsets of data. Model evaluation showed that the results of study 2 were more difficult to recreate even though study differences were taken into consideration. The difference between the simulated data and the observed data of study 2 could be due to ignoring study differences in the variance parameters, which was not investigated. Simulations of alternative study designs were made showing limited difference in moving readout to week 4 instead of week 6 and no relevant gain in starting the study with a loading dose of 800 mg.
Conclusions: Exposure-response was robustly shown under a variety of settings for all three studies.