Rebekka Fendt 1,2, Annika Schneider 1,2, Jan-Frederik Schlender 2, Ute Hofmann 3,4, Elke Schäffeler 3,4, Reinhold Kerb 3, Matthias Schwab 3,5, Lars Kuepfer 1,2
(1) Institute of Applied Microbiology, RWTH Aachen University, Aachen, Germany (2) Systems Pharmacology & Medicine, Bayer AG, Leverkusen, Germany (3) Dr. Margarete Fischer-Bosch-Institute of Clinical Pharmacology, Stuttgart, Germany (4) University of Tuebingen, Tuebingen, Germany (5) Depts. of Clinical Pharmacology and Biochemistry and Pharmacy, University of Tuebingen, Tuebingen, Germany
Objectives: Physiologically based pharmacokinetic model typically rely on parameter values that represent the mean of the study cohort or population. In reality, persons that have the same properties as average individuals hardly exist. The goal of this study was to build personalized physiologically based pharmacokinetic (PBPK) models and evaluate their performance.
Methods: Pharmacokinetic data was obtained in a drug cocktail trial [1]. Healthy volunteers (n=103) received a single subtherapeutic oral dose of caffeine, codeine, midazolam, pravastatin, talinolol and torsemide. Biometric data (height, weight, age, sex) and physiologic data (glomerular filtration rate, liver blood flow) were used for the development of a personalized caffeine PBPK model. The paraxanthine/caffeine plasma ratio at 4h was determined to inform CYP1A2 activity. PBPK models were simulated with PK-Sim, which is part of the open systems pharmacology suite (OSP) [2]. Personalization was carried out in MATLAB with the OSP toolbox.
Results: : Before model personalization, the caffeine PBPK model was adjusted to the mean caffeine plasma concentration of the study cohort. Two metrics were used to evaluate the models, the percentage of the data that fell into the 2fold range and the concordance correlation coefficient (ccc).
The base model described the mean data very well (100% within 2fold range, ccc=0.95) and served as a reference for the evaluation of model personalization. When the base model was compared to the individual caffeine plasma concentrations, 93 % of the data fell into the 2fold range and the ccc was 0.50.
Informing the PBPK model simulations with biometric data (age, height, weight, sex) yielded a similar percentage of data in the 2fold range (94%) and increased the ccc to 0.69. Addition of physiological information (glomerular filtration rate, liver blood flow) did not further improve the model prediction (% in 2fold range 95%, ccc=0.68).
Caffeine metabolism by CYP1A2 has a large influence on plasma PK. Paraxanthine is the predominant caffeine metabolite and the paraxanthine/caffeine ratio showed a good correlation to caffeine clearance (R2=0.87). The enzyme activities in the PBPK model were informed with single point measurements of paraxanthine/caffeine ratio in plasma at 4 hours. The final personalized PBPK models contained biometric, physiologic and CYP1A2 phenotype information. The percentage within the 2fold range increased to 0.96 and the ccc to 0.80.
Conclusions: We systematically analyzed the impact of biometric, physiologic and CYP1A2 phenotype data on personalized caffeine PBPK model simulations. Informing the caffeine PBPK model with of biometric data (height, weight, age, sex) of study individuals slightly improved the performance of model simulations, inclusion of physiologic information (glomerular filtration rate and liver blood flow) did not. CYP1A2 phenotyping by the paraxanthine/caffeine ratio in plasma worked well and improved the model performance. The presented approach contributes to a profound understanding of PK variability which is crucial for optimization of efficacy and safety in clinical pharmacology.
References:
[1] Kuepfer, L., R. Kerb, et al. (2014). “Clinical Translation in the Virtual Liver Network.” CPT Pharmacometrics Syst. Pharmacol. 3(7): e127.
[2] http://www.open-systems-pharmacology.org/
Reference: PAGE () Abstr 9405 [www.page-meeting.org/?abstract=9405]
Poster: Drug/Disease Modelling - Absorption & PBPK