T. Cain(1), K.Feng(1), M. Jamei(1), A. Rostami-Hodjegan(1,2)
(1) SIMCYP Limited (a Certara Company), Sheffield, UK, (2) School of Pharmacy and Pharmaceutical Sciences, the University of Manchester, UK
Objectives: To predict the compliance scenario of Rosiglitazone from a patient's last PK sampling point and to compare the predictive ability when compliance scenarios are assumed equally likely and when they occur randomly.
Methods: We applied the Barriere et al. [1] method to predict patient's compliance when taking a 4 mg daily dose of Rosiglitazone (ROS) for 5 days. This involved the use of a Bayesian approach, where the probability of a compliance scenario given the final observed concentration was calculated. Prior in vitro and physicochemical parameters for ROS and the Healthy Volunteer population of Simcyp (V12 R2) were used to generate plasma concentration profiles of 500 patients where the first three doses were taken and the final two doses varied over five scenarios: full compliance, missing the first dose, missing the second dose, taking both doses together and missing both doses. Two prior distributions were investigated for compliance and their predictive ability determined by comparing the % of true positive/negative cases. First, uniform distributed scenarios and second a distribution of scenarios generated using an algorithm written in an R script.
Results: For the uniform prior, 100 patients were assumed to have each compliance scenario, and for the second prior the numbers of patients randomly assigned each compliance scenario varied between 267 for full compliance and 32 for taking neither dose. The probabilities of a true positive and a true negative using a uniform prior were between 0.41 to 0.95, and between 0.78 and 0.98 respectively. For the randomly generated compliance these probabilities were between 0.65 and 0.99, and between 0.91 and 1 respectively. In both cases the scenarios for full compliance, two missed doses and taking both doses together had the greatest probabilities of a true positive.
Conclusions: The probability of a true positive/negative for each compliance scenario is greater when assuming the randomly generated compliance scenario, which is arguably more representative of the true population. Compliance scenarios where only one dose was taken tended to be the hardest to predict from the final concentration. The concentrations in these two cases were probably more likely to be closer to those from the three other compliance scenarios. These results, which can be expanded to response, demonstrate the value of population PBPK modelling in identifying potential predictive compliance indicators
References:
[1] Barriere O et al., J Pharmacokinet Pharmacodyn. 2011 Jun 1;38:333-351
Reference: PAGE 22 () Abstr 2842 [www.page-meeting.org/?abstract=2842]
Poster: Other Modelling Applications