Pharmacokinetic/Pharmacodynamic Modelling of Recurrent Adverse Events as Function of Drug Concentration and Time
Clémence Rigaux (1), Bernard Sébastien (1), Britta Goebel (2)
(1) SANOFI, Paris, France (2) SANOFI, Frankfurt, Germany
Objectives: To develop a PK/PD model for recurrent adverse events (AE) and perform simulations to predict the incidence of AE for a new design of an upcoming study.
Methods: : PK and AE data were obtained from two Phase 1 studies. A PK model was used to fit the plasma drug concentrations and to predict individual daily Cmax concentrations, using non-linear mixed-effects modelling implemented in Monolix (version 2016R1) [1]. The number of AE per day and per patient was modelled using non-linear mixed-effects modelling, using SAEM algorithm, exploring the relations between AE emergence and drug concentration, drug-dependent or independent time effect and demographic data. Then, the model performance was evaluated by comparing observed and predicted statistics of interest using clinical trial simulation, and the PK/PD model was used to simulate new study designs.
Results: The number of AE per day and per patient was best modeled using a Poisson regression, using a log-linear drug effect consisting of a constant drug effect and an attenuated exponential drug effect describing the improvement of drug tolerability with time. This corresponds to a reduction of the temporary drug effect by 1/3 after 5 days of treatment. Inter-individual-variability on the basal risk was used to reflect the variability of risk between patients, and no demographic covariate was identified as clearly significant. Model evaluation showed that the predicted number of events was in good agreement with the observations, despite a slight tendency to over-prediction. The simulation of new study designs predicted a 3 times higher number of AE for the high dose compared to the low dose.
Conclusions: PK/PD modelling and simulation of recurrent AE could be implemented to support dose and design selection for a new clinical study. The model described the observations well and will be updated with new data to improve its robustness.
References:
[1] MonolixSuite 2016R1 Data set guide. http://lixoft.com/.