Vincent CROIXMARIE (1), Donato TEUTONICO (1) and Marylore CHENEL (1)
(1) Clinical Pharmacokinetics and Pharmacometrics division, Servier, France
Objectives: Specific studies with renal impaired (RI) patients are usually carried out for drug exposure assessment in this special population. Patient recruitment for this class of patients is difficult, especially for severe RI patients. In this exercise a model based approach was used to simulate such patients. Available data for an anti-cancer drug in normal, mild and moderate RI patients were used to simulate severe RI patients, and an available population PK model was used to simulate the corresponding exposure of these patients. The simulated exposures were then compared to the observed data.
Methods: The covariates included in the population PK model were simulated using a multivariate normal distribution (body surface area (BSA), CRCL and serum albuminuria (ALB)). Additional relevant demographic covariates were also simulated (age and gender), in order to include also these covariates in the correlation. Continuous and categorical covariates were simulated as previously described [1], but in this case the covariate simulation was also extrapolated outside of the range of the available covariate values. The empirical distribution available was used to define a continuous multivariate normal distribution (MVND) which was used to simulate the virtual patients. Ranges of the real covariates were then used as boundaries to filter realistic sets of covariates with the exception of Creatinine Clearance (CrCl) with the objective to extend the simulation range to severe patients. For CrCl, the upper limit was set to the value of the real population, while for the lower limit, it was set to the lower limit for severe RI patients (15 ml/min/1.73m²). The available population PK model was then used to simulate the exposure of these patients. The exposures of virtual patients were then compared with the real exposure of real patients. For the severe RI patients, only 4 patients were available.
Results: The exposures of normal, mild and moderate RI “simulated” patients are in agreement with the exposures obtained in available previous studies. The exposures of severe RI “simulated” patients were compared with the limited data available (4 patients) and their exposure was in agreement with these measurements. These simulations provided as additional information an estimation of the expected variability in exposure, confirming that the dose adjustment suggested for the severe RI patients should be maintained. The covariate simulation also unveiled a relationship between BSA and renal impairment which was present in the available data but which was not in agreement with the observed severe patient data. Additional investigation will be conducted in order to clarify the origin of such correlation, which can generate a bias in the simulation.
Conclusions: In this work, a distribution based method was used to simulate virtual patients for which data were not available and difficult to obtain. A prediction in this fragile population was then possible confirming the proposed dose adjustment. Particular attention needs to be addressed when covariate correlations are used to extrapolate covariate relationship outside the observed range, since this may generate bias in the simulation.
References:
[1] Teutonico, D., Musuamba, F., Maas, H.J. et al. Pharm Res (2015) 32: 3228
Reference: PAGE 27 (2018) Abstr 8682 [www.page-meeting.org/?abstract=8682]
Poster: Methodology - Covariate/Variability Models