Giulia Lestini1, Cyrielle Dumont2, France Mentré2. 1. IAME, UMR 1137, INSERM, Université Paris Diderot, Sorbonne Paris Cité, F-75018, Paris, France. giulia.lestini@inserm.fr. 2. IAME, UMR 1137, INSERM, Université Paris Diderot, Sorbonne Paris Cité, F-75018, Paris, France.
Abstract
PURPOSE: In this study we aimed to evaluate adaptive designs (ADs) by clinical trial simulation for a pharmacokinetic-pharmacodynamic model in oncology and to compare them with one-stage designs, i.e., when no adaptation is performed, using wrong prior parameters. METHODS: We evaluated two one-stage designs, ξ0 and ξ*, optimised for prior and true population parameters, Ψ0 and Ψ*, and several ADs (two-, three- and five-stage). All designs had 50 patients. For ADs, the first cohort design was ξ0. The next cohort design was optimised using prior information updated from the previous cohort. Optimal design was based on the determinant of the Fisher information matrix using PFIM. Design evaluation was performed by clinical trial simulations using data simulated from Ψ*. RESULTS: Estimation results of two-stage ADs and ξ * were close and much better than those obtained with ξ 0. The balanced two-stage AD performed better than two-stage ADs with different cohort sizes. Three- and five-stage ADs were better than two-stage with small first cohort, but not better than the balanced two-stage design. CONCLUSIONS: Two-stage ADs are useful when prior parameters are unreliable. In case of small first cohort, more adaptations are needed but these designs are complex to implement.
PURPOSE: In this study we aimed to evaluate adaptive designs (ADs) by clinical trial simulation for a pharmacokinetic-pharmacodynamic model in oncology and to compare them with one-stage designs, i.e., when no adaptation is performed, using wrong prior parameters. METHODS: We evaluated two one-stage designs, ξ0 and ξ*, optimised for prior and true population parameters, Ψ0 and Ψ*, and several ADs (two-, three- and five-stage). All designs had 50 patients. For ADs, the first cohort design was ξ0. The next cohort design was optimised using prior information updated from the previous cohort. Optimal design was based on the determinant of the Fisher information matrix using PFIM. Design evaluation was performed by clinical trial simulations using data simulated from Ψ*. RESULTS: Estimation results of two-stage ADs and ξ * were close and much better than those obtained with ξ 0. The balanced two-stage AD performed better than two-stage ADs with different cohort sizes. Three- and five-stage ADs were better than two-stage with small first cohort, but not better than the balanced two-stage design. CONCLUSIONS: Two-stage ADs are useful when prior parameters are unreliable. In case of small first cohort, more adaptations are needed but these designs are complex to implement.
Authors: Ivelina Gueorguieva; Ann L Cleverly; Anja Stauber; N Sada Pillay; Jordi A Rodon; Colin P Miles; Jonathan M Yingling; Michael M Lahn Journal: Br J Clin Pharmacol Date: 2014-05 Impact factor: 4.335
Authors: F Mentré; M Chenel; E Comets; J Grevel; A Hooker; M O Karlsson; M Lavielle; I Gueorguieva Journal: CPT Pharmacometrics Syst Pharmacol Date: 2013-06-05