Y. Merlé, E. Schmautz, and A Mallet.
INSERM U436, SIM, 91 BD DE L'HOPITAL 75634 PARIS CEDEX 13 FRANCE
Population approaches are appealing methods for detecting drug-drug interactions in case of sparse data, however they sometime fail to reveal interactions expected on theoretical considerations. Our purpose was to evaluate, on simulated data, the ability of two strategies, conducted in the context of the NPML approach, to detect a pharmacodynamic interaction between two drugs (A and B) for various combinations of designs (number of measurements by subject and number of subjects) and levels of pharmacodynamic interindividual variability. The effect of A was supposed to be described by an Emax model. The drug B was assumed to exhibit no effect by itself but the EC50 parameter of the Emax model describing the effect of A was related to the concentration of B by a simple second stage model. The concentration of B was included as a covariate in the NPML analysis which was performed for every set of simulated data. The interaction was evaluated either by computing the relative expected decrease of variance of the estimated parameter distribution conditionally to the covariate or by a graphical analysis (i.e. plot of the means of each PD parameters conditionally to various levels of the concentration ofB). A relative decrease of the expected variance was found for every studied combination but no clear influence of the design and of the interindividual variability was found on the magnitude of this decrease. In contrast, a relationship between EC50 and the concentration of B was more often found by graphical analysis when the number of subjects and the number of samples by subject increased and the PD interindividual variability decreased. Our study illustrates the influence of the design and of the pharmacodynamic variability and shows that graphical analysis is an useful tool for detecting and characterize interactions.
Reference: PAGE 7 (1998) Abstr 290 [www.page-meeting.org/?abstract=290]
Poster: poster