PDF of poster

Objectives: Simulations are more and more performed in order to evaluate the impact of variability and to compare different trial designs as well as to challenge model assumptions and to validate models. Regarding simulation, NM presents two major drawbacks : NM input datasets are tricky to build, and regarding outputs, NM does not provide statistics and another software is necessary.

Trial Simulator 2 (TS2) is a software dedicated to simulation of clinical trials (parallel, cross-over and latin-square), outputs can be directly evaluated by graphs and statistical analyses using a special release of S-PLUS 6.2 Professional. We propose a graphical comparison to check if the simulations realized by TS2 and NM are equivalent. We performed this comparison on two different types of models.

Methods: The first model is a classical two-compartment one with a first order absorption. The second one is a drug-drug interaction drug-metabolite semi-physiological model including nine compartments, in which the two parts (drug and metabolite) are linked by the liver compartments. Each model has been built with NM using a combined analysis of single dose clinical studies (five and seven studies respectively). Both models do not take into account any potential covariate effects.

For the first model, a group of 12 subjects with 11 sampling times is simulated thousand times with NM and TS2. Regarding the second one, the original dataset is composed of 47 subjects receiving a single dose of the evaluated drug and 36 subjects receiving simultaneously a single dose of ketoconazole and the evaluated drug. It is simulated thousand times using NM and TS2. Real times were kept in the NM dataset. In order to assess intra-software relative difference, simulations are performed twice (2 ´ 1000 replicates) with each software and the median of the two simulations are compared at two different times chosen empirically corresponding to peak and trough of the concentrations. Software comparison is realised representing on the same graph median, P5 and P95 of the data simulated with both software.

Results: Different problems occurred during simulation steps. The main problems were the difficulties in TS2 to correctly implement complex models (drug-drug interaction) and also the need of a code to truncate etas distributions in NM. Indeed, extreme values of random effects making NM simulation step aborted for a large number of replicates.

Intra-software relative difference was less than 5% for each software. Regarding intesoftware comparison, simulated curves were well superposed.

Conclusion: After solving problems regarding implementation (especially for complex models in TS2), software simulations are judged graphically equivalent. Structural model and error model can be checked simulating without and with variability. Once the model implemented in TS2, it can be used to simulate other trials design or reference groups.

References :
1. Beal SL, Sheiner LB. NONMEM users guides. San Francisco : NONMEM Project Group, University of California; 1992.
2. Pharsight Trial Simulator user’s guide version 2.2. Pharsight Corporation.