Evaluation of the PK and PK-PD libraries of MONOLIX: A comparison with NONMEM
C. Laveille (1), M. Lavielle (2,3), K. Chatel (4), P. Jacqmin (1)
(1) Exprimo NV, Mechelen, Belgium. (2) INRIA Futurs, Paris, France; (3) University Paris 5, Paris, France. (4) INRIA, Saclay, France
Introduction: The statistical model for most population PK/PD analyses is the nonlinear-mixed effects model (NLMEM). As opposed to linear models, there are statistical issues to express the optimization criteria for these nonlinear models so that first approximation methods (FO and FOCE) based on linearization of the model were proposed. It is well known that these methods have several methodological and theoretical drawbacks.The SAEM (Stochastic Approximation EM) algorithm avoids any linearization and is based on recent statistical algorithms. This algorithm is a powerful tool for Maximum Likelihood Estimation (MLE) for very general incomplete data models. The convergence of this algorithm to the MLE and its good statistical properties have been proven. The SAEM algorithm is implemented in the freely available MONOLIX software that can be downloaded at http://www.monolix.org.
A new version of MONOLIX will be released soon and therefore an exhaustive evaluation should be performed. At a first stage, the PK and PK-PD library were extensively evaluated. The other features of MONOLIX are also in the process of being reviewed and will be presented later on.
Methods:
- To extensively test all the available models in the PK and PK-PD library of MONOLIX
- To perform this evaluation with NONMEM V and NONMEM VI and the latest version of MONOLIX
- To discuss and compare the results.
Results:
- All available models within the PK and PK-PD library of MONOLIX were evaluated within either NONMEM version V and VI but also with the latest version of MONOLIX.
- Trial Simulator version 2.1.2 was used to simulate all the datasets. For each data set, single dose and multiple doses were simulated. The single dose was administered at time 0 and observations were made up to 72h. Then multiple doses started at time 72h and a dose was given every day during 7 days. Trough measurements were made every day. A full profile was drawn at time 144h and after the last administration at time 216h. Overall 120 subjects were simulated for each dataset and allocated to one of the four dose group.
- For the PK and PK-PD Library, all the random effects were simulated with a log-normal distribution. In most case, the inter-individual variability was set to 30% whatever the parameter and residual errors of 20% were used.
- For NONMEM, as first trial, the First-Order Conditional Estimate (FOCE) method with Interaction was used after logarithm transformation. In case of convergence issues with the FOCE method, the First-Order (FO) method was used.
- For the PK-PD models, both PK and PD were fitted simultaneously either in NONMEM or MONOLIX.
Conclusions: Preliminary results showed encouraging results for MONOLIX with all runs successful. The covariance matrix was obtained in 100% of the cases. As expected not all runs were successful with NONMEM using FOCE with Interaction. Whatever the method implemented within NONMEM, run times were significantly reduced when using MONOLIX.