2011 - Athens - Greece

PAGE 2011: Estimation methods
Karl Brendel

A comparison of MONOLIX, NONMEM 6 and NONMEM 7 based on a simulated PK example

Karl Brendel, Sophie Peigné, and Marylore Chenel

Department of Clinical Pharmacokinetics, Institut de Recherches Internationales Servier

Introduction: From decades, NONMEM is the gold standard software for NLME modelling. However, several alternative software, like MONOLIX (using SAEM algorithm), become more and more appealing. Regarding the algorithms, NONMEM 7 includes now new methods, like ITS, IMP and SAEM. Moreover the existing estimation methods in previous NONMEM versions, like the FOCE method, were improved in NONMEM 7 by decreasing abnormal termination of estimation.
A comparison between MONOLIX 3.1, NONMEM 7.1 and NONMEM 6.2 was performed on a simulated population PK example. 

Methodology: A joint population PK model for the parent drug and its metabolite was previously built using NONMEM 6 (FOCEI) on simulated plasma concentration-time data[1].
The reference model, called NM6_FOCEI, was implemented in NONMEM 7 and in MONOLIX, and parameters were estimated with both FOCEI and SAEM methods in NONMEM 7 (called NM7_FOCEI and NM7_SAEM) and with SAEM in MONOLIX (called MX_SAEM). At last, simulations were performed for each different software/method, and 90% prediction interval (PI) and the median were plotted. Parameter estimates were compared to the reference model NM6_FOCEI using Wald tests (p<0.05). Moreover, run time duration was compared and a graphical comparison was performed for simulations in superposing 90% PI and median profiles.

Results: No differences were observed between parameter estimates obtained with NM7_FOCEI, NM7_SAEM and NM6_FOCEI. However statistical significant differences were found with MX_SAEM compared to NM6_FOCEI for several parameters. The range of run time duration was very large: 2h for MX_SAEM, 4h30 for NM7_FOCEI, 7h for NM6_FOCEI and 4 days with NM7_SAEM (1 day for SAEM only and 3 days for the IMP step).Regarding simulations performed with the different models, no differences were observed in term of simulation as P5, median and P95 were all superimposed to the ones obtained with the reference model NM6_FOCEI.

Conclusion: One of the major improvements of NONMEM 7 is the implementation of new algorithms like SAEM. However the run time seems to be very time long compared to MONOLIX. With NONMEM 7, parameters estimation seems to be very sensitive to the initial parameter values contrary to MONOLIX. The FOCE method seems to be more efficient regarding run time consumption in NONMEM 7 compared to NONMEM 6. This comparison will be further extended to other types of models.  

[1] PAGE 19 (2010) Abstr 1695 [www.page-meeting.org/default.asp?abstract=1695]

Reference: PAGE 20 (2011) Abstr 2043 [www.page-meeting.org/?abstract=2043]
Poster: Estimation methods