A. Dubois, E. Leroux, M. Chenel
Clinical Pharmacokinetic Department, Institut de Recherches Internationales Servier
Context: The high rate of failure observed in clinical trials evaluating antidepressant agents is mainly explained by a marked placebo response and a high rate of dropout. There is no universally applicable method of handling such missing data. Traditionally, they were handled by carrying forward the last observation but this method could result in biased estimates of treatment efficacy. Most recently, the joint nonlinear mixed effects modelling (NLMEM) of longitudinal and time-to-event (TTE) data was developed [1-4]. Our objective was to describe the time course of the Hamilton depression rating scale score (HAMD) under placebo for patients suffering from a major depressive disorder, using joint NLMEM approach.
Methods: Placebo data from a 8-week (with an optional 16-week extension) phase II trial were used. HAMD was measured before the start of placebo treatment, at week 2, 4, 6, 8, and optionally at week 12, 16, 20 and 24. We tested several structural models to describe the HAMD time course using NLMEM [5-6]. We also tested different random effect models for the inter-individual and residual variability. Then, we added TTE data to HAMD data in the NLMEM, using a parametric TTE model to describe them. We tested different dropout mechanisms (completely random, random and informative [3]). As adding dropout into the model may influence the HAMD parameter estimates, all parameters (for HAMD and TTE data) were estimated for joint NLMEM. Furthermore, there were 3 types of TTE data (right-censored, exact or interval time event); we distinguished them during the estimation.
Results: The inverse of the Bateman function better described the HAMD data [6]. However, as expected, the analysis of the longitudinal data alone did not allow us to correctly predict HAMD at population level. Taking into account dropout data to develop a joint model was essential to improve the prediction. The final TTE model was an exponential model which considered the last observation of the HAMD score as a covariate. A specific covariate effect was also added to take into account the study design as only well-improved patients could continue after the 8th week
Conclusion: The results of the joint modelling showed that taking into account the TTE data improved the model capacity to predict HAMD in patients under placebo. This method could also be applied to clinical data from patients under active treatment and, hence, help to better distinguish the treatment effect from the placebo one [1].
References
[1] Friberg LE., de Greef R., Kerbusch T., Karlsson MO. Modeling and simulation of the time course of Asenapine exposure response and dropout patterns in acute schizophrenia. Clinical Pharmacology and Therapeutics. 2009, 86: 84-91.
[2] Gomeni R., Lavergne A., Merlo-Pich E. Modelling placebo response in depression trials using a longitudinal model with informative dropout. European Journal of Pharmaceutical Sciences. 2009, 36: 4-10.
[3] Hu C., Sale M.E. A joint model for nonlinear longitudinal data with informative dropout. Journal of Pharmacokinetics and Pharmacodynamics. 2003, 30: 83-103.
[4] Holford N, Lavielle M (2011) A tutorial on time to event analysis for mixed effect modellers. Abstr. 2281 www.page-meeting.org/?abstract=2281.
[5] Gomeni R., Merlo-Pich E. Bayesian modelling and ROC analysis to predict placebo responders using clinical score measured in the initial weeks of treatment in depression trials. British Journal of Clinical Pharmacology. 2007, 63: 595-613.
[6] Holford N, Li J, Benincosa L, Birath M (2002) Population disease progress models for the time course of HAMD score in depressed patients receiving placebo in anti-depressant clinical trials. In: Abstracts of the XI annual meeting of the population approach group in Europe. Abstr. 311 www.page-meeting.org/?abstract=311.
Reference: PAGE 21 (2012) Abstr 2445 [www.page-meeting.org/?abstract=2445]
Poster: Other Drug/Disease Modelling