Integrated model for clinical response and dropout in depression trials: a state-space approach
A. Russu (1), E. Marostica (1), G. De Nicolao (1), A.C. Hooker (2), I. Poggesi (3), R. Gomeni (3), S. Zamuner (3)
(1) Department of Computer Engineering and Systems Science, University of Pavia, Pavia, Italy; (2) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden; (3) Clinical Pharmacology / Modelling & Simulation, GlaxoSmithKline, Verona, Italy
Objectives: GSK372475 is an equipotent reuptake inhibitor of serotonin, norepinephrine and dopamine neuronal reuptake and has been investigated as a potential treatment of major depressive disorder (MDD). In traditional modelling approaches in MDD, efficacy and dropout are rarely integrated. Using state-space models the observed depression scales (HAMD-17) can be modelled as a function of variables (states) describing the status of a patient; one or more of these states (rather than the clinical score alone) can be used for describing the dropout process, allowing a more natural integration of the study observations. In the present work, we develop a joint clinical response and dropout model for GSK372475 using a state-space approach.
Methods: A double-blind, randomized, placebo controlled, flexible dose trial was analyzed using a longitudinal model for depression scores.1 The model was expressed in algebraic equations and re-formulated as a state-space model. Flexible dose scheme was implemented as a covariate of the structural parameters. Dropout data were analysed using a parametric time to event model (Weibull hazard function)2. Completely Random Dropout (CRD), Random Dropout (RD) and Informative Dropout (ID) mechanisms were investigated3. Analyses were implemented in WinBUGS. Performances were evaluated by comparing residuals, posterior distributions of individual parameters, and the Deviance Information Criterion4 (DIC). The goodness-of-fit to dropout data was checked through the modified Cox-Snell residuals5 and by visually comparing the estimated survival curve to the usual Kaplan-Meier estimate.6
Results: Modelling the flexible dosing schedule as a covariate substantially improved the model performance in terms of goodness-of-fit and DIC. In the placebo arm, the joint analysis of DIC and residuals showed better performances of RD and ID mechanisms compared to CRD. In the treatment arm, inspection of residuals pointed out misspecification of the hazard model, suggesting that additional covariates (e.g. related to safety/tolerability) should be considered in the model development.
Conclusions: The proposed state-space approach was shown to be a valuable option to account for time-to-event data (i.e. dropouts) and discontinuities such as flexible doses. Dropout mechanism needs to be properly accounted for, together with its relationship with efficacy and/or safety. Interpretation of residual plots provided valuable suggestions on how to modify the hazard model to better describe the dropout pattern.
References:
[1]. Gomeni R, Lavergne A, Merlo-Pich E (2009), Modelling placebo response in depression trials using a longitudinal model with informative dropout, European Journal of Pharmaceutical Sciences 36, pp. 4-10
[2]. Hooker C, Gomeni R, Zamuner S (2009), Time to event modeling of dropout events in clinical trials (presentation), Population Approach Group Europe (PAGE) 18th Meeting
[3]. Hu C, Sale ME (2003), A joint model for longitudinal data with informative dropout, Journal of Pharmacokinetics and Pharmacodynamics 30, pp. 83-103
[4]. Spiegelhalter DJ, Best NG, Carlin BP, van der Linde A (2002), Bayesian measures of model complexity and fit (with discussion), Journal of the Royal Statistics Society 64, pp. 583-639
[5]. Lee ET, Wang J (2003), Statistical Methods for Survival Data Analysis (3rd ed.), John Wiley & Sons, Hoboken, NJ
[6]. Bergstrand M, Hooker AC, Karlsson MO (2009), Visual Predictive Checks for Censored and Categorical data (poster). Population Approach Group Europe (PAGE) 18th Meeting
