Laffont CM (1), Fink M (2), Gruet P (2), King JN (2), Seewald W (2) and Concordet D (1).
(1) INRA, UMR 1331, Toxalim, F-31027 Toulouse, France. Université de Toulouse, INPT, ENVT, UPS, EIP, F-31076 Toulouse, France; (2) Novartis, CH-4058, Basel, Switzerland
Objectives: Robenacoxib is a coxib non-steroidal anti-inflammatory drug approved for the treatment of osteoarthritis in dogs. In clinical trials, multiple scores were measured and as we have shown in previous work [1], a multivariate analysis is usually more appropriate than a traditional score by score analysis as it accounts for the potential correlations between ordinal responses. We have developed a new method for that purpose [1] and propose to apply this method to the analysis of robenacoxib data with two objectives: a better understanding of drug effect and the identification of possible redundancies between the scores.
Methods: Data were obtained from three clinical studies in osteoarthritic dogs where robenacoxib was given once or twice a day over 28 to 84 days at one or three different daily doses. Five scores ordered to a 3 to 4 levels were measured over time but one score was excluded from the analysis due to a high amount of missing data. In the end, the analysis was carried out on four scores documented at 1191 time points in 236 subjects. The model we used was a multivariate probit mixed effects model and model estimation was performed with a SAEM [2, 3] -like algorithm implemented in C following a pairwise approach [4] (to circumvent the computational difficulties related to the high number of scores). A principal component analysis was done to summarise the correlations between scores.
Results: The multivariate model was found to correctly predict the observed data with respect to both marginal and joint distribution. Especially, we were able to predict the percentage of dogs with complete cure at steady-state, while separate univariate analyses (assuming the independence between scores) severely underestimated this percentage by 60%. Finally, our analysis revealed that three scores were highly correlated while one was apparently independent from the others, thus giving different information on the disease.
Conclusions: Overall, the multivariate analysis appears more complex than traditional univariate analyses but provides additional insights that are worthwhile. First, it allows a better evaluation of drug effect on meaningful clinical endpoints (ex: a percentage of patients reaching the clinical target). Secondly, it provides information on the scoring system itself independently of drug effect: in the present case, what we actually measure looks more like 2 sub-scores with very different weighting 3:1.
References:
[1] Laffont CM and Concordet D. How to analyse multiple ordinal scores in a clinical trial? Multivariate vs. univariate analysis. PAGE 20 (2011) Abstr 2157.
[2] Kuhn E and Lavielle M. Maximum likelihood estimation in nonlinear mixed effects model. Computational Statistics and Data Analysis 49 (2005):1020-1038.
[3] Savic RM, Mentré F and Lavielle M. Implementation and evaluation of the SAEM algorithm for longitudinal ordered categorical data with an illustration in pharmacokinetics-pharmacodynamics. AAPS J 13 (2011):44-53.
[4] Fieuws S and Verbeke G. Pairwise fitting of mixed models for the joint modeling of multivariate longitudinal profiles. Biometrics 62 (2006): 424-431.
Reference: PAGE 21 (2012) Abstr 2548 [www.page-meeting.org/?abstract=2548]
Poster: Other Modelling Applications