A Bayesian meta-analysis of longitudinal lesion count data in multiple sclerosis patients
Francois Mercier (1), Ivan Demin, Novartis, Basel, Switzerland (2)
(1) Modeling and Simulation, Novartis, Cambridge, USA; (2) Modeling and Simulation, Novartis, Basel, Switzerland
Objectives: Multiple sclerosis (MS) is a demyelinating disease of the central nervous system, and the most common neurological disorder diagnosed in young adults. Nowadays, a number of safe and effective medicinal drugs are available for a chronic and symptomatic treatment of this disease; in addition, at least 6 new compounds (NC) are currently in clinical development for this indication. NUmber of lesions counted on MRI scans have become a widely used outcome measure for monitoring disease activity in clinical trials, in particular in phase 2 trials.
With access to literature data and using meta-analysis methodology, it now becomes possible to inform decision-making to proceed to phase 3 based on phase 2 data, not only by looking at the patient level in-house trial data, but also by comparing the performances of the study drug to the ones of competitors using group-level data (e.g. means).
The aim of this analysis is to characterize the time dynamic of lesion counts observed in MS patients treated with either one of the 6 following compounds: placebo, IFN-beta -1a, glatiramer acetate, natalizumab, fingolimod and teriflunomide, and to compare the time to reach full effect, amplitude of maximum effect and probability of reaching a given level of clinical effect.
Methods: Group-level data for mean lesion count were compiled into a comprehensive literature database, corresponding to a total of over 800 patients. A Bayesian model was fitted to these summary data. Various structural models were tested, and the performance of the models were also compared with and without introduction of time-independent covariate like the (mean) number of gadolinium-enhanced lesions at baseline.
Results: The model fitted the data well and provided estimates of effect size in line with expectations. The analysis also highlighted a large between trial variability for the placebo treated patients. Introducing the number of gadolinium-enhanced lesion at baseline as covariate significantly improved the goodness-of-fit.
Conclusions: Using literature data, it was possible to compare indirectly the effect size and overall trend of reduction of MRI lesion counts over time, in patients treated with various medicinal drugs either marketed or in development. Such type of comparison can be used to support decision-making at end of phase 2, as it provides insight in the potential added-value of a new compound as compared to the marketed drugs or those more advanced on the clinical development path.
