DX52
Effect of Oral Fingolimod Treatment on Annualized Relapse Rates in Patients with Relapsing–Remitting Multiple Sclerosis Using Bayesian Methodology

Friday, May 29, 2015
Griffin Hall
Guosheng Yin, MD , Department of Statistics and Actuarial Science, The University of Hong Kong, Pokfulam Road, Hong Kong
Xiangyi Meng, PhD , Novartis Pharmaceuticals Corporation, East Hanover, NJ
Zahur Islam, PhD , Novartis Pharmaceuticals Corporation, East Hanover, NJ
Ralph Kern, MD , Novartis Pharmaceuticals Corporation, East Hanover, NJ
PDF


Background: FREEDOMS and FREEDOMS II, both 24-month, randomized, double-blind, placebo-controlled phase 3 studies, have demonstrated the superior efficacy of fingolimod over placebo in patients with relapsing–remitting multiple sclerosis (RRMS), as assessed by relapse rates and magnetic resonance imaging outcomes. Primary analyses in each study estimated annualized relapse rates (ARRs) by means of a negative binomial model (NBM) after adjusting for covariates. The quality of current analysis of FREEDOMS II data may be enhanced by use of historical information from the prior study (FREEDOMS).

Objectives: To estimate ARRs in patients with RRMS treated with fingolimod 0.5 mg once daily in the FREEDOMS II study using a Bayesian power prior approach.

Methods: Bayesian power prior methodology with historical data from FREEDOMS as the informative prior was used to estimate ARRs for fingolimod 0.5 mg versus placebo in FREEDOMS II.

Results: Analysis of FREEDOMS II data using NBM estimated a 47.8% reduction in ARR in patients treated with fingolimod relative to placebo (ARR: 0.21; 95% confidence interval [CI]: 0.17–0.25; and 0.39; 95% CI: 0.34–0.46, respectively).The Bayesian analysis with non-informative prior (zero weight given to prior study data) estimated a 48.0% reduction in ARR in patients treated with fingolimod relative to placebo (ARR: 0.21; 95% Bayesian credible interval [BCI]: 0.15–0.28; versus ARR: 0.39; 95% BCI: 0.33–0.47, respectively). Applying equal weight to the likelihoods of prior and current study data resulted in an estimated reduction in ARR of 52.4% (ARR: 0.21; 95% BCI: 0.18–0.25; versus ARR: 0.44; 95% BCI: 0.40–0.48, respectively).  Power priors with weights from 0.1 to 0.9 in increments of 0.1 have been explored and the deviance information criterion will be applied to determine the best fitting of power prior.

Conclusions: Bayesian power prior methodology applied to FREEDOMS II data provided estimates of ARRs similar to those obtained with NBM, which were consistent with results previously reported in phase 3 fingolimod studies.