Improved Estimation of Treatment Effects: Analysis of Time to Recurrent Relapse with Fingolimod or Ifnβ-1a Using Proportional Means Model

Background: Many statistical comparisons of treatment effects on relapse outcomes in patients with multiple sclerosis (MS) do not use all available relapse information recorded during a study. TRANSFORMS,  a 12-month, randomized, double-blind, phase 3  study, showed the superior efficacy of fingolimod over intramuscular interferon β-1a (IFNβ-1a IM), with respect to relapses and magnetic resonance imaging outcomes, in patients with relapsing–remitting MS (RRMS). Primary analyses used the Cox regression model to identify a reduction in the risk of experiencing a first relapse in patients receiving fingolimod compared with those receiving IFNβ-1a IM. Here, we use multiple relapses to assess time to recurrent confirmed relapse using the proportional means model proposed by Lin, Wei, Yang and Ying (LWYY).

Objectives: To analyze further the effects of oral fingolimod (0.5 mg once daily) or IFNβ-1a IM (30 µg weekly) treatment in order to assess alternative methods of analyzing recurrent confirmed relapses in patients with RRMS.

Methods: The LWYY model was employed to assess treatment effect on time to recurrent confirmed relapse. This model is an extension of the proportional hazards model, which accounts for recurrent events within the model using robust variance estimation. Treatment group, number of relapses in the previous 2 years and baseline Expanded Disability Status Scale score are included as covariates.

Results: The Cox regression model prospectively estimated a 49.2% risk reduction (hazard ratio [HR]: 0.51; 95% confidence interval [CI]: 0.38–0.68), whereas the LWYY model estimated a 53.1% risk reduction (HR: 0.47; 95% CI: 0.35–0.62) for patients treated with fingolimod relative to those receiving IFNβ-1a IM, with better accuracy than the Cox regression model.

Conclusions: The additional relapse data accounted for by the LWYY model enables a more robust estimation of the HR for treatment effects, yielding narrower CIs, than the Cox regression model.