Uncovering a bias in estimated treatment effects on PIRA in multiple sclerosis clinical trials.

BACKGROUND

Interest in progression independent of relapse activity (PIRA) as an endpoint in multiple sclerosis (MS) clinical trials is surging. However, established definitions of PIRA may produce biased treatment effect estimates in the presence of a treatment-induced relapse reduction.

METHODS

We applied different definitions of PIRA to pooled data from the OPERA I/II clinical trials (clinicaltrials.gov identifiers: NCT01247324, NCT01412333). Treatment effects on PIRA according to different methods were quantified by hazard ratios (HRs) and risk ratios (RRs). Next, we evaluated the bias in each definition using synthetic Expanded Disability Status Scale (EDSS) data simulating a control and an experimental arm with varying treatment effects on relapses and on PIRA. We quantified the bias by comparing the estimated effect on PIRA with the known true effect.

FINDINGS

The pooled OPERA I/II population included 1656 participants. Estimated treatment effects on PIRA varied from a non-significant HR of 0.83 (CI = 0.66-1.04) to an HR of 0.73 (CI = 0.59-0.90) depending on the definition used. Follow-up analyses on simulated data (n = 800 per arm) revealed an underestimation of the true treatment effect on PIRA when using established definitions, with increasing bias as treatment effect on relapses increased. Defining PIRA as complementary to relapse-associated worsening (RAW) provided a less biased and operationally simple alternative.

INTERPRETATION

For clinical trials with PIRA as an endpoint, we suggest a "complementary" definition of PIRA, relying on accurate exclusion of RAW promoted by appropriate visit timing.

FUNDING

Italian Ministry of University and Research.