Combining randomized trials and observational studies data for comparison of treatment effects: an application to disease modifying therapies for multiple sclerosis

Background: Real-world data (RWD) on recently approved disease-modifying therapies (DMTs) in relapsing multiple sclerosis (RMS) are limited in sample size, treatment exposure and follow-up time, narrowing their use in comparative effectiveness research.

Objectives: To explore the combination of clinical trials and real-world data for comparative effectiveness analysis of multiple DMTs in RMS.

Methods: Patient-level data from ocrelizumab (OCR) Phase 3 OPERA trials and open-label extension phase, and from the German NeuroTransData (NTD) registry, were used to build 1:1 propensity score matched (PSM) cohorts to compare treatment effect of OCR on time to first relapse and 24-week confirmed disability progression vs. six DMTs from NTD over 5.5 years of follow-up. We applied intention-to-treat (ITT) method using all outcome data irrespective of treatment switch from index therapy. Datasets exchangeability was assessed by comparing outcomes of the interferon (IFN) OPERA arm with a PSM IFN NTD cohort over the 2-years OPERA double-blind period. We estimated hazard ratios (HRs) and 95% confidence intervals (CIs), controlling for seven baseline covariates, including brain magnetic resonance imaging (MRI) activity.

Results: Analyses included 611 OPERA patients and 7,141 NTD patients. For each outcome analysis, we built six separate paired-matched cohorts of patients treated with OCR in OPERA or with each DMT comparator in NTD. Standardized mean differences of covariates and PS distributions post-matching were well balanced. We found no significant difference in efficacy outcomes between IFN OPERA and NTD PSM cohorts, supporting datasets exchangeability (Relapse HR=0.76; 95% CI 0.52-1.13 or Disability HR=0.80; 95% CI 0.53–1.20). OCR treatment was associated with statistically significant risk reduction in time to first relapse (HRs 0.30 to 0.57) or disability (HRs 0.52 to 0.62) vs. any treatment pathways in NTD, regardless of each index DMT, except for one comparison with marginally significant result in disability outcome (HR=0.66; 95% CI 0.43-1.00). PSM led to small samples size, particularly including MRI as a matching factor, thus limiting generalizability of results. The ITT framework helped to take into account the dynamic of DMTs treatment pathways seen in NTD with frequent switching and discontinuation post-index DMT and to mitigate informative censoring.

Conclusions: Combining patient-level data from trials with high-quality observational RWD, applying PSM with ITT methods allowed to compare efficacy of several DMTs to control relapses and disability progression in RMS, which is particularly useful in the absence of head-to-head trials with multiple DMTs. The proposed approach may be generalized and applied to other disease areas. Study funding: Roche