Comparison of B-cell depletion versus natalizumab for treatment of multiple sclerosis: A semi-supervised causal analysis

BackgroundB-cell depletion (BCD) therapies (e.g., ocrelizumab, ofatumumab, rituximab) and natalizumab (NTZ) are highly effective disease-modifying therapies (DMTs) for multiple sclerosis (MS). However, no randomized clinical trial and only limited observational studies compared the two DMT classes. ObjectiveWe compared BCD and NTZ in managing MS patient-reported disability progression using registry-linked electronic healthcare record (EHR) data. MethodsThe study population of an EHR cohort of MS patients included a subset enrolled in a clinic-based MS registry that provided gold-standard outcome labels. To estimate average treatment effects, we applied a doubly-robust semi-supervised approach to analyze all (not only registry) patients and comprehensively adjusted for confounders that included not only a priori standard features but also knowledge graph-derived EHR features. While gold-standard disability outcomes were available in registry patients, we imputed the baseline pre-treatment and post-treatment disability status for non-registry patients. We categorized patient-reported disability progression status as "sustained worsening", "sustained improvement", or "no sustained change" based on 3 or more observations or imputations of Patient Determined Disease Steps (PDDS) scores within 3 years after target treatment initiation as the primary endpoint. ResultsIn this MS cohort (n=1,738, Age=46{+/-}13 years, Non-Hispanic White=86.71%), there was no significant difference between BCD (n=1,245, 71.63%) and NTZ (n=495, 28.37%) in mitigating sustained worsening (ATE=-0.020, 95% CI [-0.149, 0.076], p=.755) or promoting sustained improvement (ATE=-0.073, 95% CI [-0.187, 0.009], p=.114) of patient-reported disability. Sensitivity analyses using a 2-year window after treatment initiation confirmed no difference in sustained worsening (ATE=-0.013, 95% CI [-0.069, 0.074], p=.819) or sustained improvement (ATE=-0.187, 95% CI [-0.264, 0.008], p=.135) between BCD and NTZ. In power analysis, the semi-supervised approach increased statistical power compared to the standard approach of using gold-standard data alone. ConclusionThis real-world comparative effectiveness analysis based on a novel doubly-robust semi-supervised approach found no difference between BCD and NTZ in managing MS disability progression. Key Messages{blacksquare} Evaluation of sustained disability accumulation requires long-term follow-up beyond the typical clinical trials, while the scarcity of patient-reported and certainly rater-assessed disability outcomes in routine clinical care hinders analysis using real-world clinical data. {blacksquare}Using a large registry-linked electronic healthcare record cohort and a novel semi-supervised, doubly-robust method that incorporates knowledge graph-derived clinically relevant covariates from EHR, we conducted a causal inference study to compare sustained change in patient-reported disability in people with MS. {blacksquare}The semi-supervised approach effectively leverages additional data from patients without observed outcome information and increases the statistical power of the comparative effectiveness study while retaining robustness properties and achieving more consistent treatment effect estimation in the causal analysis. {blacksquare}There was no statistically significant difference between B-cell depletion therapy and natalizumab in sustained patient-reported disability outcomes up to 3-years after treatment initiation.