Comparative effectiveness in multiple sclerosis: A methodological comparison

Roos I., Diouf I., Sharmin S., Horakova D., Havrdova E. K., Patti F., ...More

Multiple Sclerosis Journal, vol.29, 2023 (SCI-Expanded) identifier identifier identifier

  • Publication Type: Article / Article
  • Volume: 29
  • Publication Date: 2023
  • Doi Number: 10.1177/13524585231151394
  • Journal Name: Multiple Sclerosis Journal
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, BIOSIS, EMBASE, MEDLINE
  • Keywords: Observational, causal inference, multiple sclerosis
  • Hacettepe University Affiliated: Yes


Background: In the absence of evidence from randomised controlled trials, observational data can be used to emulate clinical trials and guide clinical decisions. Observational studies are, however, susceptible to confounding and bias. Among the used techniques to reduce indication bias are propensity score matching and marginal structural models. Objective: To use the comparative effectiveness of fingolimod vs natalizumab to compare the results obtained with propensity score matching and marginal structural models. Methods: Patients with clinically isolated syndrome or relapsing remitting MS who were treated with either fingolimod or natalizumab were identified in the MSBase registry. Patients were propensity score matched, and inverse probability of treatment weighted at six monthly intervals, using the following variables: age, sex, disability, MS duration, MS course, prior relapses, and prior therapies. Studied outcomes were cumulative hazard of relapse, disability accumulation, and disability improvement. Results: 4608 patients (1659 natalizumab, 2949 fingolimod) fulfilled inclusion criteria, and were propensity score matched or repeatedly reweighed with marginal structural models. Natalizumab treatment was associated with a lower probability of relapse (PS matching: HR 0.67 [95% CI 0.62-0.80]; marginal structural model: 0.71 [0.62-0.80]), and higher probability of disability improvement (PS matching: 1.21 [1.02 -1.43]; marginal structural model 1.43 1.19 -1.72]). There was no evidence of a difference in the magnitude of effect between the two methods. Conclusions: The relative effectiveness of two therapies can be efficiently compared by either marginal structural models or propensity score matching when applied in clearly defined clinical contexts and in sufficiently powered cohorts.