PERSPECTIVES OPINION

Can we measure long-term treatment effects in multiple sclerosis? Maria Pia Sormani and Paolo Bruzzi Abstract | The gold standard for measuring treatment effects is the randomized controlled trial. In patients with multiple sclerosis (MS), trial durations are typically 2–3 years, and the long-term effects of drugs for MS can only be assessed through trial extensions or observational studies that take advantage of data from registries or large single-centre databases. The main limitation of observational studies is an unavoidable selection bias that is introduced through nonrandom assignment of the intervention. Propensity score methods can mitigate this bias by balancing the groups with respect to baseline covariates, but this approach cannot correct for unmeasurable confounding factors. Extensions of clinical trials are free from selection biases because of the initial randomization, but they can only provide an assessment of early versus delayed treatment effects. Here, we discuss these methodological issues and analyse how they have been managed in studies of the long-term effects of IFN‑β in patients with MS. Sormani, M. P. & Bruzzi, P. Nat. Rev. Neurol. advance online publication 23 December 2014; doi:10.1038/nrneurol.2014.237

Introduction

The efficacy of currently approved diseasemodifying therapies for patients with multiple sclerosis (MS) has been demonstrated by randomized controlled trials (RCTs), typically conducted over 1–3-year periods. The first drug approved for the treatment of MS was IFN‑β: three independent RCTs1–3 showed that three formulations of IFN‑β consistently reduced relapse rates and shortterm disability progression in patients with relapsing–remitting MS (RRMS). However, most individuals eventually transition from a relapsing–remitting phase to a secondary progressive course in which physical and cognitive disability gradually progresses with or without ongoing relapses. The critical question is whether—and to what extent—the short-term effects measured in RCTs translate into long-term clinically relevant benefits for patients with MS, in terms of both disability and other consequences of the disease. Moreover, the Competing interests M.P.S. has acted as a consultant for Actelion, Biogen Idec, Genzyme, Merck Serono, Novartis, Roche, Synthon and Teva. P.B. has consulted for Novartis, and provided teaching courses for Merck Serono and Roche.

efficacy of a drug assessed in an RCT, conducted under optimal conditions, does not necessarily reflect that drug’s ­effectiveness in ‘­real-world’ settings. Observational studies have been proposed as the appropriate tool to answer these questions, because they can provide long-term outcome data from large cohorts of patients treated using routine clinical practices. More than 20 years have passed since the first approval of IFN‑β for the treatment of MS,1 and many observational studies have tried to assess the long-term effects of this drug. However, such studies are known to be plagued by several potential biases.4 Statistical methods developed to deal with these biases are only partially effective, and are almost never sufficient to avoid controversies over the interpretation of study results. The only alternative to observational studies is the long-term extension of followup periods for patients included in RCTs. This option also has several drawbacks, the most obvious being a lack of information on the real-world e­ ffectiveness of a treatment. In this Perspectives article, we will review the different approaches used in studies assessing long-term effects of IFN‑β in patients with MS. In light of the conflicting

NATURE REVIEWS | NEUROLOGY

results of these studies, we discuss several biases that might have affected the data analysis.

Long-term observational studies

The results of RCTs collectively show a trend toward slower disability progression for patients with RRMS treated with IFN‑β compared with untreated patients.1–3 The translatability of these short-term results into clinically significant delays in progression over the long term has been assessed in several observational studies (Table 1). These studies used a large variety of methods and analyses, which might explain— ­at least in part—their conflicting results. More­over, these differences in study design could have interacted with the various potential biases that are inherent to observational studies. Of these, the most important is selection bias (Box 1), which is caused by baseline differences between treated and untreated patients: in an RCT, the randomization process minimizes the possibility of systematic differences between the experimental and the control group, but no such guarantees can be made in observational studies. As a consequence, any observed differences between the outcomes of treated and untreated patients in observational studies might be attributable to baseline differences rather than to a true treatment effect. It must be pointed out that, according to the strict definition given by Hernan,5 this situation is an example of confounding, rather than selection bias per se. Regardless of the definition, this problem can cause observational studies to yield flawed estimates of treatment effects. This potential bias can be addressed in two ways: with the choice of an appropriate control group, or with statistical methods that ‘adjust away’ differences in baseline prognostic factors.

Control groups Two main observational study designs— concurrent cohort studies and historical cohort studies—have been used for assessing the long-term effects of IFN‑β in patients with MS. Concurrent cohort studies are follow-up studies that compare outcomes between participants who have received an intervention and those who have not; parti­c ipants are studied retrospectively ADVANCE ONLINE PUBLICATION  |  1

© 2014 Macmillan Publishers Limited. All rights reserved

PERSPECTIVES Table 1 | Observational studies assessing long-term effects of IFN‑β in patients with relapsing–remitting MS Study

Treatment and control groups

Follow-up

End point

Statistical approach

Result*

Potential biases (bias direction)

Trojano et al. (2007)11

IFN‑β (n = 1,103) vs contemporary untreated (n = 401)

Median 5.7 years

Time from first visit and from date of birth to SPMS and to EDSS score of 4 or 6 points

Multivariate Cox regression model Propensity score adjustment

Highly significant benefit of IFN-β

Selection bias, favours untreated Immortal time bias, favours treated

Brown et al. (2007)22

Before vs after IFN‑β treatment (n = 590)

24-year observation period

Annualized EDSS change

Fixed effects model to estimate annual EDSS increase per treatment year

Highly significant benefit of IFN-β

Nonlinearity of progression, favours treatment period Selection bias, favours untreated periods

Trojano et al. (2009)13

Early IFN‑β (n = 2,260) vs late IFN‑β (n = 310)

Median 4.5 years

Time from IFN‑β treatment start and from date of birth to a confirmed 1-point EDSS progression and to EDSS score of 4 or 6 points

Multivariate Cox regression model Propensity score adjustment

Highly significant benefit of IFN-β

Selection bias (favours untreated), reduced by sensitivity analysis

Veugelers et al. (2009)23

Before vs after IFN‑β availability (n = 1,752)

24-year observation period

Rates of progression from MS onset to EDSS score of 4, 6, or 8 points

Cox proportional hazards model adjusted for patient characteristics Time from onset to first visit as a timedependent covariate

Highly significant benefit of IFN-β

Will Rogers phenomenon, favours treated

Shirani et al. (2012)8

IFN‑β (n = 868) vs contemporary untreated (n = 829) and historical untreated (n = 959)

Median: IFN-β 5.1 years, concurrent controls 4 years, historical controls 10.8 years

Time from IFN‑β treatment eligibility to a confirmed and sustained EDSS score of 6 points

Multivariate Cox regression model with IFN‑β treatment as a time-varying covariate Propensity score adjustment

Trend of inferiority of IFN‑β vs contemporary untreated Trend of benefit of IFN‑β vs historical untreated

Selection bias, favours treated when compared with historical control groups, but untreated when compared with concurrent control groups

Bergamaschi et al. (2012)18

IFN‑β or GA (n = 606) vs contemporary untreated (n = 478)

Median: treated 16.6 years, untreated 18.3 years

Time from diagnosis to conversion to SPMS

Multivariate Cox regression model adjusted for BREMS score

Highly significant effect of IFN‑β and GA therapies

Selection bias, favours untreated

Tedeholm et al. (2013)9

IFN‑β (n = 730) vs historical untreated (n = 186)

12-year observation period

Time from disease onset to conversion to SPMS

Cox proportional hazards model adjusted for baseline covariates, time from onset to treatment start and “period” (historical vs concurrent) effect

Trend of benefit of IFN‑β vs no treatment, lower than the period effect

Selection bias, favours treated

Drulovic et al. (2013)10

IFN-β (n = 236) vs concurrent untreated (n = 183)

7-year observation period

Time from disease onset to SPMS and EDSS score of 4 or 6 points

Cox proportional hazards model adjusted for the number of previous relapses

Highly significant benefit of IFN-β

Selection bias (favours untreated), mitigated by unavailability of treatment for untreated

*Highly significant refers to P 

Can we measure long-term treatment effects in multiple sclerosis?

The gold standard for measuring treatment effects is the randomized controlled trial. In patients with multiple sclerosis (MS), trial durations are ty...
658KB Sizes 0 Downloads 13 Views