Treatment With Interferon Beta for Multiple Sclerosis

Abstract

TotheEditor:DrShirani andcolleagues suggested that interferonbeta treatmentwasnotassociatedwithareductioninprogression of long-term disability in patients with multiple sclerosis (MS). In contrast, long-term follow-up of a randomized controlled trialof interferonbeta-1breporteda60%to70%reduction in long-term disability and a 46% to 47% reduction in mortality. Although Shirani et al did not cite these studies, both reports used bias-reduction methods. Time zero was the same for all patients (either eligibility for interferon beta treatmentortrialentry)sothattherewasnoimmortaltimebias, andbothanalysesusedpropensityscore(covariate)adjustment. Through their study design, Shirani et al introduced an important source of selection bias. Although covariate adjustment was undertaken at time zero, this method does not adjust adequately for subsequent treatment decisions. Every patient who ultimately chose not to receive treatment began this option immediately after becoming eligible. If such a patient deteriorated, however, they may have been treated. Thus, only clinically stable patients would remain untreated whereas treated patients would begin therapy at staggered times after eligibility, depending on their clinical status. This bias cannot be overcome by using time-dependent covariates or covariate adjustment because, by definition, the no treatment group must survive the observation period untreated. This bias is indicated by the reversal in the hazard ratio (HR) direction when using the historical control group. Although interferon beta treatment was not statistically significantly associated with reaching an Expanded Disability Status Scale (EDSS) score of 6, the HR was increased compared with the contemporary controls (HR, 1.30; 95% CI, 0.921.83) and decreased compared with the historical controls (HR, 0.77; 95% CI, 0.58-1.02). Similarly, this selection bias in the contemporary control group is also suggested by the observed increase in HRs for the treatment and posttreatment intervals compared with the pretreatment interval (eFigure 5 in article). This pattern suggests that patients may have been followed up without interferon beta treatment until they worsened or that treatment may have been discontinued in those who progressed further. To avoid this bias, controls without access to treatment are necessary. Unfortunately, almost 20% of the historical controls in the study were excluded for having received interferon beta therapy during follow-up. Thus, even within the historical control group, untreated patients may have been selected for doing well without therapy.