Few experienced in the long-term management of multiple sclerosis (MS) patients would disagree that the development of unremitting disability progression represents the major adverse event in the course of this illness, and as a consequence, it is the most important target for clinical trials to evaluate their real cost effectiveness. Whether long-term treatment with the current interferon-β (IFNβ) products in relapsing-remitting MS (RRMS) can alter the development of long-term disability is not yet known, but it can be surmised from short-term randomized controlled trials (RCTs)1–3 and their continuous extensions at no more than 4 years4 and 5 years.5 This is the stage at which immunomodulatory therapy is likely to be most effective, but then the long-term benefits on disability have to be projected from the results generated during a much earlier and briefer period in the natural disease course. Long-term RCTs would undoubtedly provide definitive evidence, but they are impractical and considered by many to be unethical.6
More recently, further data have been collected in noncontinuous and/or retrospective, open-label extensions of the IFNβ RCTs7–10; but in these studies, the protection against bias afforded by randomization was largely lost.11 Moreover, patients enrolled in the placebo arms of RCTs do not represent the natural history of MS because they are selected for having frequent attacks during a set interval before the study. This selection enhances the power of the trial to demonstrate a treatment effect,12 but the results of trial may be only partially reproducible in widespread clinical practice.
Currently, there is increasing recognition of the value of well-designed observational studies in evidence-based medicine.13 Longitudinal, observational clinical studies may provide good information on the long-term impact of IFNβ because continuous monitoring is relatively easy and can supply data on the behavior of large cohorts of patients over extended periods.13, 14
Therefore, even if general conclusions on efficacy should be drawn only from RCTs, effectiveness data derived from observational studies may be used as a description of the situation in clinical practice or as hypothesis generating.
In this study, we conducted a 7-year survival analysis on a large cohort of untreated and IFNβ–treated RRMS patients, which aimed to evaluate the impact of treatment on three different clinical end points: times from first visit to reach an irreversible clinical disability corresponding to a Kurtzke's15 Expanded Disability Status Scale (EDSS) scores 4 and 6 and time from first visit to reach secondary progression (SP). Moreover, using current age as survival data, we also estimated the ages at the times of assignment of these landmarks.
Patients and Methods
- Top of page
- Patients and Methods
Patients with RRMS were recruited by two Italian MS Centers (Bari and Florence) and observed from the first visit for up to 7 years. Median follow-up time was 5.7 years. Clinical and therapeutic information from this cohort of RRMS patients was recorded according to computerized and standardized protocols (EDMUS and iMED; Serono International SA, Geneva, Switzerland).16, 17 A diagnosis of MS was established according to the Poser18 and McDonald criteria.19
The date of the first visit corresponded to the date of the first admission to the MS center for untreated patients and to the date of the first IFNβ administration for the IFNβ–treated patients.
The EDSS score was recorded at baseline and at least every 6 months subsequently to determine the extent of disability.15 A score of EDSS 4 corresponds to limited walking ability for greater than 500m without the need for a walking aid or rest, and a score of 6 corresponds to the ability to walk with unilateral support no greater than 100m without rest. A disability score was defined as irreversible when persisted for at least 6 months, and all the subsequent scores assessed during the follow-up of the patient were either equal to or higher than that score. The secondary progressive phase was defined as the steady worsening of symptoms and signs for at least 6 months, regardless of whether superimposed with relapses.20 The dates of MS onset and assignment to irreversible EDSS scores 4 and 6 and when the patient entered the SP phase were systematically assessed for each patient whenever appropriate.
The IFNβ–treated group included RRMS patients who received IFNβ therapy for up to 7 years. During the study period, four preparations of IFNβ were available: IFNβ-1b (Betaferon 250μg subcutaneously every other day) and IFN β-1a (Avonex 30μg intramuscularly once weekly; Rebif 22μg subcutaneously three times weekly; Rebif 44μg subcutaneously three times weekly). Periods of treatment with IFNβ were recorded for each patient, including the start and stop dates. The time spent receiving IFNβ therapy was calculated for each patient excluding transient discontinuations. The total exposure time to treatment divided by the follow-up time was used to calculate the proportion of days covered by treatment. The duration of time spent receiving transient combination therapy (eg, IFNβ and mitoxantrone or corticosteroids) was considered to be the same as administration of IFNβ alone. In this study, we assumed that different IFNβ products or transient combinations had equivalent impacts on EDSS progression; therefore, no attempt was made to adjust for specific therapeutic interventions. Another assumption was that therapy provided the same benefit regardless of the timing of the therapy and whether the therapy was provided over consecutive months or interrupted by periods off therapy (<3 months).
The untreated control group consisted of RRMS patients who voluntarily refused immunomodulatory or immunosuppressive treatment (19%), who planned to get pregnant (15%), who had concomitant diseases (ie, neoplasm, psychiatric diseases, or severe depression) that prevented them from receiving immunomodulatory or immunosuppressive drugs (23%), who discontinued immunomodulatory or immunosuppressive treatment in the first 3 to 6 weeks because of clinical and hematological adverse events (20%), or who had low disease activity (no clinical relapses in the last 2 years and EDSS score < 3.0) at first presentation (23%).
Baseline characteristics for the IFNβ–treated group and the untreated control group were reported as frequency (percentage) and mean ± standard deviation, and compared with Pearson's χ2 and Mann–Whitney U tests for categorical and continuous variables, respectively. Three different end points were analyzed: SP and irreversible EDSS scores of 4 and 6. Time (in years) from first visit and current age (calculated as time interval from date of birth to reaching each end point) were used as survival time variables. For patients who did not reach the end point at issue, time was censored at the last follow-up visit. The analyses on EDSS score of 4 and SP were performed on different sample sizes (1,375 and 1,447 patients, respectively) because each analysis included patients who were event-free before 1 year of follow-up in both groups to allow a meaningful comparison between untreated control patients and patients treated with IFNβ for at least 1 year. Cox proportional hazards regression adjusted for propensity score (PS) inverse weighting21, 22 was used to assess the differences between the two groups for the three end points. Results are expressed as hazard ratios (HRs) and 95% corrected confidence intervals.
PS analysis is a common device to reduce bias in treatment comparisons in observational studies.21–23 Because of the different sample sizes for the three end points, separate logistic regression models were first built to predict the probability (PS) to be assigned to IFNβ treatment. These models included as covariates at treatment assignment: age at disease onset, sex, disease duration, number of relapses in the past year, EDSS score in quintiles, quadratic and cubic covariates terms, and a set of two-term and three-term interactions between the same covariates. Logistic models were selected in a stepwise fashion, and model building stopped when adequate covariate balance was reached.24 Covariate residual imbalances in PS quintiles were assessed at each step with a two-way analysis of variance, where each confounder was considered as outcome and PS quintiles and treatment as factors. Overlapping of PS between treatment and control groups was also checked, and nonoverlapping subjects were excluded from the analyses. Finally, PS weights derived from the definitive logistic models were introduced in the Cox models to allow an adjusted comparison between IFNβ–treated and untreated control patients for the three end points.
Proportional hazards assumption was checked by graphic inspection of log (−log [survival]) plot and assessing the consistency of the HRs in the PS-adjusted Cox models censored to shorter follow-up timeframes (from 6 years back to only 3 years of follow-up). This analysis also allowed for the exclusion of the differences in the dropout mechanism between IFNβ–treated and untreated patients. PS-adjusted survival curves were consistently obtained from Cox models (ie, PS adjustment allowed the exchangeability of the time scale as for a randomized design) for time from first visit and current age, and estimated probabilities together with estimated time and age delays for each end point were given for IFNβ and control groups.
Furthermore, because PS methodology addresses only imbalances caused by measured confounders, we also performed a sensitivity analysis25 on time from first visit to the three end points to account for potential residual confounding caused by an unmeasured confounder.
p values less than 0.05 were considered significant. All the analyses were performed using SAS Statistical Package Release 9.1 (SAS Institute, Cary, NC).
- Top of page
- Patients and Methods
A total of 1,504 RRMS (1,103 IFNβ–treated and 401 untreated) patients participated in the study. Forty percent (n = 444) of 1,103 IFNβ–treated patients were treated with Avonex at their first prescription, 36.9% (n = 407) received Rebif 22μg, 16.3% (n = 180) received Betaferon, and 6.5% (n = 72) received Rebif 44μg. The proportion of days covered was of 75.1% for the treated cohort. Only 34 of 1,103 (3.08%) IFNβ–treated patients were also treated with mitoxantrone (median length of exposure, 0.58 year; range, 0.13–1.0 year).
Table 1 presents the baseline characteristics according to treatment group for the overall patient sample. There were significant differences between the IFNβ–treated and untreated patients for all baseline covariates except for age at first visit and sex. The IFNβ–treated group had a longer disease duration, a higher EDSS score at the first visit, and a greater number of relapses in the previous year before the first visit compared with the untreated control group. These significant differences required the use of PS-adjusted comparisons for all of the subsequent analyses. Due to nonoverlapping PS, 8 to 9% of patients were excluded from the analyses. The current sample sizes were 1,328 patients for the SP end point, 1,246 for the EDSS score of 4 end point, and 1,378 for the EDSS 6 end point.
Table 1. Baseline Characteristics of Relapsing-Remitting Multiple Sclerosis Patients according to Treatment Group (n = 1,504)
The results of the PS-adjusted Cox models for time to each end point are summarized in Table 2. The IFNβ–treated group showed a highly significant reduction (p < 0.0001) in the incidence of SP when compared with untreated patients for the two survival times: time from first visit (HR, 0.38) and current age (HR, 0.36). For the end point EDSS score of 4, there was a significant (p < 0.02) difference of at least 30% in favor of the IFNβ–treated group in the two survival times (HR of 0.70 and 0.69 for time from first visit and current age, respectively). For the end point EDSS score of 6, there was a significant difference (p ≤ 0.03) in incidence of at least 40% reduction in favor of the IFNβ–treated group (HR of 0.60 and 0.54 for time from first visit and current age, respectively).
Table 2. Results of the Analysis of Time from First Visit and Current Age to the Three Clinical End Points Using Propensity Score–Adjusted Cox Models for the Interferon-β–Treated Group versus the Untreated Control Group
PS-adjusted estimated survival curves, which graphically translated incidence rate reductions expressed by HRs, showed that IFNβ treatment delayed the time to each of the three clinical end points (Figs 1 to 3 report survival curves for time from first visit). The percentage of patients that reached SP after 7 years of follow-up was 20.2% for the untreated patients versus 8.0% for IFNβ–treated patients. For time from first visit to SP, the 8% threshold was reached with a delay of 3.8 years (7 years for IFNβ–treated vs 3.2 years for untreated patients). As to current age at SP, the delay was 8.7 years (45.7 years for the IFNβ–treated vs 37.0 years for the untreated patients).
Figure 1. Propensity score–adjusted survival curves for end point: time from first visit to secondary progression. Cumulative probability represents the estimated proportion of patients reaching the end point. Solid line indicates untreated group; dashed line indicates treatment group.
Download figure to PowerPoint
Figure 2. Propensity score–adjusted survival curves for end point: time from first visit to Expanded Disability Status Scale (EDSS) score of 4. Cumulative probability represents the estimated proportion of patients reaching the end point. Solid line indicates untreated group; dashed line indicates treatment group.
Download figure to PowerPoint
Figure 3. Propensity score-adjusted survival curves for end point: time from first visit to Expanded Disability Status Scale (EDSS) score of 6. Cumulative probability represents the estimated proportion of patients reaching the end point. Solid line indicates untreated group; dashed line indicates treatment group.
Download figure to PowerPoint
About 28% of untreated patients compared with 20.5% of the IFNβ–treated group had reached an EDSS score of 4. The 20.5% threshold was reached with a delay of 1.7 years (7 years for IFNβ–treated vs 5.3 years for untreated patients). The delay for current age at EDSS score of 4 was 4.6 years (47.3 years for IFNβ–treated vs 42.7 years for untreated patients).
For the end point EDSS score of 6, 12.4% of untreated patients compared with 7.7% of IFNβ–treated group reached an EDSS score of 6. For the time from first visit to EDSS score of 6, the threshold of 7.7% was reached with a delay of 2.2 years (7 years for IFNβ–treated vs 4.8 years for untreated patients). The same threshold estimated for age at EDSS 6 showed a delay of 11.7 years (58.3 years for IFNβ–treated vs 46.9 years for untreated patients).
Furthermore, a sensitivity analysis was performed on time from first visit for the three end points to assess the robustness of our findings. SP end point showed that the significant effect of IFNβ treatment might be altered by an unmeasured confounder with a HR ≥ 2 and a prevalence imbalance between the treatment group and control group (P0-P1) of at least 80% or with a prevalence imbalance of at least 20% but a HR ≥ 8 (Table 3). As to end points EDSS scores of 4 and 6, an unmeasured confounder with a HR ≥ 2 and a 10% prevalence imbalance would be sufficient to alter the significant effect of IFNβ treatment. However, their hazard ratios were still suggestive of a positive effect of IFNβ (0.76 and 0.65, respectively).
Table 3. Representative Results of Sensitivity Analysis on Time from First Visit to End Point Findings: How the Magnitude of an Unmeasured Binary Confounder Might Affect the Propensity Score Adjusted Hazard Ratios of Table 2
- Top of page
- Patients and Methods
Prevention of long-term disability is ultimately the most important goal of treatment in MS. The currently available IFNβ products have shown clinical benefit in short-term RCTs.1–5
Several open-label extensions of the same RCTs7–10 tried to assess the long-term impact of IFNβ, but the numbers of patients decreased dramatically and the methodological rigor was eroded during the extension phase.11 Data were collected retrospectively unblinded, after considerable intervals in which patients were not monitored and during which they may have discontinued for long periods, switched, or added other medications to the immunomodulatory therapy under study. Moreover, long-term follow-up comprised only one visit at the end of the follow-up, and patients in the placebo group received placebo for a maximum of 2 years at the start of the study, before converting to treatment.7–10
Long-term trials that remained randomized and blinded using a matched, untreated control group would undoubtedly provide definitive evidence of whether long-term treatment with IFNβ alters MS natural history, but they are impractical.
The availability of a large sample of patients, followed up by two experienced Italian MS centers, allowed us to prospectively evaluate the impact of IFNβ on the MS natural course. The results of this observational study demonstrate that patients selected to receive IFNβ treatment respond better than those who choose not to be treated or who are not selected. IFNβ treatment was associated with a significant reduction in the incidence of SP and EDSS 4 and 6 milestones, when compared with no IFNβ treatment, by using two different survival times such as time from first visit and current age.
The percentage of patients in the untreated group who converted to SP (20.2%) up to 7 years of follow-up (about 11 years from onset) is in accordance with the estimated mean rate of conversion to SP of 2 to 3% per year evaluated in previous natural history studies,26 and the proportions of them who reached EDSS scores of 4 (27.8%) and 6 (12.4%) are in accordance with more recent natural history data showing that 10 years after onset 83% of patients with EDSS score less than 3 at baseline, like those in our group, did not require walking aids (EDSS score < 6).27, 28
The proportions of treated patients who converted to SP (8%), reached EDSS score of 4 (20.5%), reached EDSS score of 6 (7.7%) were lower than those found in the whole Prevention of Relapses and Disability by Interferon-b-1a Subcutaneously in Multiple Sclerosis (PRISMS) Long-term follow-up (LTFU)10 cohort (19.7, 26.8, 19.7%, respectively). These differences may be linked to different baseline characteristics of the two study populations and to the presence, in that cohort, of a late treatment group who received placebo for 2 years at the start of the study. Moreover, we recently demonstrated29 a significant effect of the duration of IFNβ exposure on the risk for disability progression. An untreated control group for the duration of the study may give undoubtedly a better estimate of the benefits of treatment on EDSS.
The treatment effect size (30–40%) on disability landmarks EDSS 4 and 6 was similar to the results at the end of the original RCTs2, 3 showing treatment benefits on disability progression (>1.0 EDSS worsening, which required it be sustained for 6 months) of approximately 35% and to those results observed and estimated for end point EDSS 6 (of approximately 31%) in a smaller sample of RRMS patients (n = 160) at an average of 8 years after9 entry into the original phase III trial of IFNβ-1a intramuscularly.2
In this study, we estimated the effect of IFNβ also by age at end points because recent reports30, 31 suggested that survival techniques accounting for the assessment of ages at reaching SP and irreversible disability end points may provide more accurate target outcomes for therapeutic trials. The analysis of age at the end points may be much better than the time from first visit, because the latter is much more biased by that patients might show up in the MS centers earlier or later, depending on the severity of their disease. Moreover, in a previous study,32 we demonstrated that the accumulation of irreversible disability and the onset of the progressive phase of MS appear to be, at least in part, age-dependent processes.
Our results showed a significant delay in age at SP and at the end points EDSS 4 and 6 induced by IFNβ treatment. In untreated patients, the age at the time of converting to SP (37.0 years) and the ages at the time of reaching EDSS 4 and 6 (42.7 and 46.9 years, respectively) were close to those reported in two recent natural history studies,30, 31 confirming the suitability of this control group.
Although long-term, placebo-controlled studies would undoubtedly be the definitive way to evaluate the long-term benefits of a therapy, observational studies provide useful supportive data to supplement evidence-based medicine. Starting from this point of view, this observational study has demonstrated a positive long-term impact of IFNβ therapy on the course of MS when compared with no IFNβ therapy over the same period in a clinical practice setting. The limitations of this study are well-known and are associated with all observational studies.33 The major issues are due to the lack of randomization and selection bias. The treatment and control groups were imbalanced for all of the baseline covariates, which meant that it was necessary to use statistical methods to adjust the comparisons. PS-adjusted analysis is the most common device used to reduce bias in treatment comparisons in observational studies.24, 25 This technique has already been used to test drug effects in relevant therapeutic areas34, 35 and also in MS.15, 36 Furthermore, hidden bias because of potential unmeasured confounders was assessed with a sensitivity analysis. The results of this analysis showed that the positive effect of IFNβ treatment for the end point SP still remained significant under high imbalanced scenarios because of a potential unmeasured confounder. The end points EDSS 4 and 6 appeared sensitive to small bias, but their hazard ratios were still suggestive of a slower disease progression in the IFNβ–treated group. In our study, hidden bias might reflect the inability to account for factors related to physicians. Absence of blinding, for instance, might affect EDSS assessment at each visit. However, even though findings that could be sensitive to small unmeasured confounders should be interpreted with caution in a disproof approach,37 sensitivity to small biases is not a sufficient reason to dismiss such findings.38
The results of this study suggest that IFNβ delays the inevitable and irreversible clinical worsening that is the hallmark of the late phase of this disease. These data are in accordance with recent magnetic resonance imaging (MRI) longitudinal studies showing a significant effect of IFNβ on slowing brain atrophy progression and the formation of chronic, hypointense black holes,39, 40 which are important markers of tissue destruction and are correlated with disability status41 in RRMS. Proposed mechanisms by which IFNβ might slow clinical disability and atrophy rate include increasing nerve growth factors,42 limiting immune-mediated destructive inflammation,43 or limiting toxic mechanisms such as pathological iron deposition.44 Future projects will continue to build on the results of clinical studies and will hopefully provide new insights into MS progression and management.