Clinical and cost-effectiveness of beta interferon and glatiramer acetate for treating multiple sclerosis: systematic review and economic evaluation

Genetic

Unsurprisingly, genetic polymorphisms linked to MS have been identified primarily in immune response proteins. The first and most significant genetic locus was identified in the 1970s on the human leucocyte antigen (HLA) complex. 29,30 HLAs encode part of the class II major histocompatibility complex (MHC) in humans, which presents processed foreign antigens to T cells for recognition. 30,31 Variations within the HLA region have been consistently associated with a risk of MS, with the HLA-DRB1*15:01 allele particularly implicated. 32–35 It is also thought that the HLA complex carries genetic determinants of MS clinical progression. 30

Although the HLA complex has the strongest and most long-standing linkage with MS, other genes are suspected of increasing disease susceptibility, determining age at onset and causing poorer prognoses for specific types of MS. 32 These genes have been identified based on evidence from genetic linkage studies, microarray studies and, more recently, genome-wide association studies (GWASs). 36 A seminal GWAS study performed by the International Multiple Sclerosis Consortium and the Wellcome Trust Case Control Consortium investigated 465,434 single-nucleotide polymorphisms in 9772 cases and 17,376 control subjects, implicating at least 59 non-HLA genes in MS inheritance. These genes include those involved in cytokine, immune stimulation and immunological signal transduction pathways. 32

Despite substantial data on the genetic risk for MS, the rate of concordance between monozygotic twins is modest at about 25%. 37 Additionally, a study reporting genome, epigenome and ribonucleic acid (RNA) sequences in MS-discordant monozygotic twins was able to find no substantial difference accounting for MS discordance. Such evidence points to the involvement of other causes in MS pathogenesis. 38

Viral

Among all environmental risk factors investigated in MS aetiology, EBV infection has shown the strongest consistent evidence of association. 39 EBV was first suggested as a potential causative agent of MS because of the similarity in epidemiological distribution across age, geography, ethnicity and socioeconomic status. 40 In total, 99.5% of patients with MS test seropositive for EBV antibodies, compared with 94.2% of the general population. 41 The current evidence for a role of EBV in MS is multifaceted: prospective studies note increased serum anti-EBV antibody titres before the onset of MS;42 a meta-analysis found that, for both adults and children testing negative for EBV, the odds ratio (OR) for developing MS was 0.18 [for adults, 95% confidence interval (CI) 0.13 to 0.26] compared with people who tested positive;43 and, at the molecular level, EBV can be isolated from B-cell infiltrates in meninges. 44 Although EBV is a demonstrated risk factor for MS, its role in causation remains unproven.

Other environmental risk factors

Populations living farther from the equator, both native and foreign born, have consistently shown increased MS risk. 45–50 In one meta-analysis, this correlation persisted even after adjusting for regional differences in genetic HLA-DRB1 alleles,50 although it was not replicated in a separate meta-analysis using incidence instead of prevalence. 51 One hypothesis is that this effect is mediated by sun exposure and vitamin D levels, with one supporting meta-analysis of 11 studies finding lower mean serum 25-hydroxyvitamin D levels in patients with MS. 45–49,52 Other possible explanations include confounding by socioeconomic factors or the ‘hygiene hypothesis’. Smoking is also implicated as a modest but consistent risk factor for MS, with smoking cessation suggested as an effective public health intervention that carries numerous other benefits. 39

Clinical symptoms

Although the initial signs of MS are variable between patients, MS classically presents with focal neurological symptoms and signs of CNS dysfunction around the third decade of life. Relapses may present as painful loss of vision in one eye (optic neuritis), unilateral motor or sensory disturbance (corticobulbar/spinal tract involvement), double vision/vertigo/unsteadiness (brainstem or cerebellar syndrome), Lhermitte’s phenomenon (pain down the spine/body on flexing the neck, from a cervical cord lesion) or bilateral leg and bladder dysfunction (spinal cord syndrome). Fatigue is a common but non-specific symptom. As MS progresses in severity, it can also lead to cognitive decline as well as changes in mobility, bladder/bowel function and sexual function.

Imaging features

Magnetic resonance imaging modalities have an advantage over other imaging techniques, with their ability to dampen resonance signals from the cerebrospinal fluid (CSF) and intensify signals from sites of inflammation. 53 In sites of active inflammation, disruption of the blood–brain barrier allows lesions to be enhanced with the administration (and take-up) of contrast, whereas chronic lesions are generally non-enhancing. MRI formally joined the diagnostic criteria for MS in 200154 and has rapidly become a primary tool for characterising MS severity and progression. The characteristic MRI lesion is a cerebral or spinal plaque with high T2 signal, representing a region of demyelination with axon preservation. In the brain, plaques representing perivenular inflammation (and potential blood–brain barrier disruption) are known as Dawson’s fingers, and they are seen in the periventricular regions radiating perpendicularly away from ventricles. Outside the periventricular region, plaques are also commonly found in the corpus callosum, sub-/juxtacortical region, optic nerves and visual pathway. 55 Spinal cord lesions are nearly as common, although they more likely to be noticed clinically before MRI identification.

Pathology

Early acute-stage lesions are active plaques characterised by the breakdown of myelin, which may appear oedematous and inflamed histologically. Subacute-stage lesions appear paler in colour and have higher focal regions of macrophages. Chronic-stage lesions are inactive plaques with low levels of myelin breakdown, but are characterised by gliosis, leading to the production of scar tissue. 56–58 Within the chronic stages of the lesions, attempts at remyelination occur but the process may be hampered and unsuccessful because of the scar tissue formed by gliosis. 59,60

Risk factors for disease progression

Multiple sclerosis is notoriously heterogeneous and, even when all known risk factors are combined, they provide only moderate prognostic value. Generally, observational data have found male sex, older age at onset, progressive state at onset and higher number of MRI lesions to be predictive of a poor prognosis with faster disability progression. 74,75 A recent systematic review has identified several key factors related to relapse frequency and recovery. 75 Relapse activity appears to decrease with age and disease duration and cohort studies suggest that women experience relapses more frequently. Modifiable risk factors, including smoking, exposure to infectious disease and discontinuation of DMTs, are also associated with increased relapse frequency.

Relapse rates

There is some controversy over whether increased rates of relapse events represent an independent risk for disability progression in MS. Short-term studies suggest that relapses do not entirely regress; thus, EDSS scores, which are elevated during relapses, do not return to their previous baseline level. 76 Authors of these studies would conclude that a greater number of relapses would lead to earlier increases in EDSS scores. Longer cohort studies, however, have noted that the number of relapses is not associated with time to SPMS or EDSS 6. 71,77 A study examining placebo groups from two large Phase III trials also noted that half of the patients satisfying criteria for ‘confirmed progression’ (definitions ranging from a 1.0-point EDSS increase confirmed at 3 months to a 2.0-point EDSS increase confirmed at 6 months) were erroneously diagnosed, as their EDSS scores did not sustain progression, even through the end of the trial. 78 Thus, in short-term studies, EDSS scores measured months after relapse may still be reflecting changes in active, not progressive, disease. These longer timescales for recovery from relapse may need greater recognition.

Most recently, a longitudinal cohort study by Leray et al. 79 suggested that MS may be characterised by two distinct phases, with phase 1 lasting from diagnosis until irreversible EDSS 3 and phase 2 lasting from EDSS 3 until EDSS 6. Notably, disability progression in phase 1 did not influence disability progression in phase 2 and, similarly to previous studies, increased rates of relapse during the first 2 years of MS influenced only time in phase 1. Relapses after EDSS 3 were not associated with continued disability progression. Previously characterised risk factors of sex, age at onset and relapse history were not related to disability progression in phase 2. 79 These data are in line with previous studies suggesting that, although rates of relapse early in disease predict disease progression, relapses later in RRMS or during SPMS may not significantly predict or influence disability progression. 80,81

Prognoses for mortality

Patients with MS have an average lifespan that is 7–14 years shorter than that of matched control subjects. 82 A meta-analysis of standardised mortality rates (SMRs) found that, overall, patients had a SMR of 2.81 compared with control subjects, which suggests 181% more mortality per year than anticipated at any age. 83 This was especially increased for those at an EDSS score of > 7.5, who, in a separate study, were found to have a SMR of 4.0 compared with control subjects. 84 One review notes that, in most cohort studies of people with MS, MS is cited as the cause of death for between 50% and 75% of deaths. It also notes wide variation in the proportion of deaths ascribed to MS, resulting from variations in assessment, interpretation and coding practices. In particular, death from suicide is inconsistently reported as MS related, although there is a substantially increased risk of suicide among people with MS. 82

Prevalence and incidence

An international survey including data from 92 countries estimated the median global prevalence of MS to be 33 per 100,000 or about 2.3 million people worldwide. 63 This prevalence has been increasing in the past few decades, primarily because of increased rates of survival and diagnosis, but a meta-regression analysis suggested that there is also likely a true increase in the incidence of MS. 51 This analysis also suggested that the increase is primarily in women, who already face double the burden of MS compared with men. 51,85–89

A recent systematic review reported estimates for MS prevalence in the UK ranging from 97.26 per 100,000 in England in 199890 to 230.60 per 100,000 in Northern Ireland in 2008. 85,91 Incidence estimates were less common and ranged from 4.4 to 12.2 per 100,000 person-years. 85 Analysis of the UK General Practice Research Database (GPRD) between 1990 and 20109 similarly showed an estimated prevalence of 258.5 per 100,000 women and 113.1 per 100,000 men, with an incidence of 11.52 per 100,000 women per year and 4.84 per 100,000 men per year. Incidence peaked in women at age 40 years and men at age 45 years. Although no systematic reviews of longitudinal incidence trends specifically look at the UK, analysis of the UK GPRD estimates that, although the overall prevalence of MS is increasing because of increased survival, incidence has decreased by 1.5% per year (although this may be because of a decrease in the number of false-positive diagnoses). 92 This analysis estimates that 126,669 people with MS were living in the UK in 2010, although this number may be inflated by about 20% because it includes inaccurate diagnoses. 92

Burden of disease

The effects of MS have major ramifications for patients and carers, as well as financial implications for patients and the state.

Patient expectations and perceptions of disease

The literature describing qualitative experiences of patients is not as comprehensive as that surrounding pharmacological treatments and the pathology of MS. Collectively, however, what does exist unsurprisingly describes the experience of symptom onset and diagnosis as a negative one. 110–112 Patients inevitably experience distress and anxiety as they become aware of symptoms112 and this can continue or be amplified as they learn of their diagnosis; the diagnosis can, however, also be a source of relief because it provides an explanation for symptoms. 111 Receiving adequate information from health-care professionals at the time of diagnosis can have a positive effect on patients’ well-being and self-identification of relevant support services,111 whereas a lack of information or empathy can be linked to frustration, anxiety and fear. 112 The transition from RRMS to SPMS is also a challenging time for patients, as this requires adjusting to new ‘realities’ and preparing for forthcoming challenges in a declining trajectory. 113 A recent qualitative systematic review emphasises the importance of support from health-care providers and an accessible health-care system. 114 A comprehensive care plan including patient and carer support alongside therapeutics is described as key for successful management of MS. 115

Current service provision

At present there is no cure for MS, but treatment options exist based on the stage and subtype of disease. Currently approved drugs for MS act as immunomodulators or immunosuppressants, with the aim of reducing the pathological inflammatory reactions occurring in MS and thus the frequency and severity of relapses and the rate of disease progression. 116 Management of MS also includes non-pharmacological options such as lifestyle adjustments and rehabilitation, which are also included in the NICE guidelines for MS management. 117

Treatments to reduce the risk of relapses

Drugs aimed at reducing the risk of relapses are called DMTs. In addition to the DMTs introduced in Chapter 2, several newer drugs are licensed for use in the UK. Five newer drugs are recommended by NICE118–122 for the treatment of MS: alemtuzumab (Lemtrada^®; Sanofi Genzyme, Cambridge, MA, USA), dimethyl fumarate (Tecfidera^®; Biogen Idec Ltd, Cambridge, MA, USA), fingolimod, natalizumab and teriflunomide (Aubagio^®; Sanofi Genzyme, Cambridge, MA, USA). A summary of these recommendations is provided in Table 1. DMTs are indicated in the treatment of classic RRMS, with the exception of natalizumab and fingolimod, which are recommended only in patients with highly active RRMS. Among DMTs, IFN-β-type drugs and GA are indicated for patients with CIS.

Immunosuppressive agents, such as azathioprine, cyclophosphamide, mitoxantrone and methotrexate, can also be used in the management of MS. These agents can provide potential benefit through downregulating pathogenic mediators of MS, but can also induce severe adverse effects on the immune system. Consequently, these drugs are indicated only in patients with aggressive forms of MS, including patients who experience very frequent and severe relapses. They are not currently included in any NICE guidelines, although they continue to be used for MS123 and a systematic review suggests their effectiveness in preventing relapse recurrence. 124

Treatment of acute relapses

Steroids are commonly used and recommended to treat acute relapses. Steroids are aimed at reducing the duration of relapses by shutting down the production of inflammatory cytokines and destroying activated lymphocytes that cause demyelination; these drugs are not, however, thought to induce long-term benefits with regard to the course of the disease. 125 NICE guidelines126 recommend the use of 0.5 g of oral methylprednisolone daily for 5 days in the first instance and to consider 1 g of intravenous methylprednisone daily for 3–5 days as an alternative if oral steroids are not tolerated or have failed or if hospital admission for severe relapse or monitoring is required. Patients should not be offered a supply of steroids to administer at home for prophylactic use for future relapses. Lastly, patient education should target the management of potential complications, such as mental health changes or irregularities in blood glucose levels.

Pharmacological treatment of symptoms

Current NICE guidelines offer advice to health-care professionals, patients and families on the management of MS symptoms. 117 Recommendations include amantadine use for fatigue (although it does not have marketing authorisation in this indication) and baclofen or gabapentin for spasticity, with combinations of baclofen and gabapentin possible if individual drugs cannot reach a dosage for adequate relief. 126 Other drugs such as tizanidine (Actavis UK Ltd, Devon, UK), dantrolene (Dantrium^®; Norgine Ltd, Harefield, UK) or benzodiazepines should be considered as second- or third-line options. It should also be noted that fampridine (Fampyra^®; Biogen Idec Ltd, Cambridge, MA, USA), recently approved in Europe to improve walking ability in people with MS, has not been recommended by NICE as a cost-effective treatment. A systematic review, however, concluded that the absolute and comparative efficacy and tolerability of anti-spasticity agents in MS was poorly documented and no recommendations could be made to guide prescription. 127

For treatment of psychological changes, rivastigmine, donepezil and memantine, which are classically used in Alzheimer’s disease, have been shown to improve cognitive impairment, but overall evidence for their efficacy in MS patients has proved inconclusive. 128 The treatment of depression includes consideration of both psychotherapy and antidepressant medication. Commonly used medications are selective serotonin reuptake inhibitors such as fluoxetine, paroxetine and sertraline. A recent systematic review showed that depression severity was improved in three pharmacological studies of depression treatment in MS. 129 NICE guidelines130 state that amitriptyline can be considered to treat emotional liability.

Managing disability

Non-pharmacological treatment options are directed towards a rehabilitative approach, with specialist assistance from a multidisciplinary team.

There is evidence that physical activity alone can improve fatigue and it has been linked to improvement in aerobic capacity, gait parameters and QoL. 131,132 Suggestions for an effective rehabilitation regime include progression of physical activity from basic to integrated functions,133 to utilise working muscles while avoiding muscle overload. Although randomised controlled trials (RCTs) have shown some evidence of improved mobility and QoL from exercise interventions, systematic reviews have not reached consensus on whether the studies – which are especially limited by small sample sizes and risk of bias from lack of blinding – provide enough evidence to make guided exercise prescriptions. 134–136 Urinary incontinence affects approximately 75% of patients and can substantially impact on QoL. 137 NICE guidelines on lower urinary tract dysfunction in neurological disease are available and should be used to inform treatment. 138

Care should also be taken in the management of the mental health of patients. Interventions should be aimed at regular monitoring of any depressive states and mental health services should be offered routinely to encourage participation. 139 Education for all health-care providers and patients in coping mechanisms may help improve QoL. 140

Relapse outcomes and relapse severity

We elected to meta-analyse the RR of relapses as an overall measure of relapses in RRMS and SPMS. Although we narratively synthesised analyses for time to relapse and proportion free of relapses, both measures had significant issues. In particular, time to relapse data were inconsistently presented and at times impossible to impute whereas the proportion relapse free would have been especially dependent on the duration of follow-up and would not have captured the impact of drugs on multiple relapses per person.

We elected to meta-analyse two measures for relapse severity in RRMS: steroid-treated relapses and relapses described as moderate or severe. These were the most commonly reported measures.

Disability progression

We elected to meta-analyse time to disability progression as a measure of disability progression in RRMS and SPMS. We separated estimates for disability progression confirmed at 3 months and disability progression confirmed at 6 months, as we could not establish whether measures were commensurate. Although we narratively synthesised the proportions of patients with disability progression and the magnitude of EDSS change, we elected not to meta-analyse these data as they would have been especially dependent on the duration of follow-up. In particular, data for magnitude of EDSS change would have required extensive imputation.

Discontinuation as a result of adverse events

We estimated models for discontinuation as a result of AEs. To estimate these models, we examined three outcomes as reported: discontinuation of the study drug as a result of AEs, discontinuation of the study as a result of AEs and withdrawal from the study as a result of AEs. In the few studies that reported both discontinuation of the study drug as a result of AEs and discontinuation of the study as a result of AEs, we chose discontinuation of the study drug as a result of AEs as we believed that it would be better at capturing the relationship between study drugs and discontinuation. We also estimated one model with estimates of discontinuation closest to 24 months of follow-up, as available from included studies, as risk of discontinuation as a result of AEs is not an annualised measure, such as the ARR, or an ‘instantaneous’ measure, such as the HR, and we could not reliably estimate person-years of follow-up in each arm across all studies to convert study-level estimates to RRs.

Overview

The study by Filippini et al. 160 aimed to review the clinical effectiveness of immunosuppressors and immunomodulators in all MS types and to rank them based on relapse rate, disability progression and acceptability. The study by Tramacere et al. 159 aimed to review and rank these agents in RRMS specifically. Clerico et al. 158 examined the role of IFN-β-1a, IFN-β-1b and GA in delaying the conversion of CIS into MS, although this analysis was undertaken before revised diagnostic criteria classified many CIS episodes as in fact being RRMS. 62

Diagnostic criteria used to identify studies

Tramacere et al. 159 used all four sets of diagnostic criteria54,62,64,66 to identify RCTs of treatment for RRMS in participants aged > 18 years.

Filipinni et al. 160 included RCTs only, investigating the treatment of adults aged > 18 years with MS diagnosed according to Poser et al. ,64 the original McDonald criteria54 or the 2005 modified McDonald criteria. 66 This review therefore included all types of MS. However, it did not incorporate the most recent revision of the McDonald criteria62 and so excluded CIS studies.

In contrast, Clerico et al. 158 used the Poser criteria64 to identify RCTs and pseudo-randomised double-blind trials of CIS, with reference to specific MRI findings. 169 No exclusion criterion based on participant age was specified.

Included interventions

Tramacere et al. 159 included all immunomodulators and immunosuppressors, even if unlicensed. This included the IFN-β and GA drugs specified in the NICE scope,141 as well as 11 other interventions. We noted that the review by Tramacere et al. 159 excluded the study by Calabrese et al. ,188 stating that it was non-randomised. To the best of our knowledge this study is a RCT and it has been included in our review.

The interventions studied by Filippini et al. 160 included IFN-β and GA formulations that were licensed at the time (i.e. not pegylated IFN), as well as seven other interventions. Clerico et al. 158 would have included licensed IFN-β and GA interventions (i.e. not pegylated IFN), but identified only three studies comparing IFN with placebo.

All three reviews included studies evaluating DMTs with a dose regimen currently not recommended or authorised [e.g. IFN-β-1a (Rebif) given once weekly instead of three times weekly]. The reviews did not account separately for the inclusion of studies with a DMT given under a non-recommended dose regimen in a sensitivity analysis.

Outcomes

Tramacere et al. 159 and Filippini et al. 160 examined risk of relapse over 12 months and 24 months as a dichotomous outcome, as well as the presence or absence of disability progression assessed using the EDSS. In the study by Filippini et al. ,160 which included progressive forms of MS as well as RRMS, risk of disability progression was reported as the first outcome.

Both reviews assessed AEs. Filippini et al. 160 also included the incidence of relapse over 36 months and assessments of the acceptability of treatment as measured by discontinuation as a result of AEs.

Clerico et al. 158 used the proportion converting to CDMS as the primary outcome, alongside the ARR and additional MRI outcomes.

Statistical methods

In the study by Tramacere et al. ,159 NMAs were performed for the primary outcomes. Random-effects models were used within a frequentist setting. In contrast, Filippini et al. 160 performed NMAs within a Bayesian framework. For both reviews, equal heterogeneity across comparisons was assumed and any correlations induced by multiarm studies were accounted for. Both used SUCRA to describe the ranking of treatments. 233

Review findings

Tramacere et al. 159 found that, in RRMS, the SUCRA for the chance of experiencing relapse over 12 months was 52% for GA, 36% for SC IFN-β-1a (Rebif), 33% for pegylated IFN-β-1a, 27% for IFN-β-1b and 25% for IM IFN-β-1a (Avonex). The RR for GA compared with placebo for this outcome was 0.80 (95% CI 0.68 to 0.93), whereas all other interventions of interest did not return significant results. The ranking of interventions of interest for the prevention of relapse over 24 months in RRMS was GA (most successful) followed by IFN-β-1b, SC IFN-β-1a (Rebif) and IM IFN-β-1a (Avonex).

The SUCRA plots for reducing the worsening of disability over 24 months in RRMS returned results of 58% for GA, 51% for IFN-β-1b, 36% for SC IFN-β-1a (Rebif) and 21% for IM IFN-β-1a (Avonex). The only interventions of interest with significant RRs compared with placebo were GA (0.77, 95% CI 0.64 to 0.92) and IFN-β-1b (0.79, 95% CI 0.65 to 0.97).

Thus, in the Tramacere et al. 159 review, GA performed the best of the interventions of interest. IM IFN-β-1a (Avonex) was consistently the least effective intervention. Other interventions included in the Cochrane review (but which were outside the scope of the current MTA) performed better, such as alemtuzumab (SUCRA 97%, RR vs. placebo 0.40, 95% CI 0.31 to 0.51).

Filippini et al. 160 returned similar rankings derived from SUCRA values for reducing the recurrence of relapses over 12 months. However, for reducing the recurrence of relapses at 24 months, the SUCRA values resulted in a different ranking: SC IFN-β-1a (Rebif), GA, IFN-β-1b and IM IFN-β-1a (Avonex). In terms of reducing disability progression over 24 months, GA was ranked best (SUCRA 67%) followed by IFN-β-1b (54%), SC IFN-β-1a (Rebif) (47%) and IM IFN-β-1a (Avonex) (18%).

In the study by Clerico et al. ,158 only direct treatment comparisons were performed, using conventional pairwise meta-analyses to compare IFN with placebo. No studies of GA were identified, but IFN was effective compared with placebo.

Review quality

All three Cochrane reviews scored 10 out of 11 on the AMSTAR checklist and were assessed as being of high methodological quality. Tramacere et al. 159 and Filippini et al. 160 inadequately reported grey literature searching and Clerico et al. 158 did not assess the risk of publication bias.

Risk in randomisation or allocation methods

All studies that adequately detailed their method of randomisation (n = 21/35) used a method that was judged to be at low risk of bias. 174–175,186–192,195–200,213,216,219–220,222–224 Studies that reported methods of allocation concealment (the concealment of study allocation before the beginning of assigned treatment) were also judged to be at low risk of bias (n = 22/35),171,174–175,181,188–191,195–196,198–199,207,213,216–217,219–224 with the exception of one study that used open allocation. 170 All studies citing central allocation were judged as having a low risk of bias. 171,175,190,198,207,213,216–217,219–220,222

Risk in methods of blinding

In the studies examined, 86% (n = 30/35) were at high risk of bias from either complete or partial participant unblinding. 170–172,174–175,181–186,188,190,192,195–199,209,211,213,216–217,219–224 Fourteen studies, most of which were comparisons between different active drugs, specifically did not blind participants or practitioners;170,182–186,188,190,192,195–197,199,216 in another 15 studies, participants were initially blinded but were at high risk of unblinding from increased rates of side effects. 171–172,174–175,184,209,211,213,217,219–224 In particular, the lack of blinding in comparisons between different drugs meant that risk of bias was imbalanced across different comparisons for the same outcome. We designated all studies in which the rates of side effects (in particular, injection site reactions) in one study group were double those in another to be at high risk of bias from participant unblinding. In the two studies designated as being at low risk of bias for participant blinding, side effect rates were not increased by a factor of two in one group compared with the other (one study tested active vs. active treatments). 191,200

Blinding of outcome assessment was made similarly difficult by injection site reactions. Blinding of outcome assessment was designated as low risk only if injection site reaction rates were increased by less than a factor of two in the treatment group (two studies)172,200 or if participants were specifically instructed to cover their injection sites (eight studies). 171,175,189,192,198,222–224 In nine cases, outcome assessors were otherwise blinded but injection sites were not covered and these studies were designated to be at high risk of bias. 170,174,184,209,211,213,216–217,219 Additionally, studies in which participants were unblinded were designated as being at high risk of bias for outcome assessment if they did not report that participants were given specific instructions against sharing treatment information with assessors. 170,182–186,188,190,192,195–199,216 All studies that reported MRI outcomes and detailed methods for blinding of MRI assessment were found to be at low risk of bias (n = 13/15). 171–172,184,188,196,199–200,209,213,216,219–221

Risk in data analysis and reporting

In total, 29% (n = 10/35) of studies were found to be at high risk of bias from missing data, based on large numbers of missing data, a difference in rates of loss to follow-up between arms or lack of reporting of imputation methods. 173,181,183,186,191,200,209,213,216,223 In 17% (n = 6/35) of studies, outcomes were not reported as stated, and these were designated to be at high risk of bias from selective reporting. 173,181,188,198–199,207 Finally, all studies funded by drug manufacturers were designated as being at high risk of bias under the ‘other’ category,171–173,175,183–184,186,189–190,192,195,197–200,207,209,213,216,219–224 as this was not covered by other questions in the Cochrane risk of bias tool. 165

Time to diagnosis of multiple sclerosis

Both studies reported significant differences in favour of IFN-β-1a with regard to delaying time to confirmation of CDMS, diagnosed generally by a second relapse but in some cases by progressive neurological deterioration. The CHAMPS (Controlled High Risk Avonex Multiple Sclerosis Study) trial,172 which followed up 393 patients for up to 3 years, found a reduction in hazard of more than half (HR 0.49, 95% CI 0.33 to 0.73). The study by Pakdaman et al. ,173 which followed up 202 patients for up to 3 years, found a reduction in conversion to CDMS in the IFN-β-1a group (incidence 36.6% vs. 58.2%). We converted this to a HR of 0.54 (95% CI 0.36 to 0.81).

Separate publications also presented analyses stratified by risk levels, site of first lesion176 and type of first attack. 177 In analyses comparing patients with monofocal disease and patients with multifocal disease at first demyelinating event,177 patients with monofocal disease had a similar reduction in hazard to the whole trial population (HR 0.45, 95% CI 0.27 to 0.74), whereas patients with multifocal disease had a decreased reduction in hazard (HR 0.64, 95% CI 0.32 to 1.28).

Freedom from disease activity

The CHAMPS trial176 evaluated freedom from disease activity using several composite outcomes, each of which showed a reduction in hazard associated with IFN-β-1a. Patients receiving IFN-β-1a were less likely to have a composite outcome of CDMS or more than one new or enlarging T2 lesion, although this outcome may be closer to McDonald MS (adjusted HR 0.47, 95% CI 0.36 to 0.62); of CDMS or at least one new or enlarging T2 lesion (adjusted HR 0.55, 95% CI 0.42 to 0.71); or of CDMS, at least one new or enlarging T2 lesion or at least one gadolinium-enhancing lesion (adjusted HR 0.60, 95% CI 0.47 to 0.78).

Adverse events and mortality

Full results for AEs are available on request. Mortality was not reported in these studies.

Time to diagnosis of multiple sclerosis

In the REFLEX (REbif FLEXible dosing in early MS) trial,175 340 patients in the relevant trial arms were followed for up to 2 years and a significant reduction in hazard for conversion to CDMS was found (HR 0.48, 95% CI 0.31 to 0.73). An additional analysis examined time to conversion to McDonald MS (i.e. using MRI criteria as well) and found a similar reduction in hazard (HR 0.49, 95% CI 0.38 to 0.64), corresponding to median days to diagnosis of 310 compared with 97 in the IFN-β-1a and placebo groups respectively.

Several subgroup analyses were undertaken on the study sample by risk level; key findings from Freedman et al. 178 are summarised here. In examining time to CDMS, patients with monofocal presentation (HR 0.58, 95% CI 0.40 to 0.84) and with multifocal presentation (HR 0.45, 95% CI 0.31 to 0.64) both experienced a decreased hazard of conversion to CDMS, but type of presentation did not appear to be a significant moderator. Similarly, an analysis that ‘rediagnosed’ patients as having McDonald MS or not based on the revised 2010 criteria62 found that patients who were McDonald 2010 MS negative had a significantly decreased hazard of conversion to McDonald 2005 MS (HR 0.49, p < 0.001), as did those who were McDonald 2010 MS positive at baseline (HR 0.54, p = 0.01).

Adverse events and mortality

Full results are available on request. Mortality was not significantly different between the groups, although no events occurred in the study drug arm and two deaths occurred in the placebo arm.

Time to diagnosis of multiple sclerosis

In the BENEFIT (Betaferon^®/Betaseron^®/ in Newly Emerging Multiple Sclerosis for Initial Treatment) trial,171 468 patients were followed for up to 2 years. The study drug delayed time to CDMS (HR 0.50, 95% CI 0.36 to 0.70). This reduction in hazard corresponded to a difference in days to diagnosis of 618, compared with 255 at the 25th percentile. Triallists also considered time to McDonald MS; the effect of the study drug was similar in magnitude (HR 0.54, 95% CI 0.43 to 0.67).

Analyses stratified by risk levels, site of first lesion and type of first attack were also carried out in the BENEFIT trial. 179 In analyses comparing patients with monofocal and multifocal disease at first demyelinating event, patients with monofocal disease had a similar reduction in hazard to the whole trial population (HR 0.45, 95% CI 0.29 to 0.71), whereas patients with multifocal disease had a decreased reduction in hazard (HR 0.63, 95% CI 0.40 to 0.99).

Multiple sclerosis symptoms and health-related quality of life

Patients in the BENEFIT trial were assessed for cognitive performance using the Paced Auditory Serial Addition Test (PASAT). 180 At year 2, patients receiving the study drug had greater increases in score on this test than patients receiving placebo, including under conservative assumptions (2.0 vs. 0.6; p = 0.021). Additionally, patient-reported physical health and HRQoL data were collected in this trial. 171 Scores were not different between groups and were stable over the trial.

Adverse events and mortality

Full results are available on request. No deaths were reported in the BENEFIT trial. 171

Time to diagnosis of multiple sclerosis

The PreCISe (Presenting with Clinically Isolated Syndrome) trial174 followed up 481 patients for up to 3 years, although the trial was stopped early for benefit. Participants receiving 20 mg of SC GA once daily had a reduced hazard of conversion to CDMS (HR 0.55, 95% CI 0.4 to 0.77), although CDMS was defined here as the occurrence of a second exacerbation. The corresponding difference in days to diagnosis was 722, compared with 336 at the 25th percentile.

Adverse events and mortality

Full results are available on request. Mortality was not significantly different between groups, although the PreCISe trial174 reported only one death, in the study drug arm.

Pairwise meta-analyses

Direct evidence from comparisons is shown in Figure 3. All comparisons were against placebo. Only one comparison, 30 µg of IM IFN-β-1a once a week compared with placebo, included more than one study. The pooled effect size suggested that 30 µg of IM IFN-β-1a once a week reduces time to CDMS (HR 0.52, 95% CI 0.39 to 0.68), with low heterogeneity (I² = 0%, p = 0.718).

FIGURE 3.

Pairwise meta-analyses: time to CDMS.

Network meta-analyses

The set of studies reporting HRs for time to CDMS formed a connected network (Figure 4). This network was star shaped, meaning that it contained no comparisons between active drugs. We estimated this model using random effects, as per the protocol.

FIGURE 4.

Network of studies: time to CDMS. ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day; ga20, 20 mg of SC GA once daily.

Rankings from the NMA suggested that 44 µg of SC IFN-β-1a three times weekly was ranked best, followed by 250 µg of SC IFN-β-1b every other day, 30 µg of IM IFN-β-1a once a week and 20 mg of SC GA once daily (Table 6). Placebo was ranked last.

Findings for comparisons between active drugs and placebo were identical, as expected, to those in the pairwise meta-analyses. Findings for indirect comparisons between drugs did not suggest the superiority of any one drug over another.

Because the network was star shaped, we could not test for inconsistency.

Sensitivity analysis

We also re-estimated the network with effect sizes for time to conversion to McDonald MS for those studies reporting it. Effectiveness estimates were robust to this change.

Relapse outcomes

Findings for relapse outcomes relied on three trials with different follow-up times,198–200 including two of the largest trials in this review. 198,200 All three studies suggested a beneficial effect of 30 µg of IM IFN-β-1a once a week in terms of reducing the rate of relapses. The BRAVO trial,198 which followed 887 patients in the relevant trial arms for 24 months, found that patients receiving 30 µg of IM IFN-β-1a once a week had a 26% reduction in the ARR (RR 0.74, 95% CI 0.60 to 0.92). In the study by Kappos et al. ,199 108 patients were followed up over 24 weeks and, although the ARR was lower in patients receiving 30 µg of IM IFN-β-1a once a week (ARR 0.36, 95% CI 0.22 to 0.60) than in patients receiving placebo (ARR 0.64, 95% CI 0.43 to 0.94), this difference was only marginally significant (p = 0.07). Finally, in the MSCRG trial,200 301 patients were followed up for up to 3 years, although the study was stopped early for efficacy and thus patients had variable times to follow-up. In analyses including all patients, the ARR for patients receiving the study drug was significantly less than the ARR for patients receiving placebo (0.67 vs. 0.82; p = 0.04).

Only the MSCRG trial200 reported time to first relapse. This was not presented with a HR estimate, but a log-rank test suggested that 30 µg of IM IFN-β-1a once a week did not significantly delay time to first exacerbation compared with placebo (IFN-β-1a vs. placebo: median 47.3 vs. 36.1 weeks; p = 0.34).

Finally, all three studies reported the proportion of patients who were relapse free, although the findings were somewhat heterogeneous and comparability is limited by the differential follow-up. The BRAVO trial198 found that 69% of patients receiving 30 µg of IM IFN-β-1a once a week were relapse free compared with 61% of patients receiving placebo (p = 0.023). This difference was narrower in the study by Kappos et al. 199 (30 µg of IM IFN-β-1a once a week 78% vs. placebo 76%), with a RR for experiencing any relapses of 0.92 (95% CI 0.46 to 1.84). The MSCRG trial200 reported proportions only for those patients who completed the intended 104 weeks of study, excluding those who were enrolled but who did not complete the 104 weeks before the study was stopped. For the 85 patients included who received 30 µg IM of IFN-β-1a once a week, 38% were free of relapses, compared with 26% of the 87 patients receiving placebo. A significance test was not presented.

Relapse severity

We could not locate any relevant comparisons between 30 µg of IM IFN-β-1a once a week and placebo for outcomes relating to moderate or severe relapses or steroid-treated relapses.

Disability progression

Only the BRAVO trial198 estimated time to disability progression confirmed at 3 months. Patients receiving 30 µg of IM IFN-β-1a once a week and placebo were delayed, but not significantly so, in time to progression (HR 0.74, 95% CI 0.51 to 1.09). The results for disability progression confirmed at 6 months were similar (HR 0.73, 95% CI 0.47 to 1.14). The MSCRG trial200 also reported time to progression confirmed at 6 months. Based on a Kaplan–Meier analysis, the predicted probability of progression at 2 years was 21.9% in patients receiving 30 µg of IM IFN-β-1a once a week compared with 34.9% in patients receiving placebo (log-rank p = 0.02), indicating a slowing of time to progression. 200,201 In a separate publication, the reduction in hazard was reported as 43.0% (i.e. HR 0.570; p = 0.03). 202

Empirical proportions of patients with progression confirmed at 3 months were also reported in the BRAVO trial198 (30 µg of IM IFN-β-1a once a week 11% vs. placebo 13%). The proportions with progression at 6 months were similarly low (30 µg of IM IFN-β-1a once a week 8% vs. placebo 10%). In the MSCRG trial, empirical proportions of patients with progression confirmed at 6 months were reported for the full sample in a separate publication from the main study report. 202 Patients receiving 30 µg of IM IFN-β-1a once a week had a lower probability of progression than patients receiving placebo (15% vs. 25%), although follow-up was variable. Significance tests were not presented for these proportions per se (i.e. not as part of survival analysis, discussed in the previous paragraph) in any of the three trials.

The magnitude of change from baseline in EDSS score was presented only for the MSCRG trial. 200 Among patients completing 104 weeks on the study, those receiving 30 µg of IM IFN-β-1a once a week had a smaller increase in EDSS score than those receiving placebo (0.25 vs. 0.74; p = 0.02). This finding was similar in patients examined to week 130 (0.02 vs. 0.61; p = 0.02), with the lower of the scores at week 104 or week 130 taken as a measure of ‘sustained’ change. In the BRAVO trial,198 patients receiving 30 µg of IM IFN-β-1a once a week had a smaller decrease in the Multiple Sclerosis Functional Composite score at 24 months, but this difference was not significant (z-scores –0.045 vs. –0.14; p = 0.21).

Freedom from disease activity

We could not locate any relevant comparisons between 30 µg of IM IFN-β-1a once a week and placebo for the combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

The MSCRG trial203 reported performance on both the Comprehensive and Brief Neuropsychological Batteries by examining change from baseline to 2 years and estimated models with both no covariates and with baseline performance as a covariate. Although exact effect sizes were not provided, the study found that, in patients completing 104 weeks on the study, compared with placebo, 30 µg of IM IFN-β-1a once a week improved information processing and memory (p = 0.036 unadjusted, p = 0.011 adjusted for baseline performance) and visuospatial abilities and executive functions (p = 0.005 unadjusted, p = 0.085 adjusted), but not verbal abilities and attention span (p = 0.603 unadjusted, p = 0.917 adjusted). Findings were similar for the Brief Neuropsychological Battery (p = 0.020 for both unadjusted and adjusted), although 30 µg of IM IFN-β-1a once a week did not significantly delay time to onset of deterioration confirmed at 6 months (log-rank p = 0.094). Analyses of PASAT scores indicated that, although the difference in magnitude of change did not rise to significance (p = 0.119 unadjusted, p = 0.090 adjusted), patients receiving 30 µg of IM IFN-β-1a once a week did delay time to sustained deterioration (log-rank p = 0.023).

Additionally, patients receiving 30 µg of IM IFN-β-1a once a week had a decreased hazard of sustained worsening in the timed 25-foot walk (HR 0.401; p = 0.04). However, this decreased hazard was not evidenced in the nine-hole peg test with the dominant hand (HR 0.514; p = 0.07) or the non-dominant hand (HR 0.494; p = 0.10) or the box and block test with the dominant hand (HR 0.581; p = 0.45) or the non-dominant hand (HR 0.835; p = 0.75). 202 A variety of combinations of these end points was also tested. In a separate publication, use of an instrument to examine functional independence showed that change over 104 weeks in cognitive aspects of functional independence was not significant. 204 This was the case when considered as a difference in both means (p = 0.08) and time to sustained worsening (log-rank p = 0.188), with similar findings for difference in means for motor aspects of functional independence (p = 0.10, log-rank p = 0.368). Total changes in functional independence were significant at 104 weeks (p = 0.03).

Finally, the MSCRG trial reported on the effects of 30 µg of IM IFN-β-1a once a week on the Sickness Impact Profile as a measure of QoL. 205 In the study population as a whole, there were no differences between placebo and the study drug on the overall measure, nor on its physical or psychosocial components. However, when considering patients with low HRQoL at baseline (defined as a score of ≥ 10 on the measure), patients receiving the study drug had a greater improvement on physical aspects of the measure than those receiving placebo (–3.78 vs. 3.57; p < 0.05).

Adverse events and mortality

We stratified the comparison of AEs by type of placebo, as local AEs (e.g. injection site reactions) would not apply in studies with oral or intravenous placebos. Full results are available on request.

Mortality was not different between groups for either type of placebo. However, only one death occurred in the MSCRG trial200 (in the study drug arm), no deaths occurred in the study by Kappos et al. 199 and only one death occurred in the BRAVO trial198 (in the study drug arm).

Summary of the narrative synthesis: 30 µg of interferon beta-1a intramuscularly once a week (Avonex) compared with placebo

Findings from three trials198–200 suggested that, relative to placebo, 30 µg of IM IFN-β-1a once a week reduces relapse rates, although findings were less clear for other relapse-related outcomes. Findings from two trials198,200 suggested that 30 µg of IM IFN-β-1a once a week also has a beneficial effect in terms of delaying disability progression, although only the MSRCG trial200 presented significant results. Findings from the MSCRG trial200–204 with regard to MS symptoms were inconsistent across tests. We were unable to find any relevant comparisons for relapse severity, defined as moderate/severe or steroid-treated relapses, or the combined clinical–MRI measures of freedom from disease activity. Mortality was rare and not significantly different between groups.

Relapse outcomes

Findings for relapse outcomes relied on three trials, of which the EVIDENCE trial193–195 was the largest by far. Calabrese et al. 188 analysed 141 patients randomised to either 30 µg of IM IFN-β-1a once a week (n = 47), 44 µg of SC IFN-β-1a three times a week (n = 46) or 20 mg of SC GA once daily (n = 48) over 2 years, with complete follow-up for analysed patients. Relapses were apparently analysed using a normal distribution, although formal significance tests were not presented. At 2 years, patients receiving 30 µg of IM IFN-β-1a once a week had an average ARR of 0.5 [standard deviation (SD) 0.6], whereas patients receiving 44 µg of SC IFN-β-1a three times a week had an average ARR of 0.4 (SD 0.6). We estimated a RR of 1.25 (95% CI 0.81 to 1.92). Etemadifar et al. 185 analysed 90 patients randomised 1 : 1 : 1 to either 30 µg of IM IFN-β-1a once a week, 44 µg of SC IFN-β-1a three times a week or 250 µg of SC IFN-β-1b every other day. Because relapses were analysed using a repeated measures ANOVA method with normal distributions, we re-estimated RRs based on the number of relapses in each arm. Based on a total of 57 relapses in patients receiving 30 µg of IM IFN-β-1a once a week and 66 relapses in patients receiving 44 µg of SC IFN-β-1a three times a week, we estimated a RR of 0.86 (95% 0.61 to 1.23). Finally, the EVIDENCE trial194,195 randomised 677 patients and followed them up for an intended period of at least 48 weeks, with a median follow-up of 64 weeks. Patients receiving 30 µg of IM IFN-β-1a once a week had a higher ARR (0.65) than patients receiving 44 µg of SC IFN-β-1a three times a week (0.54); this difference was statistically significant (RR 1.20; p = 0.033).

Only the EVIDENCE trial194,195 presented data for time to first relapse. The 40th percentile of patients receiving 30 µg of IM IFN-β-1a once a week had their first relapse at 6.7 months whereas the 40th percentile of patients receiving 44 µg of SC IFN-β-1a three times a week had their first relapse at 13.5 months. Relative to patients receiving 30 µg of IM IFN-β-1a once a week, patients receiving 44 µg of SC IFN-β-1a three times a week had a decreased hazard of first relapse (HR 0.70, 95% CI 0.56 to 0.88).

Both studies presenting data on the proportions of patients free of relapse were in agreement on the direction of effect. In the study by Etemadifar et al. ,185 patients receiving 30 µg of IM IFN-β-1a once a week were less likely to be free of relapses than patients receiving 44 µg of SC IFN-β-1a three times a week (20.0% vs. 56.7%), but a pairwise significance test was not presented. In the EVIDENCE trial,194,195 patients receiving 30 µg of IM IFN-β-1a once a week were less likely to be relapse free (48%) than patients receiving 44 µg of SC IFN-β-1a three times a week (56%), that is, the OR for being relapse free at the end of the study favoured patients receiving 44 µg of SC IFN-β-1a three times a week (OR 1.5, 95% CI 1.1 to 2.0).

Relapse severity

Only the EVIDENCE trial194,195 reported outcomes related to relapse severity, in this case ARRs for steroid-treated relapses. Patients receiving 30 µg of IM IFN-β-1a once a week had an ARR of 0.28 for steroid-treated relapses, whereas patients receiving 44 µg of SC IFN-β-1a three times a week had an ARR of 0.19 for steroid-treated relapses. Thus, the RR for steroid-treated relapses is 1.47 (p = 0.009).

Disability progression

The EVIDENCE trial193 was the only trial that reported time to disability progression and the proportions of patients progressing. Drawing from interim data on all patients at 48 weeks of follow-up, patients receiving 30 µg of IM IFN-β-1a once a week appeared to progress faster than patients receiving 44 µg of SC IFN-β-1a three times a week. However, this finding was not significant for either progression confirmed at 3 months (44 µg SC vs. 30 µg IM: HR 0.87, 95% CI 0.58 to 1.31) or progression confirmed at 6 months (HR 0.70, 95% CI 0.39 to 1.25). At the end of the study, there was no statistically significant difference in the proportion of patients with disability progression confirmed at 3 months between those receiving 30 µg of IM IFN-β-1a once a week and those receiving 44 µg of SC IFN-β-1a three times a week (17% vs. 16%; p = 0.710).

In the study by Calabrese et al. ,188 the magnitude of change in EDSS score did not appear to be numerically different between 30 µg of IM IFN-β-1a once a week [0.2 (SD 0.4)] and 44 µg of SC IFN-β-1a three times a week [0.2 (SD 0.5)], but formal significance testing was not reported. However, in the study by Etemadifar et al. ,185 patients receiving 30 µg of IM IFN-β-1a once a week had a reduction in EDSS score of 0.1 (95% CI –0.2 to 0.5), a numerically smaller decrease than that seen for patients receiving 44 µg of SC IFN-β-1a three times a week (0.3, 95% CI 0.03 to 0.5). Again, formal significance testing was not reported. Finally, Mokhber et al. 186,187 found no difference in EDSS score between baseline and 12 months’ follow-up for 30 µg of IM IFN-β-1a once a week (0.0, n = 20; p = 0.548), whereas a test for change was significant for 44 µg of SC IFN-β-1a SC three times a week (–1.0, n = 21; p = 0.001). Pairwise testing was not performed but an overall test was not significant.

Freedom from disease activity

We could not locate any relevant comparisons between 30 µg of IM IFN-β-1a once a week and 44 µg of SC IFN-β-1a three times a week for the combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

Mokhber et al. 186 presented tests of cognitive function, although without pairwise comparisons. Except for the symbol digit modalities test, for all tests presented (selective reminding test, spatial recall test, symbol digit modalities test, PASAT and word list generation), comparisons across all three treatment groups were not statistically significant. Post hoc tests found evidence that patients receiving 30 µg of IM IFN-β-1a once a week did not improve as much as patients receiving 44 µg of SC IFN-β-1a three times a week on the word list generation and PASAT-easy tests.

Additionally, Mokhber et al. 187 disaggregated the Multiple Sclerosis Quality of Life-54 scale into its subcomponents, including mental health (five components) and physical health (eight components). There were few significant within-group differences in this small trial and pairwise significance tests, as well as estimates of change from baseline, were not presented in a standard format, permitting only discussion of direction and significance of differences. Patients receiving 30 µg of IM IFN-β-1a once a week significantly worsened in terms of energy and fatigue compared with patients receiving 44 µg of SC IFN-β-1a three times a week, who improved. However, patients receiving 30 µg of IM IFN-β-1a once a week significantly improved in terms of physical role limitations compared with patients receiving 44 µg of SC IFN-β-1a three times a week, who also improved. Patients receiving 30 µg of IM IFN-β-1a once a week also significantly improved in terms of both emotional role limitations and cognitive function compared with patients receiving 44 µg of SC IFN-β-1a three times a week. Differences were not significant for physical function, health perceptions, pain, sexual function, social function, health distress, overall QoL or emotional well-being.

Adverse events and mortality

The EVIDENCE trial206 was the only trial that reported AEs. No studies reported mortality. Full results are available on request.

Summary of the narrative synthesis: 30 µg of interferon beta-1a intramuscularly once a week (Avonex) compared with 44 µg of interferon beta-1a subcutaneously three times a week (Rebif)

Findings from three trials, of which one was considerably larger than the others, suggested that 30 µg of IM IFN-β-1a once a week was less effective than 44 µg of SC IFN-β-1a three times a week in terms of reducing and delaying relapses. Findings from the EVIDENCE trial194,195 suggested that 30 µg of IM IFN-β-1a once a week was also less effective than 44 µg of SC IFN-β-1a three times a week at reducing steroid-treated relapses. Across disability progression outcomes, the findings did not show a clear pattern, and the largest trial193 did not find a significant difference in terms of disability progression outcomes. Findings on MS symptoms and HRQoL were poorly reported and inconsistent and relied on one small trial. We were unable to locate any comparisons for combined clinical–MRI measures of freedom from disease activity and the included studies did not report mortality.

Relapse outcomes

Findings for relapse outcomes relied on two trials, both with 24 months of follow-up. In the study by Etemadifar et al. ,185 patients receiving 30 µg of IM IFN-β-1a once a week had fewer relapses over 2 years of follow-up than patients receiving 250 µg of SC IFN-β-1b every other day (57 vs. 65; n = 30 in both groups). We estimated this as a RR of 0.88 (95% CI 0.61 to 1.25). However, in the INCOMIN trial,196 which followed up 188 patients over 24 months, patients receiving 30 µg of IM IFN-β-1a once a week had a higher ARR (0.7) than patients receiving 250 µg of SC IFN-β-1b SC every other day (0.5). Because the authors presented the effect size estimate as a standardised mean difference, we re-estimated the RR as 1.4 (95% CI 1.07 to 1.83).

Both trials suggested that the proportion of patients who were relapse free was comparatively higher in the group receiving 250 µg of SC IFN-β-1b every other day. The proportions of patients experiencing relapses were significantly different between the relevant arms in the study by Etemadifar et al. ,185 with patients receiving 30 µg of IM IFN-β-1a once a week less likely to be free of relapse (20% vs. 43.3%, p = 0.049). In the INCOMIN trial,196 patients receiving 30 µg of IM IFN-β-1a once a week were also less likely to be free of relapse than patients receiving 250 µg of SC IFN-β-1b every other day (36% vs. 51%; risk ratio 0.76, 95% CI 0.59 to 0.99).

Relapse severity

Only the INCOMIN trial196 presented findings for relapse severity, specifically ARRs for steroid-treated relapses. Although patients receiving 30 µg of IM IFN-β-1a once a week were more likely to have steroid-treated relapses than those receiving 250 µg of SC IFN-β-1b every other day (0.5 vs. 0.38), this difference was not significant (estimated RR 1.32, 95% CI 0.96 to 1.80).

Disability progression

Only the INCOMIN trial196 presented differences in time to disability progression confirmed at 6 months and proportions with disability progression. More patients receiving 30 µg of IM IFN-β-1a once a week progressed than patients receiving 250 µg of SC IFN-β-1b every other day (30% vs. 13%), with patients in the 250 µg of SC IFN-β-1b every other day group having a reduction in risk of progression of 56% (p = 0.005). In combination with a log-rank test reported as p < 0.01, this gives an estimated HR of 2.24 (95% CI 1.21 to 4.13).

Findings from all three trials suggested that 30 µg of IM IFN-β-1a once a week did not have as beneficial an effect on the magnitude of change in EDSS score as 250 µg of SC IFN-β-1b every other day. In the study by Etemadifar et al. ,185 patients receiving 30 µg of IM IFN-β-1a once a week had a reduction in EDSS score of 0.1 (95% CI –0.2 to 0.5), a numerically smaller decrease than that seen in patients receiving 250 µg of SC IFN-β-1b every other day (0.7, 95% CI 0.5 to 0.9). Again, formal pairwise significance testing was not reported. Moreover, in a comparatively small trial, Mokhber et al. 186,187 found no evidence of a significant difference in EDSS score between baseline and 12 months’ follow-up for 30 µg of IM IFN-β-1a once a week (0.0, n = 20; p = 0.548), whereas a test for change was significant for 250 µg of SC IFN-β-1b every other day (–0.6, n = 19; p = 0.028). Pairwise testing was not performed but an overall test was not significant. Finally, using ANCOVA-adjusted estimates, the INCOMIN trial196 found that patients receiving 30 µg of IM IFN-β-1a once a week had a higher EDSS score at the end of the trial than patients receiving 250 µg of SC IFN-β-1b every other day (2.5 vs. 2.1; p = 0.004).

Multiple sclerosis symptoms and health-related quality of life

Mokhber et al. 186 presented tests of cognitive function, although without pairwise comparisons. It should be reiterated that this trial was a small trial, with 39 patients analysed in total for the relevant comparisons. Except for the symbol digit modalities test, for all tests presented (selective reminding test, spatial recall test, symbol digit modalities test, PASAT and word list generation), comparisons across all three treatment groups were not statistically significant. Post hoc tests found evidence that patients receiving 30 µg of IM IFN-β-1a once a week did not improve as much as patients receiving 250 µg of SC IFN-β-1b every other day on the symbol digit modalities and PASAT-easy tests.

Additionally, Mokhber et al. 187 disaggregated the Multiple Sclerosis Quality of Life-54 scale into its subcomponents, including mental health (five components) and physical health (eight components). There were few significant within-group differences in this small trial and pairwise significance tests, as well as estimates of change from baseline, were not presented in a standard format, permitting only discussion of direction and significance of differences. Patients receiving 30 µg of IM IFN-β-1a once a week significantly improved in terms of health perceptions and pain compared with patients receiving 250 µg of SC IFN-β-1b every other day, who declined on both measures. However, patients receiving 250 µg of SC IFN-β-1b every other day improved more on overall QoL, overall mental health aspects of QoL and emotional well-being than patients receiving 30 µg of IM IFN-β-1a once a week. Differences were not significant for overall physical health aspects of QoL, physical function, energy/fatigue, physical role limitations, sexual function, social function, health distress, emotional role limitations or cognitive function.

Adverse events and mortality

Only the INCOMIN trial196 reported AEs. No studies reported mortality. Full results are available on request.

Summary of the narrative synthesis: 30 µg of interferon beta-1a intramuscularly once a week (Avonex) compared with 250 µg of interferon beta-1b subcutaneously every other day (Betaferon/Extavia)

Although the included trials were in conflict with regard to the relative effect of the drugs on relapse rates, the INCOMIN trial196 suggested that 30 µg of IM IFN-β-1a once a week was less effective than 250 µg of SC IFN-β-1b every other day in reducing relapse rates and both studies found that the proportion of patients free of relapses was lower in the group receiving 30 µg of IM IFN-β-1a once a week. The INCOMIN trial196 did not find a difference between groups with regard to relapse severity, measured as steroid-treated relapses, but both studies agreed that 30 µg of IM IFN-β-1a once a week was less effective than 250 µg of SC IFN-β-1b every other day for disability progression. Findings on MS symptoms and HRQoL relied on one small trial with inconsistent effects and poor reporting. No studies reported mortality.

Relapse outcomes

Findings for relapse outcomes relied on both trials, with substantial follow-up; one trial, the CombiRx trial,191 was considerably larger than the other. In the trial by Calabrese et al. ,188 patients receiving 30 µg of IM IFN-β-1a once a week (n = 47) did not appear to have a numerically different ARR [0.5 (SD 0.6) vs. 0.5 (SD 0.4)] after 2 years’ follow-up than patients receiving 20 mg of SC GA once daily (n = 48). A formal significance test was not reported, but we re-estimated the RR as 1.00 (95% CI 0.67 to 1.50). However, in the larger CombiRx trial,191 with 36 months’ follow-up, patients receiving 30 µg of IM IFN-β-1a once a week (n = 250) had a higher ARR than patients receiving 20 mg of SC GA once daily (0.16 vs. 0.11). This difference was tested using a Cox proportional hazards model with correction for repeated events, which found statistically significant evidence of a shorter time between relapses for 30 µg of IM IFN-β-1a once a week than for 20 mg of SC GA once daily (HR 1.43, 95% CI 1.04 to 1.95). This finding was robust to a sensitivity analysis including non-protocol-defined relapses.

However, the CombiRx trial191 did not find a significant difference between groups in time to first relapse (p = 0.19). Additional information was not reported. The CombiRx trial191 also did not find a significant difference between groups in the proportion with protocol-defined relapses at 36 months (74.0% vs. 79.5%; p = 0.14).

Relapse severity

We were unable to locate any relevant comparisons between 30 µg of IM IFN-β-1a once a week and 20 mg of SC GA once daily on outcomes relating to moderate or severe relapses or steroid-treated relapses.

Disability progression

The CombiRx trial191 reported the proportions of patients with EDSS progression at 6 months. Fewer patients receiving 30 µg of IM IFN-β-1a once a week progressed than patients receiving 20 mg of SC GA once daily (21.6% vs. 24.8%), but this difference was reported as not statistically significant.

In the trial by Calabrese et al. ,188 patients receiving 30 µg of IM IFN-β-1a once a week had a numerically lower increase in EDSS score at 2 years [0.2 (SD 0.4)] than patients receiving 20 mg of SC GA once daily [0.3 (SD 0.5)], but formal significance testing was not reported.

Freedom from disease activity

Only the CombiRx trial191 reported freedom from disease activity outcomes for this comparison. In this trial, the proportion with freedom from disease activity (defined as the absence of exacerbation, EDSS progression or combined unique lesion activity, that is, no new enhanced lesions, unenhanced T2 lesions or enlarged unenhanced T2 lesions) was not different between patients receiving 30 µg of IM IFN-β-1a once a week and patients receiving 20 mg of SC GA once daily (21.2% vs. 19.4%; p = 0.62). This finding was robust to the inclusion of non-protocol-defined exacerbations (17.1% vs. 16.1%; p = 0.762).

Multiple sclerosis symptoms and health-related quality of life

In the CombiRx trial,191 change from baseline to 36 months was measured for the Multiple Sclerosis Functional Composite and several of its components, but none of the differences between groups was statistically significant. Overall, the Multiple Sclerosis Functional Composite score improved slightly in both the 30 µg of IM IFN-β-1a once a week group [mean 0.1 (SD 0.5)] and the 20 mg of SC GA once-daily group [mean 0.2 (SD 0.5)]. Time in seconds to complete the timed 25-foot walk increased slightly in both groups [0.2 (SD 1.1) vs. 0.2 (SD 1.7)] but time in seconds to complete the nine-hole peg test decreased slightly [–0.4 (SD 3.8) vs. –0.1 (SD 4.1)]. In addition, both groups improved in terms of the number of questions answered correctly on the PASAT [3.5 (SD 8.1) vs. 4.3 (SD 7.4)].

Adverse events and mortality

Only the CombiRx trial191 reported AEs or mortality. Full results are available on request. One death occurred in each of the relevant arms of the CombiRx trial and thus differences were not significant.

Summary of the narrative synthesis: 30 µg of interferon beta-1a intramuscularly once a week (Avonex) compared with 20 mg of glatiramer acetate subcutaneously once daily (Copaxone)

Findings from the two studies were mixed with regard to relapse outcomes, but the results of the larger of the two trials suggested that 30 µg of IM IFN-β-1a once a week was less effective than 20 mg of SC GA once daily at reducing relapses. Findings for disability progression, combined clinical–MRI measures of freedom from disease activity and MS symptoms did not suggest a difference between the two drugs. We were unable to locate any evidence for relapse severity, defined as moderate or severe relapses or steroid-treated relapses. Mortality was rare and not different between treatments in the CombiRx trial. 191

Relapse outcomes

Both key studies reported relapse outcomes. The PRISMS trial,189 which tested both doses of SC IFN-β-1a three times a week, followed up 560 patients (n = 184 in the 44-µg arm, n = 189 in the 22-µg arm, n = 187 in the placebo arm) over 2 years. Relative to placebo, both the 44-µg dose (RR 0.67, 95% CI 0.56 to 0.79) and the 22-µg dose (RR 0.73, 95% CI 0.61 to 0.86) reduced the rate of relapses. The IMPROVE (Investigating MRI Parameters with Rebif imprOVEd formulation) trial,207 a comparatively short trial that followed up 180 patients over 16 weeks (n = 120 in the 44-µg arm, n = 60 in the placebo arm), also showed a substantial decrease in the rate of relapses in those receiving the study drug (RR 0.43, 95% CI 0.23 to 0.82). Time to first relapse outcomes were cursorily presented in the PRISMS trial. 189 Both the 44-µg dose and the 22-µg dose delayed time to first relapse, by 5 months and 3 months respectively, although a significance test was not presented. However, (confidential information has been removed).

Finally, the PRISMS trial189 reported the proportions of patients who were free of relapse. Higher proportions of patients were relapse free in both treatment groups than in the placebo group at 2 years of follow-up: in the placebo arm 16% of patients were free of relapses, patients receiving 44 µg of SC IFN-β-1a three times a week had a 32% chance of being free of relapses (OR 2.57, 95% CI 1.56 to 4.25) and patients receiving 22 µg of SC IFN-β-1a three times a week had a 27% chance of being free of relapses (OR 2.01, 95% CI 1.21 to 3.35).

The REMAIN trial,183 which followed up 30 patients with either RRMS or SPMS for 96 weeks, did not find a significant difference between arms in time to first relapse or the proportion of patients relapse free.

Relapse severity

The PRISMS trial189 presented data for both moderate or severe relapses and steroid-treated relapses. Patients receiving placebo had, on average, more moderate or severe relapses over the course of the study (0.99) than patients receiving 44 µg of SC IFN-β-1a three times a week (0.62) or patients receiving 22 µg of SC IFN-β-1a three times a week (0.71). We re-estimated the corresponding RR for 44 µg of SC IFN-β-1a three times a week as 0.64 (95% CI 0.53 to 0.74) and for 22 µg of SC IFN-β-1a three times a week as 0.72 (95% CI 0.61 to 0.84). Correspondingly, patients receiving 44 µg of SC IFN-β-1a three times a week were more likely to be free of any moderate or severe relapses (OR 2.32, 95% CI 1.47 to 3.37). Findings were similar for the 22 µg of SC IFN-β-1a three times a week compared with placebo (OR 2.13, 95% CI 1.41 to 3.21).

The pattern of findings in the PRISMS trial189 for steroid treatment was similar. Patients receiving placebo had, on average, more courses of steroids for MS relapses over the course of the study (1.39) than patients receiving 44 µg (0.75) or 22 µg (0.97) of SC IFN-β-1a three times a week. We re-estimated the corresponding RR for 44 µg of SC IFN-β-1a three times a week compared with placebo as 0.54 (95% CI 0.46 to 0.63) and for 22 µg of SC IFN-β-1a three times a week compared with placebo as 0.70 (95% CI 0.61 to 0.80). Correspondingly, patients receiving 44 µg of SC IFN-β-1a three times a week were more likely to be free of any steroid-treated relapses (OR 1.99, 95% CI 1.32 to 3.02), as were patients receiving 22 µg of SC IFN-β-1a three times a week (OR 1.71, 95% CI 1.14 to 2.57).

Disability progression

In the PRISMS trial,189 time to disability progression confirmed at 3 months was slowed by both doses of SC IFN-β-1a three times a week compared with placebo. The 25th percentile of the distribution of time to progression was 21.3 months for patients receiving 44 µg of SC IFN-β-1a three times a week and 18.5 months for patients receiving 22 µg of SC IFN-β-1a three times a week, compared with 11.9 months for patients receiving placebo. Corresponding HRs showed evidence of a statistically significant delay in progression (44 µg: HR 0.62, 95% CI 0.43 to 0.91; 22 µg: HR 0.68, 95% CI 0.48 to 0.98).

Both the PRISMS trial189 and the IMPROVE trial207 reported the magnitude of change in EDSS score. In the PRISMS trial,189 compared with placebo, both 44 µg and 22 µg of SC IFN-β-1a three times a week resulted in a smaller increase in EDSS score, with a difference of 0.25 EDSS points (both p < 0.05). The IMPROVE trial207 did not report a standard significance test, although the median change in EDSS score in both the 44-µg arm and the placebo arm was 0.

In the REMAIN trial,183 the magnitude of change in EDSS score, time to progression and proportions of patients with progression were not significantly different between arms.

Freedom from disease activity

We were unable to locate any relevant comparisons between 44 µg or 22 µg of SC IFN-β-1a three times a week and placebo for the combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

The PRISMS trial reported the effects of 44 µg and 22 µg of SC IFN-β-1a three times a week on various MS symptoms. 189,208 As noted in the original trial report,189 patients receiving the 44-µg dose were less likely to have a sustained worsening in ambulation than those receiving placebo (7% vs. 13%; p < 0.05); however, the corresponding value for those receiving the 22-µg dose (12%) was not significantly different from that for placebo. Subsequently, Gold et al. 231 reported that, although patients in all three groups increased their scores from baseline on the Center for Epidemiologic Studies Depression Rating Scale, these changes were not significantly different between the groups (44 µg: 0.2, 22 µg: 1.8, placebo: 0.9; p = 0.60). Similarly, the risk of exceeding the cut-off score for depression on this scale was not significantly different between either the 44-µg arm (RR 0.7, 95% CI 0.3 to 1.6) or the 22-µg arm (RR 0.8, 95% CI 0.3 to 1.8) and the placebo arm and the proportions of patients exceeding the cut-off on the Beck Hopelessness Scale were not significantly different between the placebo arm (6.9%) and either the 44-µg arm (6.9%; p = 1.0) or the 22-µg arm (10.5%; p = 0.55). Finally, data were not presented numerically, but it was reported that there was no difference between the groups in scores on the General Health Questionnaire or its subscales.

Adverse events and mortality

All studies presented AEs. Full results are available on request. None of the studies reported deaths related to the study drugs.

Summary of the narrative synthesis: 44 µg and 22 µg of interferon beta-1a subcutaneously three times a week (Rebif) compared with placebo

Findings from two trials suggested a beneficial effect of 44 µg of SC IFN-β-1a three times a week compared with placebo for relapse outcomes. Additionally, findings from the PRISMS trial189 suggested a beneficial effect of 44 µg of SC IFN-β-1a three times a week on relapse severity (both moderate/severe relapses and steroid-treated relapses) and on delaying disability progression. Findings from the PRISMS trial189,208 also suggested a beneficial effect of 44 µg of SC IFN-β-1a three times a week on ambulation, but not mental health. Findings for the 22-µg dose in the PRISMS trial189,208 were similar except for ambulation. Mortality was not reported.

Relapse outcomes

Assessment of relapse outcomes for this comparison relied on two small studies with very different follow-up times. In the study by Etemadifar et al. ,185 over 2 years of follow-up, patients receiving 44 µg of SC IFN-β-1a three times a week had 66 relapses whereas patients receiving 250 µg of SC IFN-β-1b every other day had 65 relapses (n = 30 in both groups). We estimated this as a RR of 1.02 (95% CI 0.72 to 1.43). In the REFORMS trial,197 patients receiving 44 µg of SC IFN-β-1a three times a week had an ARR of 0.15 whereas patients receiving 250 µg of SC IFN-β-1b every other day had an ARR of 0.11. This difference was statistically significant (p < 0.001), although this was a relatively small trial (n = 129), patients were followed up for only 12 weeks and patient relapses were self-reported rather than assessed by a neurologist.

In the study by Etemadifar et al. ,185 the proportion of patients without relapses at 2 years was numerically higher in the group receiving 44 µg of SC IFN-β-1a three times a week than in the group receiving 250 µg of SC IFN-β-1b every other day (56.7% vs. 43.3%), but no pairwise significance testing was performed.

Relapse severity

We were unable to find any comparisons between 44 µg of SC IFN-β-1a three times a week and 250 µg of SC IFN-β-1b every other day for outcomes relating to moderate or severe relapses or steroid-treated relapses.

Disability progression

Analysis of disability progression in both trials was by magnitude of EDSS score change, although both trials reported inadequate details of the analysis. In the study by Etemadifar et al. ,185 patients receiving 44 µg of SC IFN-β-1a three times a week had a decrease in EDSS score of 0.3 (95% CI 0.03 to 0.5), whereas patients receiving 250 µg of SC IFN-β-1b every other day had a decrease in EDSS score of 0.7 (95% CI 0.5 to 0.9). A pairwise significance test was not performed. Patients in the study by Mokhber et al. 186,187 in both treatment groups also showed a decrease in EDSS score, but in the opposite direction (44 µg: –1.0, p = 0.001; 250 µg: –0.6, p = 0.028). Again, a pairwise significance test was not performed.

Freedom from disease activity

We were unable to find any comparisons between 44 µg of SC IFN-β-1a three times a week and 250 µg of SC IFN-β-1b every other day for combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

As noted previously, analyses in the study by Mokhber et al. 186 for cognitive function found no significant differences between the groups, except for the symbol digit modalities test. Post hoc analyses indicated that patients receiving 44 µg of SC IFN-β-1a three times a week improved more than those receiving 250 µg of SC IFN-β-1b every other day on the symbol digit modalities test and the PASAT-easy.

Across the QoL domains tested in the study by Mokhber et al. ,187 44 µg of SC IFN-β-1a three times a week was not significantly different from 250 µg of SC IFN-β-1b every other day except for overall mental health aspects of HRQoL, with patients receiving 250 µg of SC IFN-β-1b every other day improving significantly more.

Adverse events and mortality

Adverse events were reported only in the AVANTAGE trial182 and the REFORMS trial. 197 Only the AVANTAGE trial182 reported mortality, with no events occurring in either study arm. Full results are available on request.

Summary of the narrative synthesis: 44 µg of interferon beta-1a subcutaneously three times a week (Rebif) compared with 250 µg of interferon beta-1b subcutaneously every other day (Betaferon/Extavia)

Findings were derived from three small trials and should thus be treated with caution. The two trials reporting relapse outcomes disagreed on the comparative effectiveness of these two drugs, although there was some evidence from the REFORMS trial197 that patients receiving 44 µg of SC IFN-β-1a three times a week had a higher ARR. Findings for disability progression, MS symptoms and HRQoL were inconsistent and poorly reported. We were unable to find comparisons for relapse severity or combined clinical–MRI measures of freedom from disease activity. No deaths were reported.

Relapse outcomes

In the study by Calabrese et al. ,188 patients receiving 44 µg of SC IFN-β-1a three times a week had a numerically lower ARR than patients receiving 20 mg of SC GA once daily after 2 years of follow-up [0.4 (SD 0.6) vs. 0.5 (SD 0.4)], but formal significance testing was not reported and relapses were analysed using a normal distribution. We re-estimated this as a RR of 0.80 (95% CI 0.52 to 1.23). In the larger REGARD trial,192 764 patients were followed up for 96 weeks. The ARR was not significantly different between patients receiving 44 µg of SC IFN-β-1a three times a week and patients receiving 20 mg of SC GA once daily (0.30 vs. 0.29; p = 0.828).

The REGARD trial192 did not find a significant difference in time to first relapse between patients receiving 44 µg of SC IFN-β-1a three times a week and those receiving 20 mg of SC GA once daily (HR 0.94, 95% CI 0.74 to 1.21); there was also no significant difference between the groups in the proportion of patients who were free of relapses at 96 weeks (62% vs. 62%; p = 0.96).

Relapse severity

In the REGARD trial,192 the ARR for steroid-treated relapses was not significantly different between patients receiving 44 µg of SC IFN-β-1a three times a week and those receiving 20 mg of SC GA once daily (0.19 vs. 0.17; p = 0.386).

Disability progression

The REGARD trial192 reported the proportions of patients with disability progression confirmed at 6 months. The proportions were not significantly different between patients receiving 44 µg of SC IFN-β-1a three times a week (11.7%) and those receiving 20 mg of SC GA once daily (8.7%) (p = 0.117).

In the study by Calabrese et al. ,188 patients receiving 44 µg of SC IFN-β-1a three times a week had a numerically lower increase in EDSS score at 2 years [0.2 (SD 0.5)] than patients receiving 20 mg of SC GA once daily [0.3 (SD 0.5)], but formal significance testing was not reported.

Freedom from disease activity

We were unable to locate any comparisons between 44 µg of SC IFN-β-1a three times a week and 20 mg of SC GA once daily for combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

We were unable to locate any comparisons between 44 µg of SC IFN-β-1a three times a week and 20 mg of SC GA once daily for MS symptoms or HRQoL.

Adverse events and mortality

Adverse events and mortality were reported in the REGARD trial. 192 Only one death occurred, in the IFN arm, and thus mortality was not significantly different between groups. Full results are available on request.

Summary of the narrative synthesis: 44 µg of interferon beta-1a subcutaneously three times a week (Rebif) compared with 20 mg of glatiramer acetate subcutaneously once daily (Copaxone)

Findings from two trials did not suggest a difference between the two drugs for relapse outcomes, relapse severity or disability progression. We could not locate comparisons relating to combined clinical–MRI measures of freedom from disease activity or MS symptoms or HRQoL. There was no difference in mortality between the groups.

Relapse outcomes

Both studies reporting ARRs suggested a beneficial effect of 250 µg of SC IFN-β-1b every other day, although only the trial by the IFNB Multiple Sclerosis Study Group (IFNB MSSG)209,210 may have been powered to detect a difference between the groups. In the IFNB MSSG trial,209,210 247 patients in the relevant arms were followed up for variable amounts of time, with the initial 2-year study phase continuing into a blinded extension; thus, some patients were followed for up to 5.5 years, with a median follow-up of 46.0 months in the placebo arm and 48.0 months in the relevant study drug arm. At the end of the study, patients receiving 250 µg of SC IFN-β-1b every other day had a lower ARR than patients receiving placebo (0.78, 95% CI 0.70 to 0.88 vs. 1.12, 95% CI 1.02 to 1.23; p = 0.0006). In a comparatively small trial, Knobler et al. 211 followed up 30 patients over 3 years, including a 6-month dose-finding period at the start of the study. The 24 patients receiving 250 µg of SC IFN-β-1b every other day had an ARR of 0.7 whereas the six patients receiving placebo had an ARR of 0.9. This difference was not significant (p = 0.33).

Both studies also reported information on time to first relapse. Knobler et al. 211 reported that the median time to first relapse was delayed, but not significantly so, in patients receiving 250 µg of SC IFN-β-1b every other day compared with patients receiving placebo (14 months vs. 2 months; log-rank p = 0.07). The comparatively larger IFNB MSSG trial209 reported a similar finding at the 3-year follow-up, albeit of a smaller magnitude and rising to statistical significance. The median time to first exacerbation was delayed in patients receiving 250 µg of SC IFN-β-1b every other day compared with placebo (264 days vs. 147 days; log-rank p = 0.028).

The proportions of patients who were free of relapse were also available only at the 3-year follow-up for the IFNB MSSG trial. 209 The proportions free of relapse were not significantly different between the groups (250 µg of SC IFN-β-1b every other day 21.8% vs. placebo 13.8%; p = 0.097). Three-year results from the study by Knobler et al. 211 showed a similar trend (42% vs. 17%), with these findings also not significant (p = 0.37).

Relapse severity

Relapse severity was reported based on both 2-year and final data from the IFNB MSSG trial,209,210 but only the 2-year data were usable. At 2 years of follow-up, patients receiving 250 µg of SC IFN-β-1b every other day had a lower ARR for moderate or severe relapses than those receiving placebo (0.23 vs. 0.45; p = 0.002). Similar findings based on final data were reported, but only a p-value (p = 0.012) for a relationship in the same direction was provided. Knobler et al. 211 did not find a significant relationship for ‘attack severity’, although the findings were reported only as a non-significant p-value (p = 0.67) and relapse severity was not defined.

Disability progression

The IFNB MSSG trial210 reported that 250 µg of SC IFN-β-1b every other day delayed disability progression confirmed at 3 months, but not significantly so, with a median time to progression of 4.79 years, compared with 4.18 years in the placebo group (log-rank p = 0.096). The proportions of patients with confirmed progression showed a similar trend (35% vs. 46%). We re-estimated this as a HR of 0.71 (95% CI 0.48 to 1.06). Knobler et al. 211 examined change in EDSS score from baseline between the groups but noted only that the difference between the groups was not statistically significant (p = 0.42).

Freedom from disease activity

We were unable to locate any relevant comparisons between 250 µg of SC IFN-β-1b every other day and placebo for combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

In the study by Schwartz et al. ,181 34 patients receiving 250 µg of SC IFN-β-1b every other day were compared with 45 patients receiving BSC. Over the course of a year, there was no difference between the groups in quality-adjusted time without symptoms and toxicity, measured in months (10.6 vs. 10.4; p = 0.50).

Adverse events and mortality

Adverse events were reported in the IFNB MSSG trial210 and the study by Knobler et al. 211 None of the studies reported mortality. Full results are available on request.

Summary of the narrative synthesis: 250 µg of interferon beta-1b subcutaneously every other day (Betaferon/Extavia) compared with placebo

Findings from two studies suggested a beneficial effect of 250 µg of SC IFN-β-1b every other day on relapse outcomes compared with placebo (although not on the proportions of patients who were relapse free). Findings from the IFNB MSSG trial209,210 suggested a reduction in the rate of moderate or severe relapses in the group receiving 250 µg of SC IFN-β-1b every other day, but findings from the study by Knobler et al. 211 were uninterpretable. Neither study found evidence of a delay in time to disability progression. One small study comparing 250 µg of SC IFN-β-1b every other day against BSC did not find differences in HRQoL over a year. We were unable to find any comparisons for combined clinical–MRI measures of freedom from disease activity. None of the studies reported mortality.

Relapse outcomes

Both the BECOME trial184 and the larger BEYOND (Betaferon Efficacy Yielding Outcomes of a New Dose) trial190 reported ARRs. In the BECOME trial,184 75 patients were followed up for up to 2 years. Patients receiving 250 µg of SC IFN-β-1b every other day did not have a significantly different ARR from patients receiving 20 mg of SC GA once daily (0.37 vs. 0.33; p = 0.68). Findings from the BEYOND trial,190 in which 1345 patients from the relevant trial arms were followed up for at least 2 and up to 3.5 years, suggested a similar trend (ARR 0.36 vs. 0.34; one-tailed p = 0.79). This was expressed using a Cox proportional hazards model with modification for repeated events (HR 1.06, 95% CI 0.89 to 1.26).

Time to first relapse was also not significantly different between arms in either study. In the BECOME trial,184 of patients who had relapses, the median time to first relapse for those receiving 250 µg of SC IFN-β-1b every other day (123 days) was not very different from the median time to first relapse for those receiving 20 mg of SC GA once daily (121 days), with a non-significant log-rank test on the whole sample (p = 0.12). In the BEYOND trial,190 there was no substantial difference in days to first relapse for patients at the 25th percentile (250 µg of SC IFN-β-1b every other day 283 days vs. 20 mg of SC GA once daily 271 days; one-sided log-rank p = 0.75). This was supported by the proportions of patients who were relapse free at 2 years, estimated from a Kaplan–Meier model, which were very similar (59% vs. 58%).

Finally, only the BECOME trial184 reported the empirical proportions of patients relapsing. Fewer patients receiving 250 µg of SC IFN-β-1b every other day were relapse free than patients receiving 20 mg of SC GA once daily, but this difference was not significant (53% vs. 72%; p = 0.10).

Relapse severity

Only the BEYOND trial190 reported ARRs for severity of relapse. ARRs for major relapse were not significantly different between patients receiving 250 µg of SC IFN-β-1b every other day and those receiving 20 mg of SC GA once daily (0.19 vs. 0.18; one-sided p = 0.36). Time to first major relapse was also not significantly different between the arms, with both arms having proportions with a relapse at 2 years of 27%, as predicted by a Kaplan–Meier model (log-rank p = 0.56).

Both studies reported the empirical proportions of patients receiving steroid treatment for MS. In the BECOME trial,184 more patients receiving 250 µg of SC IFN-β-1b every other day (44%) required steroid treatment for relapses than patients receiving 20 mg of SC GA once daily (23%), but this difference was only of marginal significance (p = 0.09). In contrast, the proportions of patients requiring steroid treatment for relapses were not significantly different in the BEYOND trial190 (34% vs. 32%; p = 0.43).

Disability progression

The BEYOND trial190 reported time to disability progression confirmed at 3 months. Because median time to progression was not reached, the time to progression at the 10th percentile was reported. The 10th percentile of patients receiving 250 µg of SC IFN-β-1b every other day progressed after 274 days, whereas patients receiving 20 mg of SC GA once daily progressed after 268 days (log-rank p = 0.35). Alternative estimates were provided based on Kaplan–Meier models, in which the probability of progression at the end of 2 years was 21% in those receiving 250 µg of SC IFN-β-1b every other day and 20% in those receiving 20 mg of SC GA once daily (log-rank p = 0.68). We estimated a HR of 1.06 (95% CI 0.81 to 1.37) from these statistics.

In a separate publication to the main trial report, the BECOME trial212 reported time to disability progression confirmed at 6 months. The empirical proportions of patients progressing in each arm were dissimilar (250 µg of SC IFN-β-1b every other day 12.1% vs. 20 mg of SC GA once daily 17.6%), but the log-rank test was non-significant (p = 0.51). Based on these statistics, we estimated a HR of 0.66 (95% CI 0.19 to 2.28). The BECOME trial212 also reported progression based on the Multiple Sclerosis Functional Composite, in which an increase of 0.2 SDs confirmed at 6 months constitutes evidence of progression. The same trend was apparent (5.7% vs. 10.3%, log-rank p = 0.39).

Freedom from disease activity

We were unable to locate any relevant comparisons between 250 µg of SC IFN-β-1b every other day and 20 mg of SC GA once daily for combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

We were unable to locate any relevant comparisons between 250 µg of SC IFN-β-1b every other day and 20 mg of SC GA once daily for MS symptoms or HRQoL. However, the BECOME trial212 did present results for the Multiple Sclerosis Functional Composite, discussed earlier in Disability progression.

Adverse events and mortality

Both studies reported AEs, but only the BEYOND trial190 reported mortality. The difference between the groups was not significant for mortality, although only one death occurred, in the GA arm. Full results are available on request.

Summary of the narrative synthesis: 250 µg of interferon beta-1b subcutaneously every other day (Betaferon/Extavia) compared with 20 mg of glatiramer acetate subcutaneously once daily (Copaxone)

Findings from two trials – one small and one large – did not suggest a difference between the two drugs in terms of relapse outcomes, relapse severity or disability progression. We were unable to locate any comparisons for combined clinical–MRI measures of freedom from disease activity. The differences between the groups was not significant for mortality.

Relapse outcomes

Participants receiving 125 µg of SC pegylated IFN-β-1a every 2 weeks showed a decrease in ARR (RR 0.644, 95% CI 0.500 to 0.831). 213 Time to first relapse was also delayed in patients receiving the active drug (HR 0.61, 95% CI 0.47 to 0.80).

Relapse severity

Publications arising from this study did not report relapse severity.

Disability progression

Participants receiving 125 µg of SC pegylated IFN-β-1a every 2 weeks experienced a delay in time to disability progression confirmed at 3 months (HR 0.62, 95% CI 0.40 to 0.97). 213 As reported in the Summary of Product Characteristics filed by the European Medicines Agency,234 the time to disability progression confirmed at 6 months was longer in patients receiving the study drug than in patients receiving placebo (HR 0.46, 95% CI 0.26 to 0.81).

Freedom from disease activity

In the ADVANCE trial, measures of freedom from disease activity included mixed clinical and MRI, clinical-only and MRI-only definitions and were reported in a separate publication214 to the main study report. As stated in the methods, we report here the mixed clinical and MRI definition, which included both the absence of relapses and the absence of onset of disability progression confirmed at 3 months, as well as no gadolinium-enhancing lesions and no new or newly enlarging T2 hyperintense lesions. Between baseline and week 48 of the trial, 33.9% of patients (n = 466 in this analysis) receiving the study drug had no evidence of disease activity, whereas 15.1% of patients (n = 484 in this analysis) receiving placebo had no evidence of disease activity (OR 2.89, 95% CI 2.11 to 3.95). This finding was robust to sensitivity analysis of data missingness.

Multiple sclerosis symptoms and health-related quality of life

In the ADVANCE trial,215 patients receiving 125 µg of SC pegylated IFN-β-1a every 2 weeks did not significantly worsen over 48 weeks on the Multiple Sclerosis Impact Scale (MSIS-29) physical subscale (mean change 0.08, 95% CI –1.10 to 1.27), whereas placebo patients did (mean change 1.24, 95% CI 0.05 to 2.44). Both groups improved on the MSIS-29 psychological subscale, with no statistically significant difference between the groups (pegylated IFN-β-1a: –2.06, 95% CI –3.58 to –0.53; placebo: –2.17, 95% CI –3.63 to –0.70). Participants also completed the Short Form questionnaire-12 items (both the physical component summary and the mental component summary), EQ-5D and EQ-5D visual analogue scale. None of the differences between groups or within groups was statistically significant (the authors did not present specific data), but patients receiving 125 µg of SC pegylated IFN-β-1a every 2 weeks did show a significant improvement on the visual analogue scale (mean change 2.06, 95% CI 0.58 to 3.54).

Adverse events and mortality

The ADVANCE trial213 reported AEs and mortality. Full results are available on request. Differences between groups for mortality were not significant, with one event occurring in the study drug arm and two events occurring in the placebo arm.

Summary of the narrative synthesis: 125 µg of pegylated interferon beta-1a subcutaneously every 2 weeks (Plegridy) compared with placebo

Findings from one study included in this comparison suggested a beneficial effect of 125 µg of SC pegylated IFN-β-1a every 2 weeks compared with placebo for relapse outcomes, disability progression and freedom from disease activity. For HRQoL measures, 125 µg of SC pegylated IFN-β-1a every 2 weeks was not different from placebo. Relapse severity outcomes were not reported. Groups were not significantly different with respect to mortality.

Relapse outcomes

All five trials comparing 20 mg of SC GA once daily against placebo reported relapse rates, as did the trial comparing 40 mg of SC GA three times a week against placebo. The study by Bornstein et al. 170 followed up 48 patients over 2 years. There were 16 relapses over 2 years in the 25 patients receiving 20 mg of SC GA once daily and 62 relapses in the 23 patients receiving placebo, which gives an estimated RR of 0.25 (95% CI 0.14 to 0.43). In another early trial by the Copolymer 1 Multiple Sclerosis Study Group (Cop1 MSSG),217,218 251 patients were followed up over at least 2 years, with an extension of up to 11 months. At 2 years, the ARR in patients receiving 20 mg of SC GA once daily was 0.59, whereas the ARR in patients receiving placebo was 0.84. 217 This difference was statistically significant (p = 0.007). Subsequent studies found similar reductions in ARR in the treatment arms. In a trial by the European/Canadian Glatiramer Acetate Study Group (ECGASG),219 which followed up 239 patients over 9 months, the ARR in the study drug group was 0.81 whereas the ARR in the placebo group was 1.21 (RR 0.67; p = 0.012). The CONFIRM (Comparator and an Oral Fumarate in Relapsing–Remitting Multiple Sclerosis) trial216 followed up 713 patients in the relevant study arms for 2 years and also found a significant difference in ARRs between the groups (20 mg of SC GA once daily 0.29 vs. placebo 0.40; RR 0.71, 95% CI 0.55 to 0.93). However, in a trial following up 357 patients receiving branded GA and 84 patients receiving placebo for 9 months,220 the ARRs were not substantially different between groups (20 mg of SC GA once daily 0.40, 95% CI 0.26 to 0.62 vs. placebo 0.38, 95% CI 0.22 to 0.66), although a standard significance test was not presented. The GALA (Glatiramer Acetate Low-Frequency Administration) trial221 compared 40 mg of SC GA three times a week against placebo (40 mg of SC GA three times a week, n = 943 vs. placebo, n = 461) over 12 months. Patients receiving the study drug had a significantly lower ARR than patients receiving placebo (40 mg of SC GA three times a week 0.33, 95% CI 0.28 to 0.39 vs. placebo 0.51, 95% CI 0.42 to 0.61), with an associated significant RR (0.66, 95% CI 0.54 to 0.80).

Two studies reported time to relapse. Including the extension phase, patients receiving 20 mg of SC GA once daily in the Cop1 MSSG trial218 had a delayed time to first relapse compared with patients receiving placebo, but this difference was not significant (median days to first relapse 287 vs. 198; p = 0.057). However, in the larger CONFIRM trial,216 patients receiving 20 mg of SC GA once daily did have a significant delay in time to relapse (HR 0.71, 95% CI 0.55 to 0.92). In the GALA trial,221 patients receiving 40 mg of SC GA three times a week also had a longer median time to first relapse (393 days vs. 377 days), with a HR of 0.61 (95% CI 0.49 to 0.74).

Finally, greater empirical proportions of patients receiving 20 mg of SC GA once daily tended to be free of relapse than patients receiving placebo, but this trend was not completely consistent. In the study by Bornstein et al. ,170 56% of patients receiving the study drug were relapse free at 2 years compared with 26% of patients receiving placebo (adjusted OR 4.6; p = 0.036). Similarly, the Cop1 MSSG trial218 found that, over the whole trial, patients receiving the study drug were more likely to be free of relapses than those receiving placebo (33.6% vs. 24.6%; p = 0.002). In the ECGASG trial219 this trend did not rise to significance (55.5% vs. 49.2%; OR 1.47, 95% CI 0.84 to 2.56) and in the GATE (Glatiramer Acetate Clinical Trial to Assess Equivalence with Copaxone) trial220 the proportions who were relapse free were not substantially different between the groups (73.9% vs. 73.8%), although a significance test was not provided. In the GALA trial,221 patients receiving 40 mg of SC GA three times a week were more likely to be free of relapses than patients receiving placebo (77.0% vs. 65.5%; OR 1.93, 95% CI 1.49 to 2.49).

Relapse severity

In the ECGASG trial,219 patients receiving 20 mg of SC GA once daily had fewer steroid-treated relapses than those receiving placebo (54 vs. 84). We estimated this as a RR for steroid-treated relapses of 0.65 (95% CI 0.46 to 0.91). The proportion of patients with steroid-treated relapses was correspondingly lower in the study drug arm (33.6% vs. 39.2%), but this was not tested for significance. In the GALA trial,221 patients receiving 40 mg of SC GA three times weekly had a lower ARR (0.30, 95% CI 0.25 to 0.36) for ‘severe’ relapses, defined as steroid-treated or hospitalised relapses, than patients receiving placebo (ARR 0.47, 95% CI 0.38 to 0.57). This translated into a RR of 0.64 (95% CI 0.53 to 0.79).

Disability progression

Three studies170,216,218 presented data on time to disability progression confirmed at 3 months, whereas only the CONFIRM trial216 presented data on time to progression confirmed at 6 months. Studies suggested a beneficial, but generally not significant, impact of 20 mg of SC GA once daily on confirmed disability progression. In the study by Bornstein et al. ,170 the median time to progression confirmed at 3 months was not reached for patients receiving 20 mg of SC GA once daily, but was 18 months for patients receiving placebo. This difference was significant (log-rank p = 0.05). Together with the proportions of patients with progression of 20% in the study drug arm and 48% in the placebo arm, we estimated the HR of progression as 0.37 (95% CI 0.14 to 1.00). In the Cop1 MSSG trial,218 the probability of non-progression was 76.8% in the 20 mg of SC GA once-daily arm and 70.6% in the placebo arm. Using the value from a related significance test (p = 0.199), we estimated the HR as 0.76 (95% CI 0.50 to 1.16). Finally, the CONFIRM trial216 did not find that 20 mg of SC GA once daily slowed time to progression confirmed at 3 months (HR 0.93, 95% CI 0.63 to 1.37). This finding was similar when disability progression was confirmed at 6 months (HR 0.87, 95% CI 0.55 to 1.38).

Only two studies presented data on the proportions of patients with confirmed disability progression in comparisons between 20 mg of SC GA once daily and placebo. As noted above, in the study by Bornstein et al. ,170 20% of patients receiving 20 mg of SC GA once daily progressed over 2 years, whereas 48% of patients receiving placebo progressed. In univariate analyses, this finding was not significant (p = 0.064), but multivariate analyses found a significant effect on the probability of progression (p = 0.033). In the Cop1 MSSG trial,218 the proportion with progression confirmed at 3 months was 23.2% in patients receiving 20 mg of SC GA once daily and 29.4% in patients receiving placebo over the whole trial. In the GALA trial,221 which compared 40 mg of SC GA three times weekly against placebo, 95.5% of patients receiving the study drug were free of confirmed progression compared with 96.3% of patients receiving placebo, but a formal significance test was not presented.

Finally, the magnitude of change in EDSS score was reported by most studies, with changes small across studies. In the study by Bornstein et al. ,170 the findings were presented as the proportions improving or worsening by magnitude of improvement/worsening. We estimated that patients receiving 20 mg of SC GA once daily improved by 0.12 EDSS points and patients receiving placebo worsened by 0.74 EDSS points, with a significant difference between groups (p < 0.05). In the Cop1 MSSG trial,218 patients receiving 20 mg of SC GA once daily did not show a significant improvement in EDSS score (–0.11, 95% CI –0.31 to 0.10) whereas patients receiving placebo showed a significant worsening (0.34, 95% CI 0.13 to 0.54). This difference was statistically significant (p = 0.006). In the ECGASC trial,219 mean EDSS change from baseline was not significantly different between groups (20 mg of SC GA once daily 0.02 vs. placebo 0.05), but a p-value or CIs were not presented. In the GATE trial,220 neither patients receiving the study drug (–0.08, 95% CI –0.19 to 0.03) nor patients receiving placebo (–0.02, 95% CI –0.17 to 0.14) showed a significant improvement in EDSS score. Change in the GALA trial221 was also negligible [40 mg of SC GA three times weekly 0.0 (SD 0.6) vs. placebo 0.1 (SD 0.6)].

Freedom from disease activity

The GATE trial220 was the only study that reported combined clinical–MRI findings for freedom from disease activity. The proportion free from disease activity was slightly greater in the arm receiving 20 mg of SC GA once daily than in the placebo arm (9.2% vs. 7.1%), with similar findings once proportions were adjusted for stratification variables (8.5% vs. 6.6%). A formal significance test was not presented.

Multiple sclerosis symptoms and health-related quality of life

The CONFIRM trial230 presented data for HRQoL disaggregated by subscale of the Short Form questionnaire-36 items (SF-36). Compared with placebo, which showed a negative trend, change from baseline in the 20 mg of SC GA once-daily group was positive for the whole scale and the two groups were significantly different on the physical component summary (p = 0.0259). However, the groups were not significantly different on the mental component summary. The group receiving 20 mg of SC GA once daily significantly improved (p < 0.05) compared with the group receiving placebo with regard to physical functioning (0.3 vs. –2.2), bodily pain (2.3 vs. –1.3) and general health (1.9 vs. –0.6), but not physical (0.3 vs. –2.2) or emotional (1.4 vs. –3.3) aspects of role limitation, vitality (1.1 vs. 0.4), social functioning (–0.6 vs. –0.1) or mental health (0.3 vs. 0.6). Changes in EQ-5D scores were not presented, but scores were stated to be stable in all groups over the course of the study. Compared with patients receiving placebo, patients receiving 20 mg of SC GA once daily were not more likely to have been stable or improved in either the physical component (OR 1.24, 95% CI 0.83 to 1.85) or the mental component (1.22, 95% CI 0.82 to 1.83) of the SF-36.

In the Cop1 MSSG trial,217 at 2 years the mean ambulation index scores were similar between patients receiving 20 mg of SC GA once daily (0.27) and patients receiving placebo (0.28).

Adverse events and mortality

We stratified comparisons by type of placebo. All studies reported AEs but only the GALA,221 GATE220 and CONFIRM216 trials reported deaths. Only one death occurred in trials with matched placebos, in the placebo arm of the GALA trial;221 in the CONFIRM trial,216 one death occurred in each arm. Full results are available on request.

Summary of the narrative synthesis: 20 mg of glatiramer acetate subcutaneously once daily and 40 mg of glatiramer acetate subcutaneously three times a week (Copaxone) compared with placebo

Taken together, findings from the five trials testing 20 mg of SC GA once daily and the one trial testing 40 mg of SC GA three times a week suggested a beneficial effect on relapse outcomes. Both trials (20 mg of GA;219 40 mg of GA221) reporting relapse severity outcomes also found that the study drug decreased the rate of steroid-treated relapses. Findings for disability progression were less convincing and studies generally did not present significant results. Only one study presented combined clinical–MRI measures of freedom from disease activity and this study did not show a large difference between groups, although significance testing was not undertaken. One study showed some effects of 20 mg of SC GA once daily on HRQoL measures. Groups were not significantly different with regard to mortality.

Pairwise meta-analyses

Direct evidence from comparisons against placebo is shown in Figure 5. All drugs had a statistically significant beneficial effect on relapse rate compared with placebo. Findings for 125 µg of SC pegylated IFN-β-1a every 2 weeks, 40 mg of SC GA three times weekly and 22 µg of SC IFN-β-1a three times weekly all relied on one study. Comparisons that relied on multiple studies were diverse in terms of heterogeneity: heterogeneity ranged from an I² of 0% (250 µg of SC IFN-β-1b every other day, 30 µg of IM IFN-β-1a once a week) to an I² of 43% (44 µg of SC IFN-β-1a three times weekly) and 73% (20 mg of SC GA once daily). However, there were too few studies in each comparison to enable exploration of heterogeneity.

FIGURE 5.

Pairwise meta-analyses: ARRs for active drug vs. placebo trials in RRMS.

Direct evidence from comparisons between active drugs is shown in Figure 6. None of the pooled comparisons showed evidence of a statistically significant effect favouring one drug over another. Although several analyses had high I² values each comparison had too few studies to permit exploration of heterogeneity.

FIGURE 6.

Pairwise meta-analyses: ARRs for active drug vs. active drug trials in RRMS.

Network meta-analyses

The set of studies reporting ratios of relapse rates formed a connected network (Figure 7). In the network, all drugs were compared against placebo, but 40 mg of SC GA three times weekly and 125 µg of SC pegylated IFN-β-1a every 2 weeks were not compared against other active drugs in the network; 22 µg of SC IFN-β-1a three times weekly was connected to the network because of its inclusion in the PRISMS trial,189 which also tested the 44-µg dose.

FIGURE 7.

Network of studies: ARRs in RRMS. ga20, 20 mg of SC GA once daily; ga40, 40 mg of SC GA once daily; ifn1a22, 22 µg of SC IFN-β-1a three times weekly; ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day; peg, 125 µg of SC pegylated IFN-β-1a every 2 weeks.

Random-effects NMA generated estimates for each drug compared with placebo and with every other drug (Table 8). Ranking of the drugs suggested that the drug with the highest cumulative probability SUCRA of being the best was 20 mg of SC GA once daily, followed by 125 µg of SC pegylated IFN-β-1a every 2 weeks and 40 mg of SC GA three times weekly, with 30 µg of IM IFN-β-1a once a week ranked second to last and placebo ranked last.

Findings derived from the NMA for comparisons between each drug and placebo substantially mirrored those of the pairwise comparisons and reflected statistically significant reductions in relapse rates in patients receiving active drugs. Pairwise comparisons between drugs mostly revealed little evidence of superiority of one drug over another, although 20 mg of SC GA once daily (RR 0.82, 95% CI 0.73 to 0.92), 44 µg of SC IFN-β-1a three times weekly (RR 0.85, 95% CI 0.76 to 0.95) and 250 µg of SC IFN-β-1b every other day (RR 0.86, 95% CI 0.76 to 0.97) all produced significant reductions in the relapse rate compared with 30 µg of IM IFN-β-1a once a week. These findings from the pairwise comparisons in the NMA, which all included direct (i.e. head-to-head) evidence, were similar in magnitude of effect to findings from the pairwise meta-analyses, but may have benefited from a ‘stabilised’ heterogeneity parameter because of the assumption of equal between-studies variance.

Tests of inconsistency in the network did not suggest disagreement between direct and indirect evidence. A Wald test for overall inconsistency derived from a design*treatment interaction model was not statistically significant (p = 0.38) and comparisons between the direct and the indirect evidence derived from the side-splitting model did not show any statistically significant differences.

Sensitivity analyses

Several characteristics of the trials included in this network suggested that additional analyses would confirm the robustness of our findings. All of these analyses were post hoc. First, we excluded the REFORMS trial197 from the analysis, as it was the only study in which relapses were self-reported by subjects instead of documented by an examining neurologist. Effect estimates remained essentially unchanged for all pairwise comparisons.

Second, we compared findings from studies with ‘true’, blinded placebos with findings from studies that did not use blinded placebos, that is, studies that did not deliver placebos using the same route of administration as for the study drugs. Specifically, the BRAVO,198 CONFIRM216 and Kappos et al. 199 trials did not administer placebo by the same route as the relevant IFN or GA arm in each trial. We found that the effects of these drugs compared with placebo were robust to inclusion of a covariate in the model for trials without a blinded placebo.

Third, we noticed that the study by Bornstein et al. 170 was an outlier in the comparison between 20 mg of SC GA once daily and placebo. When we excluded this trial from the pairwise meta-analysis, the pooled RR for relapses still suggested a reduction in ARR compared with placebo (RR 0.71, 95% CI 0.62 to 0.82), with an I² of 0%. Re-estimation of the NMA yielded a change in the SUCRA-based rankings, with 20 mg of SC GA once daily now ranked third, but point estimates and CIs were not substantially different in the new model (Table 9).

Pairwise meta-analyses

Direct evidence from pairwise comparisons is shown in Figure 8. Each comparison was informed by one study. All drugs compared with placebo had a statistically significant beneficial effect in terms of reducing the rate of moderate or severe relapses. In comparisons based on active drugs, there was no evidence that one dose of SC IFN-β-1a three times weekly was statistically better than the other (44 µg vs. 22 µg) or that 250 µg of SC IFN-β-1b every other day was different from 20 mg of SC GA once daily. The active drugs 40 mg of SC GA three times weekly, 30 µg of IM IFN-β-1a once a week and 125 µg of SC pegylated IFN-β-1a every 2 weeks were not represented in this analysis.

FIGURE 8.

Pairwise estimates: ARRs for moderate or severe relapses in RRMS.

Network meta-analyses

The set of studies reporting ratios of relapse rates for moderate and severe relapses formed a connected network (Figure 9). In the network, direct evidence for 20 mg of SC GA once daily was only available for the comparison with 250 µg of SC IFN-β-1b every other day.

FIGURE 9.

Network of studies: ARRs for moderate or severe relapses in RRMS. ga20, 20 mg of SC GA once daily; ifn1a22, 22 µg of SC IFN-β-1a three times weekly; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day.

Because of the shape of the network, in which there was no opportunity for inconsistency and in which no direct comparison was informed by more than one trial, the model was estimated using fixed effects instead of random effects, as in the protocol. Ranking of drugs suggested that 20 mg of SC GA once daily was best, followed by 250 µg of SC IFN-β-1b every other day and 44 µg and 22 µg of SC IFN-β-1a three times weekly, with placebo ranked last (Table 10).

Findings derived from the NMA for comparisons between each drug and placebo were similar to findings from comparisons between each drug and placebo from the direct evidence, as would be expected. In an indirect comparison, 20 mg of SC GA once daily reduced the rate of moderate and severe relapses compared with placebo (RR 0.48, 95% CI 0.31 to 0.76). Pairwise comparisons between active drugs did not yield evidence of the superiority of any one drug over another.

Because there was no possibility of inconsistency in the network, we did not test for it.

Pairwise meta-analysis

Direct evidence from comparisons against placebo is shown in Figure 10. Each comparison was informed by one study. All drugs that were compared with placebo had a significant effect in terms of reducing the rate of steroid-treated relapses. In head-to-head comparisons between active drugs, 44 µg of SC IFN-β-1a three times weekly produced a greater reduction in steroid-treated relapses than the 22-µg dose of the same drug (RR 0.77, 95% CI 0.67 to 0.89) and 30 µg of IM IFN-β-1a once a week (RR 0.68, 95% CI 0.51 to 0.91). Pairwise comparisons between 30 µg of IM IFN-β-1a once a week and 250 µg of SC IFN-β-1b every other day and between 44 µg of SC IFN-β-1a three times weekly and 20 mg of SC GA once daily did not show statistical evidence of superiority. The active drug 125 µg of SC pegylated IFN-β-1a every 2 weeks was not included in this analysis.

FIGURE 10.

Pairwise estimates: ARRs for steroid-treated relapses in RRMS.

Network meta-analyses

The set of studies reporting ratios of steroid-treated relapse rates formed a connected network (Figure 11). In the network, each comparison was informed by one study, but there were closed loops between studies, suggesting the possibility of inconsistency. Because in this parameterisation of the model inconsistency is regarded as a source of heterogeneity, even though there is no potential for heterogeneity in any of the comparisons informed by direct evidence, we estimated the model as both a fixed-effects and a random-effects model.

FIGURE 11.

Network of studies: ARR for steroid-treated relapses in RRMS. ga20, 20 mg of SC GA once daily; ga40, 40 mg of SC GA three times weekly; ifn1a22, 22 µg of SC IFN-β-1a three times weekly; ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day.

Numerical estimates of intervention effectiveness were not meaningfully different between the random-effects model and the fixed-effects model (Table 11). However, the random-effects model did not support a significant reduction in the rate of steroid-treated relapses with 250 µg of SC IFN-β-1b every other day (fixed-effects RR 0.62, 95% CI 0.40 to 0.98; random-effects RR 0.64, 95% CI 0.36 to 1.14). The random-effects model also did not support the superiority of any one drug over another, except for 44 µg of SC IFN-β-1a three times weekly over 30 µg of IM IFN-β-1a once a week (RR 0.68, 95% CI 0.48 to 0.97). However, in the fixed-effects model, 44 µg of SC IFN-β-1a three times weekly was superior to both 30 µg of IM IFN-β-1a once a week (RR 0.68, 95% CI 0.51 to 0.91) and 22 µg of SC IFN-β-1a three times weekly (RR 0.79, 95% CI 0.68 to 0.91), both of which comparisons were informed by direct evidence; 20 mg of SC GA once daily was also superior to both 30 µg of IM IFN-β-1a once a week (RR 0.67, 95% CI 0.47 to 0.95) and 22 µg of SC IFN-β-1a three times weekly (RR 0.77, 95% CI 0.61 to 0.98), although neither comparison was informed by direct evidence.

Because the overall Wald test for inconsistency did not provide evidence of a difference between direct and indirect evidence (p = 0.20), the fixed-effects model may be preferable.

Pairwise meta-analyses

Direct evidence from comparisons is shown in Figure 12. Only one comparison, 20 mg of SC GA once daily compared with placebo, included more than one study and 40 mg of SC GA three times weekly was not represented in this analysis.

FIGURE 12.

Pairwise meta-analyses: time to disability progression confirmed at 3 months in RRMS.

Comparison of drugs against placebo showed a mixed pattern of results. The active drugs 20 mg of SC GA once daily (HR 0.79, 95% CI 0.60 to 1.05), 30 µg of IM IFN-β-1a once a week (HR 0.74, 95% CI 0.51 to 1.08) and 250 µg of SC IFN-β-1b every other day (HR 0.71, 95% CI 0.48 to 1.06) did not show evidence of delaying disability progression. However, both 44 µg of SC IFN-β-1a three times weekly (HR 0.62, 95% CI 0.43 to 0.90) and 22 µg of SC IFN-β-1a three times weekly (HR 0.68, 95% CI 0.48 to 0.97) and 125 µg of SC pegylated IFN-β-1a every 2 weeks (HR 0.62, 95% CI 0.40 to 0.97) did show evidence of delaying disability progression. None of the three direct comparisons between active drugs suggested a benefit of one over another.

Network meta-analyses

The set of studies reporting HRs for time to disability progression confirmed at 3 months formed a connected network (Figure 13). In the network, all active drugs were compared against placebo and three comparisons between active drugs were present as well.

FIGURE 13.

Network of studies: time to disability progression confirmed at 3 months in RRMS. ga20, 20 mg of SC GA once daily; ifn1a22, 22 µg of SC IFN-β-1a three times weekly; ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day; peg, 125 µg of SC pegylated IFN-β-1a every 2 weeks.

The NMA, which was estimated with random effects as per the protocol, generated HR estimates for each drug compared with placebo and with every other drug (Table 12). Ranking of the drugs suggested that the drug with the highest cumulative probability of being the best was 44 µg of SC IFN-β-1a three times weekly, followed by 125 µg of SC pegylated IFN-β-1a every 2 weeks and 22 µg of SC IFN-β-1a three times weekly, with 250 µg of SC IFN-β-1b every other day ranked second to last and placebo ranked last.

Comparisons of active drugs with placebo were similar between the NMA and the pairwise meta-analyses. Notably, additional information from indirect comparisons yielded a more precise estimate of effectiveness for both 30 µg of IM IFN-β-1a once a week compared with placebo (HR 0.73, 95% CI 0.53 to 1.00; p = 0.0499) and 20 mg of SC GA once daily compared with placebo (HR 0.76, 95% CI 0.60 to 0.97). Comparisons between active drugs in the NMA did not indicate that any one drug was statistically better than any other, as all pairwise comparisons were not statistically significant.

Tests of inconsistency in the network did not suggest any disagreement between direct and indirect evidence. An overall Wald test derived from a design*treatment interaction model returned a non-significant result (p = 0.84) and comparisons between the direct evidence and the indirect evidence derived from the side-splitting model did not show any statistically significant differences.

Pairwise meta-analyses

Direct evidence from comparisons is shown in Figure 14. All comparisons were based on a single study, except for 30 µg of IM IFN-β-1a once a week compared with placebo. The active drug 40 mg of SC GA three times weekly was not represented in this analysis.

FIGURE 14.

Pairwise meta-analyses: time to disability progression confirmed at 6 months in RRMS.

Three drugs were compared against placebo: 20 mg of SC GA once daily did not delay confirmed disability progression compared with placebo, but 30 µg of IM IFN-β-1a once weekly (HR 0.66, 95% CI 0.47 to 0.92) and 125 µg of SC pegylated IFN-β-1a every 2 weeks (HR 0.46, 95% CI 0.26 to 0.81) did. Of the three comparisons between active drugs, only 30 µg of IM IFN-β-1a once a week yielded a significant difference, when compared with 250 µg of SC IFN-β-1b every other day.

Network meta-analyses

The set of studies reporting HRs for time to disability progression confirmed at 6 months formed a connected network (Figure 15). In the network, 250 µg of SC IFN-β-1b every other day and 44 µg of SC IFN-β-1a three times weekly are not compared with placebo, but only with other active drugs.

FIGURE 15.

Network of studies: time to disability progression confirmed at 6 months in RRMS. ga20, 20 mg of SC GA once daily; ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day; peg, 125 µg of SC pegylated IFN-β-1a every 2 weeks.

The NMA, which was estimated with random effects as per the protocol, generated HR estimates for each drug compared with placebo and with every other drug (Table 13). Ranking of the drugs suggested that the drug with the highest cumulative probability of being the best was 250 µg of SC IFN-β-1b every other day, followed by 125 µg of SC pegylated IFN-β-1a every 2 weeks, 44 µg of SC IFN-β-1a three times weekly and 30 µg of IM IFN-β-1a once a week; 20 mg of SC GA once daily was ranked second to last and placebo was ranked last.

When compared with placebo in the NMA, 20 mg of SC GA once daily had a similar estimate of effectiveness (HR 0.82, 95% CI 0.53 to 1.26) as in the direct evidence, as did 30 µg of IM IFN-β-1a once a week (HR 0.68, 95% CI 0.49 to 0.94) and 125 µg of SC pegylated IFN-β-1a every 2 weeks (HR 0.46, 95% CI 0.26 to 0.81). Both 44 µg of SC IFN-β-1a three times weekly (HR 0.47, 95% CI 0.24 to 0.93) and 250 µg of SC IFN-β-1b every other day (HR 0.34, 95% CI 0.18 to 0.63) showed evidence of delaying disability progression compared with placebo. However, both of these estimates are based solely on indirect evidence and findings from the INCOMIN trial,196 which informed the comparison between 250 µg of SC IFN-β-1b every other day and 30 µg of IM IFN-β-1a once a week and relied on a HR estimated from summary statistics.

Comparisons between active drugs in from the NMA suggested that 250 µg of SC IFN-β-1b every other day is superior to both 30 µg of IM IFN-β-1a once a week (HR 0.50, 95% CI 0.29 to 0.87) and 20 mg of SC GA once daily (HR 0.41, 95% CI 0.21 to 0.83). In particular, the result of the comparison between 250 µg of SC IFN-β-1b every other day and 20 mg of SC GA once daily was greater in magnitude than the direct evidence suggested. No other comparisons between active drugs yielded statistically significant evidence of superiority of one drug over any other.

Tests of inconsistency in the network did not suggest that direct and indirect evidence disagreed to a statistically significant level; however, the network was sparse and only one comparison included more than one study. An overall Wald test for inconsistency returned a statistically non-significant result (p = 0.38).

Summary of the adverse events meta-analyses

Full results for pairwise meta-analyses of AEs are available on request. Although the diversity and heterogeneity of AEs precluded detailed examination of each, several trends were apparent across pairwise comparisons:

• comparing 30 µg of IFN-β-1a (Avonex) with equivalent placebo, 30 µg of IFN-β-1a was associated with more chills, flu-like symptoms, NABs and myalgia

• comparing 30 µg of IFN-β-1a (Avonex) with 44 µg of SC IFN-β-1a (Rebif), 44 µg of IFN-β-1a was associated with more injection site reactions, liver disorders, NABs and white blood cell abnormalities, whereas 30 µg of IFN-β-1a was associated with more fatigue

• comparing 30 µg of IFN-β-1a (Avonex) with IFN-β-1b (Betaferon/Extavia), IFN-β-1b was associated with more injection site reactions and NABs

• comparing 30 µg of IFN-β-1a (Avonex) with GA (Copaxone), there were no significant differences in AEs

• comparing 44 µg of IFN-β-1a (Rebif) with placebo, 44 µg of IFN-β-1a was associated with more injection site reactions, flu-like symptoms, liver disorders, granulocytopenia, leucopenia, lymphopenia and NABs

• comparing 44 µg of IFN-β-1a (Rebif) with IFN-β-1b (Betaferon/Extavia), 44 µg of IFN-β-1a was associated with more alanine aminotransferase disorders and IFN-β-1b was associated with more injection site pain

• comparing 44 µg of IFN-β-1a (Rebif) with GA (Copaxone), 44 µg of IFN-β-1a was associated with more liver enzyme disorders, NABs, headache, flu-like symptoms and myalgia and GA was associated with more injection site reactions, immediate post-injection reactions and binding antibodies

• comparing IFN-β-1b (Betaferon/Extavia) with placebo, IFN-β-1b was associated with more injection site inflammation and NABs

• comparing IFN-β-1b (Betaferon/Extavia) with GA (Copaxone), IFN-β-1b was associated with more flu-like symptoms, insomnia and disordered liver enzymes and GA was associated with more injection site reactions, itching, pain, inflammation and induration and immediate post-injection reactions

• comparing GA (Copaxone) with equivalent placebo, GA was associated with more injection site induration, itching, injection site mass, erythema, pain, inflammation and reactions and more immediate post-injection systemic reactions

• comparing pegIFN-β-1a (Plegridy) with placebo, pegylated IFN-β-1a was associated with more injection site erythema, pain, itching, chills and/or fever, headache, flu-like symptoms, myalgia, pyrexia, any AE possibly related to the drug, patients who discontinued the study because of AEs and severe AEs.

Discontinuation because of adverse events: modal follow-up

Pairwise meta-analyses

Pairwise meta-analyses for discontinuation because of AEs, combined across studies at the modal follow-up, are presented in Figure 16. The modal follow-up was approximately 24 months and thus we included studies with an intended follow-up period around this point. We included 12 estimates in these meta-analyses. There was no visual evidence of a systematic difference between the groups based on the strict definition of the outcome. In every pairwise meta-analysis, CIs were wide, as would be expected. Three pooled estimates relied on multiple studies: 20 mg of SC GA once daily compared with placebo, 30 µg of IM IFN-β-1a once a week compared with placebo and 250 µg of SC IFN-β-1b every other day compared with 20 mg of SC GA once daily. There was no evidence in this analysis for 40 mg of SC GA three times weekly or 125 µg of SC pegylated IFN-β-1a every 2 weeks.

FIGURE 16.

Pairwise meta-analyses: discontinuation because of AEs at 24 months in RRMS. RR, risk ratio.

Direction of effect in pooled estimates and single studies generally suggested that discontinuation because of AEs was more likely in study arms testing active drugs than in study arms testing placebo, but these findings were generally not statistically significant. The one exception was the IFNB MSSG trial,210 from which we used 24-month data. In this study, which compared 250 µg of SC IFN-β-1b every other day with placebo, patients receiving the study drug were more likely to withdraw from the study because of an AE (risk ratio 9.92, 95% CI 1.29 to 76.32).

Network meta-analyses

The set of studies included in the NMA formed a connected network (Figure 17). All drugs were compared with placebo; 40 mg of SC GA three times weekly and 125 µg of SC pegylated IFN-β-1a every 2 weeks were not included in this analysis.

FIGURE 17.

Network of studies: discontinuation because of AEs at 24 months in RRMS. ga20, 20 mg of SC GA once daily; ifn1a22, 22 µg of SC IFN-β-1a three times weekly; ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day.

The NMA, which was estimated with random effects, generated estimates for each drug compared with placebo and with every other drug (Table 14). Because CIs were wide in pairwise, direct meta-analyses, CIs were wide in the NMA, and estimates compared with placebo were often numerically different. The NMA did not offer statistical evidence that any one drug was more likely to result in discontinuation because of AEs than any other drug. Based on SUCRAs, 250 µg of SC IFN-β-1b every other day was ranked highest for discontinuation because of AEs, followed by 44 µg of SC IFN-β-1a three times weekly. Placebo was ranked last.

TABLE 14 - Network meta-analysis: discontinuation because of AEs at 24 months in RRMSa

Findings are presented as risk ratio (95% CI).

Drug	SUCRA	IFN-β-1b 250 µg SC every other day	IFN-β-1a 44 µg SC three times weekly	GA 20 mg SC once daily	IFN-β-1a 22 µg SC three times weekly	IFN-β-1a 30 µg IM weekly	Placebo
IFN-β-1b 250 µg SC every other day	0.79		1.15 (0.20 to 6.56)	1.70 (0.50 to 5.81)	2.37 (0.22 to 25.84)	2.74 (0.56 to 13.38)	4.41 (1.07 to 18.29)
IFN-β-1a 44 µg SC three times weekly	0.76			1.48 (0.39 to 5.57)	2.07 (0.32 to 13.44)	2.39 (0.38 to 15.22)	3.85 (0.81 to 18.29)
GA 20 mg SC once daily	0.57				1.40 (0.17 to 11.76)	1.61 (0.38 to 6.91)	2.60 (0.88 to 7.64)
IFN-β-1a 22 µg SC three times weekly	0.41					1.15 (0.10 to 13.09)	1.86 (0.21 to 16.83)
IFN-β-1a 30 µg IM weekly	0.35						1.61 (0.52 to 5.02)
Placebo	0.12
Wald test for inconsistency (χ2, df, p-value)	2.38, 3, 0.50

In comparison with the direct evidence from the IFNB MSSG trial,210 the risk ratio for discontinuation because of AEs for 250 µg of SC IFN-β-1b every other day compared with placebo was lower but remained statistically significant (risk ratio 4.41, 95% CI 1.07 to 18.29). The risk ratio for 44 µg of SC IFN-β-1a three times weekly compared with placebo was lower in the NMA (risk ratio 3.85, 95% CI 0.81 to 18.29) than the pairwise estimate derived from the PRISMS trial189 (risk ratio 7.11, 95% CI 0.88 to 57.25), as was the risk ratio for 22 µg of SC IFN-β-1a three times weekly compared with placebo (NMA: risk ratio 1.86, 95% CI 0.21 to 16.83; PRISMS trial:189 risk ratio 2.97, 95% CI 0.31 to 28.28). However, the risk ratio for 20 mg of SC GA once daily compared with placebo was higher in the NMA (risk ratio 2.60, 95% CI 0.88 to 7.64) than in the pairwise meta-analysis (risk ratio 1.69, 0.51 to 5.58).

An overall test for inconsistency across the network did not suggest the presence of inconsistency (p = 0.50). However, a side-splitting test did find that direct and indirect evidence were in conflict for the comparison between 20 mg of SC GA once daily and placebo, with a suggestion that the risk of discontinuation because of AEs in the indirect evidence was higher than that presented in the direct evidence (p = 0.037). Thus, there is some evidence of inconsistency in this network.

Discontinuation because of adverse events: all follow-up times

Pairwise meta-analyses

Pairwise meta-analyses for discontinuation because of AEs across all time points are shown in Figure 18. There was no visual evidence of a systematic difference between the groups based on the strict definition of the outcome. In every pairwise meta-analysis, CIs were wide, as would be expected. Five pooled estimates relied on multiple studies: 20 mg of SC GA once daily compared with placebo, 30 µg of IM IFN-β-1a once a week compared with placebo and 250 µg of SC IFN-β-1b every other day compared with each placebo, 20 mg of SC GA once daily and 44 µg of SC IFN-β-1a three times weekly.

FIGURE 18.

Pairwise meta-analyses: discontinuation because of AEs at all time points in RRMS. RR, risk ratio.

Despite visual evidence suggesting that discontinuation because of AEs was more likely in study arms testing active drugs than in study arms testing placebo, almost all individual study estimates and pooled estimates did not suggest that discontinuation was more likely in trial arms corresponding to one drug over another to a statistically significant level. The one exception was 125 µg of SC pegylated IFN-β-1a every 2 weeks compared with placebo, in which patients receiving the study drug were more likely to discontinue the study because of AEs (risk ratio 3.49, 95% CI 1.52 to 7.99). Estimates for 40 mg of SC GA three times weekly compared with placebo were marginally non-significant (risk ratio 2.36, 95% CI 0.99 to 5.65). Again, both estimates were based on one study. Of note is that comparisons between 20 mg of SC GA once daily and placebo, which included five studies, did not suggest a substantial relationship between the study drug and discontinuation (risk ratio 1.07, 95% CI 0.64 to 1.79), but this was driven (at least in part) by the null finding from the CONFIRM trial216 (risk ratio 0.95, 95% CI 0.62 to 1.47).

Network meta-analyses

The studies included in this analysis formed a connected network (Figure 19). All drugs were compared with placebo and all drugs were included in this analysis.

FIGURE 19.

Network of studies: discontinuation because of AEs at all time points in RRMS. ga20, 20 mg of SC GA once daily; ga40, 40 mg of SC GA three times weekly; ifn1a22, 22 µg of SC IFN-β-1a three times weekly; ifn1a30, 30 µg of IM IFN-β-1a once a week; ifn1a44, 44 µg of SC IFN-β-1a three times weekly; ifn1b250, 250 µg of SC IFN-β-1b every other day; peg, 125 µg of SC pegylated IFN-β-1a every 2 weeks.

The NMA, which was estimated with random effects as per the protocol, generated estimates for each drug compared with placebo and with every other drug (Table 15). The NMA did not offer statistical evidence that any one drug was more likely to result in discontinuation because of AEs than any other drug. Based on SUCRAs, 125 µg of SC pegylated IFN-β-1a every 2 weeks was ranked highest for risk of discontinuation because of AEs (i.e. greatest risk of discontinuation), followed by 44 µg of SC IFN-β-1a three times weekly. Placebo was ranked last.

TABLE 15 - Network meta-analysis: discontinuation because of AEs at all time points in RRMSa

df, degrees of freedom.

Findings are presented as risk ratio (95% CI).

Drug	SUCRA	Pegylated IFN-β-1a 125 µg SC every 2 weeks	IFN-β-1a 44 µg SC three times weekly	GA 40 mg SC three times weekly	IFN-β-1b 250 µg SC every other day	IFN-β-1a 30 µg IM weekly	GA 20 mg SC once daily	IFN-β-1a 22 µg SC three times weekly	Placebo
Pegylated IFN-β-1a 125 µg SC every 2 weeks	0.82		1.40 (0.31 to 6.45)	1.48 (0.29 to 7.43)	1.99 (0.43 to 9.15)	2.15 (0.57 to 8.04)	2.24 (0.59 to 8.44)	2.82 (0.35 to 23.04)	3.49 (1.13 to 10.76)
IFN-β-1a 44 µg SC three times weekly	0.73			1.05 (0.22 to 4.95)	1.42 (0.61 to 3.30)	1.53 (0.65 to 3.59)	1.60 (0.76 to 3.36)	2.01 (0.45 to 9.01)	2.49 (0.89 to 6.95)
GA 40 mg SC three times weekly	0.66				1.35 (0.29 to 6.35)	1.45 (0.38 to 5.60)	1.52 (0.39 to 5.89)	1.91 (0.23 to 15.88)	2.36 (0.74 to 7.53)
IFN-β-1b 250 µg SC every other day	0.50					1.08 (0.42 to 2.79)	1.12 (0.51 to 2.49)	1.42 (0.26 to 7.71)	1.75 (0.63 to 4.89)
IFN-β-1a 30 µg IM weekly	0.45						1.04 (0.51 to 2.13)	1.32 (0.24 to 7.17)	1.62 (0.82 to 3.23)
GA 20 mg SC once daily	0.40							1.26 (0.24 to 6.50)	1.56 (0.77 to 3.14)
IFN-β-1a 22 µg SC three times weekly	0.33								1.24 (0.21 to 7.26)
Placebo	0.12
Wald test for inconsistency (χ2, df, p-value)	11.04, 6, 0.09

Because CIs were frequently wide in pairwise, direct meta-analyses, CIs were wide in the NMAs, and estimates compared with placebo were often numerically different. Compared with direct estimates from the PRISMS trial,189 evidence from the NMA suggested a numerically lower risk of discontinuation because of AEs for 44 µg of SC IFN-β-1a three times weekly compared with placebo (NMA: risk ratio 2.49, 95% CI 0.89 to 6.95; PRISMS trial:189 risk ratio 7.11, 95% CI 0.88 to 57.25), that is, the magnitude of the risk of discontinuation compared with placebo was smaller in the NMA than in the one trial informing the direct comparison. The same applied for 22 µg of SC IFN-β-1a three times weekly (NMA: risk ratio 1.24, 95% CI 0.21 to 7.26; PRISMS trial:189 risk ratio 2.97, 95% CI 0.31 to 28.28). Similarly, estimates for discontinuation because of AEs for 250 µg of SC IFN-β-1b every other day compared with placebo were lower in the NMA than in the pairwise meta-analysis (NMA: risk ratio 1.75, 95% CI 0.63 to 4.89; pairwise meta-analysis: risk ratio 4.93, 95% CI 0.76 to 32.00). Estimates of discontinuation because of AEs were higher in the NMA for 20 mg of SC GA once daily compared with placebo (NMA: risk ratio 1.56, 95% CI 0.77 to 3.14; pairwise meta-analysis: risk ratio 1.07, 95% CI 0.64 to 1.79).

An overall Wald test for inconsistency in the network did not reach significance, but suggested some conflict between direct and indirect evidence (p = 0.09). Examination of the specific design effects from the design*treatment interaction model suggested that direct estimates of discontinuation because of AEs for 250 µg of SC IFN-β-1b every other day compared with placebo could be driving this result (design effect p = 0.075). However, a side-splitting test did not suggest an obvious source of conflict between direct and indirect evidence. Thus, although there is no statistically significant evidence of inconsistency in this network, the findings should be viewed with caution.

Comparison of network meta-analyses: modal follow-up compared with all time points

Neither NMA found evidence that one drug was superior to any other.

However, estimates for discontinuation because of AEs for active drugs compared with placebo tended to be lower in the network including all time points, possibly because the majority of studies included in this analysis that were set aside in the modal follow-up analysis included shorter follow-up periods (generally of ≤ 1 year). Estimates were essentially unchanged for 30 µg of IM IFN-β-1a once a week compared with placebo (modal follow-up: risk ratio 1.61, 95% CI 0.52 to 5.02; all time points: risk ratio 1.62, 95% CI 0.82 to 3.23).

Relapse outcomes

In the SPECTRIMS trial,224 618 patients were followed up for 3 years. RRs based on ARRs were numerically identical for both active arms compared with placebo (44 µg: RR 0.69, 95% CI 0.56 to 0.85; 22 µg: RR 0.69, 95% CI 0.56 to 0.84).

Subgroup analyses stratifying by whether patients had a history of relapse showed a pattern of significant results for those previously relapsing and non-significant results for those not previously relapsing. 224 For those previously relapsing, ARRs for the 44-µg dose (0.67; p < 0.001) and the 22-µg dose (0.57; p < 0.001) were significantly different from the ARR in the placebo arm (1.08). For those not previously relapsing, ARRs for both dosages (44 µg: 0.43, p > 0.05; 22 µg: 0.36, not significant) were not significantly different from the ARR in the placebo arm (0.39).

Both active arms also showed a similar delay in time to first relapse, although only the 44-µg dose had a significant effect compared with placebo (HR 0.77, 95% CI 0.61 to 0.98), corresponding to a median time to first relapse of 494 days in the 44-µg dose group compared with 281 days in the placebo group. 224 Although the results for the 22-µg dose group compared with the placebo group were similar (476 days compared with 281 days), this did not translate into a significant effect (HR 0.87, 95% CI 0.69 to 1.10). The difference between the two active arms was not calculated in this trial, although the HR for the 44-µg dose compared with the 22-µg dose can be approximated as 0.77/0.87 = 0.89, which is not statistically different from unity.

Relapse severity

Both active arms in the SPECTRIMS trial224 showed a similar reduction in the annualised rate of moderate or severe relapses (44 µg: RR 0.68, 95% CI 0.44 to 0.81; 22 µg: RR 0.66, 95% CI 0.51 to 0.86). 224 The findings were similar for annualised rates of steroid courses used to treat relapses (44 µg: RR 0.66, 95% CI 0.49 to 0.89; 22 µg: RR 0.59, 95% CI 0.44 to 0.81).

Disability progression

In the SPECTRIMS trial,224 disability progression was confirmed at 3 months. Neither active drug was associated with a significant decrease in hazard for time to confirmed disability progression in the main analysis (44 µg: HR 0.83, 95% CI 0.65 to 1.07; 22 µg: 0.88, p = 0.305), nor were the active arms substantially different. However, an analysis controlling for disease characteristics found a significant difference in the 44-µg arm (HR 0.78, 95% CI 0.60 to 1.00) compared with the control arm.

Subgroup analyses combined the two active drug dosages into one arm and stratified models by whether patients had a history of relapse. 224 The HR for time to confirmed disability progression suggested a positive, although non-significant, effect in previously relapsing patients (0.74; p = 0.055), whereas the HR approached unity in non-relapsing patients (1.01; p = 0.934). However, among previously relapsing patients, the proportion of patients with confirmed disability progression was significantly different between those receiving 44/22 µg of the active drug and those receiving placebo (OR 0.52, 95% CI 0.29 to 0.93), whereas among those not previously relapsing, the proportion of patients with confirmed disability progression was not significantly different between those receiving 44/22 µg of the active drug and those receiving placebo (OR 1.07, 95% CI 0.64 to 1.78).

Freedom from disease activity

We were unable to locate any relevant comparisons between 44 µg or 22 µg of SC IFN-β-1a three times a week and placebo for combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

We were unable to locate any relevant comparisons between 44 µg or 22 µg of SC IFN-β-1a three times a week and placebo for MS symptoms and HRQoL.

Adverse events and mortality

The SPECTRIMS trial224 reported AEs and mortality. Full results are available on request. Mortality was not significantly different between the groups: one patient died in the placebo arm whereas two patients died in the 44-µg arm and one patient died in the 22-µg arm.

Relapse outcomes

In the European Study Group on Interferon β-1b in Secondary Progressive MS (ESG) trial,222,225 718 patients were followed for up to 2 years. Patients receiving the study drug had a significantly lower ARR than those in the placebo arm (0.42 vs. 0.57; p = 0.003). We approximated this as a RR of 0.74 (95% CI 0.65 to 0.83). Similarly, for the 623 patients enrolled in the relevant study arms in the NASG trial223 and followed for up to 3 years before early study termination, patients receiving the study drug had a significantly lower ARR than placebo patients (0.16 vs. 0.28; p = 0.009). We approximated this as a RR of 0.57 (95% CI 0.43 to 0.75).

Both studies also demonstrated a statistically significant delay in time to first relapse in the active drug arm. In interim data from the ESG trial,222 the median time to first relapse was 644 days in the study drug arm compared with 403 days in the placebo arm (log-rank p = 0.003). In the NASG trial,223 end-of-study data demonstrated a time to relapse at the 30th percentile of 1051 days in the study drug arm compared with 487 days in the placebo arm (log-rank p = 0.01). However, in the ESG trial222 the proportions relapsing were not significantly different between the groups (57.5% in the study drug arm vs. 62.0% in placebo, p = 0.083), whereas in the NASG trial there was a significant difference between the groups (29% vs. 38%; p = 0.018).

Relapse severity

Both studies showed a significant difference between the study drug arm and the placebo arm in the proportion of patients experiencing moderate or severe relapses (ESG trial interim data:222 43.6% vs. 53.1%, p = 0.0083; NASG trial:223 21% vs. 30%, p = 0.012). In the NASG trial,223 the annualised rate of moderate or severe relapses was significantly lower in the study drug arm than in the placebo arm (0.10 vs. 0.19; p = 0.022). However, it should be noted that outcome tables for the NASG trial presented two markedly different estimates of relapse severity. Under the second set of estimates, neither the proportion of patients with moderate or severe relapses (3% vs. 6%; p = 0.056) or the annualised rate of moderate or severe relapses (0.01 vs. 0.02; p = 0.052) was significantly different between the arms. Contact with the study investigators did not yield clarification.

In both studies, the percentage of patients treated with steroids also decreased significantly in the study drug arm compared with the placebo arm (ESG trial interim data:222 53.6% vs. 67.9%, p < 0.0001; NASG trial:223 37% vs. 46%, p = 0.023).

Disability progression

In the ESG trial,225 progression was measured using a variety of criteria, including progression of at least 1.0 EDSS points confirmed at 3 months and at 6 months and progression of 2.0 EDSS points confirmed at 3 months. Each of these measures was estimated both excluding data collected during relapses (the default) and including relapse data, but proportions were similar in all cases between measures including and measures excluding data collected during relapses and thus only the default measures are discussed here. The proportion of patients progressing at least 1.0 EDSS point confirmed at 3 months was significantly lower in the study drug arm than in the placebo arm (45.3% vs. 53.9%; p = 0.031). Combined with estimated probabilities from a life table model (estimated non-progression at 33 months 53% vs. 44%) and a log-rank p-value of 0.003, this yielded an approximate HR of 0.75 (95% CI 0.61 to 0.92). The proportions with confirmed progression of at least 1.0 EDSS point at 6 months (40.8% vs. 48.6%; p = 0.049) and with confirmed progression of at least 2.0 EDSS points at 3 months (16.4% vs. 22.6%; p = 0.032) showed similar trends. However, in the NASG trial,223 disability progression was confirmed at 6 months and did not show a significant difference in terms of time to progression (study drug 32% vs. placebo 34%; log-rank p = 0.61).

Similarly, although patients in the ESG trial225 showed a significant difference between arms in the average EDSS progression score (0.47 vs. 0.69 points; p = 0.003), patients in the NASG trial223 did not (0.53 vs. 0.62 points; p = 0.634).

Freedom from disease activity

We were unable to locate any relevant comparisons between 250 µg of SC IFN-β-1b every other day and placebo for combined clinical–MRI measures of freedom from disease activity.

Multiple sclerosis symptoms and health-related quality of life

In the NASG trial,223 change from baseline was not significantly different between patients in the study drug arm and patients in the placebo arm for fatigue (Environmental Status Scale score change 1.7 vs. 1.2; p = 0.125), cognition (composite neuropsychological score change –0.28 vs. –0.32; p = 0.42) or depression (Beck Depression Inventory score change –0.5 vs. –1.0; p = 0.652; percentage newly treated with antidepressants 29% vs. 29%; p = 0.987). Changes in overall Multiple Sclerosis Quality of Life Inventory scores were also not significantly different (p = 0.502).

Adverse events and mortality

Both studies reported AEs and mortality. Full results are available on request. Studies were not significantly different with regard to mortality: there were seven deaths in the combined 250 µg of SC IFN-β-1b every other day arms of the two trials and two deaths in the combined placebo arms.

Pairwise meta-analyses

Direct evidence from comparisons is shown in Figure 20. The SPECTRIMS trial224 compared 44 µg of SC IFN-β-1a three times weekly, 22 µg of SC IFN-β-1a three times weekly and placebo, whereas the other two included trials compared 250 µg of SC IFN-β-1b every other day with placebo. The pooled effect of 250 µg of SC IFN-β-1b every other day compared with placebo suggested that the drug reduces the rate of relapse (RR 0.71, 95% CI 0.63 to 0.79).

FIGURE 20.

Pairwise meta-analyses: ARRs in SPMS.

Network meta-analyses

Ranking of the drugs in the resultant network suggested that 250 µg of SC IFN-β-1b every other day was superior to the equally ranked 44 µg of SC IFN-β-1a three times weekly and 22 µg of SC IFN-β-1a three times weekly (Table 17). Placebo was ranked last. Findings for comparisons between active drugs and placebo were, as would be expected, essentially the same as in the direct evidence. Comparisons between 250 µg of SC IFN-β-1b every other day and both 44 µg of SC IFN-β-1a three times weekly and 22 µg of SC IFN-β-1a three times weekly did not suggest a statistically significant difference between the drugs in terms of effectiveness (44 µg: HR 0.97, 95% CI 0.63 to 1.50; 22 µg: HR 0.97, 95% CI 0.63 to 1.49). Because there was no possibility of inconsistency in the network, we did not test for it.

Pairwise meta-analyses

Direct evidence from comparisons is shown in Figure 21. Comparisons included two trials. 222,224,225 The findings are the same as for the individual trials.

FIGURE 21.

Pairwise comparisons: time to disability progression confirmed at 3 months in SPMS.

Network meta-analyses

Because of the shape of the network, in which there was no opportunity for inconsistency and in which no direct comparison was informed by more than one trial, the model was estimated using fixed effects instead of random effects, as in the protocol. Ranking of drugs in the resultant network suggested that 250 µg of SC IFN-β-1b every other day was superior to 44 µg of SC IFN-β-1a three times weekly and 22 µg of SC IFN-β-1a three times weekly (Table 18). Placebo was ranked last. Findings for comparisons between active drugs and placebo were, as would be expected, essentially the same as in the direct evidence. Comparisons between 250 µg of SC IFN-β-1b every other day and both 44 µg of SC IFN-β-1a three times weekly and 22 µg of SC IFN-β-1a three times weekly did not suggest a statistically significant difference between the drugs in terms of effectiveness (44 µg: HR 0.91, 95% CI 0.65 to 1.25; 22 µg: HR 0.85, 95% CI 0.62 to 1.18). Because there was no possibility for inconsistency in the network, we did not test for it.

Summary of adverse events meta-analyses

Full results for pairwise meta-analyses of AEs are available on request. Although the diversity and heterogeneity of AEs precluded detailed examination of each, several trends were apparent across pairwise comparisons. SC IFN-β-1a three times weekly was associated with more application site disorders, more cases of necrosis, higher levels of alanine aminotransferase and aspartate aminotransferase, more cases of leucopenia and lymphopenia, higher levels of NABs and a higher number of patients who discontinued study treatment because of AEs than placebo. Comparing 250 µg of SC IFN-β-1b every other day with placebo, IFN-β-1b was associated with more injection site inflammation, more cases of necrosis, pain, injection site reactions, chest pain, chills only, chills and fever, fever only, flu syndrome, hypertonia, leucopenia, lymphadenopathy, lymphopenia, higher levels of NABs, rash and a higher number of patients who discontinued study treatment because of AEs.

Meta-analyses: discontinuation because of adverse events

Pairwise meta-analyses

All three studies presented data for discontinuation of the study drug because of AEs and all studies included follow-up of 36 months. Pairwise estimates are provided in Figure 22. Compared with placebo, all drugs were associated with a significant increase in the risk of discontinuation because of AEs.

FIGURE 22.

Pairwise meta-analyses: discontinuation because of AEs in SPMS. RR, risk ratio.

Model type

Network meta-analyses models were estimated in the Bayesian framework. Both fixed-effects and random-effects models were assessed according to the relative treatment-specific effect. The fit of the fixed-effects and random-effects models was compared using the deviance information criterion (DIC). A lower DIC is indicative of better fit. The best-fitting model was identified for each analysis. When the fit was similar between the fixed-effect model and the random-effects model, the random-effects model was adopted as a conservative approach. Moreover, the NMA included a comparison of the posterior distribution of between-study SDs with the prior distributions to assess whether it was updated by the available evidence (i.e. the additional information had had an effect). Consistency was assessed using node-splitting analyses.

(Confidential information has been removed.)

Prior distributions and estimation

The models were fitted using the OpenBUGS software package version 3.2.2 (Free Software Foundation, Inc., Boston, MA, USA). Models used 100,000 burn-in simulations with 150,000 simulations used. Flat priors were used in all cases for the treatment-specific, study-specific and between-study variance terms.

Interventions

The NMA included all trials testing licensed drugs with dosages at or below the recommended dose. Interventions and comparators of interest were immunosuppressives or immunomodulators: alemtuzumab (Lemtrada), dimethyl fumarate (Tecfidera), fingolimod (Gilenya), GA (Copaxone), IFN-β-1a (Avonex), IFN-β-1b (Betaferon/Extavia), pegylated IFN-β-1a, natalizumab (Tysabri) and teriflunomide (Aubagio).

Outcomes and data preparation

The NMA included analyses for ARR and disability progression. Models for disability progression included progression confirmed at 6 months, with additional data from confirmation at 3 months when 6-month data were not available, and the converse, that is, disability progression confirmed at 3 months, with additional data from confirmation at 6 months when 3-month data were not available. One potential issue with this method is that analyses are not strictly interpretable and rely on an assumption that progression estimates from 3 months and 6 months are exchangeable, but this is unclear and may be questionable.

Authors used an optimisation algorithm to estimate person-years and number of relapses to be used with an exact Poisson likelihood. Authors also used summary HRs in estimating disability progression models.

One strength of the reporting in this NMA was the transparency about included effect sizes for each model.

Participants

The NMA included all patients with a diagnosis of RRMS or progressive relapsing MS (PRMS). The NMA included an informal assessment of the similarity of baseline characteristics across trials. The authors did not undertake meta-regression or subgroup analyses.

Included trials

Unlike the assessment group’s NMA, the NMA in the manufacturer’s submission included trials with comparators outside the NICE scope. 141 However, even though the NMA in the manufacturer’s submission did not set explicit restrictions on the duration of follow-up, several trials appeared to be missing, including the BRAVO trial,198 IMPROVE trial,207 trial by Knobler et al. ,211 trial by Kappos et al. 199 and the GATE trial. 220 Although some of these trials may have been published after the last search, it is not clear why they were excluded.

Annualised relapse rate findings

A lower ARR is indicative of a better response. Although the submitted NMA covered a variety of doses and drugs, we summarise here only those results relating to licensed doses of the drugs under consideration.

(Confidential information has been removed.)

Sustained disability progression findings

(Confidential information has been removed.)

Model type

Models were estimated in the Bayesian framework. Both fixed-effects and random-effects models were estimated and then compared on fit. The authors also estimated pairwise meta-analyses and heterogeneity statistics.

Prior distributions and estimation

The authors used non-informative prior distributions. The authors used WinBUGS version 1.4.3 software (MRC Biostatistics Unit, Cambridge, UK) in all NMAs. In each model, two parallel chains were run, with a 50,000 iteration burn-in period. A total of 20,000 iterations against a thinning fact of 10 were sampled from each of the two chains. Convergence was assessed using Brooks–Gelman–Rubin diagnostics.

Interventions

All licensed drugs were included. Dosages were not specified, which poses significant ambiguity about whether all dosages in the literature were considered or only those that correspond to the marketing authorisation. It appears that both dosages of GA were pooled into one node in the analysis, but this was not clear.

Outcomes and data preparation

For disability progression, the authors estimated the number of events and the person-years of follow-up in each study and analysed data using a binomial likelihood with a complementary log-log link. Analyses used a model in which disability progression confirmed at 6 months was preferred, with 3-month data used when 6-month data were not available. Analyses of ARR used an arm-level data approach with a Poisson likelihood.

Although the authors presented relevant arm-level data for trials including GA in the text of the submission, it was not clear what the NMA inputs were. No forest plots for individual study estimates were presented.

Participants

Only participants with RRMS were included in the NMA.

Included trials

Unlike the assessment group’s NMA, the NMA in the manufacturer’s submission included trials with comparators outside the NICE scope. 141 However, the authors also excluded studies with a follow-up of < 6 months. Within these restrictions, it appears that the authors captured all relevant trials, although the trial by Knobler et al. 211 was not included.

Annualised relapse rate findings

(Confidential information has been removed.)

Sustained disability progression findings

(Confidential information has been removed.)

Model type

Random-effects and fixed-effects models were both estimated and compared on the DIC, with random-effects models preferred throughout. Further iterations were captured if convergence was in question.

Prior distributions and estimation

Network meta-analyses were estimated in the Bayesian framework using gemtc in the R environment (The R Foundation for Statistical Computing, Vienna, Austria). After 50,000 burn-in iterations, a further 50,000 iterations were captured. Convergence was assessed using the Brooks–Gelman–Rubin diagnostic. Prior distributions were non-informative.

Interventions

All studies testing comparisons of the drugs in the NICE scope141 and at the dosages contained in the marketing authorisations were included. Thus, dosages were clearly specified.

Outcomes and data preparation

Analyses included ARR for studies with follow-up of at least 12 months; HRs for disability progression confirmed at 3 months and, separately, at 6 months, with follow-up data at 12 or 24 months; and ORs for either any AE or serious AE. Data were analysed as log RRs, log HRs or log ORs with corresponding standard errors. The authors did not provide a justification for models that were intended to be estimated at either 12 or 24 months’ follow-up or why they chose to stratify estimates in this way. There was a lack of clarity regarding study inputs and no forest plots for individual study estimates were presented.

Participants

Although the searches included patients with RRMS, CIS and SPMS, it appears that only RRMS trials were meta-analysed.

Included trials

Studies excluded from the NMA and reasons for exclusion were, overall, clearly documented. However, the Biogen Idec Ltd NMA excluded several studies on what would appear to be the basis of short-term follow-up. This was not made explicit.

Summary of economic studies comparing disease-modifying therapies for people with clinically isolated syndrome

Definition of clinically isolated syndrome

The definitions used to characterise people with CIS were consistent. The majority of the studies defined their hypothetical cohort as adults who had experienced a single demyelinating event suggestive of MS. Two studies254,256 elaborated on this definition and suggested that their cohorts consisted of adults who had experienced a single demyelinating event in one or several areas of the CNS. To our knowledge, no studies included in this systematic review defined their population based on the McDonald 2010 criteria. 62

Characteristics of clinically isolated syndrome models

Four studies254,255,257,261 modelled the longer-term impact of treating CIS with DMTs, incorporating progression to RRMS. No studies modelled conversion from RRMS to SPMS. All studies except that conducted by Iskedjian et al. 257 considered progression until death in the analysis; there was no justification for omitting this health state in the analysis by Iskedjian et al. 257 Disease progression in the RRMS health states was stratified by severity (mild, moderate and severe)255,257 or by predicting changes in EDSS scores. 254,256,258–261 In the majority of the studies the risk of death was obtained from country-specific lifetime tables for the general population. In one study,254 mortality rates were adjusted to reflect the increased risk of mortality associated with MS. Here, background mortality was multiplied by EDSS-specific adjustment factors to reflect MS-specific mortality. All other studies accounted for death by assuming that people died on progression to EDSS 10. Adjusting the background mortality and including progression to EDSS 10 leads to double counting of people who may die from MS-related causes.

Treatment effect of disease-modifying therapies in the clinically isolated syndrome health state

Three studies254,256,261 clearly stated that treatment discontinuation was considered in the analysis. One study257 assumed that people did not discontinue treatment. The remaining studies255,258–260 were unclear on whether treatment discontinuation was included in the analysis. Treatment discontinuation was assumed to be a result of AEs from drug utilisation and/or progression to an EDSS score of ≥ 6. Discontinuation rates ranged from 6% every 2 years254 to 17.7% annually. 256 It appeared that the study by Fredrikson et al. 254 assumed a constant hazard over time for discontinuation of treatment in the first 2 years and, in subsequent years, used information from a follow-on trial. In the analysis undertaken by Caloyeras et al. ,261 a Weibull parametric model was fitted to Swedish registry data to derive time-dependent transition probabilities for people discontinuing treatment. Here, discontinuation of treatment was assumed to be the same for both early and delayed treatment (waiting until people developed MS).

Quality assessment of the modelling methods in clinically isolated syndrome studies

In this section we present a summary of the reporting quality of the studies included in the current review against the Philips et al. 244 checklist, which is presented in Appendix 6.

Summary of economic studies comparing disease-modifying therapies for people with relapsing–remitting multiple sclerosis

Definition of relapsing–remitting multiple sclerosis

The definitions used to characterise people with RRMS were consistent across all studies. However, to our knowledge no studies elaborated on the definitions used to define MS from the clinical studies that were used to obtain treatment effects of DMTs.

Characteristics of relapsing–remitting multiple sclerosis

All studies considered disease progression based on the use of EDSS levels to capture disability progression in people with RRMS. All models also captured the relapsing nature of MS. Nine studies264–272 grouped EDSS health states (e.g. EDSS 1–3.5265) but authors did not provide justification for how these groupings were derived. In contrast, Palace et al. 154 modelled each EDSS level to show disease progression. One study265 clearly presented definitions for each health state included in the model. Three studies154,265,272 included the conversion of RRMS to SPMS. Only one study154 allowed people to transition to less severe health states. In studies that considered relapses in their models,154,265,272 authors assumed that relapses occurred up to an EDSS level of 5.5. At this level, authors assumed that people discontinued treatment and followed the same pathway as people who were at the same EDSS level but untreated.

In general the risk of death was obtained from country-specific lifetime tables for the general population. Two studies264,272 assumed that people were at risk of MS-related death at EDSS 8–9.5. However, it was unclear whether Sanchez-de la Rosa et al. 264 varied the risk of death by age. Nikfar et al. 265 used another method to account for death. These authors assumed that MS increased the risk of death by threefold across age- and sex-adjusted mortality rates. Pan et al. 267 modelled mortality based on extrapolating survival data from an observational study. These authors fitted a Weibull parametric model to the placebo (no treatment) group to project survival over a lifetime, then mortality for IFN-β-1b was modified by applying the HR derived from a comparison between the treatment group and the placebo group. Evidence on other parametric model fits was not presented by the authors.

Treatment effect of disease-modifying therapy in relapsing–remitting multiple sclerosis

The effect of treatment on disability progression and the frequency of relapses was considered in all studies by applying a HR/relative risk to a baseline cohort of people with RRMS. All studies drew on evidence from RCTs. However, only one study264 was clear on the meta-analytical methods used to estimate the treatment effect from clinical trials. These authors used log-linear regression to estimate the treatment effect of DMTs on disease progression and relapse frequency.

It was unclear whether studies modelled the direct impact of DMTs in the conversion to SPMS. All studies considered an indirect impact of DMTs on mortality by showing that DMTs delay disease progression.

It was not clear whether any studies accounted for the waning effect of DMTs. One study264 considered the effect of NABs on the efficacy of DMTs.

Discontinuation of treatment in relapsing–remitting multiple sclerosis

Discontinuation rates were considered in all studies except for that by Agashivala and Kim. 266 Treatment discontinuation was assumed to occur as a result of AEs from drug utilisation and/or progression to an EDSS score of ≥ 6 or a perceived lack of efficacy. 265 To our knowledge, no studies fitted a parametric model to long-term data to derive time-dependent transition probabilities for people discontinuing treatment. Studies used short-term information on discontinuation rates from trials and assumed a constant hazard over time for the duration of the model.

Quality assessment

We present a summary of the reporting quality of the studies included in the current review assessed against the CHEERS243 and Philips et al. 244 criteria, which cover model structure, information required for the model and uncertainty. Details of the quality assessment for each study are presented in Appendix 7.