The relationship between the relapsing and the progressive course of multiple sclerosis has remained ambiguous. On one hand, development of progression is the overwhelming determinant of outcome based on natural history studies; while on the other, relapses are what can be partially suppressed by currently available treatments. Widespread belief that accumulation of much unremitting disability results from successive exacerbations is not well-founded in multiple sclerosis, although it is a key pathway in neuromyelitis optica (Wingerchuk et al.
) and in Oriental multiple sclerosis (Kira, 2003
). Some recollections of devastating relapses in multiple sclerosis were surely cases of neuromyelitis optica.
Biological mechanisms leading to the development of severe disability may be different from those responsible for attacks, as demonstrated by extensive neuropathological studies (Bjartmar et al.
; DeLuca et al.
; Trapp and Nave, 2008
). Results from interferon and glatiramer acetate studies (IFNB Multiple Sclerosis Study Group, 1993
; Johnson et al.
; Jacobs et al.
; European Study Group, 1998
; PRISMS Study Group, 1998
) and most clearly from the cladribine and the alemtuzumab studies (Beutler et al.
; Coles et al.
; Rice et al.
) brought this dichotomy to attention, failing to demonstrate a clear effect of relapse reduction on delaying progression. With alemtuzamab, 90% reduction of new gadolinium-enhancing MRI lesions and concomitant reduction in new relapses failed to prevent continued deterioration in disability (Coles et al.
), highlighting previously observed dissociation between inflammatory load and disability progression (Noseworthy et al.
). It remains possible that therapeutic reduction of early relapse rate might impact disease progression and long-term disability accumulation. Therapeutic monoclonal antibodies hold promise.
Relapse frequency and its surrogates, MRI T2
or gadolinium-enhancing lesions, still represent the most common outcome measures for evaluating treatment efficacy, notwithstanding that relapses came last among 12 clinical trial outcomes ranked for credibility two decades ago by a large group of multiple sclerosis clinical trialists (Noseworthy et al.
). Initial MRI lesion number and volume predicted conversion to clinically definite multiple sclerosis but late disability only modestly (Fisniku et al.
; The Optic Neuritis Study Group, 2008
). Frequent early relapses associated with long-term disease evolution (Weinshenker et al.
; Kantarci et al.
; Confavreux et al.
), but causality remains uncertain.
Late relapses had not shown indications they would influence unremitting disability, and certainly not after onset of the progressive phase (Kremenchutzky et al.
). Predictive effects of early relapses on disease progression (Weinshenker et al.
; Eriksson et al.
) were reported not to apply once secondary progressive begins (Confavreux et al.
). Times for progressing from DSS 4 to higher disability levels (DSS 6 and 7) were independent of early relapses. However, early and later relapses were not separated and frequency of assessments and dropout rates were not enumerated (Confavreux et al.
Further evidence that secondary progressive is largely independent of preceding relapses or of those subsequent to its onset came from comparing progressive disease subtypes. Late outcomes were indistinguishable among those with none, one or many relapses preceding onset of progression, each subgroup having near identical ages when progression began. Common mechanisms in progressive multiple sclerosis subtypes were implied (Kremenchutzky et al.
). Neither the Lyons nor London, Ontario studies directly identified determinants of secondary progression probability, latency and slope that are of much practical importance.
The London Ontario database (Weinshenker et al.
) now encompasses 28 000 patient-years of observation with most patients having reached hard disability outcomes. The low percentage of censored patients gives high reliability for survival estimates of later disability.
The role of relapses
Polysymptomatic presentation was strongly associated with a worse prognosis in patients with primary progressive multiple sclerosis (Cottrell et al.
). In contrast here, the number of neurological systems involved at disease onset did not independently influence time to secondary progressive or to disability endpoints. Similarly, degree of recovery from initial exacerbation did not influence long-term outcomes (Kremenchutzky et al.
Average attack frequencies in multiple sclerosis show marked variation within and between individuals over time (Weinshenker and Ebers, 1987
). Prospective assessments yield greater frequencies (Fog and Linnemann, 1970
; Patzold and Pocklington, 1982
) and we confirm that attack rates lessen with time (Broman et al.
; Patzold and Pocklington, 1982
; Myhr et al.
). Relapse rates were high during Years 1 and 2 (0.93 attacks/year) decreasing with disease duration. Overall, mean attack frequency during the relapsing–remitting phase (0.65 attacks/year) coheres with other studies (Confavreux et al.
: rate 0.86; Patzold and Pocklington, 1982
: rate 1.1; Goodkin et al.
: rate 0.64), the higher rates being taken earlier in the disease overall or not population based, or not extending up to onset of progression. They conform to rates seen for placebo arms in relapsing–remitting multiple sclerosis trials. gives the distribution of relapse frequencies in early and later relapsing–remitting multiple sclerosis prior to onset of the progressive phase.
Relapses and the probability and the latency of progression
We confirmed that neither the risk of entering the secondary progressive phase nor the latency of onset of progression were related to total attack number during the relapsing–remitting phase (Kremenchutzky et al.
). Patients with fewer total relapses prior to progression and with fewer relapses from Year 3 up to progression () actually converted to secondary progressive multiple sclerosis significantly earlier. The size of this effect was larger for relapses from Year 3-secondary progression compared to the total number of attacks in the relapsing–remitting phase (including early relapses). These data should further discourage any direct causal relationship between clinical attack numbers and disability accumulation.
Given the predictive effects of frequent Years 1 and 2 relapses for shortened times to disability endpoints, we hypothesized that patients with higher relapse frequency in Years 1 and 2 must not only have increased probabilities of developing a progressive course but also shorter latencies from disease onset to progression. This was proven to be true, highlighting the key role of developing the progressive course. Interestingly, attacks from Year 2 exerted slightly greater predictive effects compared to attacks from Year 1; possibly Year 2 relapses are a marker for an inadequate immune regulatory response to events in Year 1.
A modest independent impact of the first inter-attack interval on probability and time to enter the secondary progressive phase was also observed. The predictive effect of Years 1 and 2 relapses and first inter-attack interval on the probability of entering the progressive phase became larger when we took into account the effect of Year 3-secondary progression relapses; five attacks versus none in the first 2 years tripled the risk (HR = 3.02) of developing secondary progressive multiple sclerosis. The relationship between relapses from Year 3 up to progression and latency to progression remained unchanged. Patients with more frequent attacks appeared to convert to secondary progressive multiple sclerosis significantly later; five attacks after Year 2 halved the risk of starting to progress. This is most unlikely to result from uncertainties in defining progressive onset in those still having relapses. Onset of progressive phase in those with fewer relapses was identified lower not higher on the DSS (Supplementary Fig. 2
Relapses and disability outcomes
We confirmed and extended the independent predictive effect for hard disability outcomes of early relapses and first inter-attack interval, observed for lesser degrees of disability after 12 years of follow-up (Weinshenker et al.
). With 16 years of additional follow up, Years 1 and 2 relapses influenced times to DSS 6, 8 and 10 from onset, from onset of the progressive phase and from DSS 3; predictive effects from disease onset were smaller than from onset of progressive phase. The analysis from onset of disease included patients with long relapsing–remitting phases or who never developed secondary progressive and therefore were less impacted by early relapses. Times from disease onset to DSS 6, 8 and 10 between 1 attack and ≥3 in Years 1 and 2 were substantially different i.e. 7.6, 12.8 and 20.3 years, respectively. Intervals between the first two attacks strongly associated with times from onset to DSS 6, 8 and 10 and from onset of progressive phase to DSS 6 and 8.
The risk of accumulating disability increased proportionally with number of attacks and inversely with time between the first and the second attack. Increased Years 1 and 2 relapses had the larger impact ( and ). In multiple analysis, the effect exerted by early relapses remained unchanged. The impact of first inter-attack interval diminished but remained significant implying that the predictive effect of a short interval between the first two attacks largely derives from having or not having the second relapse in Years 1 or 2.
Total relapses during the relapsing–remitting phase (including Years 1 and 2), exerted no significant effect on attainment of high disability levels from disease onset (). Times for reaching DSS 6–8–10 were remarkably equal, being 15, 26 and 41 ± 1 years for groups with high (≥5), intermediate (3–4) and low (1–2) numbers of attacks prior to onset of progressive phase. Times were nearly identical from DSS 3 to DSS 6, 8 and 10 based on numbers of relapses prior to progression (). The only indication that total relapses influenced any late outcome was seen in times from onset of progressive phase to DSS 6 and 8 but driven by Years 1 and 2, indicating a modest influence of these early relapses on slope of progression ().
It is important to put the relapse data in the general context provided by previous articles in this series. We had shown that among those with primary progressive multiple sclerosis, outcome did not differ by the presence (in 28%) or absence of relapses, and survival curves for those with relapsing–progressive multiple sclerosis were indistinguishable from those with progressive multiple sclerosis without relapses (Kremenchutzky et al.
). Disease course during the progressive phase was homogenous among multiple sclerosis progressive subtypes (Kremenchutzky et al.
). These findings, coherent in showing no impact of relapses on hard outcomes in progressive disease, left little rationale for considering relapses in the assessment of progressive disease or for unremitting changes in disability. These observations then permitted more focused examination of the relapsing–remitting phase in isolation. The data in this article are restricted to this phase but it will be apparent that, combined with what we have already examined, the two together total three decades of disease evolution, encompassing the relapsing–remitting and secondary progressive phases.
We have shown already that neither the location nor severity/degree of recovery characterizing the first attack nor a polysymptomatic onset were independently predictive of hard outcomes (Kremenchutzky et al.
). We show here that total relapses in relapsing–remitting phase are unrelated to hard outcomes (combined with the progressive results therefore, essentially the entire course of multiple sclerosis is spanned). These findings once again serve to discount or invalidate relapses in general, either as factors prognostic for hard outcomes in the relapsing–remitting phase overall or as therapeutic targets (but vide infra
). However, the results for Year 3-secondary progressive showed an inverse relation to hard outcomes, enfeebling relapses in this time period as they fail, as do total relapses, to attain the basic starting premise for postulating a causal relationship.
The analysis of Year 3-secondary progression relapse number isolated the impact on outcome of these later relapses, frequently counted in clinical trials. Negative regression coefficients indicated that more attacks after Year 2 correlated with significantly lower risk () and longer times to reach the endpoints from disease onset (Supplementary Table 1
). Five versus zero attacks after Year 2 reduced the hazard of attaining DSS 6, 8 and 10 by 24% (HR = 0.76), 22% (HR = 0.78) and 34% (HR = 0.66), respectively (). This was unexpected but highly significant, counterbalancing the negative impact of early relapses on outcome. Some kind of as yet undetermined interaction between the development of progression and the suppression of relapses is strongly implied, possibly analogous to what occurs in primary progressive multiple sclerosis.
Because times for Year 3-secondary progression are necessarily variable, they are not comparable to the time fixed by definition for Years 1 and 2. Therefore we assessed variation in relapse frequency in serial 2-year intervals from Year 3 up to secondary progression, making comparisons within each time interval. For each 2-year interval past Years 1 and 2, and despite considerable variation in relapse frequency and slight indication of an impact beyond Year 2, no significant effect of relapse frequency for individual 2-year blocks could be found. We cannot easily explain the apparent and counterintuitive negative association of relapses with hard outcomes coming from Year 3-secondary progressive but the results, at the very least, serve to discredit relapse outcomes in this stage of disease. It seems likely that existing models relating relapse to disability have been too simple. All that remains viable in these contexts for relapse as an outcome is frequency in the first 2 years as that does predict hard outcome, modestly overall, but strongly for higher relapse frequencies.
The effect of early and later relapses on outcomes remained unchanged after multiple analysis. Patients at higher risk of accumulating disability from disease onset had more Years 1 and 2 relapses, shorter relapsing–remitting phases, yet fewer total Year 3-secondary progression attacks. The combined risk of attaining disability endpoints decreases consistently with increasing numbers of relapses after Years 1 and 2 (). Those with larger numbers of relapses in Years 1 and 2 seem unable to suppress or generate mechanisms evolving into progressive disability accumulation with further relapses being suppressed and/or masked by earlier development of the progressive phase.
These results re-emphasize that the predictive effect of relapses is more or less restricted to their frequency in Years 1 and 2 and within this window to those having three or more attacks. Nevertheless, the relapse rate characterizing the overall relapsing–remitting phase does not detectably contribute to the long-term accumulation of disability from disease onset ().
The findings for Years 1 and 2 relapses and first inter-attack interval may serve to explain the somewhat worse outcome seen for those seen at onset versus those coming to attention later (Weinshenker et al.
). Those seen within a year of the first symptom are selected for having already had a second attack (and even additional ones) and therefore for short inter-attack intervals, an independent contributor to outcome.
It has been suggested that late progression is still related to inflammation caused by local compartmentalization of effector cells later in the disease (Meinl et al.
; Frischer et al.
). Although this would be an attractive way of linking relapsing–remitting and secondary progressive phases, this notion seems improbable. We have pointed out how progression has no predilection for initial or previous sites of exacerbation (Kremenchutzky et al.
), which might be predicted by this notion. Not only would previous sites have vulnerable partially damaged axons and oligodendrocyte loss but they would be loci where ‘compartmentalization’ would be expected to have a focal head start. There is no hint of this when secondary progressive supervenes and clinicians will know, for example, how rarely they see progressive blindness localized to the optic nerve affected with the first attack. A potential role for continued inflammation would have to be disconnected from relapsing–remitting inflammation to the extent that progression does not associate with the same concomitants as does the relapsing–remitting course. The widely differing prevalence of progressive disease in Caucasian versus Japanese Western multiple sclerosis hints at a genetic explanation for what must be a true dichotomy.
Relapses and the course of progression
The progressive phase was reported to be independent of preceding factors (Confavreux et al.
). Despite much individual variation, its age of onset and rate is remarkably homogeneous among progressive multiple sclerosis subtypes (Kremenchutzky et al.
). However, these studies did not address separately the role of early and later relapses on the evolution of the progressive course; potential effects on the probability of developing progressive multiple sclerosis or the latency of onset of progression were not examined.
We were able to address this aspect in two ways: (i) separately analysing the predictive effect of early and later relapses on the attainment of endpoints from onset of progressive phase; and (ii) analysing the predictive effect of early and later relapses on the time for progressing from DSS 3 to higher disability levels. The two approaches were methodologically different. The first analysis included only patients having entered secondary progression. The onset of progression could certainly be ambiguous and confounded by concomitant relapses but there was the advantage of long retrospect in this study, which typically clarified ambiguities at the time of evaluation. The second analysis also included patients still experiencing the relapsing–remitting phase, although >60% of secondary progressive patients in our population were deemed to have started to progress at DSS ≤ 3.
In mild contrast to previous reports (Confavreux et al.
), the slope of the progressive phase was modestly affected by early relapses. More frequent Years 1 and 2 relapses were independently related to significantly shorter times to attain DSS 6–8–10 from onset of progressive phase and the same independent predictive effect was observed on time to progress from DSS 3 to higher disability levels. Again, Year 3-secondary progression relapses had no impact on times to disability endpoints either from progression onset or from DSS 3.
The predictive effect of early relapses on long-term disability appears to be exerted primarily by increasing the probability of developing secondary progressive disease, shortening its latency and, to a lesser degree, by influencing the slope of progression. The impact of Years 1 and 2 relapses on the attainment of endpoints from progression was larger than the effect exerted from disease onset (). This probably is indirect, driven by increased probability of developing secondary progressive multiple sclerosis and by shortened latency of its onset. The analysis from onset of progressive phase excluded those patients who never entered the secondary progressive phase and therefore less impacted by Years 1 and 2 relapses.
Once progression has begun, there is consensus that outcome has been largely determined. In fact, early relapses and, similarly the first inter-attack interval, exerted a much smaller impact on times to disability from DSS 3 than on times to the same endpoints from onset of progressive phase (). Although DSS 3 is composed mostly of those already progressing, a minority (25%) reached this level through relapses, remained stable and free of progression for long periods or never entered the secondary progressive phase, explaining this result. Those with more frequent attacks after Year 2 have longer latency to progression and paradoxically better outcomes, but again this operates via an effect on progression, albeit a beneficial one.
The evolution of the progressive phase then is largely driven by mechanisms independent of the inflammatory attack frequency characterizing the entire relapsing–remitting phase. Total relapses prior to secondary progressive and number of relapses from Year 3 to secondary progressive exerted no detectable independent effect on the attainment of hard disability endpoints from DSS 3 upward. In addition, multiple analysis accounting for early relapses, first inter-attack interval and later relapses showed little impact from any of the covariates on times for progressing from DSS 3 to higher DSS levels. These results highlight the landmark status of both time to DSS 3 and even more of onset of progressive phase as predictors of disability, further emphasizing the impropriety of later relapses as surrogates for long-term outcome.
Although frequent Years 1 and 2 relapses predict shorter times to DSS 6, 8 and 10, a causal relationship between such attacks and faster disease progression cannot and should not be assumed. A higher early relapse frequency could be concomitant to a predestined, more rapid clinical course. Time to DSS 3, known to predict time to DSS 6 (Weinshenker et al.
) at 12 years follow-up, did predict risks for attaining DSS 6, 8 and 10, which increased inversely with length of interval between disease onset and DSS 3 (). The effect size remained unchanged after multiple analysis, thus independent of Years 1 and 2 relapses and first inter-attack interval.
The impact of Years 1 and 2 relapses on DSS 6, 8 and 10 lessened consistently when adjusted for initial disease progression (time to DSS 3) remaining modestly significant for times to DSS 6 and DSS 8 from DSS 3 (). DSS 3 and time to it necessarily encompasses most individuals becoming progressive over the period of observation. However, time to DSS 3, if reached, is often protracted while early relapses in Years 1 and 2 are, by definition, available early. Despite the practical predictiveness of Years 1 and 2 relapses, validation as an outcome still requires demonstration that suppression of these relapses translates into suppression of long-term disability.
These results, in sum, indicate that late disability is predetermined relatively early. Time to DSS 3 in multivariate analysis accounts for the effect of Years 1 and 2 relapses, probably by heralding the progressive course and the effect may have been underestimated by inclusion of relapse-mediated arrival at DSS3. This suggests that even frequent early relapses might be concomitant with, rather than causative of, poor outcome.