Delay From Treatment Start to Full Effect of Immunotherapies for Multiple Sclerosis

Izanne Roos; Emmanuelle Leray; Federico Frascoli; Romain Casey; J. William L. Brown; Dana Horakova; Eva K. Havrdova; Maria Trojano; Francesco Patti; Guillermo Izquierdo; Sara Eichau; Marco Onofrj; Alessandra Lugaresi; Alexandre Prat; Marc Girard; Pierre Grammond; Patrizia Sola; Diana Ferraro; Serkan Ozakbas; Roberto Bergamaschi; Maria José Sá; Elisabetta Cartechini; Cavit Boz; Franco Granella; Raymond Hupperts; Murat Terzi; Jeannette Lechner-Scott; Daniele Spitaleri; Vincent Van Pesch; Aysun Soysal; Javier Olascoaga; Julie Prevost; Eduardo Aguera-Morales; Mark Slee; Tunde Csepany; Recai Turkoglu; Youssef Sidhom; Riadh Gouider; Bart Van Wijmeersch; Pamela McCombe; Richard Macdonell; Alasdair Coles; Charles B. Malpas; Helmut Butzkueven; Sandra Vukusic; Tomas Kalincik


Brain. 2020;143(9):2742-2756. 

In This Article


In this study, we have used the two largest multiple sclerosis registries to develop and externally validate a method to quantify the duration of clinical therapeutic lag of immunotherapies for multiple sclerosis. We have then applied this method to estimate therapeutic lag with respect to two principal clinical presentations of multiple sclerosis—relapses and progression-of-disability—for the most commonly used immunotherapies. Full effect of treatment on relapses is reached within 12–30 weeks after commencing therapy, whilst the full effect on disability progression is reached within 30–70 weeks.

Therapeutic Lag: The Clinical and the Biological Perspective

When treating multiple sclerosis, timely reduction in disease activity is required to diminish inflammation, minimize neuroaxonal loss and prevent long-term disability (Trojano et al., 2009; Giovannoni et al., 2016). As treatment initiation or switch is often prompted by ongoing disease activity, understanding of the time required to manifest fully and clinically the effect of therapy is essential for further therapeutic decisions. Phase III studies conventionally report the effect of therapy on clinical and radiological end points over the follow-up period in its entirety, typically spanning 2 or more years. Post hoc analyses have explored the differences between treatment and comparator groups in time to first relapse; this, however, is not a pure measure of the magnitude of full clinical effect of therapy, as time to first event also reflects the onset of clinical effect of the compared interventions (Kappos et al., 2013, 2015, 2016; Traboulsee et al., 2018).

Unlike biological markers of treatment effect, clinical markers of the effect of multiple sclerosis therapies cannot be studied in individual patients. This is because the traditional outcomes in multiple sclerosis trials are discrete, rarely-occurring events. To identify their temporal trends, such effects require evaluation at the level of large cohorts. We have used a reproducible mathematical approach to quantifying trends in the occurrence of relapses and disability progression events and identify the time point at which the frequency of events stabilizes after a new treatment has been started (Diller et al., 2006). Invariably, commencement of new DMTs tended to be preceded by a peak in the relapse density, consistent with a relapse presently being a major driver of therapeutic decisions (Montalban et al., 2018). Subsequent to the relapse peak, all therapies resulted in stabilization of relapse frequency at a new, on-treatment level over the following 3–6-month period. This observation is consistent with results from post hoc analyses of the pivotal trials of fingolimod, dimethyl fumarate and natalizumab; the greatest reduction in ARR occurred within the first 3 months of treatment, with further reduction between months 3 and 6 and a subsequent plateau over the rest of the follow-up period (Kappos et al., 2013, 2015, 2016).

The duration of therapeutic lag is often inferred from the timing of biological changes associated with treatment initiation. Absolute T- and B-cell counts reduced 3 months and reached steady state 9 months after starting dimethyl fumarate, with an association noted between higher absolute lymphocyte counts and shorter time to first relapse (Wright et al., 2017). This is in keeping with our observation that the estimate of Tr for dimethyl fumarate (7.1 months), is longer than in other multiple sclerosis therapies. Comparatively, fingolimod results in a reduction, and subsequent steady state, in lymphocyte counts within 2 weeks after treatment begins (Francis et al., 2014), whilst in clinical trials the proportion of patients free from relapses significantly differed from placebo after 3–6 months (Kappos et al., 2016). In our analysis, a mean period of 3 months (12.7 weeks) was required for fingolimod to attain full effect on reducing relapses, suggesting that the full effect of therapies on clinical outcomes may, in some DMTs, follow the effect on biological markers with a lag. The differences between the effect of therapies on biological and clinical markers of multiple sclerosis highlight the complementary value in evaluating both aspects of treatment effect in concert.

Consistent with the increase in the incidence of relapses that leads to commencement of new DMTs, differences between pretreatment relapse frequency highlight systematic differences in the nature of the cohorts treated with different therapies. Patients with the most active disease were predominantly commenced on high-efficacy therapies, such as alemtuzumab, natalizumab, and mitoxantrone. All therapies, with the exception of intramuscular interferon beta-1a, resulted in a change in the ARR. This highlights that clinical decisions to commence new therapies have, at the group level, been successful in achieving improved control of relapse activity. The inability to detect a change in ARR for intramuscular interferon beta-1a was likely influenced by lower pre-baseline and on-treatment ARRs in this group compared to the other low-efficacy injectable DMTs, suggesting that these patients had with less active disease. Moreover, the results reflect a magnitude of change in ARR in the pre-baseline versus on-treatment period, as opposed to a comparison to placebo. Importantly, no corrections were made for differences between the cohorts starting different DMTs, one should therefore resist the temptation to compare the results among treatments.

Therapeutic Lag in Controlling Worsening of Disability

In our study DMTs resulted in a full effect on disability progression between 6 months and 1.3 years after commencement. This is in accordance with a post hoc analysis of the combined DEFINE and CONFIRM cohorts, where dimethyl fumarate reduced the risk of 12-week confirmed disability progression after 62 weeks (Kappos et al., 2015). Time to treatment effect on disability progression is, to the authors' knowledge, infrequently reported. In our analysis the commencement of DMTs resulted in a transient attenuation in the number of progression events, but did not abolish the accumulation of disability over time. For most therapies, the frequency of confirmed disability progression events resumed to increase after ~2 years from starting a DMT. In particular, disability progressions occurring independent of relapse activity were only briefly reduced by multiple sclerosis immunotherapy. This highlights the contribution of relapse-independent disability progression to reduced capacity in patients with multiple sclerosis, and is supported by studies showing that therapies delay, but do not entirely stop, disease progression (Brown et al., 2019; Lorscheider et al., 2019). Clinical relapses, however, represent the tip of the iceberg of episodic inflammatory activity; 5–10 new white matter lesions are accrued for every relapse diagnosed (McDonald et al., 1994). Radiologically apparent, yet clinically silent, episodic inflammatory activity may thus still contribute to progression independent of relapse activity. Progression-of-disability trends differed for patients receiving mitoxantrone and interferon beta-1b; these groups experienced continued reduction in the number of progression events over the 5-year period (Le Page et al., 2008). The interferon beta-1b and mitoxantrone cohorts were enriched for patients with progressive disease and with greater EDSS scores. As the probability of experiencing a progression event reduces at higher EDSS scores (Kalincik et al., 2015), these differences cannot be entirely attributable to treatment effect, but potentially to systematic differences in the rate of transition between different EDSS steps.


This study used data obtained from longitudinal observational registries, which may be subject to variable data quality. Data quality was, however, controlled through a previously published data verification process (Kalincik et al., 2017). The use of two differing sources of data (MSBase, a global registry of self-selected predominantly academic multiple sclerosis centres, and OFSEP, a national cohort from academic multiple sclerosis centres) helps further mitigate the effects of selection and reporting biases. Second, the described method is reliant on a large number of events in order to identify a stable, reproducible estimate of the duration of therapeutic lag. We combine data from the two largest multiple sclerosis registries in order to maximize the available power. We have used objective methods to identify therapies for which the available data were sufficient, including the evaluation of the estimated Tr in relation to the number of recorded events. Where the critical mass of events was not reached, analyses were discontinued. As confirmed disability progression events are less frequent than relapses, time to treatment effect on this outcome could only be calculated for seven therapies in merged progressive and relapsing multiple sclerosis cohort. Similarly, highly effective therapies with smaller patient numbers, and newer therapies, such as the B-cell therapies or cladribine, were not sufficiently represented to qualify for inclusion. Because Tr is estimated within large groups of events, it may be subject to bias and fluctuation where the density curves are multimodal. We have therefore implemented additional measures to ensure robust estimates of the lag duration, including Monte Carlo simulations to estimate the variance, sequential analyses to ensure consistency of results and Gaussian mixture models to identify the most robustly supported Tr in multimodal curves. Importantly, the Tr estimator was developed and externally validated in two large non-overlapping registry datasets, with no significant differences in Tr between registries. As all relapses within 30 days from a prior relapse constitute a single event, this refractory period, together with regression to the mean, contributes to the decline in relapse density. Third, unmatched differences remain between the cohorts on different therapies, and any comparisons between treatments should be avoided. This includes variability in the reasons for treatment switch and the use of DMTs during the 3-year pretreatment period. These differences would most likely be associated with the height of the peak of relapses preceding treatment switch, but would not directly influence the time to treatment effect. It is therefore reassuring that the height of the relapse peak did not impact the duration of therapeutic lag. Fourth, relapses did not require confirmation with EDSS or treatment with corticosteroids; this may have inflated the number of relapse events present. A sensitivity analysis with a more stringent relapse definition was however performed, with no substantial change in the results. Fifth, the EDSS was used as a measure of disability progression. The EDSS has a number of limitations including a floor and a ceiling effect, low intra- and inter-rater reliability and at higher scores is predominantly a measure of ambulation (Hobart et al., 2000). The issue of variability is partially addressed through the use of specialist neurologist EDSS raters (D'Souza et al., 2017) and the use of a robust definition of disability progression. Additionally, no objective measures of disability in cognitive domain or manual dexterity were included in the analysis as they are not routinely documented in registry data. Sixth, the requirement for 3-year pretreatment follow-up and 1-year treatment persistence precludes generalizability to patients who commence treatment early after multiple sclerosis onset and those with early treatment cessation due to intolerance or treatment failure. Seventh, the duration of treatment effect after the last dose was based on rough estimates as per Kunchok et al. (2020). Eighth, the use of re-baselining brain MRI shortly after commencement of therapy, as per the present monitoring guidelines (Wattjes et al., 2015), may have influenced the persistence on study therapy. Where MRI activity was detected, physicians may have been tempted to discontinue study medication, which would thus not fulfill the inclusion criterion of 1-year treatment persistence. This would have, in turn, led to selective exclusion of treatment epochs with early subclinical activity. On the other hand, early on-treatment MRI activity may be representative of changes immediately preceding treatment initiation or occurring before the newly commenced treatment has become fully effective, and such early MRI assessment would be requested with the aim of creating a new radiological 'baseline' rather than immediately guide continued treatment (Montalban et al., 2018). The novel information generated by our study will help clinicians in choosing the optimal time for the re-baselining brain MRI after start of a new DMT. Lastly, evaluation of the time to treatment effect on MRI activity was not included in this study; an observational cohort, with semiquantitative imaging information acquired approximately at yearly intervals does not provide a reliable marker of neuroradiological onset of treatment effect. To address this, a prospective study such as the ongoing MAGNIFY-MS trial (, with frequent prespecified imaging time points (at least one MRI per month), would be needed.