Skip to main content
European Journal of Neurology logoLink to European Journal of Neurology
. 2026 Jun 8;33(6):e70659. doi: 10.1111/ene.70659

Progression to Wheelchair in Secondary Progressive Multiple Sclerosis and Impact of Siponimod: Post Hoc Analyses From the EXPAND Study

Patrick Vermersch 1,✉, Sophie Arnould 2, Ralf Gold 3, Ludwig Kappos 4, Robert Fox 5, Amit Bar‐Or 6, Bruce A C Cree 7, Soudeh Ansari 8, Daniela Piani Meier 9, Goeril Karlsson 8, Nicolas Rouyrre 8, Gavin Giovannoni 10
PMCID: PMC13244661  PMID: 42257510

ABSTRACT

Background

Delaying time to requiring a wheelchair in people living with MS is an important treatment goal to maintain quality of life and guard against increased healthcare costs. In this post hoc analysis, the effects of siponimod versus placebo on time to sustained Expanded Disability Status Scale (EDSS) ≥ 7.0 over the core part of the Phase 3 EXPAND study were assessed.

Methods

The time to sustained EDSS ≥ 7.0 (time to requiring a wheelchair) was analysed for participants with secondary progressive multiple sclerosis (SPMS). Analyses were based on a While on Treatment (WOT) set that censored participants if entering open‐label treatment. Subgroup analyses were conducted in participants with active versus non‐active SPMS (defined based on pre‐study/baseline criteria), and in participants with baseline EDSS 6.5.

Results

In the overall EXPAND population, siponimod reduced the risk of reaching sustained EDSS ≥ 7.0 (requiring a wheelchair) by 40% (hazard ratio [HR], 0.60; 95% confidence interval [CI], 0.41–0.88; p = 0.009) versus placebo; the reduction in risk was 51% (HR, 0.49; 95% CI, 0.29–0.81; p = 0.005) versus placebo in participants with active SPMS and 22% (HR, 0.78; 95% CI, 0.42–1.45; p = 0.437) in non‐active SPMS.

Conclusion

Siponimod reduced the risk of requiring a wheelchair in participants with SPMS versus participants receiving placebo, with a greater effect in those with active disease. Time to requiring a wheelchair is a highly relevant treatment goal and an important indicator of treatment effect in patients with SPMS.

Keywords: Markov model, multi‐state model, secondary progressive multiple sclerosis, siponimod, sustained EDSS ≥ 7.0, wheelchair


Siponimod lowered the risk of wheelchair use in SPMS patients compared to placebo, especially in those with active disease. Time to wheelchair is a key metric for evaluating treatment in SPMS.

graphic file with name ENE-33-e70659-g003.jpg

1. Introduction

Up to 20% of patients with MS may require mobility devices, including a wheelchair, within 15 years after the onset of MS [1, 2]. Most patients with MS follow a relapsing–remitting disease course, with ~50% of patients reaching the secondary progressive MS (SPMS) phase by 20 years after disease onset [3, 4]. A key clinical feature of SPMS is worsening of ambulation that begins early in the disease course and progresses over time to impact the patient's overall quality of life [5, 6, 7, 8]. Patients with SPMS experience great disruption to their daily lives and employment, and an increased financial burden, which also affects the families of patients and society [5, 9, 10, 11, 12].

Given this associated reduced quality of life, the transition to wheelchair dependence, which is a widespread consequence of MS, is a major disability milestone for patients [13, 14]. While there is minimal evidence supporting the efficacy of disease‐modifying therapies (DMTs) for reversing mobility disability or restoring ambulatory function in people living with MS [13], there is an interest in delaying these disability milestones. The short‐term effect of ocrelizumab on time to requiring a wheelchair has been assessed in the ORATORIO study [15]. In this study, extrapolations were also conducted to characterise the potential long‐term impact of ocrelizumab on this outcome [16]. Additionally, multi‐state Markov models are increasingly used to model disease progression in which individuals can advance or recover between adjacent stages of disease severity. In the context of MS, this approach was applied to estimate survival curves for time to disability progression in patients with relapsing–remitting multiple sclerosis on fingolimod [17]. The flexibility of this method allowed the whole disability process to be modelled and even the short‐term effects of treatment to be estimated.

Disability progression was the primary endpoint in the Phase 3 EXPAND study, the objective of which was to demonstrate the impact of siponimod on delaying disability accrual in patients with SPMS [18]. As expected for the SPMS study population and inclusion criteria (Expanded Disability Status Scale [EDSS] 3.0–6.5), ambulation was substantially impaired at baseline, with 56% of participants requiring the use of walking aids (cane or walker). Siponimod was shown to delay the risk of both 3‐month and 6‐month confirmed disability progression (CDP) in the EXPAND study population [18]. This population, therefore, presented an opportunity to examine whether siponimod might delay the onset of wheelchair use. Here, we determine the effect of siponimod versus placebo on time to a sustained EDSS score of ≥ 7.0 (the EDSS milestone corresponding to wheelchair dependence) in the core part of the EXPAND study, and we use extrapolation techniques to assess the long‐term impact of siponimod versus placebo in the extension part.

2. Method

2.1. EXPAND Study Design and Participants

The core part of the Phase 3 EXPAND trial was a randomised (2:1), event‐ and exposure‐driven, double‐blind, placebo‐controlled study in which participants with SPMS were assigned either siponimod or placebo [18]. Given the event‐driven design, duration in the core part before transitioning to the open‐label extension part varied for individual participants (median duration 21 months: range: < 1 month to 37 months). In the extension part, participants receiving siponimod in the core part were maintained on siponimod and those receiving placebo were switched to open‐label siponimod. Those who entered the study extension part transitioned to open‐label siponimod from as early as Month 12 and up to Month 36 of the core part [19].

In this core part, participants with 6‐month CDP during double‐blind treatment were offered a switch to open‐label siponimod and to remain in the study, while participants who withdrew from double‐blind treatment for other reasons could also choose to remain in the core part. These data comprised the full analysis set (FAS).

Key inclusion criteria for this study included ages 18–60 years, a diagnosis of SPMS (which constitutes a history of gradual worsening after an initial relapsing disease course, with or without acute exacerbations during the progressive phase) [7, 20, 21], and an EDSS score of 3.0–6.5 at screening. Active SPMS was defined as the presence of at least one relapse in the 2 years prior to screening and/or at least one T1 gadolinium‐enhancing lesion at baseline. Non‐active SPMS was defined as no relapse in the 2 years prior to screening and no T1 gadolinium‐enhancing lesions at baseline.

2.2. Outcomes

Progression to wheelchair dependence was defined as progression to EDSS ≥ 7.0, sustained until the end of the observation period. EDSS of 7 means ‘unable to walk beyond approximately 5 m even with aid. Essentially restricted to a wheelchair; wheels self and able to get in and out alone. Up and about in wheelchair some 12 hours a day’. Participants with baseline EDSS ≥ 7 were excluded (n = 2 siponimod and n = 1 placebo). To reduce the direct impact of clinical relapses on EDSS scores, all analyses excluded EDSS assessments performed during relapses.

2.3. Definitions

CDP corresponds to an increase in EDSS score maintained on repeated evaluation at 3 or 6 months, whereas sustained disability progression is defined as CDP that is sustained to the end of study corresponding to irreversible worsening during the period of observation.

2.4. Statistical Methods

2.4.1. Main Analysis—Core Part

For the main analyses of the core part, data while on double‐blind treatment (either siponimod or placebo) were included in analysis. Thus, analyses for the core part were based on a while on treatment (WOT) set with participants being censored at the time they transitioned to open‐label siponimod treatment. The HR comparing sustained progression to EDSS ≥ 7.0 between siponimod relative to the placebo group was assessed in the core part using Cox proportional hazards models. Kaplan–Meier estimates were derived.

A possible concern with the conventional method used to adjust for baseline differences is that such models might produce unstable estimates when the number of events in certain strata is small [22, 23]. Only a limited number of sustained progression to EDSS ≥ 7.0 events was observed in the baseline EDSS categories < 5.5 and 5.5–6.0 (Table 1). Therefore, as an alternative to conventional covariate adjustment, inverse probability weighting (IPW) using the propensity score (PS) was used. This approach is also recommended for adjustment of survival curves see [24]. Using an IPW approach based on Therneau et al. [25], data were re‐weighted in each treatment group such that the proportion of participants in the following baseline EDSS categories (< 5.5, 5.5–6.0, 6.5) were targeted to that of the overall WOT population. IPW adjustment was used in Cox regression and in Kaplan–Meier analyses for the overall WOT population (Figure 1). Robust variance estimator was used for estimating the variance of the treatment effect [25].

TABLE 1.

Number (%) of sustained EDSS ≥ 7.0 events by baseline EDSS categories (WOT set).

Baseline EDSS category Siponimod, N = 1097 Placebo, N = 545 HR (95% CI)
n/m (%) Years at risk (EAIR) n/m (%) Years at risk (EAIR)
< 5.5 1/389 (0.3) 643 (0.2) 3/193 (1.6) 306 (1.0) 0.16 (0.02; 1.49)
5.5–6 7/416 (1.7) 656 (1.1) 8/233 (3.4) 370 (2.2) 0.49 (0.18; 1.34)
6.5 57/292 (19.5) 428 (13.3) 33/119 (27.7) 151 (21.9) 0.59 (0.38; 0.90)

Note: n = number of sustained EDSS 7.0 events; m = number of patients per category; EAIR = n/years at risk × 100.

Abbreviations: EAIR, exposure‐adjusted incidence rate; EDSS, Expanded Disability Status Scale; WOT, while on treatment set.

FIGURE 1.

FIGURE 1

Percentage of participants with progression to EDSS ≥ 7.0 sustained until last observation in the core part——while on treatment (WOT). In the derivation of the Cox regression and Kaplan–Meier estimates, data were re‐weighted using IPW, such that the proportion of participants in the following subgroups (baseline EDSS < 5.5, 5.5–6.0, 6.5) was balanced between siponimod and placebo, and matched that of the overall population. Numbers at risk are not adjusted. CI, confidence interval; EDSS, Expanded Disability Status Scale; HR, hazard ratio; IPW, inverse probability weighting; WOT, while on treatment.

The effect of siponimod versus placebo was also analysed for the subgroup of participants with baseline EDSS 6.5 (Figure 2). Treatment by log (time) interaction terms was tested to assess the proportional hazards assumption (p > 0.05 is expected).

FIGURE 2.

FIGURE 2

Percentage of participants with progression to EDSS ≥ 7.0 sustained until last observation in the core part—subgroup of participants with baseline EDSS 6.5—while on treatment (WOT). After onset of progression, an EDSS of ≥ 7.0 was required until the end of the core part in order to include a given patient in this analysis. CI, confidence interval; EDSS, Expanded Disability Status Scale; HR, hazard ratio; KM, Kaplan–Meier; WOT, while on treatment.

2.4.2. Main Analysis—Core Part + Extension Part

For the main analysis of the core part + extension part analysis, observed siponimod data were presented up to 6 years. Competing deaths which occurred in patients who did not experience sustained progression to EDSS ≥ 7.0 were treated as censored.

To characterise the potential long‐term impact of siponimod on time to requiring a wheelchair, the Weibull model (data not shown) and the multi‐state Markov model were evaluated for extrapolation of observed data from the core part (up to 37 months) into the future. Ultimately, the multi‐state model in continuous time was selected to extrapolate data (Figure 3). The time at which 50% of participants were expected to have reached EDSS ≥ 7.0 was estimated; extrapolation beyond the median survival time is not reported. This multi‐state model represents a series of successively more severe disease stages. Notably, due to the relatively small number of transition events and to increase the stability of parameter estimates, we did not consider all possible transitions between EDSS scores. Instead, we grouped EDSS scores into three intermediate disease stages (EDSS < 5.5, 5.5–6.0, 6.5) of comparable sojourn time duration and an absorbing ‘sustained EDSS ≥ 7.0’ state. Transient EDSS scores of 7.0 were included in the 6.5 disease stage. Individuals may advance into, or recover from, adjacent disease stages. Transitions to the next disease stage were considered in case the next state was confirmed at a subsequent scheduled visit. Comparison across study arms was evaluated using a bootstrap method for hypothesis testing proposed by Efron and Tibshirani [26]; 5000 bootstrap samples were drawn.

FIGURE 3.

FIGURE 3

Percentage of participants with progression to EDSS ≥ 7.0—overall population (WOT; based on extrapolated and observed data). In the derivation of the Cox regression and Kaplan–Meier estimates, data were re‐weighted using IPW, such that the proportion of participants in the following subgroups (baseline EDSS < 5.5, 5.5–6.0, 6.5) were balanced between siponimod and placebo and matched that of the overall population. CP, core part; EDSS, Expanded Disability Status Scale; EP, extension part; HR, hazard ratio; IPW, inverse probability weighting; KM, Kaplan–Meier; WOT, while on treatment set.

2.4.3. Supplementary Analyses

For the validation and comparison of the performance of the multi‐state Markov model and Weibull model (adjusted using IPW), absolute difference between extrapolation and observed data (up to 6 years) was derived from 5000 bootstrap samples. Ultimately, the multi‐state model in continuous time was used to extrapolate data. To provide support for the extrapolation, estimates based on the multi‐state model were compared graphically with observed data (i.e., Kaplan–Meier estimates) of time to EDSS ≥ 7.0 sustained until end of follow‐up in the extension part (data up to 6 years). These observed data were obtained for the subset of participants who received at least one dose of siponimod in the core part or in the extension part. The Kaplan–Meier estimates for the siponimod data up to 6 years were adjusted using IPW and targeted the overall WOT. Data from the extrapolation were also compared to the IPW Kaplan–Meier curves for placebo‐ and siponimod‐treated participants in the core part.

All analyses were conducted using SAS version 9.4 (SAS Institute Inc.) and R (version 4.1.x/4.2.x; R Core Team) [27, 28].

3. Results

In total, 1642/1651 participants (siponimod, n = 1097; placebo, n = 545) from the overall population and 411 participants (siponimod, n = 292; placebo, n = 119) from the baseline EDSS 6.5 subgroup were included in the survival analysis. The multi‐state Markov model analysis was also applied to the overall population. The analysis conducted for the FAS yielded similar results to the WOT set (data not shown).

In the baseline EDSS category < 5.5 there was 1 incident (out of 389 participants) of sustained progression to EDSS ≥ 7.0 in the siponimod group with a median (Interquartile range [IQR]) time of sustainment of 591 (450–803) days; in the placebo group 3/193 experienced sustained progression to EDSS ≥ 7.0 over a median (IQR) time of 550 (434–749) days (HR siponimod vs. placebo = 0.16; 95% CI: 0.02–1.49). In the baseline EDSS category 5.5–6 there were 7/416 incidences of sustained progression to EDSS ≥ 7.0 in the siponimod group sustained for a median (IQR) of 569 (440.5–735) days and 8/233 in the placebo group sustained over a median (IQR) of 560 (408–731) days (HR = 0.49; 95% CI: 0.18–1.34). In the baseline EDSS = 6.5 category there were 57/292 incidences of sustained progression to EDSS ≥ 7.0 in the siponimod group sustained over a median (IQR) of 522 (362–729.5) days and 33/119 in the placebo group sustained over a median (IQR) of 457 (282–625) days (HR = 0.59; 95% CI: 0.38–0.90).

Competing death events and switch to open‐label siponimod occurred, respectively, in six (siponimod n = 2; placebo n = 4) and 220 patients (siponimod n = 131; placebo n = 89) who did not reach sustained EDSS ≥ 7.0. In the main analysis (i.e., WOT), competing events were treated as censored. Results from supportive competing risks regression models are available in Table A2. Comparison of baseline characteristics between siponimod and placebo was evaluated based on standardised mean differences (SMD) in Table A1. An imbalance was noted for baseline EDSS (SMD 0.11). Differences in the mean or prevalence of other covariates were negligible (SMD < 0.1).

3.1. Adjusted Survival Analysis in the Overall Population

To deal with the imbalance noted in terms of baseline EDSS (Table 1, Table A1) and considering baseline EDSS is prognostically important on the outcome, survival analyses were adjusted for baseline EDSS categories (< 5.5, 5.5–6.0, 6.5) targeting the overall population using IPW [25]. Adjusting for additional baseline variables had little impact on estimates (Table A2). Based on adjusted Kaplan–Meier estimates, the cumulative percentage of participants who reached sustained EDSS ≥ 7.0 at Month 24 was 6.8% for those on siponimod versus 12.0% for those on placebo. The 25th percentile was not reached for the overall population.

Based on the adjusted Cox proportional hazards model, the reduction in risk of reaching sustained EDSS ≥ 7.0 was 40% (HR, 0.60; 95% CI, 0.41–0.88; p = 0.009) for participants on siponimod versus those on placebo (Figure 1). Of note, the event rates in the baseline EDSS categories of “< 5.5” and “5.5–6.0” were low (Table 1), thus the precision of HR estimates was mainly driven by the EDSS 6.5 subgroup. The direction of the siponimod effect was consistent across baseline EDSS subgroups.

For participants with active SPMS in the overall population, siponimod reduced the risk of reaching sustained EDSS ≥ 7.0 by 51% (HR, 0.49; 95% CI, 0.29–0.81; p = 0.005); for participants with non‐active SPMS, the reduction in risk of reaching sustained EDSS ≥ 7.0 was not statistically significant with siponimod treatment (22% [HR, 0.78; 95% CI, 0.42–1.45; p = 0.437]).

3.2. Survival Analysis in the EDSS 6.5 Subgroup

The Cox proportional hazards model was also used to estimate time to reaching sustained EDSS ≥ 7.0 for the baseline EDSS 6.5 subgroup, and revealed that 25% of participants assigned to placebo reached sustained EDSS ≥ 7.0 after approximately 1 year and 4 months. Siponimod delayed time to EDSS ≥ 7.0 by 1 year (2 years and 4 months). The reduction in risk of reaching sustained EDSS ≥ 7.0 in this subgroup was 41.4% (HR, 0.59; 95% CI, 0.38–0.90; p = 0.015) for participants on siponimod versus those on placebo (Figure 2). The percentage of participants requiring a wheelchair while on siponimod was 19.5% and the percentage requiring a wheelchair while on placebo was 27.7%. The proportional hazards assumption appeared reasonable (p > 0.05).

In the subgroup of participants with baseline EDSS 6.5 and with active SPMS, the reduction in risk of reaching sustained EDSS ≥ 7.0 was 49.3% (HR, 0.51; 95% CI, 0.28–0.91; p = 0.024). The percentage of active SPMS participants requiring a wheelchair while on siponimod was 20.9% and while on placebo was 32.2%. In participants with baseline EDSS 6.5 and with non‐active SPMS, the reduction in risk was 32.3% (HR, 0.68; 95% CI, 0.35–1.32; p = 0.253). The percentage of participants requiring a wheelchair while on siponimod was 17.7% and while on placebo was 22.8%. The observed reduction in time to sustained EDSS ≥ 7.0 was greatest in the active SPMS subgroup.

3.3. Main Analysis—Core Part + Extension Part: Extrapolated Markov Model Data

In total, 1642 participants (siponimod, n = 1097; placebo, n = 545) from the core part and 1481 participants treated with siponimod were included from the extension part.

The HR and 95% confidence intervals for the Markov model are provided in Table 2. Compared to extrapolation based on the Weibull model, the extrapolation based on the Markov model fitted more closely with the observed data (up to 6 years); the absolute difference between extrapolation and observed data averaged over 6 years was 4.2% (standard error [SE]: 0.59) for the Weibull model and 1.6% (SE: 0.57) for the Markov model. To further appraise the consistency between extrapolation based on the Markov model and observed data, estimates from the Markov model are presented in Figure 3. Only extrapolation up to around the median time to reaching sustained EDSS ≥ 7.0 is presented. Extrapolated siponimod delayed the median time to sustained EDSS ≥ 7.0 by 5.8 years compared with placebo (15.3 vs. 9.4 years, p = 0.0134) in the overall population (using the WOT). The median time to sustained EDSS ≥ 7.0 was delayed by 7.9 years (15.5 vs. 7.6 years, p = 0.015) in the active SPMS subgroup and by 3.0 years (15.5 vs. 12.6 years, p = 0.398) in the non‐active SPMS subgroup for participants on siponimod compared with those on placebo (Figure A1). Of note, time to sustained EDSS ≥ 7.0 progressed more slowly with placebo in the non‐active SPMS population as compared to placebo in the overall and active SPMS populations.

TABLE 2.

Transition to wheelchair (excluding assessments performed during relapse)—hazard ratio.

EDSS category HR (95% CI)
≤ 5—5.5–6 0.666 (0.52–0.853)
5.5–6—≤ 5 0.924 (0.639–1.335)
5.5–6—≥ 6.5 0.985 (0.802–1.209)
≥ 6.5—5.5–6 0.914 (0.65–1.284)
≥ 6.5—Wheelchair 0.856 (0.65–1.126)

4. Discussion

Reaching EDSS ≥ 7.0 (requiring a wheelchair) is a significant milestone for patients with SPMS and the delay in progression to walking impairment remains an unmet need in this disease. It should be noted that participants from the EXPAND study with baseline EDSS 6.5 were the subgroup most at risk of transitioning to wheelchair dependency. The adjusted survival analysis in the overall population found a 40% risk reduction (p = 0.009) in patients progressing to EDSS ≥ 7.0 for siponimod versus those on placebo. A stronger risk reduction of 51% (p = 0.005) was observed in the active SPMS subgroup, while a risk reduction of 22% (p = 0.437) was observed in the non‐active SPMS subgroup; this effect was not statistically significant. These results are in line with findings from the long‐term EXPAND study (5 years) where participants with non‐active SPMS appeared to progress more slowly than participants with active SPMS, suggesting that a longer follow‐up period may be required to see the full treatment effect of siponimod versus placebo in non‐active SPMS [19]. The reduction in risk of patients advancing to a wheelchair highlights the efficacy of siponimod in slowing disability progression in patients with SPMS [19].

Statistical models were used to extrapolate the potential long‐term impact of siponimod on time to requiring a wheelchair beyond the core part of the EXPAND study. This extrapolation may be used to contextualise the results of the core part and was necessary due to the duration of placebo treatment in the core part being too short for the analysis. For patients with lower EDSS, it would make sense to estimate the time to the subsequent intermediate disease stage (hence the rationale for applying the multi‐state Markov model). Analysis using the multi‐state model to extrapolate the data over a longer time frame indicated a median increase of 5.8 years in the time to wheelchair dependency with siponimod use versus placebo. In EXPAND, the median time to a wheelchair for placebo‐treated patients with non‐active SPMS was longer (12.6 years) than for placebo‐treated patients with active SPMS (7.6 years) and for the overall SPMS population (9.4 years). This suggests that a longer observation time is needed to see the effects of the disease in patients with non‐active SPMS.

In investigations of time to requiring a wheelchair carried out in PPMS data from the ORATORIO study, a younger population with shorter disease duration and lower levels of disability (baseline median EDSS 4.5) than a typical PPMS population [16] was assessed. This cohort also presented a higher proportion of T1 gadolinium‐enhancing lesions compared with the EXPAND population [15, 16, 18]. Risk of requiring a wheelchair was reduced by up to 46% versus placebo in PPMS patients on ocrelizumab [16]. Median time to 24‐week confirmed wheelchair dependency was estimated as 12.4 years in patients with PPMS, as compared with an untreated PPMS cohort in the online registry MSBase. Data were extrapolated from the double‐blind phase and the extended controlled period of the ORATORIO study using the Weibull model, and suggested that ocrelizumab treatment delayed the median time to requiring a wheelchair by 7.1 years versus placebo [16].

For extrapolating data beyond the follow‐up period of a trial (here: the placebo‐controlled core part of the EXPAND study), survival models typically have a parametric specification, which means that they rely on assumed distribution of probabilities of outcome events (here: requiring wheelchair). The Cox model could not be used for this purpose because it is a semi‐parametric approach. Two parametric models, that is, the Weibull and the multi‐state Markov models, were evaluated in this analysis. Since the Markov model fitted the observed data more closely in EXPAND, this model was used for extrapolating data beyond the follow‐up of the placebo‐controlled core part. In future research, it could be of interest to evaluate alternative models. The multi‐state model has the additional advantage of being able to take into account recurrences and improvement, while the Weibull model focuses on time to a single progression event (see Table A3).

The modelled siponimod arm in the overall SPMS population from the EXPAND study closely matched the observed siponimod long‐term data (core+extension), supporting the robustness of the data extrapolation method based on the multi‐state model. In future research, it would be of interest to further assess the generalisability of these extrapolations using external data sources with longer observation periods, such as disease registries.

Study limitations included: (a) the post hoc nature of this analysis, which limited the interpretation of these findings to be exploratory rather than confirmatory; (b) the limited median core study duration of 21 months; (c) the low frequency of outcome of interest—the EXPAND trial was powered to detect differences between siponimod and placebo for a minimum of 374 3‐month CDP events, but only 106 wheelchair events were observed and not all were confirmed/sustained or at least 3‐months; (d) it was assumed that the rate of disability progression was the same during the core part and the extension part of the study; and (e) the analyses presented here were only adjusted for Baseline EDSS; however, other covariate adjustments were explored with minimal contribution to the findings.

Biases may also have been introduced from participants who withdrew from the study or study drug before the end of the core part and those who changed from placebo to siponimod during the core part, after which their EDSS was censored.

Observed data for the placebo arm beyond follow‐up time in the core part were not available. While it would be of interest to explore borrowing long‐term real‐world data (RWD) for untreated patients to support these analyses, obtaining RWD that can be compared with EXPAND data poses a number of challenges: RWD include patients who declined or were denied treatment options when all patients in EXPAND had an intent to be treated, and operational differences between EXPAND and RWD may introduce an additional source of bias. It might also be complicated to obtain RWD that are contemporaneous to EXPAND and collected under comparable standards of care.

The delay in time to wheelchair reported here further corroborates the clinical efficacy of siponimod in the treatment of SPMS that was previously demonstrated by reductions in 3‐ and 6‐month CDP, reductions in worsening of cognitive processing speed, and MRI measures, including whole brain and regional grey matter volume loss [19, 29]. Furthermore, analyses demonstrate that patients with SPMS benefited more from continued siponimod treatment versus siponimod initiated later after a period on placebo, thus supporting the benefit of earlier treatment initiation [19]. In addition to the challenges of increased physical disability, transition to wheelchair dependency is linked to poorer quality of life and increased healthcare costs [5, 30]. As such, delaying wheelchair dependency is a highly relevant treatment goal and an important indicator of treatment effects in SPMS.

Progression to wheelchair dependency is associated with a relevant decline in quality of life for patients with SPMS, and therefore delaying time to requiring a wheelchair (sustained EDSS ≥ 7.0) is an important target for treatment and a meaningful endpoint for these patients. The results of these analyses showed that siponimod delayed time to becoming wheelchair‐dependent in the overall SPMS population, with a stronger effect in those with active disease.

Author Contributions

Sophie Arnould: formal analysis, visualization, writing – review and editing. Nicolas Rouyrre: methodology, writing – review and editing. Goeril Karlsson: conceptualization, writing – review and editing. Ralf Gold: investigation, writing – review and editing. Robert Fox: investigation, writing – review and editing. Soudeh Ansari: methodology, formal analysis, writing – review and editing. Patrick Vermersch: conceptualization, investigation, writing – review and editing. Daniela Piani Meier: conceptualization, writing – review and editing. Amit Bar‐Or: investigation, writing – review and editing. Bruce A. C. Cree: investigation, writing – review and editing. Ludwig Kappos: investigation, writing – review and editing. Gavin Giovannoni: conceptualization, investigation, writing – review and editing.

Funding

This work was supported by Novartis Pharma AG, Basel, Switzerland.

Conflicts of Interest

P.V. has received honoraria and consulting fees from Biogen, Sanofi, Teva, Novartis, Merck, Imcyse, Roche and AB Science, and research support from Biogen, Sanofi and Merck. R.G. has received compensation for serving as a consultant or speaker from Bayer HealthCare, Biogen Idec, Merck Serono, Novartis and Teva Neuroscience. He, or the institution he works for, has received research support from Bayer HealthCare, Biogen Idec, Merck Serono, Novartis and Teva Neuroscience. He has also received honoraria as a Journal Editor from SAGE and Thieme Verlag. L.K. has received no personal compensation. His institutions (University Hospital Basel/Stiftung Neuroimmunology and Neuroscience Basel) have received the following exclusively for research support: steering committee, advisory board and consultancy fees (Abbvie, Actelion, Auriga Vision AG, Bayer HealthCare, Biogen, Celgene, Dörries Frank‐Molnia & Pohlman, Eli Lilly, EMD Serono, Genentech, Genzyme, Glaxo Smith Kline, Janssen, Merck, Minoryx, Novartis, Roche, Sanofi, Santhera, Senda Biosciences, Shionogi and Wellmera AG); speaker fees (Bristol Myers Squibb, Celgene, Janssen, Merck, Novartis and Roche); support for educational activities (Biogen, Desitin, Novartis, Sanofi and Teva); licence fees for Neurostatus products; and grants (European Union, Innosuisse, Novartis, Roche, Swiss MS Society and Swiss National Research Foundation). R.F. has received compensation for serving as a consultant or advisory committee member for AB Science, Biogen, Celgene, EMD Serono, Genentech, Genzyme, Greenwich Biosciences, Immunic, Janssen, Novartis, Sanofi and TG Therapeutics and has received clinical trial contract and research grant funding from Biogen, Novartis and Sanofi. He, or the institution he works for, has received research support from Novartis. A.B.‐O. has participated as a speaker in meetings sponsored by and received consulting fees and/or grant support from Accure, Atara Biotherapeutics, Biogen, BMS/Celgene/Receptos, GlaxoSmithKline, Gossamer, Janssen/Actelion, Medimmune, Merck/EMD Serono, Novartis, Roche/Genentech, Sanofi Genzyme. B.A.C.C. has received personal compensation for consulting from Alexion, Atara, Autobahn, Avotres, Biogen, EMD Serono, Gossamer Bio, Hexal/Sandoz, Horizon, Neuron23, Novartis, Sanofi, Siemens, TG Therapeutics and Therini and received research support from Genentech. G.G. a steering committee member on the daclizumab trials for AbbVie, the BG12 and daclizumab trials for Biogen, the fingolimod and siponimod trials for Novartis, the laquinimod trials for Teva and the ocrelizumab trials for Roche. He has also received consultancy fees for advisory board meetings for oral cladribine trials for Merck KGaA, Sanofi Genzyme, and in relation to DSMB activities for Synthon BV, as well as honoraria for speaking at the Physicians' Summit and several medical education meetings. He is also the Co‐Chief Editor of Multiple Sclerosis and Related Disorders (Elsevier). S.Ar, S.An, G.K. and N.R. are employees of Novartis. D.P.M. was an employee of Novartis during the development of this paper.

Acknowledgements

The authors thank Brianna Fitzmaurice of Novartis Ireland Limited and Anuradha Nalli of Novartis Healthcare Private Limited, India for providing medical writing support. The funder had a role in the study design; in the collection, analysis and interpretation of data; in the writing of the manuscript; and in the decision to submit the article for publication.

Appendix A. Methods

Comparability of participants randomised to siponimod versus placebo in terms of baseline characteristics was evaluated using standardised mean differences (SMD) (Table A1). The SMD is one of the most widely used quantities to assess covariate balance although it should be interpreted with caution for non‐normal distributions. An SMD < 0.1 has been considered to indicate negligible differences [31].

Inverse probability weights were used to derive Kaplan–Meier curves. The approach described in Therneau et al. [25] was used to derive the weights. In brief:

  • in the analysis of the core part data, each observation in the data set were given a weight of wBLEDSSi=PrBLEDSSPrBLEDSSi, where PrBLEDSS is the probability of baseline EDSS categories in the target population (here: overall EXPAND population) and PrBLEDSSi is the probability of each baseline EDSS/treatment group combination. The weights obtained with this approach, using a single categorical baseline covariate, is equivalent to the conventional IPW‐ATE (inverse probability weighting‐average treatment effect) weights (for a review of propensity score methods, see Austin [31]). In the Cox‐model, such weights allow us to estimate the average treatment effect using the EXPAND population as the reference/target population. An alternative could have been to use siponimod‐treated patients as the reference population; the weights that would have been obtained using this approach are the IPW‐ATE weights (average treatment effect in the treated)

  • and for the derivation of IPW Kaplan–Meier curves for the siponimod data up to 6 years, weights were also derived using the overall EXPAND population as the reference/target population.

TABLE A1.

Comparability of baseline characteristics between treatment groups—Adjusted and unadjusted standardised mean difference (SMD) (WOT).

Baseline characteristics Siponimod N = 1097 Placebo N = 545 SMD Adjusted SMD
Age, mean (SD) 48.0 (7.83) 48.1 (7.94) 0.015 0.016
Sex = Male 435 (39.7%) 222 (40.7%) 0.022 0.018
EDSS
< 5.5 389 (35.5%) 193 (35.4%) 0.112 0.000
5.5–6 416 (37.9%) 233 (42.7%)
6.5 292 (26.6%) 119 (21.8%)
EDSS, mean (SD) 5.4 (1.08) 5.4 (1.02)
Median 6.0 6.0
[Q1; Q3] [4.5; 6.5] [4.5; 6.0]
T25FWT, mean (SD) 17.1 (20.87) 16.0 (22.10) 0.051 0.007
Active SPMS = Yes 515 (48.1) 263 (49.4) 0.027 0.028
With relapse in the last 2 years 387 (35.4) 202 (37.1) 0.037 0.033
MS disease duration since diagnosis (years), mean (SD) 12.9 (7.9) 12.1 (7.5) 0.095 0.080
Gd+ lesion ≥ 1 236 (22.2%) 114 (21.6%) 0.015 0.009
T2 volume (mm3), mean (SD) 15,598 (16,207) 14,717 (15,626) 0.055 0.042
Normalised brain volume (cc), mean (SD) 1422 (86.4) 1424 (87.7) 0.029 0.027
Previous DMT = yes 856 (78.0%) 431 (79.1%) 0.026 0.027

Note: Three subjects with baseline EDSS 7.0 were excluded from this analysis. For categorical baseline characteristics, number (%) of patients are reported. Adjusted using inverse probability weights (IPW), weights were derived based on baseline EDSS categories < 5.5, 5.5–6.0, 6.5. SMD < 0.1 indicate neglible difference, SMD ≥ 0.1 indicate imbalance. Value in bold indicates SMD above the threshold that indicates non‐comparability of the two arms, but after adjustment is back below the threshold.

Abbreviations: DMT, disease‐modifying therapy; EDSS, Expanded Disability Status Scale; Gd+, gadolinium‐enhancing; MS, multiple sclerosis; Q1, Quarter 1; Q3, Quarter 3; SD, standard deviation; SMD, standardised mean difference; SPMS, secondary progressive multiple sclerosis; T25FWT, Timed 25‐Foot Walk Test; WOT, while on treatment set.

TABLE A2.

Alternative regression models of progression to EDSS ≥ 7.0 sustained until last observation in the core part—overall population (WOT).

Endpoint model HR 95% CI p
Endpoint: EDSS ≥ 7.0 sustained until last observation
Cox model adjusted using IPW (Therneau et al. [25])
(Main analysis) PS derived based on: BL‐EDSS categories 0.60 (0.41;0.88) 0.009
(S1) PS derived based on: BL‐EDSS categories and relapse in the last 2 years 0.60 (0.41;0.89) 0.011
(S2) Cox model with standard covariate adjustment
(S2a) Covariate: BL‐EDSS 0.54 (0.37; 0.80) 0.002
(S2b) Covariates: BL‐EDSS and relapse in the last 2 years 0.55 (0.38; 0.81) 0.003
(S3) Subdistribution hazard (Fine and Grey) a
(S3a) Competing risk: death 0.55 (0.38; 0.81) 0.002
(S3b) Competing risks: death & transition to open‐label siponimod 0.56 (0.38; 0.81) 0.003
Cause‐specific hazard
(S3c) Competing risk: death 0.54 (0.37; 0.80) 0.002
(S3d) Competing risks: death & transition to open‐label siponimod 0.55 (0.37; 0.80) 0.002
(S4) Composite endpoint of EDSS ≥ 7.0 sustained until last observation and death
Cox model adjusted using IPW with PS derived based on BL‐EDSS categories 0.59 (0.41; 0.86) 0.006

Abbreviations: BL, baseline; CI, confidence interval; EDSS, Expanded Disability Status Scale; HR, hazard ratio; IPW, inverse probability weighting; PS, Propensity Score; WOT, while on treatment set.

a

Model adjusted based on baseline EDSS categories (< 5.5, 5.5–6.0, 6.5).

TABLE A3.

Disease aspects captured with the multi‐state model but not with standard survival analysis.

Multi‐state model Standard survival model (Cox, Weibull)
Event of interest (i.e., absorbing event): Wheelchair (i.e., sustained EDSS ≥ 7.0)
Recurrent events Single event
Improvement and Worsening Worsening
Path from lower EDSS scores up to wheelchair can be modelled Information from lower EDSS scores is lost
More flexible model (i.e., in terms of number of parameters estimated) Simpler model
Computationaly demanding Does not require extensive computation time

Abbreviation: EDSS, Expanded Disability Status Scale.

FIGURE A1.

FIGURE A1

Percentage of participants with progression to EDSS ≥ 7.0—overall population (WOT; based on extrapolated from multistate Markov model and observed data). The above analysis can be viewed from different perspectives: (1) observed data (KM) are useful to assess the validity of the multi‐state Markov model, (2) standard KM and the multi‐state model complement each other in that the observed data (KM) in the core part were mostly based on information from patients with EDSS 6.5 when extrapolation (Markov model) accounted for information from a lower EDSS score. In particular, extrapolation helps to infer the effect of siponimod on delaying wheelchair dependence in the overall SPMS population, including patients with a lower EDSS score. CP, core part; EDSS, Expanded Disability Status Scale; EP, extension part; KM, Kaplan–Meier; SPMS, secondary progressive multiple sclerosis.

Vermersch P., Arnould S., Gold R., et al., “Progression to Wheelchair in Secondary Progressive Multiple Sclerosis and Impact of Siponimod: Post Hoc Analyses From the EXPAND Study,” European Journal of Neurology 33, no. 6 (2026): e70659, 10.1111/ene.70659.

Sophie Arnould and Daniela Piani Meier were affiliated with Novartis when the study was conducted.

Data Availability Statement

If you are interested in accessing the data used for the analysis reported in this publication, then please contact the study sponsor. Data access and full description of parameters entered into the model as used in this study are available on request and in accordance with the determination of appropriate data usage and subject to all necessary data protection legislation and ethics approval conditions being fulfilled.

References

  • 1. Hirst C., Ingram G., Swingler R., Compston D. A. S., Pickersgill T., and Robertson N. P., “Change in Disability in Patients With Multiple Sclerosis: A 20‐Year Prospective Population‐Based Analysis,” Journal of Neurology, Neurosurgery, and Psychiatry 79, no. 10 (2008): 1137–1143. [DOI] [PubMed] [Google Scholar]
  • 2. Rolak L. A., “Multiple Sclerosis: It's Not the Disease You Thought It Was,” Clinical Medicine & Research 1, no. 1 (2003): 57–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Scalfari A., Neuhaus A., Daumer M., Muraro P. A., and Ebers G. C., “Onset of Secondary Progressive Phase and Long‐Term Evolution of Multiple Sclerosis,” Journal of Neurology, Neurosurgery, and Psychiatry 85, no. 1 (2014): 67–75. [DOI] [PubMed] [Google Scholar]
  • 4. Tremlett H., Yinshan Z., and Devonshire V., “Natural History of Secondary‐Progressive Multiple Sclerosis,” Multiple Sclerosis 14, no. 3 (2008): 314–324. [DOI] [PubMed] [Google Scholar]
  • 5. Gross H. J. and Watson C., “Characteristics, Burden of Illness, and Physical Functioning of Patients With Relapsing‐Remitting and Secondary Progressive Multiple Sclerosis: A Cross‐Sectional US Survey,” Neuropsychiatric Disease and Treatment 13 (2017): 1349–1357. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Larochelle C., Uphaus T., Prat A., and Zipp F., “Secondary Progression in Multiple Sclerosis: Neuronal Exhaustion or Distinct Pathology?,” Trends in Neurosciences 39, no. 5 (2016): 325–339. [DOI] [PubMed] [Google Scholar]
  • 7. Lublin F. D. and Reingold S. C., “Defining the Clinical Course of Multiple Sclerosis: Results of an International Survey. National Multiple Sclerosis Society (USA) Advisory Committee on Clinical Trials of New Agents in Multiple Sclerosis,” Neurology 46, no. 4 (1996): 907–911. [DOI] [PubMed] [Google Scholar]
  • 8. Kutzelnigg A., Lucchinetti C. F., Stadelmann C., et al., “Cortical Demyelination and Diffuse White Matter Injury in Multiple Sclerosis,” Brain 128, no. 11 (2005): 2705–2712. [DOI] [PubMed] [Google Scholar]
  • 9. Cattaneo D., Gervasoni E., Anastasi D., et al., “Prevalence and Patterns of Subclinical Motor and Cognitive Impairments in Non‐Disabled Individuals With Early Multiple Sclerosis: A Multicenter Cross‐Sectional Study,” Annals of Physical and Rehabilitation Medicine 65, no. 1 (2021): 101491. [DOI] [PubMed] [Google Scholar]
  • 10. Larocca N. G., “Impact of Walking Impairment in Multiple Sclerosis: Perspectives of Patients and Care Partners,” Patient 4, no. 3 (2011): 189–201. [DOI] [PubMed] [Google Scholar]
  • 11. Thompson A. J., Baranzini S. E., Geurts J., Hemmer B., and Ciccarelli O., “Multiple Sclerosis,” Lancet 391, no. 10130 (2018): 1622–1636. [DOI] [PubMed] [Google Scholar]
  • 12. Zwibel H. L., “Contribution of Impaired Mobility and General Symptoms to the Burden of Multiple Sclerosis,” Advances in Therapy 26, no. 12 (2009): 1043–1057. [DOI] [PubMed] [Google Scholar]
  • 13. Baird J. F., Sandroff B. M., and Motl R. W., “Therapies for Mobility Disability in Persons With Multiple Sclerosis,” Expert Review of Neurotherapeutics 18, no. 6 (2018): 493–502. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. van Asch P., “Impact of Mobility Impairment in Multiple Sclerosis 2—Patients' Perspectives,” European Neurological Review 6, no. 2 (2011): 115–120. [Google Scholar]
  • 15. Montalban X., Hauser S. L., and Kappos L., “Ocrelizumab Versus Placebo in Primary Progressive Multiple Sclerosis,” New England Journal of Medicine 376, no. 3 (2017): 209–220. [DOI] [PubMed] [Google Scholar]
  • 16. Butzkueven H., Spelman T., Horakova D., et al., “Risk of Requiring a Wheelchair in Primary Progressive Multiple Sclerosis: Data From the ORATORIO Trial and the MSBase Registry,” European Journal of Neurology 29 (2022): 1082–1090. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Mandel M., Mercier F., Eckert B., Chin P., and Betensky R. A., “Estimating Time to Disease Progression Comparing Transition Models and Survival Methods—An Analysis of Multiple Sclerosis Data,” Biometrics 69, no. 1 (2013): 225–234. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Kappos L., Bar‐Or A., Cree B. A. C., et al., “Siponimod Versus Placebo in Secondary Progressive Multiple Sclerosis (EXPAND): A Double‐Blind, Randomised, Phase 3 Study,” Lancet 391, no. 10127 (2018): 1263–1273. [DOI] [PubMed] [Google Scholar]
  • 19. Cree B. A., Arnold D. L., Fox R. J., et al., “Long‐Term Efficacy and Safety of Siponimod in Patients With Secondary Progressive Multiple Sclerosis: Analysis of EXPAND Core and Extension Data up to >5 Years,” Multiple Sclerosis 28, no. 10 (2022): 1591–1605. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Rovaris M., Confavreux C., Furlan R., Kappos L., Comi G., and Filippi M., “Secondary Progressive Multiple Sclerosis: Current Knowledge and Future Challenges,” Lancet Neurology 5, no. 4 (2006): 343–354. [DOI] [PubMed] [Google Scholar]
  • 21. Lublin F. D., Reingold S. C., Cohen J. A., et al., “Defining the Clinical Course of Multiple Sclerosis: The 2013 Revisions,” Neurology 83, no. 3 (2014): 278–286. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. European Medicines Agency , Guideline on Adjustment for Baseline Covariates in Clinical Trials (European Medicines Agency, 2015). [Google Scholar]
  • 23. Williamson E. J., Forbes A., and White I. R., “Variance Reduction in Randomised Trials by Inverse Probability Weighting Using the Propensity Score,” Statistics in Medicine 33, no. 5 (2014): 721–737. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Cole S. R. and Hernan M. A., “Adjusted Survival Curves With Inverse Probability Weights,” Computer Methods and Programs in Biomedicine 75 (2004): 45–49. [DOI] [PubMed] [Google Scholar]
  • 25. Therneau T. M., Crowson C. S., and Atkinson E. J., “Adjusted Survival Curves,” (2015), https://cran.r‐project.org/web/packages/survival/vignettes/adjcurve.pdf.
  • 26. Efron B. and Tibshirani R. J., An Introduction to the Bootstrap (Chapman and Hall, 1993). [Google Scholar]
  • 27. R Core Team , “R: A Language and Environment for Statistical Computing (Version 4.1.x or 4.2.x),” [Software] (2023).
  • 28. SAS Institute Inc , “SAS Software (Version 9.4),” [Software] (2023).
  • 29. Arnold D. L., Piani‐Meier D., Bar‐Or A., et al., “Effect of Siponimod on Magnetic Resonance Imaging Measures of Neurodegeneration and Myelination in Secondary Progressive Multiple Sclerosis: Gray Matter Atrophy and Magnetization Transfer Ratio Analyses From the EXPAND Phase 3 Trial,” Multiple Sclerosis 28, no. 10 (2022): 1526–1540. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Jones E., Pike J., Marshall T., and Ye X., “Quantifying the Relationship Between Increased Disability and Health Care Resource Utilization, Quality of Life, Work Productivity, Health Care Costs in Patients With Multiple Sclerosis in the US,” BMC Health Services Research 16 (2016): 294. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Austin P. C., “An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies,” Multivariate Behavioral Research 46, no. 3 (2011): 399–424. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

If you are interested in accessing the data used for the analysis reported in this publication, then please contact the study sponsor. Data access and full description of parameters entered into the model as used in this study are available on request and in accordance with the determination of appropriate data usage and subject to all necessary data protection legislation and ethics approval conditions being fulfilled.


Articles from European Journal of Neurology are provided here courtesy of John Wiley & Sons Ltd on behalf of European Academy of Neurology (EAN)

RESOURCES