Sage Journals: Discover world-class research

Abstract

Background:

Late-onset multiple sclerosis (LOMS), defined by symptom onset after age 50, is increasingly recognised as a distinct clinical entity. Evidence comparing disease-modifying therapies (DMTs) in this subgroup remains limited.

Objectives:

To compare clinical outcomes of anti-CD20 monoclonal antibodies and sphingosine-1-phosphate receptor modulators (S1PRMs) in LOMS.

Design:

Multicentre, observational cohort study based on real-world data from an international multiple sclerosis registry.

Methods:

We analysed data from the MSBase registry, including relapsing–remitting LOMS patients treated with anti-CD20 therapies (ocrelizumab, ofatumumab, rituximab) or S1PRMs (fingolimod, ozanimod, siponimod, ponesimod) for ⩾6 months. Primary outcomes were annualised relapse rate (ARR), Expanded Disability Status Scale (EDSS) change, confirmed disability worsening (CDW), progression independent of relapse activity (PIRA), and PIRA without MRI activity (PIRMA). Analyses used inverse probability of treatment weighting (IPTW). Causal forest and best linear projector (BLP) models explored effect modification.

Results:

After weighting, 347 patients (median age 53.7 years; 69% female; median follow-up 6.9 years) were included. No significant differences were found for ARR, EDSS change, CDW, PIRA, or PIRMA. Exploratory analyses suggested greater anti-CD20 benefit in patients with earlier onset (⩽55 years), shorter disease duration (⩽2 years from diagnosis), and lower disability (EDSS < 3).

Conclusions:

In this real-world LOMS cohort, no statistically significant differences were observed between anti-CD20 and S1PRM therapies. Exploratory analyses suggested anti-CD20 may be associated with better outcomes in selected subgroups; these findings are hypothesis-generating.

Trial Registration:

Not applicable.

Plain language summary

How two main treatments for late-onset multiple sclerosis compare in real life

Multiple sclerosis (MS) is a disease where the immune system attacks the brain and spinal cord. When MS begins after the age of 50, it is called late-onset multiple sclerosis (LOMS). This form of MS can progress faster and respond differently to treatment compared to cases that start earlier in life. In this study, we compared two important groups of modern MS treatments: - Anti-CD20 therapies, which target specific immune cells, and - S1P receptor modulators, which prevent immune cells from reaching the brain. We analysed real-world data from an international MS registry called MSBase, including people with relapsing LOMS who had used one of these therapies for at least six months. We looked at how often relapses occurred, how disability changed over time, and whether disability increased independently of relapses. The results showed that, overall, both treatment types worked similarly in preventing relapses and disability worsening. However, in some patients—especially those younger than 55, with early disease and lower disability—anti-CD20 therapies appeared to have a small advantage. These findings suggest that both therapies are effective for most people with late-onset MS, but that starting anti-CD20 treatment early may be beneficial for selected patients. This information can help doctors personalise treatment plans for older adults living with MS.

Keywords

anti-CD20 causal forest disability progression late-onset multiple sclerosis real-world evidence S1P receptor modulators

Introduction

Late-onset multiple sclerosis (LOMS), defined by symptom onset after the age of 50, has long been considered uncommon, representing only 4%–6% of diagnoses in earlier cohorts.^1–3 More recently, population-based registries have reported increasing recognition of LOMS, now accounting for up to 10%–12% of new MS cases.^4–6 This trend reflects a demographic shift in ageing populations, as well as improved diagnostic accuracy through broader access to MRI, updated diagnostic criteria, and greater awareness of atypical presentations in older adults.^7,8 National data from Denmark, Sweden and the UK confirm this increase and highlight important clinical distinctions: LOMS patients are more likely to present with progressive phenotypes, motor onset and faster disability accumulation compared with younger-onset cases.^4–6,9,10

Biological mechanisms also differentiate LOMS. Age-related immune remodelling, reduced lymphocyte repertoire diversity, and microglial activation contribute to a pro-inflammatory and neurodegenerative milieu that limits recovery.^11–16 Neuropathological studies report a predominance of chronic active lesions and relatively fewer gadolinium-enhancing lesions, consistent with a disease course that is less inflammatory and more insidiously progressive than in younger patients.¹⁷

These features have direct therapeutic implications. Although the expansion of high-efficacy disease-modifying therapies (DMTs) has transformed MS management, their benefit appears attenuated in older patients, with both age at onset and age at treatment initiation influencing responsiveness.^18–20 At the same time, comorbidities and immunosenescence increase the risk of treatment-related complications, making safety a critical concern in this population.²¹

To date, data are scant and the few available studies have examined treatment efficacy in LOMS by grouping DMTs into broad categories, essentially according to their level of efficacy (e.g., LET), without directly comparing agents that differ in both mechanism of action and molecular size, factors that may influence their ability to cross the blood–brain barrier. In this study, we directly evaluate two therapeutic strategies frequently used in clinical practice—anti-CD20 monoclonal antibodies and sphingosine-1-phosphate receptor modulators (S1PRMs). By analysing relapse outcomes, disability progression and safety outcomes in a real-world LOMS cohort, we aim to provide clinically relevant evidence to guide treatment decisions in this increasingly prevalent subgroup.

Methods

Study population

Demographic and clinical data were collected from patients with LOMS, defined as symptom onset after age 50, enrolled in MSBase, a large international observational registry of patients with multiple sclerosis. Prior to cohort selection, standardised quality-control procedures were applied, as routinely implemented in MSBase studies. Automated scripts were used to assess data accuracy (error rates), data density across key domains (follow-up, demographics, clinical visits, relapses, paraclinical data, therapies) and generalizability, in accordance with previously published MSBase methodology.²²

Patients were included if they had a diagnosis of relapsing–remitting MS (RRMS) established according to the McDonald criteria valid at the time of diagnosis, and if symptom onset occurred between January 1, 2000 and April 30, 2025. Additional inclusion criteria were the availability of a complete baseline EDSS assessment, at least two recorded EDSS evaluations, and treatment with DMTs belonging to either the anti-CD20 class (ocrelizumab, ofatumumab, or rituximab) or the S1PRMs (fingolimod, ozanimod, siponimod or ponesimod), for at least 6 months.

Demographic variables analysed included age at symptom onset, age at diagnosis, sex and disease duration. Clinical variables comprised the type of first clinical manifestation (classified as optic, supratentorial, brainstem or spinal), EDSS score at baseline, the annualised relapse rate (ARR) during the year preceding the initiation of anti-CD20 or S1PRM therapy, and the number of previous DMTs. To account for potential confounding in cumulative probability models, visit frequency (visits per year) and MRI monitoring (MRIs per year) were also included.

This study was designed, conducted and reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines,²³ with a completed checklist and a structured overview of the statistical models (including covariate adjustment and effect measures) provided in the Supplemental Material (Tables S1–S2).

Outcomes and definitions

The primary outcomes considered in this study were the ARR, longitudinal changes in the EDSS, the reaching of EDSS milestones 4.0 and 6.0, confirmed disability worsening (CDW), progression independent of relapse activity (PIRA), and progression independent of relapse activity with no MRI activity (PIRMA).

ARR was defined as the number of relapses divided by patient-years of follow-up.²⁴ A relapse was defined as the onset of new, or worsening of existing, neurological symptoms lasting at least 24 h, not attributable to fever or infection, and separated by at least 30 days from a previous relapse.²⁵

Longitudinal EDSS changes were assessed by evaluating the trajectory of EDSS scores across the follow-up period, based on all available assessments.²⁶ Reaching EDSS milestones was defined as the first EDSS score of at least 4.0 or 6.0, sustained for 6 months, provided that the baseline EDSS was below the respective threshold.⁴

CDW was defined as a sustained increase in EDSS confirmed at 6 months. Specifically, CDW was defined as an increase of ⩾1.5 points if the baseline EDSS was 0.0, ⩾1.0 point if the baseline EDSS was between 1.0 and 5.5, or ⩾0.5 point if the baseline EDSS was > 5.5.²⁷

PIRA was defined as a 6-month confirmed CDW event occurring in the absence of a clinical relapse within 30 days before or 90 days after the worsening event.²⁸ PIRMA was defined as a PIRA event occurring more than 30 days before or more than 90 days after a clinical relapse, in the absence of any new or enlarging T2-hyperintense lesions and/or gadolinium-enhancing lesions on brain MRI within the 12 months preceding the event.²⁹

Safety outcomes included severe infections and malignancies. Severe infections were defined as infections that were life-threatening, opportunistic, required intravenous or prolonged antimicrobial therapy, or showed a severe or prolonged clinical course. Infection events were considered treatment-associated if occurring during DMT exposure or within 6 months after treatment discontinuation. Malignancies were defined as any newly diagnosed neoplastic disease and were considered treatment-associated if occurring during DMT exposure or within 24 months after treatment discontinuation.

Statistical analysis

To minimise confounding from non-random treatment allocation, we applied inverse probability of treatment weighting (IPTW) using propensity scores estimated from baseline demographic and clinical variables, including age at onset and at diagnosis, disease duration, sex, baseline EDSS, annualised relapse rate in the year prior to treatment initiation, first symptom localisation, number of previous DMTs, follow-up time, and surveillance intensity (visits and brain MRI per year). Stabilised weights were used. To account for residual imbalance and potential non-linear relationships between baseline covariates and treatment assignment, spline terms were introduced for covariates showing residual imbalance after weighting: standardised mean differences (SMD) > 0.10. This flexible specification helps to reduce model misspecification and improve covariate balance in IPTW analyses.³⁰ All outcomes were analysed in the weighted population.

ARR was assessed with negative binomial regression (incidence rate ratio, IRR). Longitudinal EDSS change was evaluated using ANCOVA (adjusted mean difference). Time to EDSS 4.0 and 6.0 was analysed with Cox proportional hazards models (hazard ratios, HR). Recurrent disability outcomes (CDW, PIRA, PIRMA) were modelled with Andersen–Gill extensions of the Cox model (HR).

Sensitivity analyses were performed in treatment-naïve patients to avoid confounding from prior therapy exposure, and in sex-stratified subgroups given the potential influence of biological and clinical differences between men and women on treatment response. Additional sensitivity analyses were conducted by restricting follow-up to a common 2-year period to address potential survival bias due to differential follow-up duration between treatment groups.

Furthermore, we applied a causal forest for the same disability outcomes (CDW, PIRA, PIRMA) to capture potential non-linear associations that conventional proportional hazards models may not detect. The forest was fitted on the unweighted study population, in accordance with methodological recommendations, over a fixed 5-year time horizon, and included the same baseline covariates as the IPTW models. To further explore effect modification, we applied a best linear projector (BLP) to the estimated conditional effects, evaluating predefined subgroups by age at onset, sex, clinical onset phenotype, disease duration, baseline EDSS and prior treatment exposure.

All analyses were performed in R software (version 2024.12.1 + 563), and a two-sided p-value < 0.05 was considered statistically significant.

Results

Of the 117,419 people with multiple sclerosis recorded in the MSBase registry, 447 were included in the analysis. The study population flow diagram is presented in Figure 1.

Figure 1.

Study population flow diagram. Selection of late-onset multiple sclerosis (LOMS) patients from the MSBase registry and stratification by treatment group (anti-CD20 vs S1PRMs).

The median age at onset was 53.7 years (IQR 51.6–57.7), and the median age at diagnosis was 55.6 years (IQR 53.0–59.3). Median disease duration was 2.5 years (IQR 0.9–5.6). A total of 109 patients (24%) were treatment-naïve, and the median follow-up time was 6.9 years (IQR 4.2–10.2). The cohort included 307 women (68.7%). The median baseline EDSS was 3.5 (IQR 2.5–5.0), and the median number of previous DMTs was 2 (IQR 1–5).

Of the 447 eligible patients, 100 (22.4%) with missing baseline covariates were excluded prior to weighting, leaving 347 patients for the IPTW analyses. Complete baseline characteristics of both the weighted and unweighted cohorts are summarised in Table 1.

Table 1.

Baseline demographic and clinical characteristics of the study cohort.

Variables	Overall cohort	Anti-CD20	S1PRM	SMD unweighted	SMD weighted
Demographics
Age at onset, median (IQR)	53.7 (51.6–57.7)	54.8 (51.8–58.1)	52.9 (51.4–56.2)	0.171	0.056
Age at diagnosis, median (IQR)	55.6 (53.0–59.3)	56.3 (53.2–59.4)	54.4 (53.2–59.4)	0.196	0.111
Disease duration, median (IQR)	2.5 (0.9–5.6)	1.9 (0.6–5.0)	3.8 (1.5–6.4)	0.159	0.119
Female sex, n (%)	307 (68.7)	196 (66.7)	111 (72.5)	0.074	0.014
First clinical symptom, n (%)
Optic	77 (17.2)	54 (19.0)	23 (15.0)	0.040	0.022
Supratentorial	105 (23.5)	64 (21.8)	41 (26.8)	0.054	0.010
Brainstem	104 (23.3)	56 (19.0)	48 (31.4)	0.111	0.004
Spinal cord	139 (31.1)	98 (33.3)	41 (26.8)	0.064	0.006
Clinical characteristics
EDSS at baseline, median (IQR)	3.5 (2.5–5.0)	3.5 (2.5–5.5)	3.0 (2.0–4.5)	0.320	0.017
ARR prior to treatment, median (IQR)	0 (0–1)	0 (0–1)	0 (0–1)	0.016	0.036
Previous DMTs, median (IQR)	2 (1–5)	2 (0–4)	3 (1–6)	0.040	0.017
Follow-up time, years, median (IQR)	6.9 (4.2–10.2)	5.8 (3.4–8.6)	9.1 (6.6–12.2)	0.651	0.102
Visits per year, median (IQR)	1.9 (1.4–2.7)	1.8 (1.4–2.7)	2.1 (1.3–3.1)	0.083	0.059
MRI per year, median (IQR)	0.5 (0–1.0)	0.6 (0–1.1)	0.5 (0–0.9)	0.058	0.037

ARR, Annualized Relapse Rate; DMTs, disease-modifying therapies; EDSS, Expanded Disability Status Scale; IPTW = inverse probability of treatment weighting; IQR, Interquartile Range; LOMS = late-onset multiple sclerosis; MRI = magnetic resonance imaging; S1PRM, Sphingosine-1-Phosphate Receptor Modulator(s); SMD, Standardized Mean Difference.

Residual imbalance was mainly observed for continuous variables with potentially non-linear associations with treatment allocation (age at diagnosis, disease duration and follow-up time). To address this, spline terms were added for these covariates in the propensity score model, as recommended for inverse probability weighting analyses with continuous predictors. After refitting, covariate balance improved markedly (all SMD < 0.10), except for disease duration (SMD = 0.12), which remained slightly unbalanced and was therefore included as an adjustment covariate in the weighted outcome models. Graphical representations of covariate balance distributions further confirmed the quality of the weighting process (Figure S1).

Outcomes analysis in the weighted cohort

The mean ARR was 0.05 (95% CI 0.02–0.14) among patients treated with S1PRMs and 0.03 (95% CI 0.01–0.08) among those receiving anti-CD20 therapies. In the negative binomial regression model, S1PRM treatment was not associated with a significantly higher relapse rate compared with anti-CD20 (IRR 1.62, 95% CI 0.85–3.07; p = 0.14).

Regarding disability outcomes, the adjusted mean difference in EDSS change between groups (S1PRM vs anti-CD20) was −0.15 (95% CI −0.37 to 0.06; p = 0.16), indicating no significant difference in disability progression. During follow-up, 20 patients reached EDSS 4.0 and 18 reached EDSS 6.0. In Cox regression models, S1PRM therapy was not associated with a significantly different risk of reaching EDSS 4.0 (HR 0.66, 95% CI 0.26–1.66; p = 0.38) or EDSS 6.0 (HR 0.52, 95% CI 0.19–1.41; p = 0.19). Kaplan–Meier curves showed no separation between groups, with overlapping confidence intervals throughout follow-up (Figures 2–3).

Figure 2.

Kaplan–Meier curves for time to EDSS ⩾4.0. The probability of remaining free from disability progression (defined as a confirmed increase of EDSS to ⩾4.0 sustained for ⩾24 weeks) is shown for patients initiating treatment with anti-CD20 monoclonal antibodies (red) versus S1PRM (blue). Shaded areas indicate 95% confidence intervals. Numbers at risk at yearly intervals are reported below the graph. Analyses were performed in the inverse probability of treatment weighted (IPTW) population, and no statistically significant difference between treatment groups was observed in the weighted Cox proportional hazards model adjusted for baseline covariates.

Figure 3.

For CDW, 33 (15%) patients treated with anti-CD20 therapies and 25 (20%) treated with S1PRM experienced at least one event (median time to first event: 18.0 months in both groups). In the Andersen–Gill model, S1PRM treatment was not associated with a significantly different risk compared with anti-CD20 (HR 1.12, 95% CI 0.66–1.90; p = 0.67). The mean cumulative function (MCF) curves confirmed overlapping trajectories (Figure 4).

Figure 4.

Mean cumulative function of CDW. Mean cumulative function estimates for CDW events are shown for patients treated with anti-CD20 therapies (red) and S1PRM (blue). Shaded areas represent 95% confidence intervals. The MCF accounts for recurrent events, allowing visualisation of the cumulative burden of disability worsening over time. Numbers below the x-axis indicate the number of patients at risk in each treatment group at the corresponding time points. Analyses were performed in the inverse probability of treatment weighted (IPTW) population.

For PIRA, 32 patients (15%) in the anti-CD20 group and 22 patients (18%) in the S1PRM group experienced at least one event (median time to first event: 19.4 months in both groups). S1PRM treatment was not associated with a significantly different risk compared with anti-CD20 (HR 1.09, 95% CI 0.63–1.88; p = 0.76), and MCF curves were largely overlapping (Figure 5).

Figure 5.

Mean cumulative function of progression independent of relapse activity (PIRA). Mean cumulative function estimates for PIRA events are shown for patients treated with anti-CD20 therapies (red) and S1PRM (blue). Shaded areas represent 95% confidence intervals. The MCF framework accounts for recurrent events, providing an estimate of the cumulative burden of disability progression independent of relapses over time. Numbers below the x-axis indicate the number of patients at risk in each treatment group at the corresponding time points. Analyses were performed in the inverse probability of treatment weighted (IPTW) population.

For PIRMA, 19 patients treated with anti-CD20 (8.6%) and 10 treated with S1PRM (7.9%) experienced at least one event (median time to first event: 18.0 vs 26.7 months, respectively). S1PRM therapy was not associated with a significantly different risk compared with anti-CD20 (HR 0.94, 95% CI 0.46–1.89; p = 0.85). MCF curves for PIRMA were also overlapping (Figure 6).

Figure 6.

Mean cumulative function of progression independent of relapse activity with no MRI activity (PIRMA). Mean cumulative function estimates for PIRMA events are shown for patients treated with anti- CD20 therapies (red) and S1PRM (blue). Shaded areas represent 95% confidence intervals. The MCF framework accounts for recurrent events and allows visualisation of the cumulative burden of disability progression independent of relapses and in the absence of new or enlarging MRI activity. Numbers below the x-axis indicate the number of patients at risk in each treatment group at the corresponding time points. Analyses were performed in the IPTW population.

The distribution of recurrent events, including CDW, PIRA and PIRMA, is summarised in the Supplemental Material (Figure S2).

Sensitivity analysis for the risk of CDW, PIRA and PIRMA

In sensitivity analyses restricted to treatment-naïve patients, S1PRMs did not differ significantly from anti-CD20 therapies: CDW (HR 1.42, 95% CI 0.66–3.04; p = 0.37), PIRA (HR 1.38, 95% CI 0.61–3.09; p = 0.44) and PIRMA (HR 1.28, 95% CI 0.48–3.40; p = 0.62), with corresponding MCF curves shown in the Supplemental Material (Figures S3–S5).

In sex-stratified analyses, no statistically significant differences between S1PRM and anti-CD20 therapies were observed. Among women, hazard ratios for S1PRM compared with anti-CD20 were 0.78 for CDW (95% CI 0.43–1.43; p = 0.42), 0.68 for PIRA (95% CI 0.35–1.30; p = 0.24), and 0.62 for PIRMA (95% CI 0.28–1.39; p = 0.25). Among men, the corresponding hazard ratios were 1.57 for CDW (95% CI 0.65–3.82; p = 0.32), 1.58 for PIRA (95% CI 0.65–3.85; p = 0.31), and 1.14 for PIRMA (95% CI 0.35–3.68; p = 0.83). Interaction tests did not demonstrate significant sex-by-treatment effects. The corresponding MCFs by sex are presented in the Supplemental Material (Figures S6–S8).

Sensitivity analyses with restricted follow-up

In sensitivity analyses restricted to a common 2-year follow-up period, no significant differences between S1PRMs and anti-CD20 therapies were observed for CDW (HR 0.94, 95% CI 0.46–1.89), PIRA (HR 0.82, 95% CI 0.38–1.78), or PIRMA (HR 0.50, 95% CI 0.16–1.53).

Safety outcomes

During follow-up, six severe infection events were observed in six patients, all treated with anti-CD20 therapies. Severe infections included two (33.3%) cases of sepsis, one (16.7%) recurrent or complicated urinary tract infection, one (16.7%) herpes zoster, one (16.7%) opportunistic gastrointestinal infection and one (16.7%) prolonged respiratory infection requiring multiple antibiotic courses. No severe infections were observed among patients treated with sphingosine-one-phosphate receptor modulators.

In addition, 16 malignancy events were observed in nine patients. Malignancies included five (31.3%) breast cancers, one (6.3%) lung carcinoma, one (6.3%) prostate carcinoma, one (6.3%) cervical carcinoma, one (6.3%) bladder cancer, one (6.3%) bowel carcinoma, one (6.3%) lymphoma, one (6.3%) melanoma, one (6.3%) lentigo maligna, and one (6.3%) carcinoma not otherwise specified. Among these, 14 malignancy events occurred in 7 patients treated with anti-CD20 therapies, while 2 events occurred in 2 patients treated with sphingosine-1-phosphate receptor modulators, both breast cancers.

Exploratory machine-learning analyses

In the causal forest analysis and BLP projections, we observed signals favouring anti-CD20, particularly in patients with shorter disease duration (⩽2 years from diagnosis; CDW HR 0.965 (95% CI 0.933–0.998), p = 0.039; PIRA HR 0.947 (95% CI 0.909–0.987), p = 0.010), lower baseline disability (EDSS < 3; CDW HR 0.937 (95% CI 0.904–0.972), p = 0.001; PIRA HR 0.891 (95% CI 0.852–0.932), p < 0.001; PIRMA HR 0.958 (95% CI 0.941–0.976), p < 0.001), and earlier age at onset (50–55 years) compared with > 55 years (CDW HR 0.917 (95% CI 0.891–0.943), p < 0.001; PIRA HR 0.904 (95% CI 0.873–0.936), p < 0.001; PIRMA HR 0.984 (95% CI 0.970–0.998), p = 0.028). These findings are hypothesis-generating and should be interpreted with caution; full results are presented in Table 2 and Figure 7.

Table 2.

Best linear projector (BLP) estimates of treatment effect modification.

Endpoint	Modifier	HR (95% CI)	p-Value
CDW	Naïve	1.019 (0.981–1.059)	0.329
	Female	0.981 (0.954–1.009)	0.181
	Spinal onset	1.009 (0.978–1.042)	0.561
	Brainstem onset	0.995 (0.960–1.032)	0.783
	Disease duration ⩽2 years	0.965 (0.933–0.998)	0.039
	Baseline EDSS < 3	0.937 (0.904–0.972)	0.001
	Previous DMT < 2	1.030 (0.933–1.137)	0.555
	Age at onset 50–55 years	0.917 (0.891–0.943)	0.000
PIRA	Naïve	1.015 (0.969–1.064)	0.526
	Female	0.972 (0.940–1.006)	0.108
	Spinal onset	0.987 (0.950–1.026)	0.513
	Brainstem onset	0.984 (0.941–1.029)	0.478
	Disease duration ⩽2 years	0.947 (0.909–0.987)	0.010
	Baseline EDSS < 3	0.891 (0.852–0.932)	0.000
	Previous DMT < 2	1.051 (0.931–1.186)	0.425
	Age at onset 50–55 years	0.904 (0.873–0.936)	0.000
PIRMA	Naïve	1.002 (0.982–1.022)	0.851
	Female	0.989 (0.975–1.003)	0.123
	Spinal onset	1.002 (0.986–1.019)	0.770
	Brainstem onset	0.991 (0.973–1.010)	0.352
	Disease duration ⩽2 years	0.987 (0.971–1.004)	0.146
	Baseline EDSS < 3	0.958 (0.941–0.976)	0.000
	Previous DMT < 2	1.027 (0.976–1.080)	0.303
	Age at onset 50–55 years	0.984 (0.970–0.998)	0.028

Hazard ratios (HR) with 95% confidence intervals (CI) and p-values are reported for CDW, PIRA, and PIRMA. Values < 1 indicate associations favoring anti-CD20 over S1PRM. Bold values indicate statistically significant coefficients (p < 0.05).

Figure 7.

Forest plot showing the effect modifiers of treatment response (CD20 vs S1P) across the three outcomes (CDW, PIRA, PIRMA). Points represent hazard ratios (HR) with 95% confidence intervals. HR < 1 indicates a relatively greater treatment benefit in the subgroup of interest compared to the reference group (Male, Spinal onset absent, Brainstem onset absent, Disease duration >2 years, Baseline EDSS ⩾3, ⩾2 previous DMTs, Age at onset >55 years, Non-naïve).

Discussion

In this large, real-world cohort of LOMS, we did not observe significant differences in relapse or disability outcomes between patients treated with S1PRMs and those treated with anti-CD20 monoclonal antibodies. These results are consistent with previous real-world studies suggesting an overall attenuation of DMT efficacy in older patients.^31,32

A particularly relevant observation is that, among treatment-naïve patients, the direction and magnitude of treatment effects were similar to those seen in the overall cohort, with no statistically significant difference between anti-CD20 and S1PRM therapies. However, treatment-naïve status should not be equated with early treatment initiation. In our exploratory analyses, patients who commenced therapy within 2 years from diagnosis—regardless of prior treatment exposure—tended to derive greater benefit from anti-CD20 therapies. This aligns with evidence from randomised trials and large registries showing that initiating high-efficacy treatments early, rather than after switching, is associated with improved long-term outcomes in younger populations (28–30). Our findings suggest that this advantage may be attenuated in LOMS, although the limited number of events and follow-up duration in this subgroup warrant cautious interpretation.^33–35

Sex-stratified analyses also did not show clear treatment differences. Neuropathological studies have shown that men with MS tend to experience faster neurodegeneration and greater grey matter atrophy, while women more often display stronger inflammatory activity.³⁶ However, these biological differences have not been consistently associated with divergent responses to disease-modifying therapies, as systematic reviews of sex-stratified outcomes have failed to demonstrate clear treatment effects by sex.³⁷ In our cohort, hazard ratios for disability outcomes diverged in direction, suggesting a relative advantage of anti-CD20 therapies in men and of S1PRM in women, but none of these results reached statistical significance. Taken together, these findings should be regarded as exploratory and underline the need for dedicated sex-stratified analyses in larger LOMS populations.

In addition to the IPTW analyses, the exploratory causal forest and BLP indicated a favourable association of anti-CD20 therapies over S1PRMs across all disability outcomes (CDW, PIRA, and PIRMA), particularly driven by subgroups of patients with earlier age at onset (50–55 years compared with >55 years), shorter disease duration (⩽2 years from diagnosis), and lower baseline disability (EDSS < 3). Given the limited sample size, low event rates, and multiple subgroup dimensions explored, these findings should be regarded as strictly hypothesis-generating and interpreted with caution.

Safety considerations are particularly important in late-onset multiple sclerosis, as older patients are generally more susceptible to treatment-related adverse events due to immunosenescence, comorbidities, and cumulative treatment exposure.^38–40 Anti-CD20 therapies have been consistently associated with age- and exposure-dependent risks of infections and hypogammaglobulinemia,⁴¹ as well as attenuated vaccine responses,⁴² whereas S1P receptor modulators are characterised by distinct safety concerns, including cardiovascular adverse events,⁴³ macular oedema,⁴⁴ and the risk of disease rebound following treatment discontinuation.⁴⁵ In the present registry-based analysis, safety outcomes were assessed descriptively. Severe infections and malignancies were observed predominantly among patients treated with anti-CD20 therapies, while fewer events were recorded in patients receiving sphingosine-1-phosphate receptor modulators. However, given the limited number of events, the heterogeneity of reported outcomes, and the potential for underreporting inherent to real-world registries, no causal inference or comparative risk assessment can be drawn.

Several limitations should be acknowledged. The most prominent limitation of this study is its limited statistical power, resulting from the relatively small number of patients included in the weighted analysis and the low number of clinical events, particularly for disability outcomes and subgroup analyses. Given that disability progression typically accrues slowly in older multiple sclerosis populations, the study may have been underpowered to detect modest but clinically meaningful differences between treatment groups, with a substantial risk of type II error.

In addition, the classification of PIRMA relies on consistent MRI surveillance; heterogeneity in MRI availability and protocols across centres may have led to misclassification of subclinical inflammatory activity and biased results toward null differences between treatment groups.

Accordingly, the absence of statistically significant differences should not be interpreted as evidence of therapeutic equivalence, and clinically relevant effects cannot be excluded, as reflected by the wide confidence intervals around most estimates. Despite careful covariate balancing with IPTW and additional spline adjustments, residual confounding cannot be excluded. In particular, treatment allocation in LOMS is likely influenced by unmeasured factors such as comorbidity burden, prior infection risk or malignancy history, physician treatment preference, and local prescribing policies, which are not fully captured within the registry. In addition, the exclusion of patients with missing baseline covariates may have reduced generalizability by preferentially retaining more closely monitored or less clinically complex cases. These factors may have contributed to indication bias and should be considered when interpreting the results. Lastly, safety outcomes were reported descriptively only, and the risk of underreporting and heterogeneity in adverse event capture inherent to registry-based data should be acknowledged.

Taken together, our findings indicate a possible more favourable profile of anti-CD20 therapies compared with S1PRMs in specific subgroups of LOMS, namely individuals with an age at onset not exceeding 55 years, a shorter disease duration (⩽2 years from diagnosis), and lower baseline disability (EDSS < 3). By contrast, in patients with later onset, longer disease duration or higher baseline disability, the effectiveness of a high-efficacy therapy such as anti-CD20 appears to converge with that of moderate-efficacy treatments.

Collectively, these findings support the hypothesis that anti-CD20 therapies may be most beneficial when introduced early in carefully selected patient populations. Prospective, age-stratified clinical trials and biomarker-informed approaches will be essential to validate these observations and guide individualised therapeutic strategies in LOMS.

Conclusion

In this real-world study of LOMS, we did not observe statistically significant differences in relapse or disability outcomes between anti-CD20 therapies and S1P receptor modulators. Exploratory analyses suggested potential heterogeneity of treatment effects, with signals favouring anti-CD20 in selected subgroups (age at onset ⩽ 55 years, disease duration ⩽ 2 years from diagnosis, and baseline EDSS < 3); these findings should be considered hypothesis-generating and interpreted with caution. Given the limited sample size and low event rates, clinically meaningful differences cannot be excluded. Overall, these results highlight the need for individualised treatment decisions in LOMS, and for prospective studies integrating both effectiveness and safety outcomes to guide therapy selection over the disease course.

Supplemental Material

sj-docx-1-tan-10.1177_17562864261430084 – Supplemental material for Comparative effectiveness of anti-CD20 therapies and S1P receptor modulators in late-onset multiple sclerosis: real-world evidence from the MSBase registry

Supplemental material, sj-docx-1-tan-10.1177_17562864261430084 for Comparative effectiveness of anti-CD20 therapies and S1P receptor modulators in late-onset multiple sclerosis: real-world evidence from the MSBase registry by Andrea Surcinelli, Tomas Kalincik, Izanne Roos, Emanuele D’Amico, Jeannette Lechner-Scott, Serkan Ozakbas, Pavel Hradilek, Dana Horáková, Marta Vachova, Ivan Panzera, Stefano Ruzza, Maria Grazia Piscaglia, Oliver Gerlach, Jose Eustasio Meca-Lallana, Gabriel Valero-López, Allan G. Kermode, Marzena J. Fabis-Pedrini, Julie Prevost, Radek Ampapa, Suzanne Hodgkinson, Francois Grand’Maison, Samia J. Khoury, Nevin A John, Marek Peterka, Jana Houskova, Eva Recmanova, Zuzana Rous, Vahid Shaygannejad, Raed Alroughani, Jens Kuhle, Gregor Brecl Jakob, Pierre Grammond, Francesco Patti, Anneke Van der Walt, Helmut Butzkueven and Matteo Foschi in Therapeutic Advances in Neurological Disorders

Footnotes

Acknowledgements

We thank the patients, clinicians, and data managers participating in the MSBase registry. We also acknowledge the MSBase Foundation for data governance and infrastructure support. MSBase Study Group collaborators and their affiliations are listed in the Supplementary Acknowledgements.

Declarations

ORCID iDs

Andrea Surcinelli

Dana Horáková

Marta Vachova

Jose Eustasio Meca-Lallana

Allan G. Kermode

Nevin A. John

Vahid Shaygannejad

Raed Alroughani

Francesco Patti

Anneke Van der Walt

Helmut Butzkueven

Matteo Foschi

Supplemental material

Supplemental material for this article is available online.

Reporting guideline

This observational cohort study was designed and reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) and RECORD (REporting of studies Conducted using Observational Routinely collected health Data) statements. A completed STROBE checklist is available in the Supplementary Materials.

References

Roohani

Emiru

Carpenter

, et al. Late onset multiple sclerosis: Is it really late onset. Mult Scler Relat Disord 2014; 3: 444–449.

Nasiri

Naseri

Abbasgholizadeh

, et al. Clinical and epidemiological aspects of late onset multiple sclerosis in east-Azerbaijan, Iran; a population-based study. Arch Iran Med 2022; 25: 725–729.

Polliack

Barak

Achiron

Late-onset multiple sclerosis. J Am Geriatr Soc 2001; 49: 168–171.

Mouresan

Mentesidou

Berglund

, et al. Clinical characteristics and long-term outcomes of late-onset multiple sclerosis: a Swedish Nationwide Study. Neurology 2024; 102: e208051.

Cicero

Chisari

Salafica

, et al. Incidence of late onset multiple sclerosis in Italy: a population-based study. Sci Rep 2024; 14: 29649.

Holm

Wandall-Holm

Magyari

Multiple sclerosis in Denmark (1950–2023): mean age, sex distribution, incidence and prevalence. Brain 2025; 148: 626–634.

Thompson

Banwell

Barkhof

, et al. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet Neurol 2018; 17: 162–173.

Vaughn

Jakimovski

Kavak

, et al. Epidemiology and treatment of multiple sclerosis in elderly populations. Nat Rev Neurol 2019; 15: 329–342.

Lewis

Galetta

Editors’ note: Incidence of MS has increased markedly over 6 decades in Denmark particularly with late onset and in women. Neurology 2019; 92: 396–396.

10.

Lorefice

Ferraro

Fenu

, et al. Late-onset multiple sclerosis: disability trajectories in relapsing–remitting patients of the Italian MS Registry. J Neurol 2024; 27: 1630–1637.

11.

Fitzgerald

Salter

Tyry

, et al. Aging, disability, and progressive multiple sclerosis. Mult Scler Relat Disord 2020; 38: 717–725.

12.

De Biasi

Tartari

Gibellini

. The aging of the immune system and its impact on diseases. Immunol Lett 2013; 149: 12–20.

13.

Cepok

Rosche

Grummel

Short-lived plasma blasts are the main B cell effector subset during the course of multiple sclerosis. Brain 2005; 128: 1667–1676.

14.

Gao

Jiang

Tan

, et al. Microglia in neurodegenerative diseases: mechanism and potential therapeutic targets. Sig Transduct Target Ther 2023; 8: 359.

15.

Stork

Stephan

Kutllovci

, et al. Impaired remyelination in late-onset multiple sclerosis. Acta Neuropathol 2025; 149: 30.

16.

Steinegger

Kana

Nierobisch

, et al. Distinct clinical, imaging, and cerebrospinal fluid profiles in people with late-onset multiple sclerosis. Mult Scler Relat Disord 2025; 97: 106399.

17.

Nociti

Romozzi

Mirabella

Challenges in diagnosis and therapeutic strategies in late-onset multiple sclerosis. J Pers Med 2024; 14: 400.

18.

Faissner

Gold

Progressive multiple sclerosis: latest therapeutic developments and future directions. Drugs 2019; 79: 903–918.

19.

Spelman

Kalincik

Zhang

Comparative efficacy of first-line disease-modifying therapies in older versus younger patients with multiple sclerosis. Mult Scler J 2021; 27: 880–891.

20.

Silva

Casales

Donoso

, et al. Safety of high efficacy therapies in older people with multiple sclerosis: a real-world evidence study. Mult Scler Relat Disord 2024; 90: 105830.

21.

DiMauro

Swetlik

Cohen

JA.

Management of multiple sclerosis in older adults: review of current evidence and future perspectives. J Neurol 2024; 271: 3794–805.

22.

Kalincik

Kuhle

Pucci

, et al. Data quality evaluation for observational multiple sclerosis registries. Mult Scler 2017; 23: 647–655.

23.

Von Elm

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol 2008; 61: 344–349.

24.

Akaishi

Ishii

Aoki

, et al. Calculating and comparing the annualized relapse rate and estimating the confidence interval in relapsing neurological diseases. Front Neurol 2022; 13: 875456.

25.

Galea

Ward-Abel

Heesen

Relapse in multiple sclerosis. BMJ 2015; 350: h1765–h1765.

26.

Kurtzke

JF.

Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology 1983; 33: 1444–1444.

27.

Kappos

Wolinsky

Giovannoni

, et al. Contribution of relapse-independent progression vs relapse-associated worsening to overall confirmed disability accumulation in typical relapsing multiple sclerosis in a pooled analysis of 2 randomized clinical trials. JAMA Neurol 2020; 77: 1132.

28.

Müller

Sharmin

Lorscheider

, et al. Standardized definition of progression independent of relapse activity (PIRA) in relapsing-remitting multiple sclerosis. JAMA Neurol 2025; 82: 614.

29.

Kapica-Topczewska

Collin

Tarasiuk

, et al. Assessment of disability progression independent of relapse and brain MRI activity in patients with multiple sclerosis in Poland. JCM 2021; 10: 868.

30.

Austin

Stuart

EA.

Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med 2015; 34: 3661–3679.

31.

Al-Araji

Moccia

Jha

, et al. Effectiveness of disease-modifying therapies in patients with late-onset relapsing-remitting multiple sclerosis. Neurology 2025; 105: e213967.

32.

Weideman

Tapia-Maltos

Johnson

, et al. Meta-analysis of the age-dependent efficacy of multiple sclerosis treatments. Front Neurol 2017; 8: 577.

33.

Filippi

Danesi

Derfuss

, et al. Early and unrestricted access to high-efficacy disease-modifying therapies: a consensus to optimize benefits for people living with multiple sclerosis. J Neurol 2022; 269: 1670–1677.

34.

Oreja-Guevara

Martínez-Yélamos

Eichau

, et al. Beyond lines of treatment: embracing early high-efficacy disease-modifying treatments for multiple sclerosis management. Ther Adv Neurol Disord 2024; 17: 17562864241284372.

35.

Freeman

Longbrake

Coyle

, et al. High-efficacy therapies for treatment-naïve individuals with relapsing–remitting multiple sclerosis. CNS Drugs 2022; 36: 1285–1299.

36.

Alvarez-Sanchez

Dunn

SE.

Potential biological contributors to the sex difference in multiple sclerosis progression. Front Immunol 2023; 14: 1175874.

37.

Sun

Shu

, et al. Sex differences in outcomes of disease-modifying treatments for multiple sclerosis: a systematic review. Mult Scler Relat Disord 2017; 12: 23–28.

38.

Zecca

Gobbi

Long-term treatment with anti-CD20 monoclonal antibodies is untenable because of risk: YES. Mult Scler 2022; 28: 1173–1175.

39.

Smets

Giovannoni

Derisking CD20-therapies for long-term use. Mult Scler Relat Disord 2022; 57: 103418.

40.

Yang

Yan

Liu

, et al. Potential adverse events associated with sphingosine-1-phosphate (S1P) receptor modulators in patients with multiple sclerosis: an analysis of the FDA adverse event reporting system (FAERS) database. Front Pharmacol 2024; 15: 1376494.

41.

Smolik

Camilli

Panzera

, et al. Hypogammaglobulinemia and severe infections in Multiple Sclerosis patients on anti-CD20 agents: a multicentre study. Mult Scler Relat Disord 2025; 93: 106191.

42.

Schraad

Mäurer

Salmen

, et al. Cellular and humoral vaccination response under immunotherapies—German consensus on vaccination strategies in neurological autoimmune diseases. Ther Adv Neurol Disord 2025; 18: 17562864251396006.

43.

Zhao

Z-C

, et al. Risk for cardiovascular adverse events associated with sphingosine-1-phosphate receptor modulators in patients with multiple sclerosis: insights from a pooled analysis of 15 randomised controlled trials. Front Immunol 2021; 12: 795574.

44.

Coyle

Freedman

Cohen

, et al. Sphingosine 1-phosphate receptor modulators in multiple sclerosis treatment: a practical review. Ann Clin Transl Neurol 2024; 11: 842–55.

45.

Hatcher

Waubant

Nourbakhsh

, et al. Rebound syndrome in patients with multiple sclerosis after cessation of fingolimod treatment. JAMA Neurol 2016; 73: 790.