Real-World Efficacy and Safety of Neuromyelitis Optica Spectrum Disorder Disease-Modifying Treatments

Abstract

Background and Objectives

Neuromyelitis optica spectrum disorder (NMOSD) is characterized by inflammatory relapses that result in severe disability, including blindness and paralysis. Relapse prevention with safe and effective treatments is key to reducing long-term disability. We aim to compare the efficacy and safety of rituximab—the most commonly used treatment—with recently approved NMOSD-specific treatments—eculizumab, inebilizumab, satralizumab—and other common off-label NMOSD treatments—mycophenolate mofetil (MMF) and azathioprine. The primary outcomes are relapse-free survival and annualized relapse rate. Secondary outcomes are serious infectious adverse event (SIAE) and treatment-limiting adverse event (TLAE)-free survival.

Methods

A retrospective cohort study of NMOSD was conducted on patients at the Mass General Brigham hospital network. Patients meeting 2015 NMOSD diagnostic criteria, who were seen between 2000 and 2024 were included. Poisson regression, frequentist negative binomial analysis with inverse probability of treatment weighting, and Cox proportional hazard models were used to assess relapse rates, relapse-free survival, and SIAEs.

Results

A total of 176 patients with NMOSD were followed for a median of 9 years (interquartile range: 5–14), contributing 691 relapse assessments. The median age at first attack was 42 years, and 83% were female. Compared with rituximab, relapse risk was significantly lower with C5 inhibitors (HR 0.12, 95% CI 0.07–0.24), inebilizumab (HR 0.22, 95% CI 0.12–0.65), and satralizumab (HR 0.19, 95% CI 0.11–0.42). Annualized relapse rates were the lowest for C5 inhibitors (0, 95% CI 0–0.063) and the highest for azathioprine (0.34, 95% CI 0.18–0.56). A composite outcome of relapse, SIAE, and TLAE favored C5 inhibitors (HR 0.22, 95% CI 0.05–0.67), while azathioprine (HR 2.33, 95% CI 1.08–4.86) and MMF (HR 1.75, 95% CI 1.02–2.95) showed increased risk compared with rituximab. C5 inhibitors had the lowest incidence of serious infections (incidence rate ratio 0.16, 95% CI 0.05–0.42 vs rituximab).

Discussion

Clinicians should consider using NMOSD-approved treatments given their favorable efficacy and safety profiles in the real-world setting. MMF and azathioprine should be avoided. We caution against rituximab as a default first-line given the cumulative risk of relapse, SIAEs, and TLAEs over time.

Classification of Evidence

This study provides Class III evidence that, in patients with NMOSD, FDA-approved disease-modifying therapies are associated with lower relapse rates and fewer serious adverse events compared with rituximab.

Introduction

Neuromyelitis optica spectrum disorder (NMOSD) is a rare neuroinflammatory condition with a prevalence of ∼4 per 100,000.^1,2 In NMOSD, a misdirected immune response against the aquaporin-4 (AQP4) water channel targets AQP4-dense regions, including the optic nerves, spinal cord, brainstem, and area postrema. These inflammatory attacks can result in blindness, paralysis, and even death.^3-8 Unlike in MS, disability in NMOSD does not typically accrue independently of relapses,^9,10 and most relapses cause disability worsening and deterioration of quality of life,¹¹ underscoring the importance of optimized immunotherapy to suppress antibody-mediated inflammation and prevent cumulative neurologic injury.

Historically, NMOSD relapse prevention has relied on various off-label treatment approaches. Among these, rituximab emerged as a preferred option in clinical practice, with a small randomized controlled trial demonstrating its superiority in reducing annualized relapse rates (ARRs) and improving disability in patients with NMOSD.^12,13 Mycophenolate mofetil and azathioprine are also used, but azathioprine carries a risk of lymphoproliferative disorders,¹⁴ and mycophenolate mofetil (MMF) carries a risk of skin cancer, particularly at higher doses.¹⁵ Recently, decisional dilemmas have rapidly emerged following the approval of NMOSD-specific disease-modifying treatments (DMTs). Eculizumab, a terminal complement (C5) inhibitor, was approved in the United States in 2019 for the treatment of AQP4+ NMOSD based on results of a phase 3 trial in AQP4 antibody-positive (AQP4+) patients showing a 94% lower risk of relapse compared with placebo.¹⁶ Inebilizumab and satralizumab were approved the following year after studies showed a 77%¹⁷ and 79%¹⁸ lower risk of relapse, respectively, compared with placebo in AQP4+ patients. In early 2024, a longer-acting complement (C5) inhibitor, ravulizumab, was approved in the United States on the basis of a phase 3 trial that had no relapses (98.6% relapse risk reduction) in the treatment group compared with the historical placebo group from the PREVENT trial.¹⁹ Few publications have explored the comparative efficacy of NMOSD therapies. A 2022 network meta-analysis of FDA-approved treatments suggested a lower risk of relapse with C5 inhibitors (eculizumab and ravulizumab)^20,21 compared with inebilizumab and satralizumab. However, this analysis was done post hoc using trials with different baseline patient characteristics, had limited follow-up time, and did not include rituximab.

Varying safety profiles also contribute to the complexity of treatment decision-making. Direct comparisons of safety profiles are challenging, as data for off-label treatments are often derived from long-term, real-world observational cohorts, whereas data for newer agents typically originate from the more selected populations of controlled clinical trials. Nonetheless, long-term studies of rituximab in NMOSD have shown a greater than 70% risk of hypogammaglobulinemia, which can lead to symptomatic infections in up to 50% of patients and serious infections in up to 20%.^22,23 In some cases, lifelong infusions with IV immunoglobulin (IVIg) may be required to decrease infection risk.²⁴ By contrast, clinical trials and long-term extension studies found that satralizumab is welltolerated, with similar rates of adverse events (AEs) compared with placebo and no increased infection risk.²⁵ Inebilizumab showed rare infusion reactions and <3% rates of serious AEs despite associated hypogammaglobulinemia.²⁶ C5 inhibitors showed lower rates of serious AEs than placebo, but the trials were short. Although a recent study by the German Neuromyelitis Optica Study Group (NEMOS) found that 88% of patients on eculizumab remained attack-free and 13% had severe infections,²⁷ little is known about its long-term safety relative to current off-label treatments.

Here, we aimed to characterize the efficacy, tolerability, and safety outcomes of patients with NMOSD in a large real-world cohort.

Methods

Study Population

Patients were eligible if they met the 2015 International Panel for NMO Diagnosis (IPND) criteria for NMOSD and had been evaluated in person by a Mass General Brigham neurologist between January 1, 2000, and June 30, 2024. Two parallel ascertainment streams were used: (1) all patients with a positive AQP4 cell-based assay in our laboratory information system and (2) all patients with an NMOSD ICD code who tested negative for AQP4. Charts from the second group were reviewed and excluded if IPND criteria were not met. AQP4 antibody status was determined by a commercially available cell-based assay; patients with negative AQP4 results were labeled “seronegative.” Myelin oligodendrocyte glycoprotein (MOG) IgG was assayed in 20 of 25 seronegative patients; all 20 were negative.

Treatment Exposure

Dates of maintenance therapies (rituximab, eculizumab/ravulizumab, inebilizumab, satralizumab, methotrexate, mycophenolate mofetil, and azathioprine) were collected for each patient. Eculizumab and ravulizumab data were combined and are hereafter referred to as C5 inhibitors.

Time zero was defined as the date of first maintenance therapy administration. Patients were censored at the date of the last clinic visit. For each treatment period, we recorded patient demographics, disease history (number of prior attacks and prior therapies), safety outcomes, treatment exposure (time on treatment), and EDSS and mRS within 30 days of treatment start.

Relapse Definition and Classification

A relapse was defined as new or worsening CNS symptoms lasting >24 hours and occurring ≥30 days after the prior attack. Events were classified by clinical certainty as:

1. Definite—a core phenotype documented by a neurologist with an objective change on examination (e.g., new visual acuity deficit or motor weakness). For area postrema syndrome, intractable hiccups and/or vomiting were accepted as sufficient for “definite” classification because no examination change is expected;

2. Probable—a core phenotype described retrospectively in outside records without contemporaneous neurologic examination;

3. Possible—nonspecific symptoms potentially attributable to NMOSD (e.g., isolated numbness) without confirmatory examination findings;

4. Unlikely—symptom worsening clearly explained by a non-NMOSD trigger (e.g., infection) without new examination or imaging findings.

Only definite or probable relapses were included for treatment efficacy and relapse outcome analyses. Relapses were classified as radiologically confirmed if there was a new gadolinium-enhancing lesion or new T2 lesion corresponding to symptoms.

Safety End Points

For the secondary end point, serious infectious adverse events (SIAEs) were defined as infections requiring hospitalization. All hospital admissions were reviewed and those with a primary diagnosis of infection were counted as SIAEs. Treatment-limiting adverse events (TLAEs) were defined as any adverse event leading to permanent discontinuation of therapy. This included infusion reactions, recurrent infections (e.g., urinary tract infections, pneumonia), cytopenias, etc.

Data Quality and Validation

Data were entered by one and reviewed by a second neuroimmunologist. Disagreements were resolved by the senior author. Completeness was assessed for each variable, and the proportion of missing data is reported in the corresponding tables. Complete-case analysis was applied for all primary and secondary end point models.

Statistical Analysis

All analyses were performed in R Studio (Version 2024.09.0+375). Relapses were excluded from treatment effect analyses if they occurred before the end of the treatment loading period. The demographic characteristics of the cohort, characteristics of relapses, nonrelapse hospitalizations, and treatment groups were compared. The Wilcoxon rank sum test or Kruskal-Wallis rank sum test was used to compare groups for continuous variables. The Fisher exact test was applied to count data for categorical variables, with a simulated p value calculated if necessary for small sample sizes. The Pearson χ² test was used when sample sizes were sufficient. Categorical variables were summarized as counts and percentages (n [%]), and continuous variables were summarized with medians and interquartile ranges (Q1–Q3). A Cox proportional hazards model with Firth penalized regression was used to assess the effect of maintenance therapy on 2 end points: relapse-free survival and a combined endpoint consisting of relapse-free survival, SIAE-free survival, and TLAE-free survival. SIAEs were defined as infections requiring hospitalization. TLAEs were defined as AEs leading to treatment discontinuation. Firth penalized regression was chosen because of its robust handling of small sample sizes and sparse events. The model included a frailty term to account for random effects at the patient level. Hazard ratios were computed compared with rituximab for each treatment group. Confidence intervals were obtained by taking the 2.5th and 97.5th percentiles of 1,000 nonparametric bootstrap resamples of the fitted Cox-Firth model. Only the first treatment course was considered if there were multiple treatment courses. Kaplan-Meier survival curves were generated for both end points to compare the time to relapse for patient groups treated with different maintenance therapies.

Next, the ARR was calculated by dividing the total number of relapses by the total time on treatment in years. The upper limit of the 95% confidence interval was determined using an exact Poisson method that adjusts for treatment duration. A Poisson test was performed to determine whether the rates were significantly different. A frequentist negative binomial model was also used to model relapse counts. Inverse probability of treatment weighting (IPTW) was applied to adjust for confounding across treatment groups, with the propensity scores calculated from a logistic regression adjusting for age and number of prior attacks. The number of prior attacks was chosen as the IPTW covariate because it differed significantly between treatment groups, whereas baseline EDSS scores did not. Relapse count serves as a practical surrogate for disease aggressiveness in NMOSD and may modulate therapeutic response.

Predicted ARRs were computed using the marginal means for each treatment group, and patient ID as a random effect, to account for the correlation between repeated measurements within individuals. IPTW weights were applied. Confidence intervals could not be calculated given that there were zero events in the C5 inhibitors, inebilizumab, and satralizumab groups. The rate of infections requiring hospitalization was determined using a negative binomial regression using the same methodology as above. Incidence rate ratios compared with rituximab were computed.

Standard Protocol Approvals, Registrations, and Patient Consents

Institutional review board approval was obtained from Mass General Brigham. Consent was not required given the retrospective nature of this study.

Data Availability

Deidentified data may be made available upon reasonable request to the corresponding author and pending approval by the Mass General Brigham Institutional Review Board and applicable data use agreements.

Results

Of 293 patient charts analyzed, 176 met our study inclusion criteria, of whom 151 (86%) were AQP4+ and 25 were AQP4-negative patients (eFigure 1). The most common exclusion was for positive MOG antibodies (N = 37). The median age at the first attack was 42 years (interquartile range: 32, 56), and the median follow-up was 9 years (5, 14). There were 83% female patients, 87% of patients were non-Hispanic, and 65% were White. A total of 702 possible relapse events were analyzed, of which 48% were deemed clinically definite, 15% clinically probable, 24% clinically possible, and 13% unlikely (Table 1). Definite and probable relapses were most likely to be radiologically confirmed (94% and 91%, respectively). The most common relapse phenotypes were isolated myelitis (46%) and isolated optic neuritis (29%) (Table 1). The number of prior therapies was significantly different between the treatment groups, with patients on C5 inhibitors, MMF, and satralizumab being less likely to be treatment-naïve at the time of treatment initiation compared with those on rituximab and inebilizumab (Table 2). EDSS at treatment start was not significantly different between the treatment groups. Patients treated with rituximab were most commonly AQP4+. One seronegative patient received a C5 inhibitor, and inebilizumab and satralizumab were used exclusively in seropositive patients. Four patients on ravulizumab and 2 patients on eculizumab were previously in the phase 3 clinical trials. Regarding insurance approval, eculizumab was initially denied and approved on appeal for 3 patients. Inebilizumab was initially denied for 2 patients, approved on appeal for 1, and required a switch to satralizumab for 1. Satralizumab was initially denied for 3 patients, approved on appeal for 2, and required a switch to rituximab for 1. Ravulizumab was denied and approved on appeal for 1 patient.

Table 1.

Characteristics of Relapses

Characteristic	Overall N = 702a	Definite n = 338a	Probable n = 107a	Possible n = 167a	Unlikely n = 90a	p Valueb
Radiologic confirmation						<0.001
Yes	371 (81)	267 (94)	61 (91)	37 (82)	6 (9.5)
No	87 (19)	16 (5.7)	6 (9.0)	8 (18)	57 (90)
Unknown	244	55	40	122	27
Relapse phenotype (s)						<0.001
Isolated transverse myelitis	325 (46)	152 (45)	49 (46)	83 (50)	41 (46)
Isolated optic neuritis	203 (29)	89 (26)	34 (32)	57 (34)	23 (26)
Area postrema syndrome	38 (5.4)	17 (5.0)	5 (4.7)	13 (7.8)	3 (3.3)
Optic neuritis, transverse myelitis	31 (4.4)	21 (6.2)	5 (4.7)	3 (1.8)	2 (2.2)
Noncore syndrome	27 (3.8)	1 (0.3)	0 (0)	7 (4.2)	19 (21)
Transverse myelitis, area postrema syndrome	25 (3.6)	17 (5.0)	7 (6.5)	1 (0.6)	0 (0)
Other multifocal	24 (3.4)	20 (5.9)	3 (2.8)	0 (0)	1 (1.1)
Isolated cerebral syndrome	19 (2.7)	12 (3.6)	3 (2.8)	3 (1.8)	1 (1.1)
Transverse myelitis, cerebral syndrome	10 (1.4)	9 (2.7)	1 (0.9)	0 (0)	0 (0)
If myelitis, was it longitudinally extensive?	150 (62)	119 (73)	13 (45)	15 (63)	3 (12)	<0.001
Unknown	30	28	0	0	2
DMT at time of relapse
C5 inhibitors	8 (1.1)	0 (0)	0 (0)	0 (0)	8 (9.0)	<0.001
MS injectables	42 (6.0)	21 (6.3)	3 (2.8)	15 (9.0)	3 (3.4)	0.14
Rituximab	119 (17)	45 (13)	21 (20)	15 (9.0)	38 (43)	<0.001
Antimetabolites	117 (17)	49 (15)	14 (13)	29 (17)	25 (28)	0.024
Inebilizumab	2 (0.3)	0 (0)	0 (0)	0 (0)	2 (2.2)	0.017
Satralizumab	1 (0.1)	0 (0)	0 (0)	0 (0)	1 (1.1)	0.13

^an (%).

^bFisher exact test for Count Data with simulated p value (based on 2000 replicates); Kruskal-Wallis rank sum test.

Table 2.

Characteristics of DMT Treatment Groups

Characteristic	Overall N = 331a	Azathioprine n = 23a	C5 inhibitors n = 21a	Inebilizumab n = 11a	MMF n = 65a	Rituximab n = 192a	Satralizumab n = 19a	p Valueb
Age at first attack	40 (31, 54)	39 (32, 45)	37 (15, 49)	54 (40, 61)	40 (32, 56)	40 (30, 55)	43 (33, 56)	0.3
Sex								0.4
Male	44 (13)	2 (8.7)	3 (14)	0 (0)	5 (7.7)	30 (16)	4 (21)
Female	287 (87)	21 (91)	18 (86)	11 (100)	60 (92)	162 (84)	15 (79)
Race								0.4
White	195 (59)	13 (57)	15 (71)	8 (73)	32 (49)	113 (59)	14 (74)
Black/African American	67 (20)	3 (13)	2 (9.5)	1 (9.1)	16 (25)	40 (21)	5 (26)
AAPI	26 (7.9)	3 (13)	0 (0)	0 (0)	7 (11)	16 (8.3)	0 (0)
Other	43 (13)	4 (17)	4 (19)	2 (18)	10 (15)	23 (12)	0 (0)
Ethnicity								0.5
Hispanic	40 (12)	2 (8.7)	3 (14)	0 (0)	8 (12)	27 (14)	0 (0)
Non-Hispanic	291 (88)	21 (91)	18 (86)	11 (100)	57 (88)	165 (86)	19 (100)
No. of prior attacks	2.0 (1.0, 3.0)	1.0 (1.0, 3.0)	3.0 (2.0, 4.0)	1.0 (1.0, 2.0)	2.0 (1.0, 3.0)	2.0 (1.0, 3.0)	1.5 (1.0, 4.5)	0.005
No. of prior therapies	1.0 (0.0, 1.0)	0.0 (0.0, 2.0)	1.0 (1.0, 2.0)	0.0 (0.0, 1.0)	1.0 (0.0, 2.0)	0.0 (0.0, 1.0)	2.0 (0.0, 2.0)	<0.001
EDSS at treatment start	3.5 (2.5, 6.0)	5.0 (3.0, 7.0)	4.0 (3.5, 6.5)	3.0 (3.0, 5.0)	3.5 (2.5, 6.5)	3.5 (2.0, 6.0)	2.8 (2.3, 4.0)	0.2
Unknown	23	9	0	0	7	7	0
mRS at treatment start								0.071
mRS 0–2	183 (61)	5 (33)	12 (57)	5 (63)	35 (60)	112 (61)	14 (88)
mRS 3–5	118 (39)	10 (67)	9 (43)	3 (38)	23 (40)	71 (39)	2 (13)
Unknown	24	8	0	1	7	8	0
Complications
Common infections	95 (29)	4 (17)	2 (9.5)	1 (9.1)	20 (31)	66 (34)	2 (11)	0.019
Opportunistic infections	81 (25)	4 (17)	2 (9.5)	2 (18)	16 (25)	54 (28)	3 (16)	0.4
Hypogammaglobulinemia	26 (7.9)	0 (0)	0 (0)	1 (9.1)	0 (0)	25 (13)	0 (0)	0.001
Hypersensitivity	13 (3.9)	0 (0)	0 (0)	0 (0)	0 (0)	12 (6.3)	1 (5.3)	0.2
Time on treatment (y)	2.5 (0.7, 5.7)	0.9 (0.2, 3.1)	2.9 (0.7, 4.7)	1.7 (0.6, 2.7)	2.1 (0.6, 5.9)	3.4 (1.1, 7.0)	0.6 (0.4, 2.0)	<0.001
No. of relapses
Definite	92	11	0	0	35	46	0	0.050
Probable	35	3	0	0	11	21	0	0.8
Possible	43	7	0	0	21	15	0	0.083
Unlikely	72	2	8	2	21	38	1	0.4
Previously in clinical trial	6 (1.8)	N/A	6 (28)	0 (0)	N/A	0 (0)	0 (0)	N/A

Abbreviations: AAPI = Asian American or Pacific Islander; MMF = mycophenolate mofetil.

^aMedian (Q1, Q3); n (%); Sum.

^bKruskal-Wallis rank sum test; Fisher exact test; Fisher exact test for count data with simulated p value (based on 2000 replicates).

Efficacy of NMOSD DMTs

We analyzed relapse-free survival on rituximab, C5 inhibitors, inebilizumab, satralizumab, MMF, and azathioprine by determining the time to the first clinically definite or probable relapse. Rituximab had a 69.7% relapse-free probability at 5 years compared with 100% for C5 inhibitors, inebilizumab, and satralizumab, and 51.1% for MMF and 35.3% for azathioprine (Figure 1A). C5 inhibitors had a lower risk of relapse than rituximab with an HR of 0.12 (95% CI 0.07–0.24), as did inebilizumab (HR 0.22, 95% CI 0.12–0.65) and satralizumab (HR 0.19, 95% CI 0.11–0.42). Restricting the analysis to AQP4+ patients produced similar results (Figure 1, C and D).

Figure 1. Efficacy of NMOSD DMTs.

(A) Kaplan-Meier curve showing relapse-free survival for all patients with NMOSD. (B) Adjusted hazard ratios (aHRs) for relapse in all patients, estimated using a Cox proportional hazards model with Firth penalized likelihood. (C) Kaplan-Meier curve for relapse-free survival in AQP4-IgG seropositive patients. (D) aHRs for relapse in AQP4-IgG seropositive patients using the same Cox model. Only definite and probable relapses were included in all analyses. AQP4 = aquaporin-4; DMT = disease-modifying treatment; NMOSD = neuromyelitis optica spectrum disorder.

ARR was 0.08 (95% CI 0.062–0.1) for rituximab, 0 (95% CI 0–0.062) for C5 inhibitors, 0 (95% CI 0–0.06) for inebilizumab, 0 (95% CI 0–0.17) for satralizumab, 0.19 (95% CI 0.14–0.26) for MMF, and 0.34 (95% CI 0.18–0.56) for azathioprine (Figure 2). A frequentist negative binomial analysis with IPTW was performed to allow for causal inference, which produced similar results (eFigure 2). CD19 and CD20 counts were available for 32% of relapses on rituximab and were 0 in 14/14 and 12/14 relapses, respectively.

Figure 2. Annualized Relapse Rates on NMOSD DMTs.

Confidence intervals are computed using the Poisson method. For all analyses, only definite and probable relapses are included. DMT = disease-modifying treatment; NMOSD = neuromyelitis optica spectrum disorder.

Eight events were deemed unlikely relapses on C5 inhibitors, 6 of those events occurred in 2 patients with severe spasticity and disability at baseline. In all cases, there was no new deficit on the neurologic examination, and the predominant symptom was pain. None of those events were associated with a new MRI lesion, and 4 of 8 had an identifiable trigger (UTI and COVID-19). A sensitivity analysis for imaging-confirmed relapses showed similar results (eFigure 3).

Safety of NMOSD DMTs

We determined the frequency of AEs for each treatment period (Table 2). 13% of rituximab treatment periods were complicated by hypogammaglobulinemia, 34% by mild infections (e.g., urinary tract infections, upper respiratory tract infections), 28% by serious infections (e.g., bacterial pneumonia, infection requiring IV antibiotics), and 6.3% by hypersensitivity reactions (Table 2). There was a significantly lower incidence of SIAEs on C5 inhibitors compared with rituximab, with an incidence rate ratio of 0.17 (95% CI 0.06–0.43, p value 0.0002) (Figure 3A).). Kaplan-Meier curve of SIAE and TLAE-free survival showed a substantial failure rate on rituximab (Figure 3B). Cox analysis showed no significant difference between the treatment groups (Figure 3C). Restricting the analysis to seropositive patients showed similar results (eFigure 4).

Figure 3. Safety of NMOSD DMTs.

(A) Incidence rate ratios for serious infections on NMOSD DMTs. Incidence rate ratios are derived from a negative binomial model with inverse probability of treatment weights. (B) Kaplan-Meier curve for SIAE and TLAE-free survival. (C) aHRs for SIAE and TLAE-free survival. aHR = adjusted hazard ratios; DMT = disease-modifying treatment; NMOSD = neuromyelitis optica spectrum disorder; SIAE = serious infectious adverse event; TLAE = treatment-limiting adverse event.

Combined End Point

We combined efficacy, safety, and tolerability into a composite end point of time to first relapse, SIAEs, or TLAEs. Relapse, SIAEs, and TLAEs-free probabilities at 5 years were 55% for rituximab, 91% for C5 inhibitors, 19% for azathioprine, and 35% for MMF (Figure 4A). Because of lower follow-up time, data were only available at 3 years for satralizumab and inebilizumab and showed 3-year event-free probabilities of 79% and 38%, respectively. C5 inhibitors were associated with a lower risk of relapse, SIAEs, or TLAEs than rituximab, with an HR of 0.22 (95% CI: 0.05–0.67) (Figure 4, A and B). Azathioprine had a higher risk of relapse, SIAEs, or TLAEs, with an HR of 2.33 (95% CI: 1.08–4.86), and MMF also showed an increased risk with an HR of 1.75 (95% CI: 1.02–2.95). By contrast, there was no statistically significant difference in relapse, SIAEs, or TLAEs risk for inebilizumab (HR 1.23, 95% CI: 0.24–3.12) or satralizumab (HR 1.01, 95% CI: 0.16–2.68) compared with rituximab. Restricting the analysis to seropositive patients showed similar results (Figure 4, C and D).

Figure 4. Safety and Efficacy of NMOSD DMTs.

(A) Kaplan-Meier curve showing relapse, SIAE, and TLAE-free survival for all patients with NMOSD. (B) Hazard ratios (HRs) for relapse, SIAE, and TLAE-free survival in all patients. (C) Kaplan-Meier curve for relapse, SIAE, and TLAE-free survival in AQP4-IgG seropositive patients. (D) aHRs for relapse, SIAE, and TLAE-free survival in AQP4-IgG seropositive patients. Only definite and probable relapses were included in all analyses. HRs are from a Cox proportional hazards model with Firth penalized likelihood. AQP4 = aquaporin-4; DMT = disease-modifying treatment; NMOSD = neuromyelitis optica spectrum disorder; SIAE = serious infectious adverse event; TLAE = treatment-limiting adverse event.

Classification of Evidence

This study provides Class III evidence that, in patients with NMOSD, FDA-approved disease-modifying therapies are associated with lower relapse rates and fewer serious adverse events compared with rituximab.

Discussion

This study provides critical insights into comparative treatment effectiveness and the adverse event burden experienced by patients with NMOSD. Overall, we found that C5 inhibitors are more effective and carry a lower risk of serious adverse events than rituximab, while MMF and azathioprine carry a higher risk of relapse, SIAEs, or TLAEs.

We found an ARR of 0.08 for patients on rituximab. This finding is similar to results of an Australian study in which rituximab had an ARR of 0.136 (95% CI 0.088–0.201).²⁸ In contrast to the Johns Hopkins University cohort, which had a median ARR of 0,²⁹ we found that rituximab failure continues to occur well beyond 6 months, with only 38% of rituximab failures occurring within the first 6 months. This may reflect different patient characteristics or different relapse definitions. Although there were no relapses among patients on rituximab in the 2-year RIN-1 trial, most RIN-1¹³ participants received concurrent prednisone,¹³ while patients in our cohort were largely on rituximab monotherapy and were followed for a median of 3.2 years. We also found a similar rate of rituximab failure (30%) as the previous study,³⁰ although our cohort had a longer time on treatment, and some patients relapsed beyond the 5-year time point. Although most patients did not have early B-cell reconstitution, our findings warrant further investigation into factors associated with rituximab failure, including longitudinal CD19 and CD20 monitoring.

In our cohort, azathioprine and MMF performed particularly poorly, with relapses occurring in most patients despite azathioprine being the second-most commonly used NMOSD treatment.³¹ This is similar to older studies that also found a substantial failure rate on azathioprine and MMF.^30,32,33 The failure rate was even higher in our cohort, likely due to longer follow-up times.

C5 inhibitors were more efficacious and associated with fewer serious adverse events than rituximab. Inebilizumab and satralizumab were effective and had no relapses in our cohort, but a lower number of patients and shorter time on treatment limited the statistical power. There were 8 unlikely relapses on C5 inhibitors, as we suspect that patients who were started on these treatments were more likely to have severe disability at baseline and, therefore, may be more likely to have pseudorelapses. There was a lower rate of SIAEs and TLAEs on C5 inhibitors compared with rituximab.

A composite end point of relapse-free survival, SIAEs, and TLAEs showed that almost half of patients will have at least one relapse, serious infection, or severe adverse event after 5 years on rituximab, and a vast majority of patients on MMF or azathioprine will experience at least one relapse, serious infection, or severe adverse event within 5 years. There were several serious adverse events on inebilizumab, although the lower number of patients likely biases the composite end point, and this needs to be confirmed with more patients and longer follow-ups. By contrast, most patients remain relapse and serious infection-free on C5 inhibitors and, to a lesser extent, on satralizumab. This finding held although patients on C5 inhibitors had a higher number of prior relapses, indicating higher disease activity and disability.

Treatment costs for NMOSD vary dramatically across available therapies. Complement inhibitors (eculizumab, ravulizumab) can cost over $500,000 in the first year; satralizumab can cost over $250,000 in the first year; B-cell depletion costs vary from inebilizumab (over $400,000 in the first year) to rituximab biosimilars that cost as little as $10,000–$20,000 yearly. Actual costs to patients and payers vary widely on the basis of individual payer contracts, patient assistance programs, and global region. Given these disparities, comprehensive cost-effectiveness studies are imperative though difficult to do. Such analyses should integrate real-world data on relapse rates, serious adverse events, infusion logistics, and health care utilization to inform treatment guidelines and ensure sustainable access to optimal therapies worldwide. Overall, these findings highlight the need for comparative effectiveness trials, such as platform trials,^34,35 to facilitate the comparison of rituximab with the FDA-approved NMOSD DMTs and a comparison between the FDA-approved NMOSD DMTs.

The strength of this study is its real-world cohort design, which allows for evaluating treatment effectiveness and safety in a real-world setting reflective of clinical practice. The long follow-up period, particularly for rituximab, allows us to determine long-term outcomes with more certainty. Moreover, the use of different statistical models to evaluate efficacy, including a negative binomial analysis with IPTW and Cox proportional hazards models, allows us to evaluate both the time to first relapse and the relapse rates over the course of treatment. Our composite endpoint combines relapse prevention with safety and tolerability, thus evaluating the overall impact of a treatment on a patient's health. It could serve as a valuable outcome measure in clinical trials. This is important because most studies of NMOSD DMTs have focused on relapse prevention.

There were several limitations. First, this is a retrospective study, and the treatment groups had significant differences in potentially important variables, such as the number of prior relapses. We adjusted for this using IPTW, but there can be residual confounding, given the lack of randomization. In addition, retrospective relapse assessment is inherently subjective; although dual-reviewer adjudication was used, subtle or mild relapses may have been underrecognized or incompletely documented. Similarly, the availability of MRI confirmation varied. Our cohort was also subjected to ascertainment bias, as tertiary referral patients often have more severe or refractory disease, potentially limiting generalizability. Historically inconsistent monitoring practices meant we lacked detailed longitudinal laboratory data (e.g., CD19/CD20 counts, IgG levels), limiting mechanistic insights into treatment responses and adverse events like hypogammaglobulinemia.

Although the cohort includes a substantial number of patients, the small sample sizes in some treatment arms, particularly for inebilizumab and satralizumab, limit our statistical power. Moreover, the zero relapses in some treatment groups made frequentist regression analysis challenging and required using other statistical techniques to determine confidence intervals.

Given their similar mechanisms of B-cell depletion, inebilizumab and rituximab would be expected to have comparable safety profiles, particularly regarding hypogammaglobulinemia. Selection bias may contribute to observed differences, as patients with comorbidities or those ineligible for clinical trials may have been preferentially prescribed rituximab. Moreover, the substantially shorter duration of follow-up and lower cumulative exposure to inebilizumab relative to rituximab may have limited our ability to detect infrequent or late-onset adverse events.

Finally, while composite endpoints provide holistic treatment comparisons, they inherently combine events of varying clinical importance, potentially masking key differences between efficacy and toxicity outcomes.

In conclusion, this study provides insights into the comparative effectiveness and safety of NMOSD treatments in a real-world setting. Among the FDA-approved drugs, C5 inhibitors had the highest patient-years in our cohort and showed both superior relapse prevention and lower rates of serious adverse events (including serious infection requiring hospitalization) compared with rituximab, whereas MMF and azathioprine had higher rates than rituximab. We also found that nearly half of the patients on rituximab experienced a relapse or significant side effect within 5 years. In jurisdictions where other treatments are available, we caution against using rituximab first-line in NMOSD. We also recommend against the use of MMF and azathioprine. These findings emphasize the importance of optimizing treatment selection to reduce the burden of relapses and serious complications. Comparative effectiveness trials with composite endpoints that incorporate relapses and safety are needed to comprehensively compare NMOSD treatments.

Glossary

AE

adverse event

AQP4

aquaporin-4

ARR

annualized relapse rate

DMT

disease-modifying treatment

IPTW

inverse probability of treatment weighting

MMF

mycophenolate mofetil

NMOSD

neuromyelitis optica spectrum disorder

SIAE

serious infectious adverse event

TLAE

treatment-limiting adverse event

Author Contributions

P.A. Bilodeau: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. M. Wruble Clark: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. A. Ganguly: major role in the acquisition of data. J.B. Harowitz: major role in the acquisition of data. J.V. Mahler: major role in the acquisition of data. M. Jiang: major role in the acquisition of data. S.S. Narasimhan: major role in the acquisition of data. D.K. Pua: major role in the acquisition of data. B.C. Healy: analysis or interpretation of data. F.J. Mateen: drafting/revision of the manuscript for content, including medical writing for content; study concept or design. M. Levy: drafting/revision of the manuscript for content, including medical writing for content; study concept or design. S. Bhattacharyya: drafting/revision of the manuscript for content, including medical writing for content; study concept or design.

Study Funding

Alexion (AstraZeneca).

Disclosure

P.A. Bilodeau has received research support from the US Department of Defense, the Canadian Institutes of Health Research, and the Ann Theodore Foundation. P.A. Bilodeau's institution has received grants from Alexion Pharmaceuticals and Sanofi. S. Bhattacharyya's institution has received funding from Alexion (AstraZeneca) M. Levy has received personal compensation for advising the following companies: Alexion, Horizon, Genentech/Roche, UCB, Sanofi and Mitsubishi. Through the Massachusetts General Hospital, M. Levy received grants from Alexion/AstraZeneca Rare Disease, Horizon/Amgen, Genentech/Roche, UCB and Sanofi for research projects. Go to Neurology.org/NN for full disclosures.