Late-Onset Neutropenia After Rituximab in Multiple Sclerosis

Abstract

Background and Objectives

Late-onset neutropenia (LON) is a rare, transient reduction in absolute neutrophil count (ANC) associated with anti-CD20 therapies. In multiple sclerosis (MS), data regarding severity, infectious outcomes, and predictors remain limited. We aimed to determine the incidence, characteristics, and predictors of LON in rituximab-treated people with MS (pwMS).

Methods

We conducted a registry-based retrospective multicenter cohort study of pwMS in Sweden exposed to rituximab between 2008 and 2024. Any LON was defined as ANC <1,500 cells/μL occurring at least 30 days after rituximab initiation. Grades 3–4 LON was defined as ANC <1,000 cells/μL, recording infectious complications for these events. Incidence rates (IRs) were calculated per 1,000 person-years, and event times were summarized as medians (interquartile range [IQR]). Predictors of LON events were evaluated using Cox regression with time-varying covariates and binary logistic regression.

Results

Among 2,686 rituximab-exposed pwMS, 6.2% (n = 165) experienced ≥1 LON event during a median follow-up of 7.8 years (IQR 5.7–9.8). The IR for any LON (grades 2–4) was 10.1/1,000 person-years (95% CI 8.6–11.8); the median time-to-first event was 360 days (IQR 112–1,017). Most events were mild, 5.3% (grade 2, n = 142). Grades 3–4 LON occurred for 1% (23 individuals; 36 events) and IR 2.1/1,000 person-years (95% CI 1.4–2.9). Ten infections occurred among LON grades 3–4 events, classifying 8 as severe. All individuals with infectious LON grades 3–4 events (48%) recovered after hospitalization, antibiotics/antivirals, and/or granulocyte colony-stimulating factor. Rituximab was rechallenged in 89% (32/36) of grades 3–4 events: 14 recurred and 2 complicated by infections. In time-dependent Cox models, higher accumulated rituximab exposure (HR 0.87, 95% CI 0.81–0.93) and prior injectable therapy (HR 0.66, 95% CI 0.45–0.96) were associated with a lower hazard of any LON. Exploratory binary analyses of LON grades 3–4 showed associations with prior natalizumab (odds ratio [OR] 2.8) and higher body mass index (OR 0.87), not confirmed in time-to-event models.

Discussion

In this large real-world cohort, high-grade LON (grades 3–4) events were rare and frequently complicated by infections. Mild (grade 2) LON events were often fluctuating and generally clinically uneventful. Rechallenge with rituximab was rarely complicated by serious adverse events. Prediction of LON remains challenging, highlighting the relevance of infection vigilance during anti-CD20 therapy.

Introduction

Late-onset neutropenia (LON) is an uncommon but clinically relevant adverse effect of rituximab (RTX) therapy, characterized by a delayed decline in neutrophil count after RTX exposure.¹ Although often transient, severe LON can predispose to serious infections requiring hospitalization, antimicrobial therapy, or granulocyte colony-stimulating factor (G-CSF).² Owing to its self-limiting nature, delayed onset, and fluctuating course, neutropenic events may remain undetected without systematic laboratory monitoring.³

In multiple sclerosis (MS), B-cell–depleting therapies (BCDTs) are a cornerstone of disease-modifying therapies (DMTs).^4,5 Compared with hematologic or rheumatologic indications, persons with MS (pwMS) differ substantially in age, baseline immune status, cumulative treatment exposure, prior immunomodulatory therapies, and monitoring routines.^6,7 These factors may influence both the occurrence and clinical significance of treatment-related neutropenia. BCDTs are associated with increased susceptibility to infections in MS through several mechanisms: sustained B-cell depletion,⁸ reductions in immunoglobulin levels,⁹ and impaired vaccine response.¹⁰

LON is considered a secondary neutropenia contributing to the risk of severe infections such as sepsis.^11,12 Severity grading can assist in determining clinical relevance, using LON grades 2–4 from the Common Terminology Criteria for Adverse Events (CTCAE).¹³ Proposed mechanisms for LON include transient disruption of granulocytopoiesis during immune depletion and reconstitution,^14,15 infection-related immune perturbations,^16,17 and genetic factors, such as antineutrophil antibodies.¹² However, the relevance of these mechanisms in MS populations remains incompletely understood.

Reported incidence estimates of LON after BCDT vary across disease settings. In multiple sclerosis (MS), available data are limited, with reported risks ranging from 1% to 4.4% for RTX, ocrelizumab, and ofatumumab from cohorts and registration studies.^2,18,19 Similar incidences are reported for RTX, ranging between 2% and 5%, in rheumatoid arthritis (RA) and neuromyelitis optica spectrum disorders (NMOSDs),^20,21 while about 30% have been reported in SLE (systemic lupus erythematosus) and lymphomas.^12,22 Owing to differences in patient characteristics, treatment regimens, and background immunosuppression, estimates cannot be directly extrapolated to MS therapy. This heterogeneity complicates clinical decisions regarding monitoring intensity, infection vigilance, and treatment continuation or rechallenge after neutropenia.

There is insufficient MS-specific knowledge regarding the incidence, predictors, recurrences, and clinical implications of late-onset neutropenia in multiple sclerosis. Although several potential risk factors for LON during rituximab therapy have been proposed, e.g., low baseline IgG/IgM levels, longer RTX exposure, older age, and prior use of immunosuppressants,²³ it remains unclear why certain MS patients develop severe LON events. Evidence to guide clinicians regarding the identification of patients at increased risk in real-world MS populations treated with BCDT is limited.

This study aimed to determine the incidence of LON in a large real-world cohort of RTX-treated pwMS. In addition, we aimed to characterize the severity and clinical features of LON and to identify patient-related and treatment-related predictors of LON with particular emphasis on high-grade LON (grades 3–4).

Methods

We conducted a retrospective registry-based observational cohort study collecting clinical data on pwMS treated with rituximab at 6 Swedish neurology centers with long experience in registry-based clinical studies. A previous real-world MS RTX cohort²⁴ was followed for an additional 3 years, identifying participants through the Swedish Neuro Registries, a national Quality Registry for MS care with coverage of almost 85%, high accuracy, and completeness.^6,25 Registry data were validated for treatment data, and laboratory results were extracted from the journals through manual entry at all sites, except for electronic extraction followed by curation used at one center (Stockholm CFN Karolinska).

Study Cohort and Follow-Up

We included 2,713 patients with a diagnosis of MS or clinically isolated syndrome according to the current McDonald criteria,²⁶ having received at least 1 dose of RTX from March 14, 2008, to December 31, 2023. The inclusion date was the first RTX infusion date. The follow-up ended either 365 days after RTX discontinuation or at data extraction on November 24, 2024. From the initial cohort (n = 2,713), exclusions were applied at baseline and during follow-up based on center and complete data (Figure 1). Exclusions were made if not being followed at a participating center (n = 13) or lacking data on absolute neutrophil count (ANC) or RTX dosing (n = 8). Thirteen participants transferring to nonparticipating centers were retained because data coverage was deemed sufficient. After exclusions, the final study cohort comprised 2,686 patients. The 6 participating MS centers were grouped into 4 geographical study centers, accounting for differences in cohort size and treatment practices: Northern (n = 493; Umeå University Hospital and Östersund Hospital), Gothenburg (n = 424; Sahlgrenska University Hospital), Stockholm CFN/Karolinska (n = 1,441; Center for Neurology (CFN) at Academic Specialist Center and Karolinska University Hospital Huddinge), and Stockholm Danderyd (n = 328; Danderyd Hospital). The study center was included as an adjustment variable in all regression models, not evaluated as a causal predictor.

Figure 1. Flowchart for the Inclusion and Exclusion of Study Participants.

A flowchart depicting the number of people with MS eligible for study inclusion, with information on sex and study center. All patients fulfilling inclusion criteria (n = 2,713), being treated with rituximab, were included in the initial study cohort. There are 23 individuals in the initial study cohort (15 female, 65.2%) initially followed at a nonparticipating center who later transferred to a participating center. CFN = Center for Neurology.

Exposures and Outcomes

The exposure was at least 1 dose of RTX, administered IV according to local routine (usually 500–1,000 mg once or twice yearly), with standard premedication (oral paracetamol, antihistamines, and steroids). The main outcome was the proportion and incidence rate of late-onset neutropenia (LON). Secondary outcomes included predictors of neutropenia and descriptive analyses of recurrent events. For grades 3–4 events, we recorded clinical infectious presentations, duration of neutropenia, management of infections (hospitalization, use of antimicrobial treatment with antibiotics, antivirals, antifungals, and G-CSF), RTX rechallenges (reintroduction of RTX after a LON event), and recurrences of neutropenia (more than 1 episode of neutropenia). Other possible contributing causes of neutropenia, not fulfilling exclusion or censoring criteria, were evaluated. In addition, cases of preexisting or early-onset neutropenia were described.

Events of Neutropenia

LON was defined as an ANC <1,500 cells/µL from normal pretreatment levels, occurring ≥30 days after rituximab infusion. Neutropenic events were recorded as the nadir value within each 30-day period, using the most severe event per patient for the main outcome. Preexisting neutropenia was defined as neutropenia 0–90 days before and early-onset as neutropenia 1–30 days post-RTX initiation. Grading of neutropenia severity followed the CTCAE, version 5.0.¹³ In our analyses, CTCAE grades 0–1 was grouped as no neutropenia (ANC ≥1,500 cells/µL). Analysis grades were defined as grade 2 (mild), 1,000–<1,500 cells/µL; grade 3 (moderate), 500–<1,000 cells/µL; and grade 4 (severe), <500 cells/µL. For the main outcomes, any LON (grades 2–4) and high-grade LON (grades 3–4) were used.

Variables and Data Collection

We collected data on age, sex, MS type, center, body mass index (BMI), prior immunomodulatory therapy, and RTX doses and dates. Laboratory data on ANC and immunoglobulin were collected retrospectively from routine samples for preinfusion, 0–90 days pretreatment. Data cleaning for logical checks of consistency and completeness was performed by reviewing medical records. Expanded Disability Severity Score (EDSS),²⁷ registry data were available, but not consistently time aligned with rituximab exposure or LON events. A baseline EDSS measurement was defined as ±2 years of inclusion, therefore used for descriptive purposes and not included as a covariate in regression analyses. BMI data were obtained from the patient charts or registry as health care–collected or self-reported measurements, calculating mean individual BMI values to account for differences in available measurements per individual, and differences in data sources. A sensitivity analysis assessed the stability of mean BMI measurements over time. Data on infectious complications for LON grades 3–4 were collected from medical records and validated against registry output, using variables for RTX dose and rechallenge data, clinical workup, and neutrophil dynamics (eTable 4, A and B). Data checks were not performed for grades 3–4 neutropenic events not fulfilling LON criteria.

Statistical Analyses

Baseline characteristics were summarized using means with SD and medians with interquartile ranges (IQR) at a 95% confidence level. Cumulative incidences of any LON (grades 2–4) and for grades 3–4 were calculated as proportions. The number of participants who experienced a LON event was divided by the population at risk, individuals without preexisting or early-onset neutropenia. For preexisting or early-onset neutropenia, percentages were calculated for the total cohort. For follow-up time, as person-years, the total time at risk of LON started 30 days after initiating RTX, contributing time until the first event or until the end of follow-up. Incidence rates for any grades 3–4 LON were calculated as the number of events per 1,000 person-years. Exact 95% CIs were calculated assuming a Poisson distribution for the number of events.

Predictors and covariates were selected a priori based on clinical relevance and data availability. Age and sex were included as adjustment variables in regression models. The exposures of interest variables, such as BMI, IgG, baseline ANC, MS phenotype, and prior DMT, were evaluated in separate multivariable models, adjusting for confounders and study center, included to account for differences in treatment practices and data structure. Data on medical comorbidities were not available as covariates. Separate multivariable models were fitted for each exposure variable of interest. Effect modification was not formally assessed through interaction terms because the study was not powered for interaction analyses, with a risk of overfitting given the number of events.

Time-to-first LON event was analyzed using Cox proportional hazards regression, assuming independent data, with rituximab exposure modelled as a time-dependent variable. A counting-process (start-stop) data structure was used to account for changes in RTX exposure over time, with follow-up time split into multiple intervals per patient, either as the accumulated number of infusions or cumulative dose (in grams). Models were adjusted for age at RTX start, sex, mean BMI, MS phenotype, prior DMT, and study center. The accumulated number of rituximab doses was selected based on model stability and precision.

Analyses using cumulative rituximab dose were conducted as sensitivity analyses at years 2, 3, and 5.

Binary outcomes for any LON and grades 3–4 LON compared with no neutropenia were analyzed using logistic regression to estimate odds ratios (ORs) with 95% confidence intervals. Models were adjusted for demographic and clinical covariates. BMI was modeled as a continuous covariate; category-based BMI analyses were considered exploratory because of limited precision.

For binary and Cox regression modelling, pre-RTX treatment groups (natalizumab vs oral/injectables) were used to obtain comparable groups. For time-to-event Cox analyses, the tmerge function in the survival package in R was used. Robust variance estimations were applied to account for clustering by center. Hazard ratios (HRs) with 95% CIs are reported. Data were stored with a pseudonymized ID code. Statistical analyses were performed using R Studio (version Apple Blossom 2026.01.0 + 392). Statistical significance was set at p < 0.05.

Censoring and Missing Data

Treatment-related neutropenia was censored from outcome analyses, excluding 29 cases. Patients undergoing autologous stem cell therapy for MS (n = 28) only contributed data until 90 days before conditioning, and one patient receiving cytostatic therapy before a grade 3 LON event did not contribute further ANC measurements. Missing mean BMI values were imputed using multiple imputation by chained equations with 5 data sets, analyzed separately, and pooled estimates reported (mice package in R).

Grants and Funding

The study was financed by grants from the Swedish state under the agreement between the Swedish government and the county councils, the ALF agreement B5069.

Standard Protocol Approvals, Registrations, and Patient Consents

Ethical approval for the study was obtained from the Swedish Ethical Review Authority (DNR: 2019-01187). The study was based on data extracted from the Swedish Neuro Registries, together with verification through patient medical records. In accordance with national regulations governing registry-based research, written informed consent from individual patients was not required.

Reporting Guidelines

This manuscript adheres to the STROBE reporting guidelines for observational studies.

Data Access

The study's underlying data consist of individual-level, linked entries from the Swedish MS Registry and routinely collected electronic health records. These data sets are protected under Swedish law and the General Data Protection Regulation (GDPR) and therefore cannot be directly shared by the authors. Researchers may request access from the registry's data controllers, contingent on (1) an approved ethics application within Sweden and (2) a data-sharing agreement with the Swedish MS Registry and relevant health care authorities.

Data Availability

Owing to Swedish legislation and the GDPR governing registry-based health data, deidentified individual-level data cannot be made publicly available. To support reproducibility, the statistical analysis code, analytic specifications, and variable definitions used in this study will be made available from the corresponding author upon reasonable request and subject to institutional and ethical approvals.

Results

The study population was derived from a large multicenter cohort comprising 2,713 pwMS patients from 6 MS centers with catchment areas of about half of Sweden's population. The final study cohort included 2,686 pwMS treated with RTX (68% female) with a mean age of 41.9 years (SD 11.9). The baseline characteristics are presented in Table 1, with minor differences in age, sex, and prior DMT. The median follow-up time was 7.8 years (IQR 5.7–9.8; range 16.7 years). The dose range was 100–1,000 mg, with a mean number of doses of 8.4 (range 1–28). The mean accumulated RTX dose was 4,850 mg (SD 2350). The median interval between doses was 192 days (IQR 182–362).

Table 1.

Baseline Characteristics of the Study Population

	Gothenburg (n = 424)	Northern (n = 493)	Stockholm CFN Karolinska (n = 1,441)	Stockholm Danderyd (n = 328)	Total (n = 2,686)
Sex, n (%)
Female	282 (66.5)	331 (67.1)	993 (68.9)	232 (70.7)	1838 (68.4)
Male	142 (33.5)	162 (32.9)	448 (31.1)	96 (29.3)	848 (31.5)
Age
Mean (SD)	43.5 (12.3)	40.7 (11.6)	41.9 (11.6)	41.7 (10.8)	41.9 (11.6)
Median [range]	43.0 (9.70–74.6)	40.8 (13.5–75.1)	41.8 (14.4–77.1)	41.7 (15.7–69.6)	41.9 (9.70–77.1)
BMI
Mean (SD)	25.6 (4.94)	27.4 (6.94)	25.3 (5.11)	25.8 (5.47)	25.7 (5.46)
Median [range]	24.7 (17.5–46.6)	25.2 (16.3–49.8)	24.3 (14.5–61.6)	25.2 (13.7–49.9)	24.6 (13.7–61.6)
Missing, n (%)	252 (59.4)	288 (58.4)	477 (33.1)	111 (33.8)	1,128 (42.0)
MS phenotype, n (%)
RRMS	278 (65.6)	381 (77.3)	1,006 (69.8)	266 (81.1)	1931 (71.9)
SPMS	107 (25.2)	84 (17.0)	361 (25.1)	55 (16.8)	607 (22.6)
PPMS	38 (9.0)	28 (5.7)	72 (5.0)	6 (1.8)	144 (5.4)
NA	1 (0.2)	0 (0)	2 (0.1)	1 (0.3)	4 (0.1)
EDSS
Median [range]	2.50 (0–8.75)	2.00 (0–8.00)	2.50 (0–8.75)	1.75 (0–9.00)	2.00 (0–9.00)
Missing	55 (13.0)	186 (37.7)	279 (19.4)	38 (11.6)	558 (20.8)
Previous DMT, n (%)
Cladribine	23 (5.4)	1 (0.2)	16 (1.1)	2 (0.6)	42 (1.6)
Fumarate	36 (8.5)	20 (4.1)	125 (8.7)	44 (13.4)	225 (8.4)
Glatiramer acetate	12 (2.8)	20 (4.1)	61 (4.2)	13 (4.0)	106 (3.9)
Interferon	41 (9.7)	118 (23.9)	297 (20.6)	60 (18.3)	516 (19.2)
Natalizumab	118 (27.8)	104 (21.1)	242 (16.8)	76 (23.2)	540 (20.1)
Other	24 (5.7)	32 (6.5)	54 (3.7)	10 (3.0)	120 (4.5)
Ocrelizumab/ofatumumab	8 (1.9)	1 (0.2)	30 (2.1)	10 (3.0)	49 (1.8)
S1PR inhibitors	30 (7.1)	21 (4.3)	147 (10.2)	15 (4.6)	213 (7.9)
Teriflunomide	15 (3.5)	1 (0.2)	22 (1.5)	5 (1.5)	43 (1.6)
Treatment naïve	117 (27.6)	175 (35.5)	447 (31.0)	93 (28.4)	832 (31.0)
Baseline IgG
Median [range]	9.40 (2.30–17.0)	9.41 (4.26–19.0)	9.84 (4.37–20.6)	9.69 (4.98–16.3)	9.67 (2.30–20.6)
Baseline ANC
Median [range]	3,950 (1,120–1,140)	3,900 (1,830–9,380)	3,870 1,220–15,200)	3,720 (1,030–1,000)	3,870 (1,030–15,200)

Abbreviations: AHSCT = autologous stem cell therapy; ANC = absolute neutrophil count; DMT = disease-modifying therapy; EDSS = Expanded Disability Severity Scale; IgG = immunoglobulin G; PPMS = primary progressive MS; RRMS = relapsing-remitting MS; RTX = rituximab; SPMS = secondary progressive MS.

Baseline characteristics by study center. Continuous variables are summarized as mean (SD) or median (range). BMI includes self-reported and clinic-measured values. Prior DMTs before RTX are summarized by treatment class: fumarates (dimethyl fumarate, n = 223; diroximel fumarate, n = 2), S1P receptor modulators (fingolimod, n = 204; siponimod, n = 9), other DMTs (unspecified/other, n = 44; mitoxantrone, n = 38; AHSCT n = 19; alemtuzumab, n = 15; daclizumab, n = 2; and azathioprine and ciclosporin, n = 1 each), and ocrelizumab/ofatumumab (ocrelizumab n = 24; ofatumumab n = 25). Baseline laboratory values for ANC in cells/µL; IgG in g/L). Missing values are reported as NA.

Laboratory testing of ANC and IgG followed standardized clinical practice with a median time from sampling to a subsequent RTX infusion of 12 days (IQR 6–34 days). EDSS progression was slow at the population level, with a median annual decline of 0.04 points (IQR 0.17), and did not differ meaningfully between patients who developed LON and the overall cohort. Neither age at the start of RTX, sex, MS type, nor EDSS was associated with the outcomes of any or high-grade LON. EDSS trajectories did not indicate more aggressive disease among LON cases (data not shown).

Events of Late-Onset Neutropenia

The proportion of individuals experiencing at least 1 episode of LON was 6.2% (165; population at risk 2,641) during follow-up (Table 2). The incidence rate of any LON (grades 2–4) was 10.1 per 1,000 person-years (95% CI 8.6–11.8), with a median time-to-first event of 360 days (IQR 112–1,017). In most cases (86%), the LON event was mild (grade 2), occurring in 142 individuals (5.3%). High-grade LON (grades 3–4) was uncommon, with 36 events affecting 23 individuals (0.9%): 17 (0.6%) with moderate (grade 3) and 6 (0.2%) with severe (grade 4) neutropenia. The IR of grades 3–4 was 2.1 per 1,000 person-years (95% CI 1.4–2.9), with a median time-to-first event of 404 days (IQR 145–941). Data on the median duration of LON events were limited. For any LON, data on repeated testing until recovery were not available. The median duration of grades 3–4 events was 6 days (range 2–570), and ANC measurements until recovery were available for 22% (n = 8) of cases.

Table 2.

Incidence and Severity of Neutropenia After RTX Initiation

Category	No. of individuals	Percent	Proportion	IR per 1,000 PY (IQR)	Median days to event (IQR)
Any LON (grades 2–4)	165	6.14	0.062	10.12 (8.64–11.77)	360 (112–1,017)
Grade 2 (mild)	142	5.29	0.054
LON grades 3–4	23	0.86	0.009	2.06 (1.43–2.88)	404 (145–941)
Grade 3 (moderate)	17	0.63	0.006
Grade 4 (severe)	6	0.22	0.002
Preexisting/early-onset neutropenia	45	1.68
No neutropenia	2,476	92.18
Population at risk	2,641
Study cohort	2,686
Any LON (grades 2–4)	165	6.14	0.062	10.12 (8.64–11.77)	360 (112–1,017)
Grade 2 (mild)	142	5.29	0.054
High-grade LON (grades 3–4)	23	0.86	0.009	2.06 (1.43–2.88)	404 (145–941)
Grade 3 (moderate)	17	0.63	0.006
Grade 4 (severe)	6	0.22	0.002
Preexisting/early-onset neutropenia	45	1.68
No neutropenia	2,476	92.18
Study cohort	2,686

Abbreviation: IQR = interquartile range; RTX = rituximab.

Number and percentage of individuals with late-onset neutropenia (LON) of any grade and by severity and incidence rates (IRs) per 1,000 person-years (PY), median time to a LON event, and total accumulated person-time (16,508 person-years). The population at risk includes 2,641 individuals without preexisting or early-onset neutropenia.

Recurrent LON

There were 165 first LON events, after which 28 discontinued RTX. Of the 137 who received repeated RTX therapy, 41 experienced recurrent LON events. Of these, 26 individuals had 2 events, and 15 had 3 or more episodes during follow-up (Figure 2, eFigure 1).

Figure 2. Timeline Displaying Neutrophil Dynamics After the First Grades 3–4 Late-Onset Neutropenia (LON) Event During Rituximab (RTX) Therapy for pwMS.

ANC trajectories for individuals with LON grades 3–4 (n = 21). Time is centered at the first LON event (day 0). Shapes indicate the individual's highest neutropenia grade. The red dashed line indicates ANC <1,500 cells/µL, and color-coded background bands show ANC thresholds (0 green, 2 yellow, 3 orange, and 4 red). ANC = absolute neutrophil count; pwMS = people with MS.

Rechallenge of RTX in grades 3–4 LON, see eTable A-B: Overall, RTX was rechallenged in 89% of grades 3–4 episodes (32 of 36 events). A recurring neutropenia pattern was observed in grade 3 (n = 13) and grade 4 (n = 1) LON events (Figure 2). Two events were symptomatic: one severe infectious event and one fluctuating neutropenia complicated by a mild infection. All 6 patients with a first event of severe neutropenia grade 4 were rechallenged with RTX. On 3 occasions, recurring asymptomatic LON grade 3 complicated the RTX treatment for 2 individuals.

Infectious complications during grades 3–4 LON events are summarized in eTable A-B. In total, 10 infections (8 severe) occurred during grades 3–4 neutropenia, corresponding to an infectious complication rate of 28%. Overall, 22 individuals (47.9%) with high-grade LON were symptomatic. Among grade 4 events (ANC 0–0.4 cells/µL), half were symptomatic (3 of 6 events), and all symptomatic cases represented severe infections. These episodes resulted in hospitalization for 4–7 days (n = 3), broad-spectrum antibiotic treatment (n = 3), and G-CSF administration (n = 2). Neutropenia resolved within 7 days. Three asymptomatic, self-resolving grade 4 events (ANC 0.2–0.3 cells/µL) were also observed. Grade 3 LON was more common, occurring in 19 individuals across 30 events. During these episodes, 7 infections were documented, including 5 severe infections, requiring hospitalization for up to 12 days.

Predictors of LON were evaluated for any LON (grades 2–4) and for high-grade LON (grades 3–4). For any LON (Figure 3A, eTable 2A), prior injectable therapy was associated with a lower hazard of LON (HR 0.66, 95% CI 0.45–0.96). The number of accumulated rituximab infusions was also associated with a lower hazard of LON (HR 0.87, 95% CI 0.81–0.93). Analyses using cumulative rituximab dose (per 1,000 mg) yielded similar estimates, although with wider confidence intervals and reduced precision (data not shown). In sensitivity analyses with time-censored follow-up, progressive MS showed the strongest association with increased LON hazard, while accumulated rituximab exposure and prior disease-modifying therapy remained associated with a reduced early hazard of LON (eTable 3).

Figure 3. Forest Plots.

(A) Predictors of any late-onset neutropenia (LON). Forest plot showing hazard ratios (HRs) and 95% confidence intervals (CIs) from a time-dependent time-to-event Cox proportional hazards regression model evaluating predictors of any LON event (grade 2–4, absolute neutrophil count [ANC] <1,500 cells/mL) vs no neutropenia (grade 0; ANC ≥1,500 cells/mL). Covariates included accumulated number of rituximab (RTX) infusions, age at RTX initiation, sex (male vs female), body mass index (BMI), MS phenotype, prior treatment group (natalizumab; oral disease-modifying therapies [DMTs], injectable DMTs; treatment-naïve as reference), and study center (Stockholm CFN/Karolinska, Stockholm Danderyd, Northern region; Gothenburg as reference). HRs (95% CIs) are displayed on a logarithmic scale. (B) Predictors of high-grade LON. Forest plot showing HRs and 95% CIs from a time-dependent time-to-event Cox model evaluating predictors of grade 3-4 LON (ANC <1,000 cells/mL) compared with grade 0-2 neutropenia status (ANC ≥1,000 cells/mL). Covariates included accumulated number of RTX infusions, age at RTX initiation, sex (male vs female), BMI, MS phenotype, prior treatment group (natalizumab; oral DMTs, injectable DMTs; treatment-naïve as reference), and study center (Stockholm CFN/Karolinska, Stockholm Danderyd, Northern region; Gothenburg as reference). HRs (95% CIs) are displayed on a logarithmic scale. Color key: green = lower odds (significant), brown = higher odds (significant), orange = nonsignificant, and blue = study-center effects. ANC = absolute neutrophil count; BMI = body mass index; CFN = Center for Neurology; DMT = disease-modifying therapy; LON = late-onset neutropenia.

For high-grade LON (grades 3–4), effect estimates were directionally similar but less precise, reflecting the smaller number of severe events (Figure 3B, eTable 1B). In the time-dependent Cox models, the hazard ratio for accumulated rituximab infusions was 0.87 and did not reach statistical significance. Trends toward a protective association were observed for prior injectable therapy; however, confidence intervals were wide.

In binary logistic regression analyses of any LON (eTable 2A, eFigure 2A), older age at rituximab initiation was associated with lower odds of LON with an odds ratio of 0.98 per year (95% CI 0.97–0.996). BMI as a continuous variable showed no independent association with the outcome. Sex, MS phenotype, region, and treatment group were not significantly associated with any LON. Additional adjustments for baseline IgG levels and neutrophil counts did not materially alter the results and were not retained in the binary models. For high-grade LON (grades 3–4; eTable 2B, eFigure 2B), prior natalizumab treatment was associated with higher odds of severe LON (OR 2.87, 95% CI 1.31–6.13), while higher BMI was associated with lower odds (OR 0.87, 95% CI 0.78–0.96). The models were adjusted for study center effects. Given the limited number of high-grade events, effect estimates for grades 3–4 LON should be interpreted with caution.

Preexisting and Early-Onset Neutropenia

A subset of 45 individuals (1.7%) in the full cohort had neutropenic events not meeting the definition of LON, either preexisting (33 individuals; 1.2%) before the first RTX infusion or early-onset neutropenia 12 (0.4%) within 30 days of initiation. This subset (Figure 4) exhibited fluctuating and recurrent patterns (17 individuals; 0.6%) in 65 events over follow-up. Most cases fluctuated as grade 2 (n = 12), despite about half (5 individuals, 11 events) progressing to grades 3–4 neutropenia. The last prior DMT was injectables (50%; n = 20), oral DMTs (27%; n = 12), natalizumab (18%; n = 8), and treatment naïve (11%; n = 5); none had previous anti-CD20 exposure.

Figure 4. Timeline of Neutrophil Dynamics During RTX Therapy for Preexisting and Early-Onset Neutropenia.

ANC trajectories for individuals with preexisting or early-onset neutropenia occurring less than 30 days of RTX initiation (n = 45), faceted by previous DMTs. Shapes indicate the individual's highest neutropenia grade. The red dashed line indicates ANC <1,500 cells/µL, and color-coded background bands show ANC thresholds (0 green, 2 yellow, 3 orange, and 4 red). ANC = absolute neutrophil count; DMT = disease-modifying therapy; RTX = rituximab.

Discussion

In this large, longitudinal, multicenter cohort of RTX-treated pwMS, we provide comprehensive characteristics of LON in a real-world setting. We observed a cumulative LON incidence of 6.2% over a median follow-up of 7.8 years, higher than previously reported estimates (1.3–4%) in MS cohorts^2,28 but comparable with rates (6–7%) in longitudinal studies of BCDT in autoimmune diseases.^1,12 These differences likely reflect longer follow-ups, systematic laboratory monitoring, and broader capture of mild events. Most LON events were mild (grade 2), affecting 5.3%, where those with fluctuating patterns rarely progressed to clinically relevant neutropenia after rechallenging RTX. The clinically relevant risk of having a high-grade event (grades 3–4) was rare (1%) and was more aligned with previous reports, displaying sudden onset, rapid resolution, and high rates of severe infections.^18,29

Data on infectious complications in LON in MS and NMOSD are limited, and report infection rates of 60–70% in BCDT using different ANC cutoffs, dosing protocols, and sample sizes,^19,29,30 not correlating outcomes with LON severity.^2,28 We studied infectious complications occurring in LON grades 3–4 events, affecting nearly half of individuals, and most were clinically severe. These rare and mostly transient LON events resolved with clinical infectious management. Data on grade 2 infections were not collected in this study, based on published evidence indicating a low risk of symptomatic grade 2 LON.¹ Therefore, no conclusions can be drawn on infection rates for the mild LON events in this study.

Given this retrospective setting, causality between neutropenia and infection cannot be firmly established.¹⁷ Neutropenia may predispose to infection, but infections themselves—particularly severe systemic infections—can suppress granulocytopoiesis and precipitate transient neutropenia.^16,31 In our study, a few grades 3–4 LON cases occurred in temporal proximity to a severe COVID-19 infection, suggesting a possible infectious trigger of neutropenia and bidirectional severity of the infection. Viral infections may interfere with neutrophil tissue distribution, causing milder neutropenia.^17,32 BCDT can increase the risk of infection, often bacterial, while masking symptoms of subclinical infections.^8,14,29 Many grades 3–4 LON events were asymptomatic and identified through laboratory monitoring because clinically relevant neutropenia may remain undetected in the absence of overt infectious symptoms. In selected cases, G-CSF was used; notably, there is limited evidence of an association with increased relapse risk in MS.^33,34 Consequently, monitoring of neutrophil counts is warranted both in routine preinfusion assessments for rituximab and during the workup of severe or atypical infections.

Rechallenge of RTX after LON was generally feasible in this cohort, performed for 83% overall and for 89% with high-grade events, and severe recurrent events uncommon. However, data from other studies imply that severe LON events may recur; therefore, individual risk management and clinical monitoring are advised rather than routine discontinuation of BCDT.^28,35

We did not identify strong or consistent predictors of LON. Accumulated RTX exposure was not associated with an increased hazard of LON, suggesting that continued treatment will not increase risk, although clinical selection effects and healthy survivor bias cannot be fully excluded.

Prior natalizumab treatment was associated with increased odds of high-grade LON in logistic regression but not in time-to-event analyses, indicating that any excess risk may be transient and related to treatment-switching effects. This pattern is consistent with previous observations linking secondary neutropenia to periods of rapid immunologic change.^30,36 In addition, depletion of susceptibles because of higher detection rates in prior natalizumab health care–seeking for infections may attenuate observed hazard ratios during long-term follow-up.

Prior injectable therapy was associated with a lower hazard of any LON, which may reflect patient selection with lower disease activity. Compared with escalation therapies, injectables are less likely to induce profound immune redistribution or bone marrow stress, potentially reducing susceptibility to treatment-related neutropenia.³⁷ For grades 3–4 events, associations were in similar directions as for any LON but with less precision, consistent with the limited numbers of events. There was no evidence for any clear difference in predictors compared with mild LON cases. The mean BMI showed an association with LON events, where higher BMI was protective in the time-to-event analysis for any LON. This should be interpreted with caution because the BMI data had methodological issues regarding consistency and density.

LON in BCDT for MS has primarily been reported in association with rituximab and ocrelizumab,^4,38 whereas neutropenia appears uncommon with ofatumumab. In extension studies of ofatumumab, reported incidence has been low (0.9%), and few case reports describe late-onset neutropenia.^28,35,39 In our cohort, no grades 3–4 LON events were observed after treatment with ocrelizumab or ofatumumab. Differences in dosing regimens between BCDT have been proposed for the lower rates of ofatumumab.^40,41

Several mechanisms have been proposed to explain LON during BCDT, including transient disruption of granulopoiesis during B-cell depletion and immune reconstitution.^14,21 Our findings support a heterogeneous pathophysiology in MS, in which mild or fluctuating neutropenia likely reflects transient immune instability, whereas rare high-grade events may occur during periods of increased vulnerability, such as after treatment switching.

In line with reports from MS and other autoimmune conditions,^1,12,35 we observed fluctuating neutropenia patterns predominantly limited to low-grade events. These episodes were generally clinically benign and not associated with hospitalization or progression to severe neutropenia, suggesting that fluctuating LON represent a more low-key immune instability.^14,17 These findings emphasize that the heterogeneous presentation of LON in MS likely reflect overlapping mechanisms.

In conclusion, we demonstrate in this large longitudinal MS study that LON during RTX therapy is a heterogeneous phenomenon, in which severity grade and ANC patterns are associated with distinct clinical implications. Although LON was relatively common, most events were mild, nonprogressing, and of limited clinical consequence. Fluctuating and recurrent neutropenia was more frequent but largely uneventful, whereas clinically relevant risk was concentrated in rare high-grade LON events, for which early detection and management of severe infections were important. Rechallenge after LON was generally manageable, but decision making should account for individual patient risks and emphasize vigilance for infectious complications rather than routine treatment discontinuation. Overall, the lack of clear predictors of LON remains challenging, suggesting awareness after treatment sequencing.

This large retrospective cohort study collected data from several major MS centers in Sweden and provides robust real-world estimates of a rare adverse outcome in RTX-treated pwMS. The participating centers have extensive experience with RTX treatment, and clinical and laboratory data are systematically recorded in the national MS registry, allowing reliable identification of LON. The longitudinal design can capture delayed adverse events that may be underestimated in smaller cohorts or studies with shorter follow-up times.

This study has several limitations. Detection of rare and delayed events may be incomplete in an observational study setting, and LON is therefore likely to be underascertained. Although laboratory testing followed standardized protocols with relatively uniform timing, detection bias may occur because neutrophil counts are more likely to be measured in symptomatic patients. Infectious symptoms were assessed only for grades 3–4 neutropenia. There is an obvious bias for detecting neutropenia when it occurs during a severe infection, which may not only underestimate the occurrence of severe asymptomatic LON but also overestimate the risk of developing such an event. Residual heterogeneity between centers may persist despite adjustment for treatment center. In addition, as with all observational treatment studies, immortal time bias cannot be completely excluded. We cannot draw conclusions on causal associations because more extensive modelling using target trial emulations with G-modelling was not within the scope of this article. Registry data were subject to incomplete recording and occasional input errors, and reliable information on alternative causes of neutropenia was limited.

Late-onset neutropenia after rituximab treatment in multiple sclerosis was predominantly mild and clinically uneventful, affecting 6.2%. Clinically meaningful high-grade LON affected 1% and were frequently associated with infectious complications, several requiring hospitalizations. Time-dependent analyses did not indicate an increased long-term hazard with accumulated rituximab exposure. No predictors with sufficient strength or consistency could be identified, emphasizing the need for neutrophil monitoring in BCDT and clinical vigilance for rare high-grade events rather than reliance on baseline risk stratification.

Glossary

ANC

absolute neutrophil count

BCDT

B-cell depleting therapy

BMI

body mass index

CFN

Center for Neurology

CTCAE

Common Terminology Criteria for Adverse Events

DMT

disease-modifying therapy

EDSS

Expanded Disability Severity Score

G-CSF

granulocyte colony-stimulating factor

HR

hazard ratios

IQR

interquartile range

IR

incidence rate

LON

late-onset neutropenia

MS

multiple sclerosis

NMOSD

neuromyelitis optica spectrum disorders

OR

odds ratio

pwMS

people with MS

RA

rheumatoid arthritis

RTX

rituximab

Author Contributions

S. Hallberg: 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. S. Kallin: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data. B. Evertsson: 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. Boremalm: major role in the acquisition of data. P. de Flon: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data. J. Salzer: 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. J. Lycke: 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. K. Fink: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data. A. Svenningsson: 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.

Study Funding

The study was financed by grants from the Swedish state under the agreement between the Swedish government and the county councils, the ALF agreement B5069.

Disclosure

S. Hallberg, S. Kallin, B. Evertsson, M. Boremalm, P. de Flon, J. Salzer, and A. Svenningsson report no disclosures relevant to the manuscript. J. Lycke has received travel support and/or lecture honoraria and has served on scientific advisory boards for Alexion, Biogen, Bristol Myers Squibb, Celgene, Janssen, Merck, Novartis, Roche, and Sanofi; has received unconditional research grants from Biogen and Novartis; and has received financial support from Sanofi for an investigator-initiated study. K. Fink has served on scientific advisory boards for Alexion, Biogen, Janssen, Merck, Novartis, Takeda, and Roche; and has received financial research support from Merck and Amicus. Go to Neurology.org/NN for full disclosures.