Remnant cholesterol and lipid ratios predict the relapse of neuromyelitis optica spectrum disorder

Abstract

Background

Neuromyelitis optica spectrum disorder (NMOSD) is a demyelinating condition in the central nervous system whose relapses cause severe disability progression. Although conventional blood lipid markers are linked to disease course and outcomes, the predictive value of emerging lipid indicators, such as remnant cholesterol (RC) and lipid ratios, for NMOSD relapse remains unclear.

Methods

The single-centre retrospective study enrolled a total of 245 patients diagnosed with NMOSD, based on the availability of clinical and laboratory data. To evaluate RC and lipid ratios in predicting NMOSD relapse, multivariate Cox proportional hazards models and restricted cubic spline evaluations were applied. Predictive performance was assessed using the concordance index (C-index), continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Subgroup evaluations examined the stability of the observed RC-relapse connection across diverse patient strata. Cumulative hazard function curves illustrated the clinical relevance of the RC inflection point. Additionally, mediation analyses tested whether inflammatory markers mediated the RC effect on relapse.

Results

Among 245 NMOSD patients, 55.10% of the patients relapsed during follow-up. RC, non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio, total cholesterol and low-density lipoprotein cholesterol levels emerged as independent determinants of relapse across both continuous and categorical Cox models, after adjusting for demographic, clinical and therapy-associated factors. An “S”-shaped nonlinear relationship was observed between RC values and relapse risk, with a turning point at 0.46 mmol/L: protective below, risk factor above. Performance metrics (C-index, NRI, IDI) indicated that RC significantly improved relapse prediction. The RC-relapse association persisted across subgroups, with the inflection point effectively distinguishing relapsing patients in the anti-aquaporin 4-immunoglobulin G seropositive and monoclonal antibody treatment group. Mediation analysis revealed increased neutrophil ratio and decreased lymphocyte ratio partially mediated RC’s effect on relapse.

Conclusions

RC was identified as the most robust lipid metabolism indicator for predicting NMOSD relapse, displaying an inflection at 0.46 mmol/L. Neutrophil ratio and lymphocyte ratio may partially mediate the relationship between elevated RC and relapse. These findings aid timely recognition of patients at elevated risk and provision of individualised therapeutic interventions to reduce disability and improve long-term outcomes in this debilitating disease.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12944-025-02846-6.

Introduction

Neuromyelitis optica spectrum disorder (NMOSD) represents a serious autoimmune condition characterised by demyelination within the central nervous system (CNS), in which pathogenic mechanisms are largely driven by antibodies targeting aquaporin‑4 (AQP4) [1]. NMOSD is marked by repeated episodes of optic neuritis and longitudinally extensive transverse myelitis; besides, it often results in irreversible neurological disability. Given its relapsing nature, the early identification of factors predisposing patients to relapse is essential for improving patient outcomes.

Previous studies have suggested that female sex, serum AQP4-Immunoglobulin G (IgG) positivity, renal dysfunction, high free thyroxine levels and dyslipidaemia may be related to the relapse of NMOSD [2–6]. Lipids account for more than 70% of the myelin structure and are crucial for myelin formation and maintenance. Dyslipidaemia affects the stability and repair capacity of myelin [7]. Cholesterol metabolism gives rise to oxysterols, including hydroxycholesterol and 7-ketocholesterol, which compromise membrane fluidity and activate inflammatory, oxidative and apoptotic pathways. These mechanisms collectively contribute to neurotoxicity [8, 9]. Moreover, triglyceride (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) levels also correlate with relapse risk in NMOSD [4, 10]. On the contrary, inflammation may promote dyslipidaemia through alterations in lipid metabolism [11].

Given the involvement of aberrant lipid metabolism in NMOSD, the thorough examination of blood lipid markers in relation to NMOSD relapse is essential. Recently, remnant cholesterol (RC) levels and lipid ratios have been proposed as new markers of lipid metabolic status [12, 13]. RC represents cholesterol that is not cleared by high-density lipoprotein and could exert a proinflammatory influence on the pathophysiology of stroke [14]. By reflecting the relative balance between LDL-C and HDL-C, lipid ratios including the TG/HDL-C ratio, LDL-C/HDL-C ratio, and non-HDL-C to HDL-C ratio (NHHR) may serve as surrogate markers of systemic inflammation and immune dysregulation [15–18]. However, whether RC and lipid ratios are associated with relapse risk in patients with NMOSD remains unknown. Accordingly, the present study sought to clarify how RC and lipid ratios may serve as prognostic indicators for NMOSD relapse, with the intention of identifying novel predictive biomarkers and improving individualised disease management.

Research design and methods

Study participants

This investigation, based on a retrospective cohort, was conducted in accordance with the ethical standards established by the Declaration of Helsinki for medical research with human participants and received approval from the ethics committee at the First Hospital of Shanxi Medical University. By reviewing the hospital’s case database, all of the patients who were initially diagnosed with NMOSD at the centre between January 2013 and May 2024 were continuously included. Prior to data collection in 2024, written informed consent was obtained from all participants. AQP4-seropositive and AQP4-seronegative NMOSD were diagnosed according to the 2015 international consensus criteria for NMOSD [1]. In addition, all of the patients who were initially diagnosed with NMOSD at the centre underwent spinal and cerebral magnetic resonance imaging (MRI), serum MOG antibody testing, cerebrospinal fluid oligoclonal band (OB) analysis and rheumatic autoimmune antibody screening. These assessments (combined with the core clinical symptoms of the patients) were utilised to discriminate NMOSD from other autoimmune neurological disorders, notably multiple sclerosis (MS) and myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD). The study applied the following criteria for participant inclusion: (1) fulfilment of the diagnostic criteria for NMOSD; (2) availability of clinical information and blood samples collected during the acute attack phase; (3) completeness of the clinical data; and (4) provision of informed consent by the patient. The following exclusion criteria were utilised: (1) a follow-up period shorter than 12 months after the initial NMOSD diagnosis at the centre; (2) serum MOG-IgG positivity or cerebrospinal fluid OB positivity; (3) incomplete clinical data due to missing information or loss to follow-up; and (4) hepatic or renal insufficiency or failure.

Data collection

Baseline data included patients’ demographic features (age at onset, sex and body mass index [BMI]), clinical features (phenotype, AQP4-IgG serostatus, acute-phase treatment and maintenance therapy after enrolment at the centre), and medical history (hypertension, diabetes and prior use of lipid-lowering agents). Notably, the majority of the NMOSD patients had experienced prior relapses before their initial presentation at the centre. Accordingly, the annual relapse rates (ARRs), maintenance therapies prior to enrolment, and corticosteroid use within the preceding year were documented. Clinical phenotypes of NMOSD may manifest independently or concurrently and were determined based on clinical presentation and MRI findings [1]. Given the limited number of cases presenting with acute brainstem syndrome, area postrema syndrome, diencephalic syndrome or cerebral syndrome, these phenotypes were collectively classified as brain lesions to reduce bias. Patients who exhibited two or more core features were defined as having a mixed phenotype. Acute treatment consisted of high-dose corticosteroids and intravenous immunoglobulin (IVIg), whereas maintenance therapy included immunosuppressive agents (mycophenolate mofetil, azathioprine and cyclophosphamide) and monoclonal antibodies (rituximab, satralizumab, inebilizumab and eculizumab). The standard treatment routine at the centre for the acute attack phase consisted of high-dose corticosteroid therapy or IVIg administered for 5 to 7 consecutive days, followed by sequential maintenance therapy.

Blood marker measurements

In accordance with the standardised protocols of the centre, patients in the acute phase who have not yet received treatment from the centre underwent intravenous blood sampling after a 12-hour fasting period. Serum was separated by centrifuging the samples at room temperature at 3,500 rpm for 10 min prior to biochemical analysis, and total cholesterol (TC), TG, LDL-C and HDL-C levels were quantified. Blood samples obtained from patients at the centre were sent to a commercial institution (KingMed Diagnostics, Guangzhou, China) for AQP4-IgG testing within 2 h of collection under cold chain conditions. The utilised test method was the fixed cell-based assay. Additional laboratory evaluations, namely, complete blood count, lymphocyte subset distribution and systemic immune response markers, were also performed, as well as assessments of erythrocyte sedimentation rate (ESR), serum cytokine (tumour necrosis factor [TNF], interferon-γ [IFN-γ], interleukin-2 [IL-2], IL-4, IL-6, IL-10 and IL-17), complement component (C3 and C4) and Ig (IgM, IgG, and IgA) levels.

Calculation of lipid indices

RC and lipid ratios were derived from standard laboratory measurements of TC, TG, LDL-C and HDL-C levels. RC was quantified as the component of TC excluding contributions from LDL-C and HDL-C (defined as subtracting LDL-C and HDL-C from TC). The TG/HDL-C ratio was calculated as the quotient of TG levels over HDL-C levels, whereas the LDL-C/HDL-C ratio was defined as the proportion of LDL-C to HDL-C. To obtain NHHR, HDL-C was subtracted from TC, and the resulting value was divided by HDL-C.

Follow-up and endpoint events

Longitudinal monitoring of NMOSD patients is systematically performed at the centre. Patients were recommended to return for clinical evaluation every 3 to 6 months and to seek immediate medical attention if symptoms worsened or if new symptoms emerged. For patients who missed their scheduled follow-ups, the centre contacted them via telephone every three months to assess any changes in their condition. The endpoint event was defined as the first relapse after enrolment, which was characterised by either an exacerbation of previously existing clinical manifestations or the onset of novel symptoms occurring for more than 24 h. For patients with suspected relapse, MRI scans of the brain and spinal cord (including both plain and contrast-enhanced scans) were performed, along with evaluations of serum AQP4 antibodies, MOG antibodies, and cerebrospinal fluid-specific OB. These evaluations help to confirm disease relapse. Follow-up and relapse data were obtained from the medical records database. Inclusion in the statistical analysis required patients to possess a minimum of 12 months of follow‑up data following their initial diagnosis at the centre.

Statistical analysis

The statistical analyses and visualisation of figures were conducted with R (version 4.3.2) and SPSS (version 27.0). Categorical outcomes were expressed in terms of counts (percentages), with group differences assessed via the chi-square (χ²) test. Continuous variables are presented as mean ± standard deviation (SD), and between-group differences for normally distributed data were assessed using Student’s t-test. To explore the prognostic value of RC and lipid ratios in identifying relapse risk among patients with NMOSD, multivariate Cox proportional hazards analyses were constructed. Continuous and categorical models were both applied; specifically, the continuous model treated RC as a linear variable, whereas the categorical model stratified RC into quartiles to explore potential threshold effects. Three hierarchical Cox models were constructed to incorporate adjustments for confounding variables. Model 1 included unadjusted variables. Model 2 was adjusted for age at onset, sex, and BMI. Model 3 additionally accounted for clinical and therapeutic variables in its adjustments, including clinical phenotype, acute and maintenance immunotherapy, serum anti-AQP4 antibody status, ARR before enrolment, corticosteroid therapy within 1 year before enrolment, maintenance therapy before enrolment, hypertension, diabetes and use of lipid-lowering medications. Nonlinear patterns were further explored by using a restricted cubic spline (RCS) to visualise dose-response curves. For lipid indicators demonstrating a nonlinear association with NMOSD relapse in the RCS analysis, a hazard ratio (HR) = 1 was used as the cutoff value to determine the inflection point. The concordance index (C‑index) was employed to evaluate both the discriminatory ability of the model and its predictive accuracy. Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were applied to evaluate incremental prognostic utility. The consistency of the RC-relapse association was tested through subgroup evaluations partitioned by age at onset, sex, BMI, hypertension status, diabetes status, treatment of lipid-lowering drugs, clinical phenotype, serum anti-AQP4 antibody status and immunotherapy status. Patients were stratified into high-risk and low-risk groups based on the cutoff value of RC, and cumulative hazard function curves were plotted to illustrate the clinical significance of the inflection point in predicting relapse risk in the overall NMOSD population and in subgroups stratified by serum AQP4-IgG status and maintenance therapy. Potential immunological mechanisms were investigated through mediation analyses, which assessed the extent to which inflammatory markers mediated the influence of RC on NMOSD relapse. Peripheral white blood cell (WBC) count, neutrophil ratio, monocyte ratio, eosinophil ratio, basophil ratio, lymphocyte ratio, lymphocyte subset distribution, ESR, serum cytokines, complement components and immunoglobulin levels were assessed as mediators by using the causal steps approach and bootstrapping methods. All of the statistical tests used two-sided tests, with significance defined as P < 0.05.

Results

Baseline characteristics of NMOSD patients

By the time of the final follow-up in May 2024, a total of 383 NMOSD patients were included in this study. In accordance with the study design, 38 patients were initially excluded because of loss to follow-up. For the final analysis, 100 patients were excluded as serum lipid indicators, basic clinical information or treatment regimen data were not recorded. Consequently, the analysis incorporated 245 individuals diagnosed with NMOSD, all of whom had comprehensive and dependable clinical data available.

Table 1 presents the baseline demographic and clinical features of the enrolled patients. The study population demonstrated a mean age at onset of 41.93 ± 16.10 years, with female patients accounting for the majority of the patients (84.49%). During the observation period (with a median follow-up of 691 days), 135 patients (55.10%) experienced at least one clinical relapse, whereas the remaining 110 patients experienced no relapse events. The follow-up duration of nonrelapsing patients was significantly longer than that of relapsing patients (1120.13 ± 634.04 days vs. 591.76 ± 521.31 days, P = 0.048). The relapse group demonstrated a significantly younger age at disease onset compared with the non-relapse group (40.10 ± 16.39 years vs. 44.17 ± 15.59 years, P = 0.048), thus suggesting that younger patients may be more prone to disease relapse. With respect to maintenance treatment strategies, there were also statistically significant distinctions observed when comparing the two groups (P = 0.040). The nonrelapse group contained a greater percentage of patients who received monoclonal antibody therapy, whereas the absence of maintenance therapy was more prevalent in the relapse group. Among the 245 patients, 80 were diagnosed with NMOSD at the centre during their first disease manifestation, of whom 21 experienced a relapsing course during follow-up. However, there was no statistically significant difference detected in the ARR prior to enrolment between patients who experienced relapse and those who did not. Before enrolment, 59 patients had been receiving continuous maintenance therapy; additionally, following a definitive diagnosis at the centre, 11 of these patients had transitioned to monoclonal antibody-based maintenance therapy.

Table 1.

Baseline characteristics

Characteristic	All (245)	Non relapse (110)	Relapse (135)	P value
Follow-up duration, days	828.98 ± 631.02	1120.13 ± 634.04	591.76 ± 521.31	< 0.001
Age at onset, years	41.93 ± 16.10	44.17 ± 15.59	40.10 ± 16.39	0.048
Sex, n(%)				0.297
Male	38 (15.51)	20 (18.18)	18 (13.33)
Female	207 (84.49)	90 (81.82)	117 (86.67)
BMI, kg/m2	23.57 ± 3.83	23.10 ± 3.67	23.96 ± 3.93	0.079
Hypertension status, n(%)				0.080
No	230 (93.88)	100 (90.91)	130 (96.30)
Yes	15 (6.12)	10 (9.19)	5 (3.70)
Diabetes status, n(%)				0.510
No	234 (95.51)	104 (94.55)	130 (96.30)
Yes	11 (4.49)	6 (5.45)	5 (3.70)
Lipid-lowering therapy, n(%)				0.632
No	232 (94.69)	105 (95.45)	127 (94.07)
Yes	13 (5.31)	5 (4.55)	8 (5.93)
TC, mmol/L	4.69 ± 1.30	4.41 ± 1.04	4.91 ± 1.45	0.002
TG, mmol/L	1.33 ± 0.81	1.23 ± 0.67	1.42 ± 0.89	0.060
LDL-C, mmol/L	2.96 ± 0.91	2.78 ± 0.77	3.10 ± 0.99	0.005
HDL-C, mmol/L	1.22 ± 0.33	1.20 ± 0.27	1.25 ± 0.37	0.211
TG/HDL-C	1.18 ± 0.85	1.09 ± 0.65	1.25 ± 0.98	0.126
LDL-C/HDL-C	2.50 ± 0.75	2.39 ± 0.64	2.60 ± 0.82	0.026
NHHR	2.94 ± 0.98	2.78 ± 0.84	3.08 ± 1.06	0.014
RC, mmol/L	0.51 ± 0.34	0.44 ± 0.26	0.57 ± 0.38	0.002
ARR before enrolment	0.95 ± 1.98	0.81 ± 2.20	1.06 ± 1.77	0.340
Corticosteroid therapy within 1 year before enrolment, n (%)				0.304
No	205 (83.67%)	95 (86.36%)	110 (81.48%)
Yes	40 (16.23%)	15 (13.64%)	25 (18.52%)
Maintenance therapy before enrolment, n (%)				0.894
Immunosuppressant	39 (15.92)	18 (16.36)	21 (15.55)
Monoclonal antibody	20 (8.16)	8 (7.27)	12 (8.89)
No treatment	186 (75.92)	84 (76.36)	102 (75.56)
Clinical phenotype, n(%)				0.232
ON	47 (19.18)	21 (19.09)	26 (19.26)
Myelitis	137 (55.92)	66 (60.00)	71 (52.59)
Brain lesions	24 (9.80)	12 (10.91)	12 (8.89)
Mixed phenotype	37 (15.10)	11 (10.00)	26 (19.26)
Serum AQP4-IgG status, n(%)				0.709
Negative	86 (35.10)	40 (36.36)	46 (34.07)
Positive	159 (64.90)	70 (63.64)	89 (65.93)
Acute therapy, n (%)				0.221
High-dose corticosteroids	161 (65.71)	74 (67.27)	87 (64.44)
IVIg	8 (3.27)	4 (3.63)	4 (2.96)
High-dose corticosteroids + IVIg	34 (13.88)	10 (9.09)	24 (17.78)
No treatment	42 (17.14)	22 (20.00)	20 (14.81)
Maintenance therapy, n (%)				0.040
Immunosuppressant	80 (30.20)	40 (31.82)	40 (31.11)
Monoclonal antibody	89 (32.24)	45 (40.91)	44 (31.11)
No treatment	76 (37.55)	25 (27.27)	51 (37.78)

Immunosuppressants included mycophenolate mofetil, azathioprine and cyclophosphamide. Monoclonal antibodies included rituximab, satralizumab, inebilizumab and eculizumab

Values were expressed either as n (%) for categorical variables or Mean ± SD for continuous variables

BMI body mass index, TC total cholesterol, TG triglyceride, LDL-C low-density lipoprotein cholesterol, HDL-C high-density lipoprotein cholesterol, NHHR non-HDL-C to HDL-C ratio, RC remnant cholesterol, ARR annualized relapse rate, ON optic neuritis, AQP4 aquaporin 4, Ig immunoglobulin, IVIg intravenous immunoglobulin. Bold values indicate statistical significance (P < 0.05).

In terms of lipid metabolism-related factors, patients in the relapsed group exhibited multiple abnormalities. TC levels were significantly elevated in the relapse group relative to the nonrelapse group (4.91 ± 1.45 vs. 4.41 ± 1.04 mmol/L, P = 0.002), and the levels of LDL-C were also markedly greater (3.10 ± 0.99 vs. 2.78 ± 0.77 mmol/L, P = 0.005). Additionally, the TG levels tended to increase in the relapsed group (P = 0.060). Notably, compared with those in the nonrelapse group, the RC levels in the relapse group were significantly greater (0.57 ± 0.38 vs. 0.44 ± 0.26 mmol/L, P = 0.002). Lipid ratios were markedly greater among patients in the relapse group versus those in the nonrelapse group (NHHR: 3.08 ± 1.06 vs. 2.78 ± 0.84 mmol/L, P = 0.014; LDL-C/HDL-C: 2.60 ± 0.82 vs. 2.39 ± 0.64 mmol/L, P = 0.026). Other factors, such as sex, BMI, hypertension status, diabetes status, serum AQP4-IgG status, clinical phenotype and acute-phase treatment, showed no statistically significant variation between patients in the relapse cohort and those in the nonrelapse cohort (P > 0.05).

In addition, given the heterogeneity between NMOSD patients with AQP4-IgG seropositivity and seronegativity, the two subgroups may involve distinct pathogenic mechanisms. Comparisons of the clinical characteristics between seropositive and seronegative patients are provided in Table S1. A total of 159 seropositive and 86 seronegative patients were included. The two groups were comparable, showing no significant differences in follow-up duration, demographic characteristics, clinical features, treatment-related variables, and relapse rates (P > 0.05). Therefore, subgroup analysis based on serum AQP4-IgG status was not performed in the subsequent multivariate regression analysis.

Correlations between lipid ratios, RC and relapse risk in patients with NMOSD

A weighted multivariate Cox proportional hazards model was employed to evaluate the associations among RC, various lipid ratios and relapse risk in patients with NMOSD.

When the confounding variables outlined in Model 3 were taken into consideration, in the continuous variable analysis, the lipid ratios (TG/HDL-C, LDL-C/HDL-C, and NHHR) and RC were significantly associated with an increased risk of NMOSD relapse. The strongest predictor was RC (HR = 2.277, 95% confidence interval [CI]: 1.532–3.384, P < 0.001). Grouping lipid ratios and RC by quartiles as categorical variables suggested an increased risk of relapse from the first quartile (Q1) to Q4, with the risk for Q4 for the NHHR being 2.221 times greater than the risk of the Q1 group (95% CI: 1.292–3.819, P = 0.024). A similar trend was also observed for Q4 of RC compared with Q1 (HR = 2.581, 95% CI: 1.516–4.395, P < 0.001) (Table 2). In both the continuous and categorical models adjusted for all of the confounding variables in Model 3, TC (continuous model: HR = 1.212, 95% CI: 1.077–1.365, P = 0.001; categorical model: P for trend = 0.005) and LDL-C (continuous model: HR = 1.281, 95% CI: 1.062–1.545, P = 0.010; categorical model: P for trend = 0.011) emerged as independent risk factors for NMOSD relapse. Additionally, TG levels were independently associated with relapse risk in the continuous model (HR = 1.338, 95% CI: 1.062–1.545, P = 0.010) (Table S2).

Table 2.

Multivariate Cox regression analysis of lipid ratios and RC in predicting relapse risk of NMOSD

TG/HDL-C
	Continuous model	Categorical model				P for trend
		Q1 (< 0.66)	Q2 (0.66–0.97)	Q3 (0.97–1.36)	Q4 (≥ 1.36)
Model 1 HR (95% CI) P value	1.215 (1.002, 1.472) 0.047	1	1.112 (0.683, 1.811) 0.668	1.163 (0.715, 1.892) 0.543	1.117 (0.694, 1.799) 0.649	0.939
Model 2 HR (95% CI) P value	1.206 (0.986, 1.477) 0.069	1	1.100 (0.675, 1.793) 0.701	1.070 (0.645, 1.773) 0.794	1.040 (0.632, 1.712) 0.876	0.984
Model 3 HR (95% CI) P value	1.237 (1.014, 1.510) 0.036	1	1.121 (0.671, 1.873) 0.664	1.197 (0.710, 2.020) 0.500	1.171 (0.699, 1.964) 0.549	0.911

LDL-C/HDL-C
	Continuous model	Categorical model				P for trend
		Q1 (< 1.99)	Q2 (1.99–2.49)	Q3 (2.49–2.90)	Q4 (≥ 2.90)
Model 1 HR (95% CI) P value	1.209 (0.965, 1.515) 0.098	1	0.902 (0.544, 1.498) 0.691	0.978 (0.594, 1.610) 0.931	1.482 (0.926, 2.370) 0.101	0.135
Model 2 HR (95% CI) P value	1.194 (0.951, 1.500) 0.127	1	0.904 (0.540, 1.512) 0.700	0.969 (0.578, 1.623) 0.903	1.452 (0.893, 2.360) 0.132	0.168
Model 3 HR (95% CI) P value	1.286 (1.008, 1.642) 0.043	1	0.951 (0.549, 1.646) 0.857	1.079 (0.623, 1.871) 0.785	1.639 (0.972, 2.761) 0.064	0.124

NHHR
	Continuous model	Categorical model				P for trend
		Q1 (< 2.30)	Q2 (2.30–2.87)	Q3 (2.87–3.44)	Q4 (≥ 3.44)
Model 1 HR (95% CI) P value	1.201 (1.013, 1.425) 0.035	1	1.320 (0.794, 2.193) 0.284	1.196 (0.710, 2.017) 0.501	1.835 (1.124, 2.997) 0.015	0.081
Model 2 HR (95% CI) P value	1.193 (1.004, 1.418) 0.045	1	1.299 (0.766, 2.175) 0.320	1.157 (0.673, 1.991) 0.598	1.798 (1.086, 2.975) 0.022	0.096
Model 3 HR (95% CI) P value	1.287 (1.073, 1.543) 0.006	1	1.378 (0.796, 2.388) 0.252	1.344 (0.756, 2.387) 0.314	2.221 (1.292, 3.819) 0.004	0.024

RC
	Continuous model	Categorical model				P for trend
		Q1 (< 0.28)	Q2 (0.28–0.46)	Q3 (0.46–0.69)	Q4 (≥ 0.69)
Model 1 HR (95% CI) P value	1.793 (1.226, 2.621) 0.003	1	1.000 (0.587, 1.703) 0.999	1.593 (0.972, 2.610) 0.065	1.891 (1.175, 3.045) 0.009	0.014
Model 2 HR (95% CI) P value	1.764 (1.204, 2.585) 0.004	1	0.947 (0.550, 1.631) 0.845	1.521 (0.914, 2.529) 0.106	1.873 (1.153, 3.044) 0.011	0.014
Model 3 HR (95% CI) P value	2.277 (1.532, 3.384) < 0.001	1	1.050 (0.588, 1.875) 0.868	1.760 (1.014, 3.055) 0.045	2.581 (1.516, 4.395) < 0.001	0.001

Model 1: unadjusted variables

Model 2: adjusted for age at onset, sex, and BMI

Model 3: Model 2 + adjusted for clinical phenotype, acute and maintenance therapy, serum AQP4-IgG status, ARR before enrolment, corticosteroid therapy within 1 year before enrolment, maintenance therapy before enrolment, hypertension status, diabetes status and use of lipid-lowering medications

TG triglyceride, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, NHHR non-HDL-C to HDL-C ratio, RC remnant cholesterol, HR hazard ratio, CI confidence interval, Q quartile. Bold values indicate statistical significance (P < 0.05).

Nonlinear correlations between lipid ratios, RC and relapse in patients with NMOSD

Restricted cubic spline regression was applied to assess potential nonlinear correlations between RC, lipid ratios and NMOSD relapse. No nonlinear correlations were observed between TG/HDL-C, LDL-C/HDL-C, NHHR, TC, TG, LDL-C, HDL-C and NMOSD relapse (P for nonlinear > 0.05) (Fig. 1A-C, Figure S1). A significant “S”-shaped correlation was identified between RC and NMOSD relapse (P for nonlinear = 0.047), where the curve demonstrated an inflection at 0.46 mmol/L (Fig. 1D). Below this threshold, RC was negatively correlated with NMOSD relapse, whereas above the threshold, RC was a risk factor.

Fig. 1.

Nonlinear correlation analysis between lipid ratios, RC and relapse of NMOSD. A-D present nonlinear correlation analyses between NMOSD relapse and the following lipid indices: TG/HDL-C ratio, LDL-C/HDL-C ratio, NHHR, and RC, respectively. TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; NHHR, non-HDL-C to HDL-C ratio; RC, remnant cholesterol; CI, confidence interval

Comparison of RC and lipid ratios in predicting NMOSD relapse

The predictive performance of RC and lipid ratios for NMOSD relapse was evaluated by using the C-index, NRI, and IDI. As shown in Table S3, the baseline model (C-index = 0.632, 95% CI: 0.619–0.645) included multiple recognized factors associated with NMOSD relapse, including age at onset, sex, BMI, clinical phenotype, acute and maintenance therapy, serum AQP4-IgG status, ARR before enrolment, corticosteroid therapy within 1 year before enrolment, maintenance therapy before enrolment, hypertension status, diabetes status and use of lipid-lowering medications.

When RC was further incorporated into the model, significant improvements in predictive performance were observed; specifically, the C-index value increased to 0.652 (95% CI: 0.639–0.664, P = 0.030), and the NRI value was observed at 0.199 (95% CI: 0.033–0.354, P = 0.022), thus indicating a noteworthy enhancement in individual risk reclassification accuracy, whereas the IDI value was measured at 0.025 (95% CI: 0.002–0.061, P = 0.016), thereby demonstrating a marked increase in the model’s overall discriminative capacity. In contrast, the three other lipid ratio indicators (TC, TG, and LDL-C) showed no added predictive benefit to the basic model for NMOSD relapse (P > 0.05).

Subgroup evaluation of the correlation between RC levels and NMOSD relapse

To further elucidate the association between RC and NMOSD relapse risk, multiple subgroup analyses were conducted (Fig. 2). After stratification by age at onset, sex, BMI, hypertension status, diabetes status, lipid-lowering therapy, clinical phenotype, serum AQP4-IgG status, acute-phase therapy and maintenance therapy, subgroup analyses demonstrated that the association between RC and NMOSD relapse was stable across all strata, with no statistically significant interactions identified (P for interaction > 0.05).

Fig. 2.

Subgroup analysis of the associations between RC and NMOSD relapse. HR was used to evaluate the impact of RC on disease relapse risk across different subgroups. BMI, body mass index; ON, optic neuritis; AQP4, aquaporin 4; IVIg, intravenous immunoglobulin; HR, hazard ratio; CI, confidence interval; NA, not applicable

The relationship between the RC levels and disease relapse followed a nonlinear trend, with a cutoff value of 0.46 mmol/L being determined. Patients were stratified into two groups: those presenting RC levels ≥ 0.46 mmol/L and those presenting RC levels < 0.46 mmol/L. The clinical significance of this cutoff value in predicting NMOSD relapse risk was evaluated in the overall cohort, as well as in subgroups defined by serum AQP4-IgG status and maintenance therapy type, by using cumulative hazard function curves (Fig. 3). In the overall analysis, patients in the subgroup defined by RC concentrations ≥ 0.46 mmol/L experienced significantly earlier relapses in contrast to participants with RC levels < 0.46 mmol/L (P = 0.002). Subgroup analyses revealed that this association was evident only among patients who were serum AQP4-IgG positive (P = 0.002) and those receiving monoclonal antibody-based maintenance therapy (P = 0.005), with earlier relapse times being observed in the RC ≥ 0.46 mmol/L group. Relapse timing did not differ significantly across the remaining subgroups.

Fig. 3.

Cumulative hazard function curves for NMOSD relapse, contrasting patients with elevated RC (≥ 0.46 mmol/L) against those with lower RC (< 0.46 mmol/L). A-F represent the following groups, respectively: all patients, the AQP4-IgG seronegative subgroup, the AQP4-IgG seropositive subgroup, the immunosuppressant treatment subgroup, the monoclonal antibody treatment subgroup, and the subgroup without maintenance therapy. RC, remnant cholesterol; AQP4, aquaporin 4; Ig, immunoglobulin

Mediating variables in the effect of RC on NMOSD relapse

To preliminarily investigate the mechanism by which RC influences NMOSD relapse, a mediation analysis was performed to assess whether immune-related parameters (Table S4) serve as mediators. These parameters included the peripheral WBC count, neutrophil ratio, basophil ratio, eosinophil ratio, monocyte ratio, lymphocyte ratio, lymphocyte subset distribution, ESR, serum cytokines, complement components, and Ig levels. The analysis revealed that the neutrophil ratio (mediating effect: 12.17%, P = 0.036) and lymphocyte ratio (mediating effect: 15.09%, P = 0.032) partially mediated the relationship between RC and NMOSD relapse (Fig. 4).

Fig. 4.

Inflammatory indicators appear to mediate the relationship between RC and NMOSD relapse. A and B illustrate that the neutrophil ratio and lymphocyte ratio serve as mediating variables in the relationship between RC and NMOSD relapse, respectively. IE, the estimate of the indirect effect; DE, the estimate of the direct effect; RC, remnant cholesterol; NMOSD, neuromyelitis optica spectrum disorder; CI, confidence interval

Discussion

Based on available evidence, this is the first investigation into the predictive role of RC and lipid ratios (including the NHHR, TG/HDL-C ratio and LDL-C/HDL-C ratio) in relapse risk of patients with NMOSD. After adjustments were made for essential covariates, including age, BMI, sex, clinical phenotype, treatment regimen and serum anti-AQP4 antibody status, RC measured during the acute attack phase demonstrated superior predictive performance compared to both traditional lipid profiles and other novel lipid markers. Mediation analysis further suggested that an elevated neutrophil ratio and a reduced lymphocyte ratio in the peripheral blood may partially mediate the relationship between RC and NMOSD relapse.

Lipids are highly unevenly distributed in the CNS and are primarily enriched in the myelin sheath of white matter (which is rich in cholesterol, glycolipids and phospholipids), whereas grey matter predominantly contains phospholipids from synaptic membranes and neuronal cell membranes [19]. Their core functions include forming the structure of myelin sheaths to ensure efficient neural signal transmission; acting as signalling molecules to regulate neuroinflammation and survival; and participating in synaptic plasticity and energy storage. Lipids are essential for maintaining both the structural integrity and functional activity of the CNS, and dyslipidaemia is increasingly recognised as a contributor to the pathogenesis of multiple CNS diseases [20, 21].

Lipid metabolism is crucial for modulating inflammatory responses [22]. HDL-C, together with apolipoprotein A (apoA), demonstrate anti-inflammatory properties through regulation of immune cell activity, particularly affecting macrophages, monocytes, neutrophils, dendritic cells and lymphocytes [23]. In contrast, elevated levels of LDL-C, together with lower HDL-C and apoA concentrations, have been linked to increased production of proinflammatory cytokines, including TNF-α, IL-1β, IL-6, and transforming growth factor-β (TGF-β). These cytokines further influence antibody secretion, B-cell differentiation and T-cell activation, thus promoting immune system dysregulation [24–27].

Previous studies have suggested that lipid-related indicators (covering TC, TG, LDL-C and HDL-C) may be linked to relapse risk alongside clinical outcomes among patients with CNS autoimmune demyelinating disorders. Wu et al. reported that elevated TG levels were an independent predictor of unsatisfactory recovery outcome in patients with first-attacked NMOSD [10]. In addition, Ding’s study revealed that higher serum LDL-C concentrations demonstrated an independent and positive correlation with disease relapse in patients diagnosed with NMOSD [4]. Wang et al. demonstrated that decreased levels of apoA-I and HDL-C showed correlation with aggravated neurological dysfunction and an elevated probability of relapse among patients with NMOSD [28]. Dyslipidaemia can also be observed in MS patients, and the progression of disability is associated with high levels of LDL-C, TC and TG, whereas HDL-C serves an anti-inflammatory function within acute clinical attack [29–31]. In recent years, RC and lipid ratio indicators represented by the NHHR have emerged as new markers reflecting lipid metabolism [12]. Aberrations in RC and lipid ratios have been reported in autoimmune conditions exemplified by Guillain-Barré syndrome (GBS) and systemic lupus erythematosus [32, 33]. However, the potential relevance of RC and lipid ratio indices to the onset and prognosis of NMOSD remains unclear.

Herein, RC exhibited the strongest predictive value for NMOSD relapse when compared with other lipid parameters. RC is characterised as the cholesterol fraction contained in triglyceride-rich lipoproteins (TRLs), encompassing very low-density lipoprotein cholesterol (VLDL-C), celiac microparticles (CM) in the non-fasting state, and intermediate-density lipoprotein cholesterol (IDL-C) under the fasting condition. RC has increasingly been recognised as a potential factor in the pathogenesis of cardiovascular diseases, metabolic disorders, and some autoimmune diseases [13, 32, 34, 35]. The strong predictive value of RC for NMOSD relapse highlights the potential importance of cholesterol fractions (specifically including VLDL-C, IDL-C and CM) in modulating the immune system. RC elevation showed an independent relationship with heightened levels of multiple markers of systemic inflammation, namely high-sensitivity CRP (hs-CRP), gamma-glutamyl transferase (GGT) and the C-reactive protein-to-albumin ratio (CAR). Additionally, individuals with high RC levels often exhibit elevated peripheral blood counts of neutrophils, monocytes and platelets [36, 37]. Restricted cubic spline analysis further revealed an “S”-shaped nonlinear association between RC levels and relapse risk, with 0.46 mmol/L serving as the inflection point. The observed inflection point may reflect a biological threshold beyond which RC begins to exert proinflammatory or immunomodulatory effects that contribute to disease activity. Although the correlation between RC and NMOSD relapse risk persisted uniformly across the subgroups stratified by age, sex, and treatment regimen, further subgroup analyses based on relapse timing revealed that the identified RC inflection point significantly distinguished early relapse from late relapse only in patients who were AQP4-IgG seropositive or those receiving monoclonal antibody therapy as maintenance treatment after enrolment. These observations point to the importance of prioritising RC levels in these two patient populations. The monitoring of lipid profiles during acute attacks and the maintenance of RC levels at relatively low levels may have clinical relevance in reducing relapse risk. Moreover, the prognostic value of the RC inflection point warrants validation in larger, independent cohorts.

Over the past several years, researchers have devoted increasing attention to the clinical heterogeneity and pathogenic differences comparing AQP4-IgG seropositive and seronegative NMOSD groups [38, 39]. These differences were further examined by performing a comparative study of the clinical characteristics between seropositive and seronegative subgroups. Consistent with the findings of previous studies, AQP4-IgG seropositive patients in this cohort were more likely to be females and to use monoclonal antibody therapy, along with demonstrating relapsing disease courses following enrolment [38, 40]. In contrast, seronegative patients demonstrated a higher prevalence of brain lesions [39]. However, these differences did not exhibit statistical significance, which is potentially due to the limited sample size. Notably, the proportion of seronegative patients at the investigated centre is approximately 35%, which is higher than the 20% reported in previous studies [41]. Such inconsistency could be explained partially by the fact that the utilised centre is the largest referral facility for central nervous system demyelinating diseases in the province, where a greater number of seronegative patients with previously unclear diagnoses ultimately receive a confirmed diagnosis. In addition, the identified RC cutoff point was effective in distinguishing early-relapse patients from late-relapse patients only within the AQP4-IgG seropositive subgroup, whereas no such distinction was observed among seronegative individuals. When considering the differing pathogenic mechanisms between these subgroups, it is postulated that RC predominantly contributes to NMOSD relapse via the promotion of humoral immune responses rather than through cellular immunity or alternative nonimmune pathways [39, 42].

To examine how RC contributes to relapse risk in NMOSD, a mediation analysis was performed, which revealed that elevated RC levels may partially contribute to NMOSD relapse by upregulating peripheral blood neutrophil levels and decreasing the lymphocyte ratio. Similar associations have been reported in atherosclerotic cardiovascular disease, diabetes, and other inflammatory conditions, wherein RC is positively correlated with inflammatory cellular indices, specifically the neutrophil count and neutrophil-to-lymphocyte ratio (NLR). These findings suggest that elevated RC may play a role in promoting systemic inflammatory responses [36, 43–45]. Multiple studies have shown that an elevated NLR in the peripheral blood functions as an indicator of clinical activity in NMOSD and demonstrates a significant association with heightened relapse risks [46–48]. Consistent with these reports, the present findings suggest that the increased number of neutrophils and the decreased number of lymphocytes partially mediate the effect of elevated RC levels on NMOSD relapse. A neuropathological study revealed that neutrophils are activated and accumulate at lesion sites in the initial stage of disease activity among NMOSD patients, with the degree of activation being positively correlated with lesion volume [49]. Neutrophil activation within the CNS and cerebrospinal fluid is fundamentally involved in the pathogenic mechanisms underlying NMOSD-related neurological damage [50, 51]. An elevated neutrophil proportion in peripheral blood generally reflects an acute inflammatory response. Although the blood-brain barrier (BBB) limits the direct infiltration of peripheral immunocytes into the CNS, investigations have revealed that neutrophils in NMOSD patients are peripherally activated and release extracellular free deoxyribonucleic acid (DNA). This process promotes the activation of the IFN-1 signalling pathway and is associated with greater neurological impairment and severe disability in patients with NMOSD [52, 53]. Lymphocytes are pivotal in immune regulation throughout the course of NMOSD. In the acute clinical stage, heightened systemic inflammation can induce lymphocyte apoptosis, thereby contributing to immune imbalance and disease exacerbation [54]. As patients progress to the recovery phase, lymphocyte levels gradually increase, thus reflecting the partial restoration of immune homeostasis and resolution of inflammatory activity [47]. In conclusion, during disease exacerbation in NMOSD patients, RC appears to function as a surrogate indicator reflecting systemic inflammation, thereby contributing to an elevated risk of relapse. More importantly, dyslipidaemia may serve as a potential intervention target to reduce NMOSD relapse. Although neutrophils and lymphocytes mediate an aspect of the effect of RC on NMOSD relapse, additional mechanisms may be involved. RC levels may also directly contribute to disease relapse by impairing myelin regeneration or altering BBB permeability [7]. These potential pathways suggest a broader role for RC in CNS pathology, although further experimental research is needed to validate these hypotheses.

Study strengths and limitations

This study comprehensively evaluated the value of RC and lipid ratios in risk prediction for NMOSD relapse. Advanced statistical methods ensured methodological rigour and enabled assessment of both linear and nonlinear associations. Subgroup and mediation analyses confirmed consistent links between RC and NMOSD relapse and provided mechanistic plausibility, which bridged clinical and biological evidence. Several limitations should be acknowledged in this study. Given the retrospective, single-centre design, the results may be prone to selection and information biases. The relatively restricted number of participants limited statistical power, especially in subgroup analyses. Despite controlling for several potential confounders, unmeasured factors including diet, physical activity and genetic background may still exist. Blood samples were not uniformly collected during the first attack or before treatment initiation; many patients were already receiving immunotherapy, which could have influenced lipid levels. Data on Expanded Disability Status Scale (EDSS) scores and detailed anti-AQP4 antibody titers were also incomplete, limiting assessment of disease severity and immune status. Inconsistent treatment regimens and the lack of an external validation cohort further restrict the external validity of the results.

Conclusions

In brief, the present study demonstrates that dyslipidaemia (particularly elevated RC concentrations) is linked to a higher likelihood of relapse among individuals with NMOSD. RC outperformed traditional lipid parameters and ratios as a predictor of NMOSD relapse, and its relationship with relapse risk is nonlinear, with an inflection point at 0.46 mmol/L being observed. The incorporation of RC levels into existing clinical prediction models significantly improves their ability to estimate the probability of NMOSD relapse. These observations stress the necessity of considering lipid metabolism (particularly RC) as a new dimension in the disease management and risk assessment of NMOSD patients. By advancing the understanding of lipid-mediated mechanisms in NMOSD, this research supports the development of accessible and individualised approaches to relapse prevention, thereby ultimately contributing to improved neurological health and reduced long-term disability in diverse patient populations. These results underscore the need for additional studies to validate the results and to examine possible therapeutic interventions aimed at lipid modulation.

Authors’ contributions

ZW: the conception, interpretation of data, data analysis, drafted the work; TH: the data acquisition, data analysis, drafted the work; BC, JC and XP: the data acquisition; MZ: review & editing the work; JG: design of the work, review & editing the work; HM: the conception, design of the work, funding acquisition, writing-review & editing the work.

Funding

Financial support was provided to the authors for the purposes of conducting the research, preparing the manuscript, and facilitating its publication. Sanjin Elite Youth Top notch Talent Project fund (Grant No. SJYC2024439).

Data availability

The datasets generated and analysed during the current study are available from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

Ethical clearance for all procedures involving human subjects was granted by the ethics committee of the First Hospital of Shanxi Medical University, Shanxi, China (KYLL-2024-266). Participation in this study required written informed consent, which was secured either directly from the patients or, when necessary, from their legal guardians or next of kin.

Competing interests

The authors declare no competing interests.