Optic neuritis at the nexus of peripheral autoimmunity and central nervous system demyelination: a nationwide cohort study

Long-term course of optic neuritis is heterogeneous and varies across populations. We aim to investigate immune-related determinants that predict conversion of optic neuritis (ON) to multiple sclerosis (MS) or neuromyelitis optica (NMO) in a nationwide cohort. Methods We conducted the population-based cohort study using data from Taiwan’s National Health Insurance Research Database. Incident ON cases during 2003~2014 were followed until the end of 2015. Pediatric and adult sub-cohorts were examined separately. The associations between immune-related comorbidities or treatment and outcomes were analyzed using Cox proportional hazards models. A total of 11923 adult and 1365 pediatric ON patients were enrolled. The rates of conversion to MS were 2.7% for adult and 3.1% for pediatric ON with median follow-up duration of 6.3 and 7.3 years respectively, while 1.2% of pediatric and adult ON evolved to NMO. Comorbid systemic lupus erythematosus was associated with increased risks of subsequent development of MS in adult (adjusted hazard ratio aHR, 2.80; 95% CI, 1.04-7.49) and pediatric ON patients (aHR, 21.65; 95% CI, 1.29-363.4). Adult ON patients were at increased risks of NMO if comorbid with myasthenia gravis (aHR, 9.13; 95% CI, 1.20-69.45) or Sjogren’s syndrome (aHR, 4.71; 95% CI, 1.74-12.76). ON could be the sentinel event linking several peripheral autoimmune comorbidities to distinct forms of central nervous system demyelination. The clinical context in which ON occurs should be taken into account in the care and counseling of these patients.


Introduction
Optic neuritis (ON) often represents the harbinger of several neuroimmune disorders of central nervous system, most notably multiple sclerosis (MS) and neuromyelitis optica (NMO) [1,2]. Previous studies reported variable rates of conversion to MS after ON, ranging from 8.3-75% (7-40.6% in pediatric ON) [1,[3][4][5][6][7][8]. The disparity between studies could be attributed to case definition, ethnic 3 background, source of patient population (community versus hospital-based), follow-up duration, and secular trend. In a landmark study of the natural history of ON spanning more than 50 years, the rate of conversion to MS proved to be a function of time and remained increasing beyond 30 years after ON [1]. Nevertheless, a significant fraction of ON patients remain event free for prolonged periods, underscoring the clinical heterogeneity of this condition. On the other hand, the interval between initial and second relapses in patients with NMO is generally shorter compared to that in MS.
Therefore, it is anticipated that when ON signifies the initial event of NMO, the next clinical neuroimmune attack will happen sooner. However, this issue has seldom been specifically examined [9]. Furthermore, a common caveat in previous studies is the mixing-up of MS and NMO, in which NMO was viewed as a subset of MS [7,10]. With the recognition that NMO and MS are clinically and immunologically distinct entities, it is prudent to study their relationships with ON respectively.
The standard treatment for ON, intravenous methylprednisolone, has been established on the basis of the Optic Neuritis Treatment Trial, which showed that high-dose methylprednisolone treatment was associated with accelerated visual recovery and delayed onset of MS [11,12]. Steroid treatment is believed to improve visual outcome in NMO-associated ON [13][14][15], whereas the relationship of steroid treatment and subsequent development of NMO remains to be clarified. Interferon beta agents have been used in selected ON patients to delay the progression to clinically definite MS [16], yet they may be potentially detrimental for NMO [17]. The differential treatment responses suggest that the immunological mechanisms of ON that later evolves to NMO could be distinct from that of MS-associated ON. Therefore, differentiation between various subtypes of ON are therapeutically relevant [14].
With these questions in mind, we conduct this nationwide population-based cohort study to better understand the long-term course of ON in Taiwan and the associated risk factors in relation to MS and NMO.

Data Source
This retrospective study used data retrieved from the Taiwan National Health Insurance (NHI)  [13,15]. Therefore, the pediatric (0-19 years) and adult (20 years and above) subcohorts were examined separately.

Exposures
Comorbidities, with an emphasis on immune-mediated diseases, were identified using respective ICD-9-CM codes (Supplementary Table 1, Additional File 1). Treatment with systemic steroid was identified using Anatomical Therapeutic Chemical (ATC) codes (H02AB and H02B), and it was stratified to examine whether there were dose-response relationships between systemic steroid use after ON and subsequent risk of MS or NMO. To ensure the robustness of the results, the stratification was implemented in three different ways: (1) cumulative defined daily dose (DDD), stratified into quartile; (2) duration of systemic steroid use, stratified into 1-6 days, 7-14 days, and 15 days and above; (3) methylprednisolone use or not. Subjects not treated with systemic steroid after ON were used as the reference.
Study Outcomes and Follow-up 5 The study outcomes were MS and NMO, respectively. The diagnosis of MS and NMO were considered to be valid if they were reported in at least 3 outpatient visits within 2 years or once during hospital admission with the respective diagnostic codes. When a patient met both the criteria for MS and NMO, the diagnosis later made and preserved was adopted; otherwise NMO was assigned if both diagnoses were kept. All patients were monitored from the index date until the earliest occurrence of MS or NMO (defined as the event date), death, withdrawal from NHI, or the end of the study (December 31, 2015), whichever came first.

Statistical Analysis
The cumulative incidences for various clinical outcomes by study groups among the overall population were plotted using Fine and Gray's subdistribution method to estimate cumulative incidence function [18]. The Cox proportional-hazards regression model was used to investigate if comorbidities and treatment could be associated with subsequent development of MS or NMO, with adjustment for other potential confounders. A substantial fraction of subsequent MS is expected to be diagnosed within three months after the index date (i.e., the initial date of ON diagnosis) [19]. These subjects were excluded in Cox regression models in order to focus on patients presumably having clinically isolated ON, and to evaluate the longer-term predictive utility of demographic and clinical variables. The final model included age, sex, timing of diagnosis, comorbidities, systemic steroid use at baseline (defined as prescription of systemic steroid in the 3 months preceding the index date), and systemic steroid use after ON (defined as prescription of systemic steroid within one month after the index date). The relative risk of conversion to MS or NMO was expressed as a hazard ratio (HR) and 95% confidence interval (CI). Statistical tests were two-sided, and a P value of less than 0.05 was considered statistically significant. Statistical analyses were performed using SAS 9.4 (SAS Institute, Inc., Cary, North Carolina).

Data Availability
This study was based on datasets from Taiwan's NHI Research Database. Taiwan's Ministry of Health and Welfare owns the database and reviews application for data use for research purposes. All aggregate data and summary statistics in this study are presented in the article or uploaded as  Table 1.
The follow-up data were presented in Table 2 Table 2, Additional File 1), attesting to the robustness of this finding. No significant interaction between systemic steroid use at baseline and after ON was found with regard to subsequent development of MS (see footnotes of Table 3 and Supplementary Table 2 in Additional File 1). Therefore, this interaction term was not incorporated into the final model.
Regarding conversion of adult ON to NMO, similar analysis showed that male sex was associated with a much decreased risk (aHR, 0.16; 95% CI, 0.08-0.32) (  [1,19], while the rate of conversion to MS was much lower compared with that reported from Caucasian populations. Together these findings were consistent with the low incidence rate of MS in Taiwan [20]. Although Sjogren syndrome and MG were uncommon (< 2% each) in ON patients in our study, they were significantly associated with subsequent development of NMO in adult patients with ON. Indeed, the association of Sjogren syndrome and NMO has been established [21,22]. Coexistence of MG and NMO was also recognized to be more than chance, and MG usually preceded NMO [23]. Our study corroborates these observations by providing population-level evidence of the associations. To date, cancer was the only comorbidity formally listed as one of the "red flag" features in patients with ON [14,24]. Together with prior knowledge, our findings suggest that comorbid Sjogren syndrome or MG are also worth consideration as red flags that prompt further investigations such as aquaporin-4 serostatus in these patients. On the other hand, although SLE is also a common comorbidity of NMO, we found that it is associated with an increased risk of MS, rather than NMO, in both pediatric and adult ON. Another study in Taiwan revealed that female first-degree relatives of SLE patients are at Despite that high-dose methylprednisolone has been established as the standard treatment for ON, variations in real-world practice continue to be noted [33]. This also appeared to be the case in our population, in which more than 20% of ON patients were given only corticosteroids other than methylprednisolone (Table 1). Our pharmacoepidemiological analysis showed that systemic steroid use after ON was associated with an increased risk of conversion to MS and NMO in a dose-dependent manner in adult population. At first sight this suggests that systemic steroid treatment for ON is detrimental in the long term, yet this interpretation incurs the risk of reverse causation, and it apparently contradicts existing knowledge [12]. The more plausible explanation is that prescription of systemic steroid reflects the judgment of clinicians about the immune-mediated nature of ON in the particular patient, presumably based on clinical or paraclinical features. The association of immunomodulatory medication with subsequent MS was also observed in other retrospective studies of pediatric and adult ON [4,7], which could be similarly explicated.
It is intriguing to note that use of systemic steroid during the 3 months preceding ON was associated with a modestly decreased risk of MS in adults. This phenomenon has not been reported, and the reason is open to speculation. One possibility is that ON developing in the immunological milieu sculpted by corticosteroid has a different pathogenesis, or MS-associated ON might have been more readily ameliorated by the fortuitous use of systemic steroid. More research is needed to clarify this issue.
Previous studies showed that pediatric ON has been distinguished from ON in adults in terms of sex ratio, laterality, associated diseases and rate of conversion to MS [13,15]. We found that female patients in this population was associated with much increased risks of MS and NMO. Besides, comorbid SLE and higher cumulative dosage of systemic steroid use after ON were independently associated with increased risks of conversion of pediatric ON to MS. All of these findings were qualitatively similar to that in adults. In contrary to general notion, the rate of conversion to MS or NMO in pediatric ON was not lower than that in adults in our study. This could be partly accounted for by the age cutoff (20 years) used here, since adolescent ON may be more akin to adult as opposed to prepubertal ON [34], while they were included in the pediatric group in our study. On the other hand, if the prepubertal ON is specifically examined, analytical results would be less robust given the much smaller sample size.
The strength of this study lies in its national representativeness and unprecedented size of the cohort, making statistical modeling feasible. Although the findings are not directly generalizable to other populations, the clinical relevance herein deserves further research, as discussed above. Several limitations of this study should be considered. First, the diagnosis of individual case cannot be ascertained given the nature of datasets and the policy of the database provider. Second, although the temporal scale (spanning 15 years, median follow-up duration 6-7 years) is acceptable, it is insufficient to capture all cases of MS because some patients could run a more indolent course [1].
Third, certain features of ON, such as laterality and ophthalmologic findings, were associated with differential risks of MS or NMO [24,35], and these features could affect clinicians' decision regarding steroid use. In other words, they were the potential sources of confounding-by-indication. However, these information were not available for this study. Similarly, relevant serologic data and neuroimaging findings were also unavailable. Therefore, caution should be exercised in the interpretation of treatment effects with regard to subsequent risks of MS or NMO. Fourth, recurrent ON is associated with an increased likelihood of developing MS or NMO in children and adults [1,31,35,36]. However, we were unable to address this issue because previous diagnosis code was often retained in subsequent visits in our electronic health information systems, making differentiation between inactive and relapsing ON difficult. Fifth, the awareness of NMO as a distinct clinical entity was relatively recent, and NMO could be underdiagnosed in the earlier years of our study period. Therefore, the association of timing of ON diagnosis with subsequent development of NMO in adults (        Median (Q1,Q3) 14 (6,29) Abbreviations: MS, multiple sclerosis; NMO, neuromyelitis optica. Cumulative incidence rate of NMO and MS. (A) age 0-19 (B) age 20 and above.

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.
Additional file 1.pdf