Article Text


Research paper
Short delay to initiate plasma exchange is the strongest predictor of outcome in severe attacks of NMO spectrum disorders
  1. Mickael Bonnan1,
  2. Rudy Valentino2,
  3. Stéphane Debeugny3,
  4. Harold Merle4,
  5. Jean-Louis Fergé2,
  6. Hossein Mehdaoui2,
  7. Philippe Cabre5
  1. 1 Service de neurologie, Centre Hospitalier de Pau, Pau, France
  2. 2 Réanimation médicale, Hopital Pierre Zobda-Quitman, Fort-de-France, Martinique
  3. 3 Unité de Recherche Clinique, Centre Hospitalier de Pau, Pau, Aquitaine-Limousin-Poitou, France
  4. 4 Service d’ophtalmologie, Hopital Pierre Zobda-Quitman, Fort-de-France, Martinique
  5. 5 Service de Neurologie, Hopital Pierre Zobda-Quitman, Fort-de-France, Martinique
  1. Correspondence to Dr Mickael Bonnan, Service de Neurologie, Centre Hospitalier de Pau, 4 Bd Hauterive, 64000 Pau, France; mickael.bonnan{at}


Introduction Severe attacks of neuromyelitis optica spectrum disorder (NMO-SD) are improved by plasma exchange (PLEX) given as an adjunctive therapy. Initial studies failed to demonstrate a delay of PLEX treatment influenced clinical outcome; however PLEX was always used late. We examine the clinical consequences of delay in PLEX initiation on severe optic neuritis and spinal cord attacks in NMO-SD.

Methods All of our patients who suffered attacks of NMO-SD, treated in our centre by PLEX, were retrospectively considered for inclusion. Primary outcome was defined as complete improvement. Secondary poor/good outcomes were respectively defined to be the higher/lower third of Delta-Expanded Disability Status Scale (EDSS) (late minus baseline EDSS). Delays from clinical onset to PLEX initiation were categorised for multivariate analysis.

Results Of the 60 patients included, NMO-SD criteria (2015) were fulfilled in 92%. One hundred and fifteen attacks were included and received PLEX with a median of 7 days (0–54) after clinical onset. The probability to regain complete improvement continuously decreased from 50% for PLEX given at day 0 to 1%–5% after day 20. Through multivariate analysis, the baseline impairment and PLEX delay were associated with the probability to complete improvement (OR 5.3; 95% CI 1.8 to 15.9). Reducing the PLEX delay also influenced the good secondary outcome but not the poor secondary outcome.

Conclusions These results confirm an improved clinical benefit of early initiation of PLEX during severe attacks of NMO-SD. Perceiving PLEX as a rescue therapy only after steroid failure could be deleterious.

  • Neuromyelitis optica
  • plasma exchange
  • transverse myelitis
  • optic neuritis

Statistics from


Neuromyelitis optica spectrum disorder (NMO-SD) is a heterogeneous autoimmune disorder characterised by the inflammation and demyelination of the optic nerve and spinal cord (SC), leading to devastating outcomes. The increasing burden of impairment is driven by fixed sequelae following relapses. NMO-SD attacks are medical emergencies, and ventilatory failure may occur during severe cervical attacks sometimes resulting in death. The mainstay therapy for acute relapses is high dose steroids that target the cellular part of inflammation; which is just one aspect of NMO-SD pathophysiology. This gold standard treatment is clearly insufficient considering the high probability to accumulate residual impairment.1 2 Plasma exchange (PLEX), which targets specific antibodies, complement and several proinflammatory proteins,3 appears very promising; however, PLEX is not readily available in many centres and can be costly.4 Consequently, PLEX is often offered as a second line ‘rescue therapy’ for severe relapses resistant to steroids.5 6 Moreover, since the first study to demonstrate PLEX efficiency in severe attacks only included patients who were refractory to treatment for a minimum of 2 weeks,7 a delay of weeks is often recommended before initiating PLEX therapy.

Many studies failed to demonstrate an impact of delaying PLEX on outcomes,7–11 with three studies confirming a positive influence of shorter delays on patient outcomes12–14; however, these studies were biased by inclusion of a small number of patients with NMO-SD among other various demyelinating disorders. As such, the delay to initiate PLEX remains a thorny question, and designing a study to compare early to late PLEX initiation in severe relapses would now be unethical.

In our centre, since 2002, we have always offered PLEX to patients suffering from severe NMO-SD relapses as fast as we can.

We took advantage from the high variability of delay from relapse onset to admission to our centre to evaluate outcomes based on when PLEX was initiated in a large cohort of optic and spinal NMO-SD relapses.


Patient sample

We included all the patients consecutively followed up in our ward or referred from 2002 to July 2015 and suffering from monophasic or relapsing NMO-SD, extensive transverse myelitis or severe optic neuritis (ON) highly suggestive of NMO-SD. The diagnosis of NMO-SD was based on the latest criteria.15

Data collection

Medical records were retrospectively reviewed, and data were collected for demographic details (place of residence, sex, age at relapse and preceding and final number of SC/ON relapses), anti-aquaporin-4 (AQP4) status, final clinical diagnosis, delay of treatment initiation (steroids and PLEX), Expanded Disability Status Scale (EDSS) before, during nadir and late follow-up (≥6 months) and procedural adverse events.

Anti-AQP4 status

Sera were tested for anti-AQP4 and antimyelin oligodendrocyte glycoprotein (MOG) at the INSERM U842 laboratory of Lyon, France, by a cell-based assay on live cells transfected respectively with AQP4 and MOG.16

Inclusion and exclusion criteria

All data from relapses treated by PLEX were thoroughly examined and only included if the following criteria were met: detailed clinical status data before attack, during clinical nadir and during a ≥6-month follow-up; EDSS/visual acuity (VA) availability of delay from onset to PLEX. All data were excluded if attacks occurred in a subintrant pattern (<6 months) or with incomplete clinical data. Non-severe attacks, defined by acute EDSS <4.0 for SC attacks or VA >20/200, were also excluded. Since these clinical data are prospectively recorded, they were available in all cases.

Clinical assessment

Clinical status in SC attacks and ON were based on EDSS and VA, respectively. Baseline EDSS and VA before the first attack were assumed to be normal. Data were directly available from clinical records. EDSS was reported by a senior neurologist (PC). VA was assessed by an ophthalmologist (HM) using a Snellen scale. Low VA was descriptive (counting fingers, hand motion, light perception and no light perception). We categorised VA, which is a continuous parameter, into a line scale of nine steps extending from 0 (normal VA) to 8 (no light perception) (online supplementary table 1). Based on previous works, low VA levels were converted to decimal acuity for statistical convenience.17 18

Supplementary file 1

Outcome assessment

Late clinical scores (≥6 months) were extracted from medical records. Late cumulative impairment was defined by: Δclinical score=(late clinical score)−(baseline clinical score), with clinical score being either EDSS or VA linear scale. Acknowledging the difference between EDSS and VA line scale, combining the visual and SC data remains problematic and suitable outcome criteria designed for both clinical scales are limited. We therefore considered two outcomes. The primary outcome is the proportion of patients improving to the baseline states during the follow-up (‘complete improvement’), which is the more stringent, unbiased outcome for both clinical types, whatever the baseline impairment (figure 1). We further calculated the proportion of clinical score improvement: (acute−late)/(acute−baseline), which provides the advantage of an interscore comparability. Our secondary outcome was the ‘good response’ and ‘poor response’ outcome assigned to the extreme third ends for late assessments, respectively, for 66%–100% and 0%–33% improvement, and captured 64 and 34 of attacks, respectively, among the cohort.

Figure 1

Clinical outcome criteria. These criteria are applied to both clinical scales (EDSS and visual acuity line scale). Left panel. Quantitative outcome criterion is based on the difference between late and basal levels. The main bias of this outcome relates to the non-linearity of clinical scale. Although this outcome is major in the null basal impairment groups, a ‘ceiling’ effect is observed for high basal level of impairment. Consequently, higher basal levels of impairment are mechanically associated with lower delta clinical scores implying better outcome. Therefore, this criterion is not suitable to compare groups with different basal impairment or using different clinical scales. Right panel. Two objective binary outcome criteria were designed. The main outcome is a complete late improvement, which is the primary goal of the treatment and a more stringent criterion. Moreover, this criterion gives the opportunity to collate results from different clinical scales (optic and spinal). As a secondary outcome, the ‘good outcome’ criterion was defined to be a better improvement than two-third of the symptoms measured either in EDSS or optic scale. EDSS, Expanded Disability Status Scale.

We here preferred to pragmatically use the objective outcome ‘good response’ as previously defined during the earliest clinical assessment available, usually in the month following PLEX.

Stratification of PLEX delays

PLEX delay data were available in days following clinical onset. Stratification of primary and secondary outcomes was therefore focused on the very early phase: 1 day after clinical onset (day 0–1); up to 5 days (days 2–5); up to 10 days (days 6–10); days 11–20; and up to 20 days and later. In order to obtain at least six attacks by strata, the subgroups of null and minimally impaired patients at baseline were stratified using larger delay intervals.



Intravenous methylprednisolone was given in most relapses, with a very short delay, predominately the day of patient admission. We calculated the cumulated steroid dosage whatever the schedules used (typically over 3–10 days).

Plasma exchange

PLEXs are offered systematically in all severe NMO-SD attacks since 2002. PLEX is not undertaken as a rescue therapy after steroid failure: decision to offer PLEX is raised at admission based on clinical severity, and the procedure is initiated as soon as possible (typically the next day). PLEX procedures were performed in an intensive care ward daily for 5 days with biological monitoring until the central line was removed. PLEX was performed on a continuous flow filtration device (Prismaflex) with a central line; one volume of plasma was exchanged against 5% human albumin solution. Delay on clinical onset to procedure initiation and possible adverse events were recorded.

Statistical analysis

Data are n (%), mean (SD) or median (IQR). We used the Kruskal-Wallis two-sided test for comparisons between clinical characteristics or scores. Binary values of outcome were compared by the Cochran-Armitage one-sided test for trend, among the five levels of PLEX delays.

We performed an explanatory analysis using univariate and multivariate logistic regression approach. We considered p values of <0.20 at univariate analysis as candidate for multivariate modelling. Collinearity and interaction between variables were checked. A backward stepwise multivariate logistic regression was used for modelling nominal outcome variables selected during the previous step. Data are reported with ORs and 95% CIs. Goodness of fit, for the different models, was evaluated by the Pearson test. All statistical analyses were performed using SAS V.9.4. p Values of ≤0.05 were considered significant.


Patient characteristics

A total of 60 patients were included in the study (39 in the myelitis group overlapping with 32 in the ON group). Baseline characteristics are described in (table 1).

Table 1

Population characteristics

The majority of patients were women (sex ratio: 7.5:1), and the mean age was 39 years (14–85). Almost all patients had Afro-Caribbean ancestry (98%). Geographic origin was French Caribbean Islands Martinique in 35 patients (58%). NMO-SD criteria (2015) were fulfilled in 55 patients (92%), including 40 seropositive (anti-AQP4) patients (67%); unfortunately, one typical NMO patient died before anti-AQP4 testing. Seronegative patients included: 14 (23%) satisfying the core clinical characteristics for NMO-SD, 2 (3%) with recurrent longitudinal extensive myelitis and 3 (5%) with recurrent bilateral severe ON. Other autoimmune diseases were associated in 17%. Various immunosuppressive drugs were used at the time of PLEX. The mean duration of disease was 5.7 (6.6) years. Among the 60 patients, 115 attacks met the inclusion criteria for this study; of these, a quarter were the onset event and 110 (96%) were in confirmed patients with NMO-SD. Nine of the included attacks were optico-spinal, but inclusion criteria for optic and spinal attacks were always considered separately.

Attack characteristics

PLEX procedures were completed in 96% of attacks. Median delay prior to PLEX was 7 days (IQR 10), slightly shorter in optic attacks (p=0.08). Delay of PLEX after clinical onset was very short in our cohort since 63% (n=73) and 34% (n=39) received treatment in 10 days and 5 days, respectively, or less after clinical onset. Complications of PLEX occurred in 8.6%: sepsis, local problems (postpuncture aneurysm, arteriovenous shunt, haematoma and thrombosis) and severe bradycardia.

Baseline, acute and follow-up clinical data of SC attacks and ON are summarised in table 2. Previous relapses were more frequent in spinal than in optical attacks (p<0.001), but history of attack treatment was similar for the SC and ON attacks: age at attack, time from clinical onset to hospital, delay from onset to steroid and median doses, delay from onset to PLEX, number of procedure and delay of follow-up.

Table 2

Clinical outcome after PLEX in myelitis and ON

Clinical outcome depending on PLEX delays

The probability to regain the main clinical outcome (complete improvement) continuously decreased as the delay in receiving PLEX increased, from 50% at day 0–1 to approximately 5% improvement after day 20 (p=0.02) (figure 2).

Figure 2

Clinical outcome following attack (SC+ON) is dependent on delay initiating PLEX. Left: probability to regain a late clinical score equal to pretreatment baseline (‘no impairment’). The probability of regaining pretreatment baseline score is greatest in the first 2 days and declines as PLEX initiation is delayed (p=0.02). Right: trend to regain the highest (p<0.01) or the lowest (p=0.02) strata of improvement over delay to PLEX. ON, optic neuritis; PLEX, plasma exchange; SC, spinal cord.

The trend to regain highest third strata of improvement significantly decreased along with PLEX delay from approximately 70% at day 0–1% to 30% at day ≥20 (p<0.01). Conversely, the trend to maintain the lowest strata of improvement increased from 20% to 47% with an increased delay to PLEX (p=0.02).

Similar results were obtained after exclusion of patients not treated by steroids (data not shown).

Predictive factors of PLEX outcome

We performed an exploratory analysis of 13 potentially valuable independent predictors of outcome. In univariate analysis, complete improvement was predicted by PLEX delay (p=0.01), age of onset (p<0.01), baseline impairment (p=0.05), whereas areflexia (p=0.16), sex (p=0.16) and admission delay (p=0.14) demonstrated a moderate influence but were selected for multivariate modelling. Type of attack, residence, disease duration, associated autoimmune diseases, number of previous relapses and steroid use, delay of steroid initiation or cumulative dose were not associated with the primary outcome. After adjustment by multivariable logistic regression, complete improvement was associated to baseline impairment (p=0.05) and PLEX delay (p=0.01) (table 3). The shorter strata of PLEX delay (days 0–5) demonstrated a higher probability to regain a complete improvement than the longer strata (≥day 11) with OR 5.3 (1.8–15.9, p<0.01), whereas the probability was not significantly different between the middle (days 6–10) and longer strata (p=0.12).

Table 3

Adjusted logistic regression models of parameters associated with PLEX outcomes

In univariate analysis for secondary outcomes, highest third improvement was influenced by PLEX delay (p=0.01), optic attacks (p=0.04), residence (p<0.01), age of onset (p=0.06) and admission delay (p=0.08), whereas the lower third improvement was only affected by residence (p=0.04). In multivariate modelling, the probability to obtain the highest third improvement tended to be influenced by PLEX delay (pglobal=0.10), but the difference was only significant for the shorter PLEX delay with OR 2.8 (1.1 to 7.3, p=0.04), whereas no significant difference was demonstrated between the middle and longest strata of delay (table 3). Foreign residence was associated with a lower probability to reach this outcome with OR 0.4 (0.2 to 0.9, p=0.03), but this variable strongly correlated with PLEX delay (p=0.01).

The probability to reach the lowest third improvement was associated with a foreign residence, with OR 2.3 (1.0 to 5.2, p=0.04). PLEX delay was not associated with this poor secondary outcome (pglobal=0.29), and shorter delay was only associated with a trend to avoid this outcome (p=0.12).

Similar results were obtained if the models were run after exclusion of the relapses occurring in patients not fulfilling criteria for NMO-SD.


Although immunosuppressive drugs prevent most NMO-SD relapses, onset attacks and residual relapses are still a great medical concern. Currently, the best medical treatment only decreases mean relapse rates to 0.4/year and the mean increase of impairment after a relapse is approximately +1 EDSS point.19–21 It was estimated that paraplegia is reached after a mean of three spinal attacks, whereas blindness is obtained after a mean of 1.5 optic attacks, at least in French West Indian NMO-SD patients.22 23

Pathophysiologically, NMO-SD attacks are characterised by specific antibodies directed against AQP4, activating the complement cascade and leading to a rapid destruction of astrocytes and neuronal tissue throughout the lesion. These NMO-SD lesions are prototypical of pattern II MS lesions, which are improved by PLEX.24 25 Interestingly, in vitro experiments argued that early lesions might undergo transient phenomenon of compensation by downregulating and internalising AQP4 targeted by antibodies and complement.26 27 This could be conceptualised like a necrotic core of the lesion surrounded by a ‘penumbra’ where AQP4 is lost without necrosis.28 Therefore, irreversible necrosis could be prevented a short time after insult, and the initiation of an early aggressive treatment may be more appropriate, such as acute stroke revascularisation procedures.29 This hypothesis is supported by an improved visual outcome in patients with NMO-SD ON who receive steroid treatment very early.30

Among the studies dedicated to PLEX in inflammatory disorders, most are retrospective and based on subjective clinical outcome, whereas the clinical score preceding the attacks is often missing. Although our study is retrospective, clinical scores were prospectively obtained (as EDSS and VA are always prospectively collected for patients with NMO in our centre) to ensure a reliable analysis. A few studies demonstrated an improved outcome in patients receiving the PLEX+steroids regimen when compared with steroids alone,7 31 32 or receiving PLEX alone compared with steroids.33 In our previous study,34 we gathered a sufficient cohort to stratify patients by baseline impairment and to demonstrate that the stepwise impairment was lower in patients treated by PLEX given as adjunctive therapy. This preventive effect was higher in the baseline null impairment subgroup (∆EDSS 2.1 vs 5.9, p<0.01), with a reduced statistical trend in the non-null baseline impairment subgroups due to the intrinsic non-linearity of the EDSS. Unfortunately, heterogeneity of baseline impairment is unavoidable due to small size cohorts, introducing a bias in intergroup comparison since the severity of the attack is mechanically rated as low as patients endure higher baseline impairment. Therefore, we here attempt to overcome these former limitations by designing our main outcome, the proportion of patients fully improved. Using this criterion, it was previously demonstrated that the proportion of patients regaining baseline state at late follow-up was double in groups receiving adjunctive PLEX.31 35 This outcome criterion also gives the opportunity to gather, for the first time, optic and spinal attacks. The strength of our study is the ability to analyse, again for the first time, a large number of severe attacks with objective outcomes.

The influence of time to PLEX treatment is highly suspected based on the pathophysiology of NMO-SD, yet multiple studies reported negative results.7–10 A short delay was associated with a better outcome in only two studies.12 13 It is prudent to note that initiation of PLEX treatment was always delayed by a median of 3 weeks in many previous studies, which may have hampered any beneficial effect of early PLEX. Our cohort gathered, to the best of our knowledge, for the first time a majority of early treated patients: a third received PLEX before day five and another third before day 10. We here report that maximal improvement is achieved when the delay in initiating PLEX treatment is minimised (≤5 days), with clinical benefits gradually diminishing the longer the delay in initiating PLEX. Therefore, frequently negative or negligible positive results obtained in prior studies using PLEX as a late add-on treatment are not unexpected and relate to the virtual absence of early treated patients. Interestingly, a complete improvement was only attained in 5%–20% of the patients treated with PLEX delay exceeding 10 days, which is roughly similar to the proportion of our historical cohort of spinal attacks treated by steroids only.35

Can attack outcome be improved when PLEX is initiated on day one?36 Unfortunately, we have not gathered sufficient numbers of patients to definitely demonstrate that acutely given PLEX (≤1 day) is better than early PLEX (≤5 days). However, the few convincing cases of Lazarus effects (immediate back to baseline clinical line after a severe attack) were only observed in patients treated at day 1. Our study demonstrates that early initiation of PLEX (≤5 days) is more beneficial than delayed PLEX and suggests a better outcome if PLEX is started before day 2. Consequently, perceiving PLEX as a rescue therapy only after steroid failure could be deleterious: steroids infusion may take 3–5 days before considering potential improvement, and this delay is then added to the other delays that are onset-to-hospital and admission-to-treatment (which may be a few days at first attack). We demonstrate that the probability to regain baseline level of impairment is reduced by a half where the delay in initiation of PLEX from day 1 to ≥5 days occurs. Therefore, PLEX benefits are partially lost when it is initiated as an add-on after steroids failure. Considering the low probability to regain baseline level impairment after steroids treatment alone (0%–35%31 34), we thoroughly advocate considering PLEX as soon as materially possible during severe attacks.

Although our early treated patients received optimal care, about 20% remained deeply impaired, and this proportion increased to half in patients treated late. This high proportion suggests that attack severity may be driven by a yet unknown idiosyncratic variable (ie, unknown anti-AQP4 B cell clone displaying an enhanced antibody-dependent cell toxicity,37 other associated antibodies, role of eosinophils and so on), meaning that dedicated strategies should be further envisioned in add-on to early steroid+PLEX treatment.

The proportion of our seropositive (IgG-NMO) patients is two-thirds, which is in the expected range of NMO-SD cohorts. Since NMO-IgG is a key player of the NMO disease pathophysiology, one could expect that antibody removal may prevent or treat relapses. Interestingly, positive results of PLEX are obtained in seropositive patients with NMO-SD as well as in seronegative patients, without influence of IgG-NMO status on PLEX outcome,9 11 34 38 39 as we also observed in our cohort. This result suggests a potential role of yet unknown antibodies in seronegative patients and a major role of complement factors in NMO pathophysiology, all of them may be targeted by PLEX.

From a pragmatic point of view, we recommend considering PLEX in every severe attack of the NMO-SD. PLEX should be started as soon as possible in addition to steroids. Although less than 1% of the circulating steroids are removed by PLEX, when used the same day as PLEX procedure, steroids should be preferentially infused at the end of each PLEX session.40 41


We are indebted to M Beraud and B Maru for their fruitful comments and copyediting. We would like to thank R Marignier for testing serum samples for anti-AQP4 and anti-MOG.


View Abstract


  • Contributors The authors’ contributions are as follows: PC and MB: study concept or design and collection of clinical and radiographic data. PC, HM, RV, J-LF and HM: clinical management. MB, PC and SD: data analysis and interpretation. SD performed statistical analysis. MB wrote this draft and all authors critically evaluated the manuscript.

  • Funding None.

  • Competing interests None declared.

  • Ethics approval Local hospital ethic commitee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.