ARTICLE OPEN ACCESS

Blood neurofilament light levels segregatetreatment effects in multiple sclerosisBenedicte Delcoigne PhD Ali Manouchehrinia PhD Christian Barro MD Pascal Benkert PhD

Zuzanna Michalak PhD Ludwig Kappos MD David Leppert MD Jon A Tsai MD Tatiana Plavina PhD

Bernd C Kieseier MD Jan Lycke MD Lars Alfredsson PhD Ingrid Kockum PhD Jens Kuhle MD

Tomas Olsson MD and Fredrik Piehl MD

Neurologyreg 202094e1201-e1212 doi101212WNL0000000000009097

Correspondence

Dr Delcoigne

benedictedelcoignekise

AbstractObjectiveTo determine factors (including the role of specific disease modulatory treatments [DMTs])associated with (1) baseline (2) on-treatment and (3) change (from treatment start to on-treatment assessment) in plasma neurofilament light chain (pNfL) concentrations in relapsing-remitting multiple sclerosis (RRMS)

MethodsData including blood samples analyses and long-term clinical follow-up information for 1261Swedish patients with RRMS starting novel DMTs were analyzed using linear regressions tomodel pNfL and changes in pNfL concentrations as a function of clinical variables and DMTs(alemtuzumab dimethyl fumarate fingolimod natalizumab rituximab and teriflunomide)

ResultsThe baseline pNfL concentration was positively associated with relapse rate Expanded Dis-ability Status Scale score Age-Related MS Severity Score and MS Impact Score (MSIS-29)and negatively associated with Symbol Digit Modalities Test performance and the number ofpreviously used DMTs All analyses which used inverse propensity score weighting to correctfor differences in baseline factors at DMT start highlighted that both the reduction in pNfLconcentration from baseline to on-treatment measurement and the on-treatment pNfL leveldiffered across DMTs Patients starting alemtuzumab displayed the highest reduction in pNfLconcentration and lowest on-treatment pNfL concentrations while those starting teri-flunomide had the smallest decrease and highest on-treatment levels but also starting fromlower values Both on-treatment pNfL and decrease in pNfL concentrations were highlydependent on baseline concentrations

ConclusionChoice of DMT in RRMS is significantly associated with degree of reduction in pNfL whichsupports a role for pNfL as a drug response marker

RELATED ARTICLE

EditorialTracking therapies in MSMore evidence in favor ofneurofilament

Page 465

MORE ONLINE

PodcastDr David Lapides talkswith Dr Frederik Piehlabout his paper discussinghow blood neurofilamentlight levels segregatetreatment effects inmultiple sclerosis

NPuborg8nky7e

These authors contributed equally to this work

From the Department of Medicine Solna Clinical Epidemiology Division (BD) The Karolinska Neuroimmunology amp Multiple Sclerosis Centre Department of Clinical Neuroscience(AM IK TO FP) and Institute of EnvironmentalMedicine (LA) Karolinska Institutet Centre forMolecularMedicine (AM IK TO FP) Karolinska University Hospital StockholmSweden Neurologic Clinic and Policlinic Departments ofMedicine Biomedicine and Clinical Research (CB ZM LK DL JK) and Clinical Trial Unit Department of Clinical Research(PB) University Hospital Basel University of Basel Switzerland Sanofi Genzyme (JAT) Stockholm Sweden Biogen (TP BCK) Cambridge MA Department of Neurology MedicalFaculty (BCK) Heinrich-Heine University Duesseldorf Germany and Institution of Neuroscience and Physiology (JL) Sahlgrenska Academy University of Gothenburg Sweden

Go to NeurologyorgN for full disclosures Funding information and disclosures deemed relevant by the authors if any are provided at the end of the article

The Article Processing Charge was funded by the Swedish Research Council

This is an open access article distributed under the terms of the Creative Commons Attribution License 40 (CC BY) which permits unrestricted use distribution and reproduction in anymedium provided the original work is properly cited

Copyright copy 2020 The Author(s) Published by Wolters Kluwer Health Inc on behalf of the American Academy of Neurology e1201

Accumulating evidence supports the notion that permanent lossof neurologic functions in multiple sclerosis (MS) is primarilycorrelated with the degree of damage to nerve tracts rather thandegree of demyelination1ndash3 However due to the reserve ca-pacity of the CNS critical levels of nerve damage may take yearsto appear as clinical disability The observation that disease-modifying therapies (DMT) used in relapsing-remitting MS(RRMS) differently affect important long-term clinical out-comes underscores a need for more sensitive measures of coredisease pathologic mechanisms4 MRI is the only acknowledgedbiomarker for disease progression and different volumetric at-rophy measures have been associated with risk of developingincreasing disability5ndash7 However such measures are insensitiveto changes over shorter time periods in individual patientsMoreover spinal cord pathology a major driver of clinical dis-ability is not routinely assessed Among different solublemarkers for neuroaxonal damage neurofilaments have emergedas promising candidates in a range of diseases8 Although notspecific for disease processes operating solely in MS the po-tential value in this condition is especially high since it may beused to monitor treatment effects Most published studies onneurofilament light (NfL) and effects of DMTs have measuredconcentrations of NfL in CSF focusing on a single or a fewDMTs9ndash12 More recently improvements in assay sensitivityhavemade it possible to reliably determineNfL in serum (sNfL)or plasma (pNfL) at concentrations seen in healthy controlsSuch studies have reported a correlation between baseline levelsof pNfLsNfL andmeasures of clinical disease activity includingdevelopment of sustained disability brain atrophy signs ofnerve tissue damage and long-term clinical disabilityoutcomes13ndash15 Treatment effects have been reported by severalauthors1416 Disanto et al14 studied 2 Swiss cohorts of patientswith MS in which the effects of a limited number of DMTs onNfL were reported In this study the decrease in sNfL afterinitiation of DMT was of similar magnitude across all DMTsbut confidence intervals (CIs) were large due to the small size ofthe study population Similarly Novakova et al16 reporteda SwedishMS cohort in which start of DMT resulted in loweredsNfL levels also correlating with CSF NfL concentrationsacross all different DMTs but with low power to address effectsize of specific DMTs Thus so far there is a relative paucity ofwell-powered studies specifically addressing treatment effectsacross multiple DMTs in real-world cohorts of patients Theaim of this studywas to address treatment effects acrossmultipleDMTs through the measurement of blood NfL at 2 time pointsin patients selected within a large cohort of patients with RRMS

initiating DMT in context of a nationwide population-basedfollow-up program for all newer MS DMTs

MethodsPatient selection and sample collectionThe Immunomodulation and Multiple Sclerosis Epidemiologystudy (IMSE) is a comprehensive nationwide Swedish post-approval program of patients starting newer MS DMTs coupledwith sampling of blood at initiation of therapy and at follow-upSamples were collected from patients included in IMSE as well asin the Epidemiologic Investigation of MS and Stockholm Pro-spective Assessment of MS We analyzed data for 1139 patientswith RRMS initiating alemtuzumab (ALM n = 89) dimethylfumarate (DMF n = 339) fingolimod (FGL n = 275) natali-zumab (NTZ n = 284) or teriflunomide (TFL n = 152) In-clusion criteria comprised a baseline sample within a month priorto day of initiation of DMT and a subsequent treatment durationof gt4 months Most patients (1052) provided 2 samples (attreatment start and on treatment [absolute range 4ndash24 months])Seventeen patients (4) contributed samples for more than 1DMT A follow-up program similar to IMSE was recently startedfor rituximab (RTX) however only 11 of 122 analyzed patientshad a sample before starting therapy The total number of patientsincluded in this study is thus 1261 A total of 1026 population-based controls included in the study by Manouchehrinia et al15

was used to calculate age-adjusted pNfL reference curves

NfL analysespNfL concentrations were determined using antibodies fromUmanDiagnostics (Umearing Sweden) and the SIMOA Immu-noassay using the Quanterix Kit (Quanterix Lexington MA)All samples from different DMTs were analyzed with blindingfor treatment or clinical information The lower limit ofquantification (LLoQ) was 195 pgmL All measurementswere duplicated and were above the LLoQ with interassayand intra-assay coefficients of variation of le10

Clinical variables collectionAll IMSE patients attend regular medical visits where clinicalassessments are carried out and recorded through the SwedishMS registry In addition to general demographics (age at DMTstart age at MS onset and sex) we had access to the dates ofrelapses (if any) before DMT start the type of previous DMTs(if any) with start and stop dates and the reason for stopping aswell as clinical assessments Expanded Disability Status Scale

GlossaryALM = alemtuzumab ARMSS = Age-Related MS Severity ScoreCI = confidence intervalDMF = dimethyl fumarateDMT =disease-modifying therapies EDSS = Expanded Disability Status Scale FGL = fingolimod GA = glatiramer acetate IFN =interferon IMSE = Immunomodulation and Multiple Sclerosis Epidemiology LLoQ = lower limit of quantification MS =multiple sclerosisMSIS-29 = MS Impact ScoreMSSS = Multiple Sclerosis Severity Score NfL = neurofilament light NTZ =natalizumab pNfL = neurofilament light in plasma PS = propensity scores RRMS = relapsing-remitting multiple sclerosisRTX = rituximab SDMT = Symbol Digit Modalities Test sNfL = neurofilament light in serum TFL = teriflunomide

e1202 Neurology | Volume 94 Number 11 | March 17 2020 NeurologyorgN

(EDSS) further transformed into the Age-Related MS SeverityScore (ARMSS an alternative to the Multiple Sclerosis SeverityScore [MSSS] based on the patientrsquos age at the time of as-sessment17) the MS Impact Score (MSIS-29) divided into itsphysical and psychological domains and the Symbol DigitModalities Test (SDMT) score

Standard protocol approvals registrationsand patient consentsThe study was approved by the regional vetting board ofStockholm under permits 2006845-311 2011641-31420092017-312 and 04-2521-4 with written informedconsent from all participants

Statistical analyses

Variables preparationFor all analyses we log-transformed pNfL levels to increasethe normality of the distribution We also normalized the log-pNfL values to age 40 (log-pNfLN40) by using the linearrelationship between increasing log-pNfL and age in a largepopulation-based control sample (ie log-pNfLN40 = log[pNfL] minus 002115 [age at DMT start minus 40])15 This nor-malization implies that a difference between 2 pNfLN40measures cannot be attributable to a difference in ages Wecalculated the number of relapses in the year preceding DMTstart and the number of previous DMTs (β-interferonsglatiramer acetate [IFNGA] ALM DMF FGL NTZ RTXand TFL) since disease onset for each patient and these 2variables were considered as numerical We also created a 3-category variable denoting treatment status at start of the newDMT by including a washout period (time span between stopdate of previous DMT and start of new DMT) of at least 1month for IFNGA DMF and TFL at least 3 months forFGL and NTZ and 6 months or more for RTX (none of thepatients had switched from ALM) Patients were di-chotomized as being treated with IFNGA or with one of theother DMTs if washout periods had been shorter

Baseline log-pNfL levels analysisWe analyzed the log-pNfL levels at baseline (without age nor-malization) with linear models Initially we used univariable lin-ear models with log-pNfL levels as the dependent variables andeach of the variables measured at baseline (ie DMT start) as theindependent variables to explore the correlation among log-pNfLlevels clinical variables and patient characteristics In a secondstep we used a best subset selection approach to determinewhichsubset of the baseline variables contributedmost to explaining thevariability of the pNfL levels18 The tested variables included thenumber of previous DMTs treatment status just before DMTstart sex age at disease onset disease duration age at DMT startnumber of relapses during the year before DMT start EDSSARMSS MSIS-29 (physical and psychological scales) andSDMT all these being measured at DMT start

Propensity score estimationIn order to balance the DMT groups we calculated individualDMTpropensity scores (PS) ie the probability to be treated

with a specific DMT1920 We used a multinomial logisticmodel with ALM DMF FGL NTZ and TFL as the de-pendent variable while the independent variables included allvariables measured at DMT start including log-pNfLN40Several combinations of these variables were tested includinginteraction terms or transformed scales of variables Theability of the inverse of the PS in reducing differences betweenDMT groups in baseline log-pNfLN40 values assessed bymeasuring the standardized differences between the meanlog-pNfLN40 values of each DMT group and the overall meandepended on the input variables Among different modelstested the one resulting in the smallest average of the stan-dardized differences was selected19ndash21 In the subsequentanalyses we used weights that were calculated by using theinverse of the PS However individual weights were limited tothe 0995th percentile of their distribution in order to preventdisproportionate effects on the analytical model20 We ex-cluded RTX from PS analyses since baseline pNfL values wereavailable only for a small minority

Changes in log-pNfLN40 levels analysisWe used a graphical approach to describe changes in log-pNfLN40 levels from DMT start to follow up (4ndash24 monthslater) using unweighted means of the log-pNfLN40 acrossdifferent DMTs and subsequently values weighted by theinverse of the PS As the main question was to assess if dif-ferent DMTs were significantly associated with degree of re-duction of pNfLN40 concentrations we calculated the deltapNfLN40 (ie change in log-pNfLN40 levels) We useda weighted linear model with delta as the dependent variableand the DMTs as the independent variable using weightsobtained by inversing the PS and further adjusted for otherbaseline covariates to remove potential residual confound-ing22 Criteria to retain a variable included percentage of theexplained variance and how much the additional variablemodified the estimates for the DMTs In an additional sen-sitivity analysis we stratified on the quintiles of the PS insteadof using weights We also analyzed how the changes in log-pNfLN40 correlated with the changes in EDSS ARMSSMSIS-29 and SDMT using univariable models

Additional supporting analysesAs RTX was excluded from the analyses using PS we alsomodeled the log-pNfLN40 on treatment without using PS butadjusting the analyses for patient characteristics using linearmodels In parallel the log-pNfLN40 on treatment without RTXbut using PS was also modeled

Data availabilityData related to the current article are available from TomasOlsson Karolinska Institutet To share data from the SwedishMSregistry a data transfer agreement needs to be completed be-tween Karolinska Institutet and the institution requesting dataaccess This is in accordance with data protection legislation inEurope (General Data Protection Regulation) Persons in-terested in obtaining access to the data should contact TomasOlsson at tomasolssonkise

NeurologyorgN Neurology | Volume 94 Number 11 | March 17 2020 e1203

ResultsBaseline characteristicsData on baseline patient characteristics at therapy initiationare presented in table 1 There were large differences betweenDMT groups where for example those starting TFL were

older both at disease onset and at therapy initiation had lowerMSIS-29 and ARMSS values and had a longer disease dura-tion compared to other DMT groups (table 1) From a dis-ease severity perspective NTZ starters were characterized byboth higher EDSS and MSIS-29 scores as well as higher re-lapse activity compared to other groups These differences

Table 1 Baseline and on-treatment characteristics of patients in the 6 disease-modifying therapy (DMT) groups

Variables

Median (p25ndashp75) or n ()pValueALM DMF FGL NTZ RTX TFL

No 89 339 275 284 122 152

Womena n () 56 (63) 251 (74) 187 (68) 204 (72) 84 (69) 112 (74) 018

Baseline values

Age yab 33 (28ndash39) 42 (34ndash49) 38 (32ndash44) 37 (30ndash44) 39 (32ndash48) 45 (40ndash50) le0001

Age at MS onset ybc 26 (23ndash32) 32 (25ndash40) 28 (23ndash35) 28 (23ndash35) 315 (24ndash37) 345 (29ndash40) le0001

Disease duration ybc 5 (2ndash10) 7 (2ndash14) 8 (4ndash13) 7 (3ndash13) 3 (3ndash8) 10 (4ndash15) 0002

No of DMTsad 2 (1ndash3) 1 (1ndash1) 1 (1ndash2) 1 (1ndash1) 1 (1ndash1) 1 (1ndash2) le0001

EDSSe 2 (15ndash3) 15 (1ndash25) 2 (1ndash25) 25 (2ndash35) 2 (1ndash3) 2 (1ndash25) le0001

ARMSSe 49 (28ndash64) 28 (10ndash45) 39 (19ndash58) 54 (38ndash70) 37 (19ndash50) 27 (11ndash42) le0001

MSIS-29 physicalc 16 (13ndash26) 15 (12ndash21) 15 (12ndash22) 21 (15ndash25) 14 (11ndash22) 14 (11ndash21) le0001

MSIS-29 psychologicalc 21 (16ndash31) 20 (13ndash27) 20 (14ndash29) 24 (17ndash31) 21 (16ndash28) 17 (12ndash24) le0001

SDMTf 605 (52ndash69) 51 (45ndash59) 51 (43ndash58) 50 (42ndash57) 53 (49ndash62) 52 (46ndash57) le0001

Relapsesfg ( yes) 50 42 44 55 34h 29 le0001

Treatment history

Previous treatment (independent ofwashout)

No previous treatment 13 (15) 69 (20) 21 (8) 28 (10) 25 (37)i 28 (18) le0001

1st line (IFNGA TFL DMF)a 13 (15) 265 (78) 231 (84) 248 (87) 28 (41)i 117 (77) le0001

2nd line (FGL NTZ RTX)a 63 (71) 5 (1) 23 (8) 8 (3) 15 (22)i 7 (5) le0001

NfL values

NfLN40 baselinea 105

(63ndash248)111(82ndash156)

123(87ndash169)

155(99ndash269)

123j

(97ndash182)90(70ndash122)

le0001

NfLN40 on treatmentc 69(54ndash88)

83(68ndash107)

96(76ndash118)

87(73ndash118)

96(79ndash115)

100(72ndash13)

le0001

Time DMT start to 2nd NfLassessmentcd

369(357ndash377)

366(364ndash388)

375(359ndash395)

370(353ndash389)

379(361ndash543)

357(201ndash381)

le0001

Abbreviations ALM = alemtuzumab ARMSS = Age-Related MS Severity Score DMF = dimethyl fumarate EDSS = Expanded Disability Status Scale FGL =fingolimod GA = glatiramer acetate IFN = interferon MS =multiple sclerosis MSIS-29 =MS Impact Score NfL = neurofilament light NTZ = natalizumab RTX =rituximab SDMT = Symbol Digit Modalities Test TFL = teriflunomideThe last column informswhether the patients significantly differ between groups for each variable (1-way analysis of variance for continuous variables and χ2

for categorical variables)a 0 missingb In rounded yearsc Less than 10 missingd Previous to baselinee 26 to 40 missingf 11 to 25 missingg During the year before DMT starth Calculated on 29 valuesi Calculated on 68 valuesj Calculated on 11 values

e1204 Neurology | Volume 94 Number 11 | March 17 2020 NeurologyorgN

were mirrored in both baseline pNfL (data not shown) andbaseline pNfLN40 concentrations (table 1 and figure 1)

Modeling baseline log pNfLThe pNfL values displayed a skewed distribution and were logtransformed We then modeled log-pNfL levels at baseline(without age normalization) with a linearmodel Asmost of thevariables displayed a fluctuating degree of association with thepNfL values and also interacted we used a best subset selectionto model pNfL variability across groups The back transformedestimates (exp[β]) are given in table 2 for both the univariableand multivariable models The pNfL levels increased withEDSS ARMSS MSIS-29 (physical and psychological scales)and number of relapses before DMT start and decreased withSDMT scores and number of previous DMTs

Propensity scoresThe variables retained for modeling the PS through themultinomial logistic model of the 5 DMTs excluding RTX

were selected after testing several combinations of the base-line variables retaining the model with the smallest averagestandardized difference This model included the baselinepNfLN40 level ARMSS EDSS SDMT age at disease onsetthe number of previous DMTs the treatment status just be-fore starting the newDMT and the number of relapses duringthe year before DMT start With these variables the averageof the standardized absolute distances for log-pNfLN40dropped from 024 before weighting to 005 after weighting(figure 2)

Changes in log-pNfLN40 levels analysisThe changes in log-pNfLN40 levels between baseline and ontreatment are presented in figure 3 both for the unweightedvalues (figure 3A) and the values weighted with the inverse ofthe PS (figure 3B) Despite PS weighting some differencesbetween DMTs remained suggesting residual effects of fac-tors not accounted for The estimates from both the un-weighted and weighted linear regression models with delta

Figure 1 Baseline logndashneurofilament light in plasma (pNfL)N40 levels in groups starting different disease-modifying ther-apies (DMTs) (with median and 25th and 75th percentiles)

Box and whisker plots show the distributions of thelog-pNfLN40 concentrations in each groupof patientsat DMT start

NeurologyorgN Neurology | Volume 94 Number 11 | March 17 2020 e1205

(ie change in log-pNfLN40) as the dependent variable andthe DMTs as explanatory variables are presented in table 3The estimates (β) were back-transformed to the original scale(exp[β]) so that for example a value of 080 translates intoa 20 reduction of the baseline pNfLN40 level The meanchange was affected by the type of DMT with the largestmean reduction for ALM in both analyses (54 reduction[95 CI 43ndash62 reduction] and 48 reduction [49ndash56reduction] respectively for the unweighted and weightedanalyses) and the smallest change for TFL for which thesignificance level of 005 was not reached (12 increase [3reduction to 29 increase] and 7 reduction [16 reductionto 4 increase] respectively for the unweighted and weightedanalyses) A post hoc analysis highlighted similarities anddifferences between DMT groups the mean delta betweenDMF and FGL and between NTZ and ALM were not sta-tistically different for the unweighted model In the weightedmodel the mean delta of NTZ did not differ significantly fromDMF and FGL (data not shown) To remove any residualconfounding we further adjusted our model with severalbaseline covariates While this dramatically increased thepercentage of the variance explained it did not change the

pattern observed with our first (weighted) model The esti-mates were only slightly modified when including the log-pNfLN40 at baseline in the model (table 4) Similar limitedchanges also occurred with inclusion of additional baselinecovariates or stratification on PS quintiles (instead ofweighting) (table 4) In order to explore the effect of previousDMTs we further stratified on previous treatment and onbaseline pNfLN40 level (data not shown) This provided ad-ditional insights without modifying our previous observationsFinally we also observed that the changes in log-pNfLN40values EDSS ARMSS and MSIS-29 were all significantlycorrelated to each other though often with low correlationcoefficients (ie around 03 or below)

On-treatment log-pNfLN40 levelsThe analysis of the log-pNfLN40 on treatment with eithera weighted linear model (without RTX group) or with anunweighted model showed that all DMT groups had on av-erage lower values than TFL (table 5) Adjusting for thebaseline log-pNfLN40 improved the model substantially in-creasing the percentage of the explained variance from 21 to40 but did not affect overall estimates Additional

Table 2 Univariable and multivariable estimates and associated p values from a linear model of the baselinelogndashneurofilament light in plasma (pNfL) levels

Variables Univariable exp (β) p Value Multivariable exp (β) p Value

Age 1003 011 Not included mdash

Sex (ref = male) 0917 0054 Not included mdash

Age at disease onset 1004 0057 Not included mdash

Disease duration 09997 091 Not included mdash

No of previous DMTs 0922 le0001 0951 005

DMT just before DMT start

No DMT Ref mdash Ref mdash

IFNGA 0865 le0001 0873 0009

DMF FGL NTZ RTX TFL 0733 le0001 0784 0002

EDSS 1078 le0001 Not included mdash

ARMSS 1035 le0001 Not included mdash

MSIS-29 physical 1172 le0001 1141 le0001

MSIS-29 psychological 1078 le0001 Not included mdash

SDMT 0989 le0001 0991 le0001

No of relapses year before 1199 le0001 1134 le0001

Relapses year before (yesno) 1304 le0001 Not included mdash

Relapses 3 months before (yesno) 1344 le0001 Not included mdash

Abbreviations ARMSS = Age-Related MS Severity Score DMF = dimethyl fumarate DMT = disease-modifying therapies EDSS = Expanded Disability StatusScale FGL = fingolimod GA = glatiramer acetate IFN = interferon MSIS-29 = MS Impact Score NTZ = natalizumab RTX = rituximab SDMT = Symbol DigitModalities Test TFL = teriflunomideUnivariable estimates are given for all tested variables whilemultivariable estimates are only provided for the variables used in the finalmodel The estimatesβwere obtained on a log scale and were back transformed (ie exp[β]) for ease of interpretation Hence exp(β) of 110means an increase of 10 in the pNfLlevel while 090 for exp(β) means a 10 decrease in the pNfL level

e1206 Neurology | Volume 94 Number 11 | March 17 2020 NeurologyorgN

adjustments did not substantially modify these estimatesfurther Treatment duration was tested but did not havea significant contribution

DiscussionDisease pathogenesis in RRMS evolves over years and theavailability of a growing number of treatment options createsa need for additional means to assess disease activity andresponse to treatment including body fluid biomarkers823 Inaddition real-world studies conducted in unselected patientpopulations can provide important information on questionsthat cannot be addressed with existing data from randomizedcontrolled trials24 Along these lines we explored how pNfLconcentrations were distributed in patients with RRMSstarting newer DMTs how this distribution was associatedwith clinical measures and patient characteristics and howpNfL concentrations evolved under treatment Strengths ofthe study include the possibility to simultaneously compare

across multiple treatments in nonrestricted patient groupsbut this approach also entails major challenges in balancingout differences in baseline characteristics since DMT se-lection is heavily influenced by clinical disease character-istics Nevertheless by modeling on relevant variables wedemonstrate that the reduction in pNfL concentrationsdiffers across DMTs with the largest reduction for ALMand the smallest for TFL This result is largely in agreementwith the perceived effectiveness of the studied DMTs Stillreductions in pNfL with DMF FGL and NTZ were similareven if NTZ generally is considered to have a superior effecton relapses and focal MRI lesions of the 3 This observationmay be partly explained by indication bias (ie patientswith more active disease are started on highly effectivedrugs) however an interesting feature with pNfL is that itreflects both diffuse and focal neuroaxonal damage where itmay be speculated whether different DMTs affect these 2aspects differently for example based on their capacity topenetrate into the CNS This will need longer follow-up

Figure 2 Unweighted and weighted baseline logndashneurofilament light in plasma (pNfL)N40

The ability of the propensity scores to correct theimbalance between disease-modifying therapygroups is shown graphically and numerically forlog-pNfLN40 levels The distances are standard-ized (ie they do not depend on the unit in whichthe variable was measured) The effect of thepropensity score is to decrease the standardizeddistances where a standardized distance largerthan 020 can be considered as evidence of im-balance and a potential source of bias21 Herethere is some small residual imbalance for di-methyl fumarate The average of the absolutestandardized distances was 024 before weight-ing and 005 after weighting

NeurologyorgN Neurology | Volume 94 Number 11 | March 17 2020 e1207

studies that also integrate quantitative MRI measures Alsothe kinetics of how pNfL is affected might differ acrossDMTs necessitating longer follow-up with repeated sam-pling Finally comorbidities affecting the peripheral ner-vous system or CNS may act as confounders For exampleleflunomide which is related to TFL has been shown toaffect the peripheral nervous system25 An additional im-portant finding is that we show how essential the baselinepNfL concentration is for correctly predicting the pNfLconcentration on treatment In fact the percentage of thevariance explained by the baseline concentration (gt20)outsized all other factors Accordingly inclusion of thebaseline pNfL value affected estimates increasing the dif-ferences in pNfL concentrations between the DMTs Wealso find that reductions in pNfL concentrations correlatedwith improvements in clinical variables such as EDSSMSSS and MSIS-29 though correlation coefficientswere low (between 010 and 030) replicating earlierfindings1314 Importantly as shown by recent studies pNfLconcentrations at diagnosis also predict important long-term outcomes such as brain atrophy and risk to achieveclinical disability milestones1526

Whereas our data reveal differences in pNfL dynamicsacross the studied DMTs we cannot rule out that differ-ences had been achieved with a more complete model forthe PS even if our additional adjustments did not lead tomajor changes in the estimates Notably however we didnot have access to sufficiently precise MRI data which areknown to affect pNfL13 A further weakness is impreciseinformation on some measures eg the lack of coding forswitching from NTZ due to positive JC virus serology in theSwedish MS registry On the other hand the high generalvalidity of data entered into the Swedish MS registry re-garding treatment episodes and relapses was recently con-firmed by a large-scale national validation against medicalrecords27 Furthermore most patients in the RTX grouplacked a baseline sample which meant that this group wasexcluded from analyses involving PS and that other analysesincluding baseline log-pNfLN40 became less precise Alsothe proportion of patients missing information for somevariables that were used in the adjustment (or in the PSestimation) could also have hampered the power of ourstudy The observational design of the study implies thatpatients were not randomized to treatment nor were they

Figure 3 Baseline and on-treatment mean neurofilament light in plasma (pNfL)N40 levels in the disease-modifying therapygroups

(A) Crude mean pNfL levels at baseline and on treatment (B) Weighted mean pNfL levels at baseline and on treatment The weights are the inverse of thepropensity scores

e1208 Neurology | Volume 94 Number 11 | March 17 2020 NeurologyorgN

randomly selected within the IMSE cohorts and thereforesome selection bias could have occurred It is thereforeimportant to relate these findings to studies exploring pNfLconcentrations in the context of randomized control trialseven if such studies rarely include more than 2 DMTs28 Asa final note the extent different DMTs affected pNfL largelymimic their effect on the long-term risk to convert toa secondary progressive disease course as observed ina large recent real-world study4 The implementation ofsoluble but also novel imaging biomarkers that can com-plement current clinical and imaging monitoring likely willlead to an increased use of more effective DMTs and reducethe risks for patients to be exposed to insufficient treatmentresponses in turn improving important long-term clinicaloutcomes1229

We demonstrate that dynamics of pNfL are significantly influ-enced by specific DMTs and that the degree of pNfL reductionis correlated to clinical and patient-reported outcomes but alsothat the baseline pNfL concentration exerts an unproportionedeffect on on-treatment values in the medium term In order tounderstand if pNfL can be used as a drug response biomarker atthe individual level further studies are needed to address thecorrelation of pNfL changes to long-term clinical outcomes withdifferent DMTs as well as if modeling of pNfL dynamics can beimproved further by including additional variables such as MRIdata or more frequent measurements

Study fundingThe IMSE cohorts received grant support from Biogen(IMSE natalizumab and dimethyl fumarate) Genzyme

Table 4 Mean changes in logndashneurofilament light in plasma (pNfL)N40 values between baseline and on-treatmentmeasures provided by the analyses using propensity scores weighting by their inverse and adjusting forbaseline log-pNfLN40 and additional baseline variables or stratifying on quintile of the propensity scoredistribution

Weighted model adjusted forbaseline pNfLN40

pValue

Weighted model adjusted for baselinepNfLN40 and additional variables

pValue

Stratified model(not adjusted)

pValue

TFL(reference)

101 (094ndash109) Ref 086 (073ndash101) Ref 082 (064ndash104) Ref

DMF 066 (060ndash073) le0001 068 (062ndash075) le0001 066 (056ndash079) le0001

FGL 065 (058ndash071) le0001 067 (061ndash074) le0001 057 (048ndash068) le0001

NTZ 065 (059ndash072) le0001 066 (060ndash073) le0001 065 (055ndash078) le0001

ALM 050 (046ndash056) le0001 052 (047ndash057) le0001 054 (046ndash063) le0001

Abbreviations ALM = alemtuzumab DMF = dimethyl fumarate FGL = fingolimod NTZ = natalizumab TFL = teriflunomideAll estimates were calculated with a linearmodel with the disease-modifying therapies (DMTs) as themain explanatory variable and using propensity scoresweighting by their inverse and adjusting for baseline log-pNfLN40 (column 2) weighting by their inverse and adjusting for baseline log-pNfLN40 sex ageExpanded Disability Status Scale Age-Related MS Severity Score Symbol Digit Modalities Test age at disease onset disease duration and treatment statusjust before DMT start (column 4) or stratifying on the quintile of the propensity score distribution without adjustment (column 6) The estimates have beenback transformed to the original scale Hence a value of 065 means that the on-treatment pNfLN40 value is 065 times the baseline value The significancelevels indicate how the DMT groups differ from TFL

Table 3 Mean changes in logndashneurofilament light in plasma (pNfL)N40 values between baseline and on-treatmentmeasures provided by the unweighted and weighted analyses

Delta Changes in pNfLN40 values unweighted analysisExp(β) (95 CI)

pValue

Delta Changes in log-pNfL values weightedanalysis Exp(β) (95 CI)

pValue

TFL(reference)

1119 (0970ndash1291) Ref 0931 (0840ndash1044) Ref

DMF 0652 (0547ndash0777) le0001 0739 (0628ndash0870) le0001

FGL 0615 (0516ndash0734) le0001 0644 (0547ndash0758) le0001

NTZ 0487 (0410ndash0578) le0001 0671 (0570ndash0791) le0001

ALM 0462 (0375ndash0570) le0001 0517 (0440ndash0608) le0001

Abbreviations ALM = alemtuzumab CI = confidence interval DMF = dimethyl fumarate FGL = fingolimod NTZ = natalizumab TFL = teriflunomideAll estimates are calculated with a linear model with the disease-modifying therapies (DMTs) as the only explanatory variable The estimates have been backtransformed to the original scale Hence a value of 065means that the on-treatment pNfLN40 value is 065 times the baseline value The estimates in column2were obtained from an unweighted model while the estimates in column 4 were obtained from the weighted analysis where the data were weighted by theinverse of the propensity scores With the exception of TFL all drugs were associated with statistically significant mean reductions TFL was non-significantlyassociated with either an increase (unweighted model) or a reduction (weighted model) of the pNfLN40 level The significance levels indicate how the DMTgroups differ from TFL

NeurologyorgN Neurology | Volume 94 Number 11 | March 17 2020 e1209

(IMSE teriflunomide and alemtuzumab) and Novartis(IMSE fingolimod) Research grants with partial support toIMSE projects have been received from the Swedish ResearchCouncil the Swedish Research Council for Health WorkingLive and Welfare the AFA Foundation the Swedish BrainFoundation and the Knut and Alice Wallenberg FoundationThe cost for NfL analysis of the Epidemiologic Investigationof MS cohort was supported by a grant from EU Horizon2020 (MultipleMS grant 733161) The costs for analysis ofbaseline samples from the IMSE projects were defrayed byunrestricted MS research grants from Biogen Novartis andSanofi

DisclosureB Delcoigne reports no disclosures A Manouchehrinia hasreceived speaker honoraria from Biogen C Barro receivedconference travel grant from Novartis and Teva P Benkertand ZMichalak report no disclosures L Kappos has served inthe last 24 months as international or local principal in-vestigator for the following drug studies BOLD EXT EX-PAND (Siponimod Novartis) DECIDE DECIDE EXT(Daclizumab HYP Biogen) ENDORSE (DMF Biogen)FINGORETT FTY-UMBRELLA INFORMS INFORMSEXT LONGTERMS (Fingolimod Novartis) MOMEN-TUM (Amiselimod Mitsubishi) OCRELIZUMAB PHASEII EXT OPERA ORATORIO and extensions (OcrelizumabRoche) REFLEXION (IFN β-1a Merck) STRATA EXTTOP (Natalizumab Biogen) TERIFLUNOMIDE EXTTERRIKIDS (Teriflunomide Sanofi-Aventis) and ASCLE-PIOS III (Ofatumumab Novartis) The Research of MSCenter in Basel has been supported by grants from Bayer

Biogen Novartis the Swiss MS Society the Swiss NationalResearch Foundation and the European Union In the last 24months the institution also received grants for patient serv-ices from Bayer Merck and CSL-Behring L Kappos isa member of the editorial boards of Journal of NeurologyMultiple Sclerosis Journal Neurology and Clinical NeuroscienceMultiple Sclerosis and Related Disorders and Clinical andTranslational Neuroscience Honoraria and other payments forall these activities have been exclusively used for funding ofresearch at the department L Kapposrsquo institution (UniversityHospital Basel) received the following in the last 3 years usedexclusively for research support at the Department steeringcommittee advisory board and consultancy fees from Acte-lion Almirall Bayer Biogen CelgeneReceptos df-mpExcemed Genzyme Japan Tobacco Merck Minoryx Mit-subishi Pharma Novartis Roche sanofi-aventis SantheraTeva and Vianex and royalties for Neurostatus-UHB prod-ucts For educational activities the institution received pay-ments and honoraria from Allergan Almirall Baxalta BayerBiogen CSL-Behring Desitin Excemed Genzyme MerckNovartis Pfizer Roche Sanofi-Aventis and Teva D Leppertreports no disclosures JA Tsai is an employee of SanofiGenzyme T Plavina is an employee of and holds stockstockoptions in Biogen BC Kieseier is an employee of and holdsstockstock options in Biogen J Lycke has received travelsupport andor lecture honoraria from Biogen NovartisTeva and GenzymeSanofi-Aventis has served on scientificadvisory boards for Almirall Teva Biogen Novartis andGenzymeSanofi-Aventis serves on the editorial board ofActa Neurologica Scandinavica and has received unconditionalresearch grants from Biogen Novartis and Teva L

Table 5 Comparison of on-treatment logndashneurofilament light in plasma (pNfL)N40 between treatment groups withseveral statistical linear models unweighted and adjusted or weighted without and with adjustment

Unweighted Weighted (propensity score)

(1) p Value (2) p Value (3) p Value (4) p Value

TFL (ref) 139 Ref 139 Ref 139 Ref 139 Ref

DMF 105 le0001 111 le0001 88 le0001 96 le0001

FGL 110 le0001 119 0012 90 le0001 95 le0001

NTZ 105 le0001 123 0042 89 le0001 94 le0001

ALM 82 le0001 87 le0001 69 le0001 76 le0001

RTX 119 04 137 08 Group not included

24 8 21 46a

Abbreviations ALM = alemtuzumab DMF = dimethyl fumarate FGL = fingolimod NTZ = natalizumab RTX = rituximab TFL = teriflunomide = Percentage of the variance explained (ie r2 of the model) (1) pNfLN40 value on treatment (pgmL) unweighted adjusted model (2) same as (1) withoutincluding the baseline log-pNfLN40 (3) pNfLN40 value on treatment (pgmL) weightedmodel without adjustment (4) same as (3) with the same adjustment as(1) The on-treatment log-pNfLN40 value in the disease-modifying therapy (DMT) groups were calculated in reference to TFL (for which the on-treatmentweighted mean pNfL value was 139 pgmL) (1) The unweighted model adjusted for several baseline covariates (baseline log-pNfLN40 age at DMT start sexage at disease onset disease duration and treatment status just before DMT start) provided estimates for on-treatment log pNfLN40 which were lower for allDMTs compared to TFL though not significantly for RTX (2) The differences with TFL were attenuated when removing the baseline log-pNfLN40 from themodel (3 and 4) In contrast the differences with TFL were exacerbated in the weighted models Adjusting or not for covariates in these models slightlymodified the estimates Note that RTXwas not included in theweighted analyses The significance levels indicatewhether the values are significantly differentfrom the reference (TFL)a 21 for DMT 19 for baseline log-pNfLN40 6 for all other covariates

e1210 Neurology | Volume 94 Number 11 | March 17 2020 NeurologyorgN

Alfredsson has received lecture honoraria from Biogen andTEVA I Kockum reports no disclosures J Kuhle served onscientific advisory boards for Novartis PharmaceuticalsMerck Biogen Sanofi Genzyme Roche and Bayer has re-ceived funding for travel andor speaker honoraria fromBiogen Sanofi Genzyme Novartis Merck Serono RocheTeva and the Swiss MS Society and has received researchsupport from Bayer Celgene Biogen Merck Sanofi Gen-zyme Novartis Roche ECTRIMS Research FellowshipProgramme University of Basel Swiss MS Society and SwissNational Research Foundation (320030_160221) T Olssonhas received unrestricted MS research grants advisory boardandor lecture honoraria from Biogen Novartis SanofiRoche Merck TEVA and Allmiral F Piehl has received re-search grants from Biogen Novartis and Genzyme and feesfor serving as Chair of DMC in clinical trials with Parexel Goto NeurologyorgN for full disclosures

Publication historyReceived by Neurology May 20 2019 Accepted in final formOctober 21 2019

References1 Filippi M Bar-Or A Piehl F et al Multiple sclerosis Nat Rev Dis Primers 20184432 Tallantyre EC Bo L Al-Rawashdeh O et al Clinico-pathological evidence that

axonal loss underlies disability in progressive multiple sclerosis Mult Scler 201016406ndash411

3 Trapp BD Stys PK Virtual hypoxia and chronic necrosis of demyelinated axons inmultiple sclerosis Lancet Neurol 20098280ndash291

4 Brown JWL Coles A Horakova D et al Association of initial disease-modifyingtherapy with later conversion to secondary progressive multiple sclerosis JAMA 2019321175ndash187

5 Steenwijk MD Geurts JJ Daams M et al Cortical atrophy patterns in multiplesclerosis are non-random and clinically relevant Brain 2016139115ndash126

6 Sormani MP Arnold DL De Stefano N Treatment effect on brain atrophy correlateswith treatment effect on disability in multiple sclerosis Ann Neurol 20147543ndash49

7 Eshaghi A Prados F Brownlee WJ et al Deep gray matter volume loss drivesdisability worsening in multiple sclerosis Ann Neurol 201883210ndash222

8 Khalil M Teunissen CE Otto M et al Neurofilaments as biomarkers in neurologicaldisorders Nat Rev Neurol 201814577ndash589

9 Gunnarsson M Malmestrom C Axelsson M et al Axonal damage in relapsingmultiple sclerosis is markedly reduced by natalizumab Ann Neurol 20116983ndash89

10 Kuhle J Disanto G Lorscheider J et al Fingolimod and CSF neurofilament lightchain levels in relapsing-remitting multiple sclerosis Neurology 2015841639ndash1643

11 Novakova L Axelsson M Khademi M et al Cerebrospinal fluid biomarkers asa measure of disease activity and treatment efficacy in relapsing-remitting multiplesclerosis J Neurochem 2017141296ndash304

12 de Flon P Gunnarsson M Laurell K et al Reduced inflammation in relapsing-remitting multiple sclerosis after therapy switch to rituximab Neurology 201687141ndash147

13 Barro C Benkert P Disanto G et al Serum neurofilament as a predictor of diseaseworsening and brain and spinal cord atrophy in multiple sclerosis Brain 20181412382ndash2391

Appendix Authors

Name Location Role Contribution

BenedicteDelcoigne PhD

KarolinskaInstitutetStockholmSweden

Author Conceived and designedthe study performedthe data managementand the statisticalanalyses interpretedthe results drafted themanuscript

AliManouchehriniaPhD

KarolinskaInstitutetStockholmSweden

Author Interpreted the resultscritically reviewed themanuscript

Christian BarroMD

UniversityHospitalBaselSwitzerland

Author Performed the dataacquisition interpretedthe results criticallyreviewed themanuscript

Pascal BenkertPhD

UniversityHospitalBaselSwitzerland

Author Interpreted the resultscritically reviewed themanuscript

ZuzannaMichalak PhD

UniversityHospitalBaselSwitzerland

Author Performed the dataacquisition interpretedthe results criticallyreviewed themanuscript

Ludwig KapposMD

UniversityHospitalBaselSwitzerland

Author Interpreted the resultscritically reviewed themanuscript

David Leppert UniversityHospitalBaselSwitzerland

Author Interpreted the resultscritically reviewed themanuscript

Jon A Tsai MD SanofiGenzymeSweden

Author Interpreted the resultscritically reviewed themanuscript

Appendix (continued)

Name Location Role Contribution

Tatiana PlavinaPhD

BiogenCambridgeMA

Author Interpreted the resultscritically reviewed themanuscript

Bernd C KieseierMD

BiogenCambridgeMA

Author Interpreted the resultscritically reviewed themanuscript

Jan Lycke MD University ofGothenburgSweden

Author Interpreted the resultscritically reviewed themanuscript

Lars AlfredssonPhD

KarolinskaInstitutetStockholmSweden

Author Interpreted the resultscritically reviewed themanuscript

Ingrid KockumPhD

KarolinskaInstitutetStockholmSweden

Author Performed the dataacquisition and theinitial datamanagementinterpreted the resultscritically reviewed themanuscript

Jens Kuhle MD UniversityHospitalBaselSwitzerland

Author Performed the dataacquisition interpretedthe results criticallyreviewed themanuscript

Tomas Olsson KarolinskaInstitutetStockholmSweden

Author Conceived and designedthe study interpretedthe results criticallyreviewed themanuscript

Fredrik Piehl MD KarolinskaInstitutetStockholmSweden

Author Conceived and designedthe study interpretedthe results drafted themanuscript

NeurologyorgN Neurology | Volume 94 Number 11 | March 17 2020 e1211

14 Disanto G Barro C Benkert P et al Serum neurofilament light a biomarker ofneuronal damage in multiple sclerosis Ann Neurol 201781857ndash870

15 Manouchehrinia A Stridh P Khademi M et al Plasma neurofilament light levels areassociated with risk of developing sustained disability in multiple sclerosis ECTRIMSOnline 2018228223P378

16 Novakova L Zetterberg H Sundstrom P et al Monitoring disease activity inmultiple sclerosis using serum neurofilament light protein Neurology 2017892230ndash2237

17 Manouchehrinia A Westerlind H Kingwell E et al Age related multiple sclerosisseverity score disability ranked by age Mult Scler 2017231938ndash1946

18 Zhang Z Variable selection with stepwise and best subset approaches Ann TranslMed 20164136

19 Austin PC An introduction to propensity score methods for reducing the effectsof confounding in observational studies Multivariate Behav Res 201146399ndash424

20 Austin PC Stuart EA Moving towards best practice when using inverse probability oftreatment weighting (IPTW) using the propensity score to estimate causal treatmenteffects in observational studies Stat Med 2015343661ndash3679

21 McCaffrey DF Griffin BA Almirall D Slaughter ME Ramchand R Burgette LF Atutorial on propensity score estimation for multiple treatments using generalizedboosted models Stat Med 2013323388ndash3414

22 Nguyen TL Collins GS Spence J et al Double-adjustment in propensity scorematching analysis choosing a threshold for considering residual imbalance BMCMed Res Methodol 20171778

23 Comabella M Montalban X Body fluid biomarkers in multiple sclerosis LancetNeurol 201413113ndash126

24 Trojano M Tintore M Montalban X et al Treatment decisions in multiple sclerosisinsights from real-world observational studies Nat Rev Neurol 201713105ndash118

25 Kim HK Park SB Park JW et al The effect of leflunomide on cold and vibratorysensation in patients with rheumatoid arthritis Ann Rehabil Med 201236207ndash212

26 Canto E Barro C Zhao C et al Association between serum neurofilament light chainlevels and long-term disease course among patients with multiple sclerosis followedup for 12 years JAMA Neurol Epub 2019 Aug 12

27 Alping P Piehl F Langer-Gould A on behalf of the COMBAT-MS Study Group et alValidation of the Swedish multiple sclerosis register further improving a resource forpharmacoepidemiologic evaluations Epidemiology 201930230ndash233

28 Kuhle J Kropshofer H Haering DA et al Blood neurofilament light chain as a bio-marker of MS disease activity and treatment response Neurology 201992e1007ndashe1015

29 Granqvist M Boremalm M Poorghobad A et al Comparative effectiveness of rit-uximab and other initial treatment choices for multiple sclerosis JAMA Neurol 201875320ndash327

e1212 Neurology | Volume 94 Number 11 | March 17 2020 NeurologyorgN

DOI 101212WNL0000000000009097202094e1201-e1212 Published Online before print February 11 2020Neurology

Beacuteneacutedicte Delcoigne Ali Manouchehrinia Christian Barro et al Blood neurofilament light levels segregate treatment effects in multiple sclerosis

This information is current as of February 11 2020

ServicesUpdated Information amp

httpnneurologyorgcontent9411e1201fullincluding high resolution figures can be found at

References httpnneurologyorgcontent9411e1201fullref-list-1

This article cites 28 articles 3 of which you can access for free at

Citations httpnneurologyorgcontent9411e1201fullotherarticles

This article has been cited by 1 HighWire-hosted articles

Subspecialty Collections

httpnneurologyorgcgicollectionmultiple_sclerosisMultiple sclerosis

httpnneurologyorgcgicollectioncohort_studiesCohort studiesfollowing collection(s) This article along with others on similar topics appears in the

Permissions amp Licensing

httpwwwneurologyorgaboutabout_the_journalpermissionsits entirety can be found online atInformation about reproducing this article in parts (figurestables) or in

Reprints

httpnneurologyorgsubscribersadvertiseInformation about ordering reprints can be found online

ISSN 0028-3878 Online ISSN 1526-632XWolters Kluwer Health Inc on behalf of the American Academy of Neurology All rights reserved Print1951 it is now a weekly with 48 issues per year Copyright Copyright copy 2020 The Author(s) Published by

reg is the official journal of the American Academy of Neurology Published continuously sinceNeurology