Clinical validation of a novel in vitro diagnostic neurofilament light chain assay for the prognostication of disease activity in people with relapsing multiple sclerosis

Abstract

Background:

Neurofilament light chain (NfL) is a promising marker for predicting disease activity in relapsing multiple sclerosis (RMS). To date, however, there has been no commercially available NfL assay validated in MS and intended for routine clinical use.

Objective:

To identify and validate a single threshold for NfL in blood that differentiates RMS patients, aged 18–55 years, at a higher versus lower risk of disease activity over 2 years, using the Atellica^® IM NfL assay.

Methods:

The optimal NfL threshold for this assay/use case was identified and independently validated using ASCLEPIOS I and II data, respectively. The primary endpoint (annualized number of new/enlarging T2 (neT2) lesions) was analyzed using negative binomial models. Threshold optimization used maximum likelihood methodology. Generalizability analyses used data from ASCLEPIOS II, FREEDOMS, and TRANSFORMS.

Results:

NfL concentration of 12.9 pg/mL was validated as the optimal cutoff for prognosticating disease activity as measured by neT2 lesion over 2 years. This threshold prognosticated individual patient risk for persistent disease activity (>3 neT2 lesions/year over 2 years) and showed prognostic value across relevant subgroups and clinical scenarios. Findings for relapses were similar.

Conclusion:

The CE-marked Atellica^® IM NfL assay is now validated for prognostic use in RMS patients.

Introduction

Disease activity levels are typically highest in early multiple sclerosis (MS), ¹ making early prognostication particularly important for guiding treatment decisions and initiating appropriate interventions before central nervous system damage becomes irreversible.^{2 –5} According to treatment guidelines, clinical decisions should consider the level of disease activity—as indicated by relapses and/or magnetic resonance imaging (MRI) lesions—and include a discussion with the patient on the benefits and risks of disease-modifying therapy (DMT). ⁶ Although no validated thresholds have been established, it is widely accepted that patients who have a higher lesion load, increased brain atrophy, or greater disease activity are generally at higher risk of progression, especially if not treated with high-efficacy therapy. However, there is currently no consensus definition on who should be considered at higher versus lower risk of future disease activity, and no validated clinical, radiological, or biological thresholds are available. This highlights the need for a minimally invasive prognostic tool that can aid physicians in stratifying patients according to their likelihood of disease activity from the earliest stages of MS.^2,3,7

Neurofilament light chain (NfL) is a neuron-specific cytoskeletal protein that has emerged as a promising marker of neuroaxonal injury in neurological diseases.^{8 –10} In MS, NfL has shown potential value for monitoring disease activity and response to treatment, ¹¹ and its implementation has been shown in a recent study to have considerable impact on clinical decision-making. ¹² Expert panels are increasingly recommending the inclusion of NfL assessment in patient care.^11,13 Approximately half of neurologists surveyed by Garcia-Dominguez et al. ¹⁴ indicated they would escalate treatment for clinically stable patients with MS exhibiting cognitive decline and elevated NfL levels. The study further revealed that neurologists less engaged in MS care, as well as those with limited awareness of NfL’s clinical significance, were more likely to make suboptimal therapeutic choices in this scenario. The utility of NfL as a prognostic indicator for focal MS disease activity has also been shown in numerous post hoc studies and has been favorably discussed in several review articles.^{15 –24} In the APLIOS study, wherein people with relapsing MS (pwRMS) received subcutaneous ofatumumab 20 mg every 4 weeks, baseline NfL had prognostic value even in individuals who appeared free of gadolinium-enhancing lesions and would therefore have been considered “free of disease activity” based on the current standard of care. ¹⁷

An international expert consensus supports the use of NfL as a biomarker of short- and long-term prognosis in pwRMS, together with existing clinical and imaging tools.^11,16 However, to support wide acceptance in routine practice, an in vitro diagnostic (IVD) assay that demonstrates stringent and reproducible analytical performance requirements for its intended use in the clinical practice setting is needed.

Here, we report the clinical validation analyses for the Atellica^® IM NfL assay, which has undergone conformity assessment for the CE mark to comply with the European Union’s (EU) IVD medical device regulations.

Materials and methods

Atellica^® IM NfL assay

The Atellica^® IM NfL assay is a fully automated, two-step sandwich immunoassay by Siemens Healthineers for quantifying NfL in human serum and plasma (EDTA) using the Atellica^® IM Analyzer and ADVIA Centaur XP/XPT Systems, and it is CE-marked, demonstrating compliance with the IVD regulations, for use in the European Union. ²⁵ The blood test, in conjunction with clinical, imaging, and laboratory findings, is intended to be used as an aid in identifying patients between 18 and 55 years of age with RMS, who are at a higher versus lower risk of MS disease activity, as defined by new or enlarging T2 (neT2) MRI lesions, within a 2-year period. ²⁶

Clinical trials

Clinical data (combining both treatment arms) were used from previously published trials (Supplemental Table 1), including ASCLEPIOS I (NCT02792218) for data exploration and threshold identification and ASCLEPIOS II (NCT02792231) for independent threshold validation of the assay that was CE-marked. ²⁷ Data from ASCLEPIOS II, FREEDOMS (NCT00289978), ²⁸ and TRANSFORMS (NCT00340834) ²⁹ were used for additional analyses. NfL data were blinded until analysis.

Outcome measures and analysis overview

The primary outcome was the annualized rate of neT2 lesions, measured from baseline to the last follow-up MRI, standardized to 1 year. These neT2 lesions indicate new or expanded areas of demyelination and reflect the intensity of MS disease activity between two scans. Lesion rates were estimated using radiological data collected up to 2.3 years following baseline NfL sampling.

Following identification and validation of the optimal NfL threshold for dichotomizing pwRMS into higher-risk versus lower-risk groups for neT2 lesions, pooled data (i.e. from ASCLEPIOS II, FREEDOMS, and TRANSFORMS) were used to explore the prognostic value of the NfL threshold on the individual patient’s risk of developing >3 neT2 lesions per year over 2 years. The occurrence of >3 neT2 lesions was selected as the threshold for radiological activity to reflect a persistent level of lesion formation over the medium term. This degree of sustained activity is considered clinically actionable and, in a clinical setting, should prompt a physician to consider a treatment switch. ³⁰

Figure 1 and Table 1 present an overview of the approach taken to identify and validate the NfL prognostic threshold for neT2 lesions as well as additional analyses performed to confirm its prognostic value. As a supplementary post hoc analysis to further assess the impact of age, the prognostic value of a single threshold as provided by our model was compared to an age-adjusted approach based on Z-scores derived from a reference population as previously proposed by Benkert et al. ²²

Figure 1.

Overview of approach taken to identify and validate the NfL prognostic threshold for neT2 lesions as well as additional analyses performed to confirm its prognostic value.

DMT, disease-modifying therapy; ne, new or enlarging; NfL, neurofilament light chain; pwRMS, people with relapsing multiple sclerosis.

Table 1.

Overview of analyses performed to identify and validate the NfL prognostic threshold for neT2 lesions and confirm its prognostic value.

Steps	Performed analyses
Step 1: Identification of NfL prognostic threshold	The optimal baseline NfL threshold for dichotomizing pwRMS into higher-risk versus lower-risk groups for neT2 lesions over the next 2 years was identified from a comprehensive set of candidate cutoff points ranging from the minimum to maximum NfL value observed within the ASCLEPIOS I dataset, with an increment step of 0.1 pg/mL between candidate cutoffs. For each candidate cutoff point within the selection interval, patients were dichotomized as lower risk (< candidate cutoff point) and higher risk (⩾ candidate cutoff point) using a negative binomial regression model with log-link function. The candidate cutoff point associated with the largest maximum log likelihood was identified as the optimal cutoff point for dichotomizing pwRMS into those with “higher” versus “lower” on-study lesion numbers and thus for prognostication of future radiological disease activity.
Step 2: Independent clinical validation	Previously unseen NfL concentrations from the independent validation dataset (ASCLEPIOS II) were transferred to the data analysis environment to perform the predefined validation analysis using a similar negative binomial model.
Step 3: Generalizability and contextualization	Generalizability of the prognostic threshold was assessed in various pwRMS subgroups, including those defined by the last DMT received before study entry (ASCLEPIOS II data) and across a range of on-study treatments with various degrees of efficacy. Generalizability was also assessed in placebo-treated pwRMS from FREEDOMS to approximate the natural history of the disease. In addition, pooled data were used to assess generalizability across clinically relevant scenarios (treatment-naive patients, patients initiating/switching to a new DMT, and patients on active treatment). Clinical relapse outcome was an exploratory endpoint, as the study was not powered for this purpose.

For each step, the analyzed population comprised patients in the full analysis set of each study who had valid NfL values at baseline (IVD-evaluable). Patients with available data for the particular endpoint (neT2 lesions or relapse data) and required covariates were included in the analysis; data imputation was not performed. DMT, disease-modifying therapy; NfL, neurofilament light chain; pwRMS, patients with relapsing multiple sclerosis.

The following two criteria for success were prespecified in the analysis plan before any samples were analyzed: (1) the null hypothesis (equal number of lesions, irrespective of the baseline NfL level) was rejected at the one-sided alpha of 0.025 and (2) the estimated lesion rate in the higher-risk group was ⩾ 40% larger than that for the lower-risk group. A 40% difference in lesion rates was considered the minimal clinically relevant difference for this device. Based on an exploratory analysis, this corresponds to an estimated difference of at least three additional lesions per year under placebo. Both criteria needed to be independently met in ASCLEPIOS I and ASCLEPIOS II for the single NfL threshold to be considered clinically validated.

See Supplementary Materials for further information on the assay, the trial data sets, blood samples, and statistical analysis.

Results

Overall, 582 (ASCLEPIOS I) and 570 (ASCLEPIOS II) pwRMS had baseline serum samples and neT2 measurements post-baseline and were included in the primary analyses. Baseline demographics and disease characteristics of the IVD-evaluable patients were similar to the overall patient population in both ASCLEPIOS I and ASCLEPIOS II (Table 2). Supplemental Figure 1 presents a consort diagram of patient disposition.

Table 2.

Baseline demographics and disease characteristics for IVD-evaluable and IVD-unevaluable patients in ASCLEPIOS I and ASCLEPIOS II (Full Analysis Set).

Characteristic	ASCLEPIOS I			ASCLEPIOS II
	IVD-evaluable a N = 596	IVD-unevaluable N = 331	Total N = 927	IVD-evaluable a N = 592	IVD-unevaluable N = 363	Total N = 955
Age (years)
N	596	331	927	592	363	955
Mean (SD)	38.2 (8.87)	38.6 (8.89)	38.4 (8.87)	37.5 (9.3)	39.1 (9.5)	38.1 (9.4)
Median (min, max)	39.0 (18, 55)	39.0 (19, 56)	39.0 (18, 56)	37.0 (18, 56)	40.0 (18, 56)	38.0 (18, 56)
Gender
Female, n (%)	399 (66.9)	236 (71.3)	635 (68.5)	392 (66.2)	246 (67.8)	638 (66.8)
Male, n (%)	197 (33.1)	95 (28.7)	292 (31.5)	200 (33.8)	117 (32.2)	317 (33.2)
BMI (kg/m2)
N	596	330	926	592	363	955
Mean (SD)	25.5 (5.8)	27.4 (6.9)	26.2 (6.3)	25.0 (5.4)	26.6 (6.8)	25.6 (6.0)
Median (min, max)	24.5 (14.4, 54.5)	25.8 (15.9, 57.9)	24.9 (14.4, 57.9)	24.0 (16.0, 55.4)	25.3 (14.6, 54.5)	24.3 (14.6, 55.4)
EDSS
N	596	330	926	592	363	955
Mean (SD)	2.9 (1.4)	3.1 (1.3)	3.0 (1.4)	2.8 (1.3)	3.0 (1.4)	2.9 (1.4)
Median (min, max)	2.5 (0.0, 6.5)	3.0 (0.0, 6.0)	3.0 (0.0, 6.5)	2.5 (0.0, 6.0)	3.0 (0.0, 6.0)	2.5 (0.0, 6.0)
Previous DMT status
Treatment naive, n (%)	259 (43.5)	114 (34.4)	373 (40.2)	236 (39.9)	140 (38.6)	376 (39.4)
Previously treated, n (%)	337 (56.5)	217 (65.6)	554 (59.8)	356 (60.1)	223 (61.4)	579 (60.6)
No. of patients with ⩾ 1 Gd-enhancing lesions at baseline
n (%)	195 (32.7)	126 (38.1)	321 (34.6)	251 (42.4)	127 (35.0)	378 (39.6)
Mean (SD)	1.2 (3.1) b	1.9 (5.1) c	1.5 (3.9) d	1.6 (4.4) e	1.5 (3.5) f	1.5 (4.1) g
Median (min, max)	0.0 (0.0, 42.0) b	0.0 (0.0, 47.0) c	0.0 (0.0, 47.0) d	0.0 (0.0, 63.0) e	0.0 (0.0, 25.0) f	0.0 (0.0, 63.0) g
T2 lesion volume at baseline (mm3)
N	591	327	918	587	359	946
Mean (SD)	12,771.2 (13,452.3)	13,830.3 (14,893.6)	13,148.5 (13,983.9)	13,356.6 (13,461.4)	12,737.8 (13,976.0)	13,121.8 (13,654.9)
Median (min, max)	7900.0 (52, 85,903)	8781.0 (163, 93,520)	8190.0 (52, 93,520)	8787.0 (40, 112,332)	8074.0 (77, 81,857)	8512.0 (40, 112,332)
MS duration since first symptom (years)
N	596	331	927	591	363	954
Mean (SD)	8.3 (7.2)	8.2 (6.7)	8.3 (7.0)	7.8 (7.1)	8.8 (7.8)	8.2 (7.4)
Median (min, max)	6.4 (0.1, 38.7)	6.9 (0.2, 31.5)	6.5 (0.1, 38.7)	5.9 (0.1, 31.8)	6.6 (0.2, 36.1)	6.2 (0.1, 36.1)
Number of relapses in the year before the study
N	596	331	927	592	363	955
Mean (SD)	1.3 (0.7)	1.2 (0.7)	1.2 (0.7)	1.3 (0.7)	1.3 (0.8)	1.3 (0.7)
Median (min, max)	1.0 (0, 5)	1.0 (0, 4)	1.0 (0, 5)	1.0 (0, 5)	1.0 (0, 7)	1.0 (0, 7)
Time since onset of the most recent relapse (months)
N	596	331	927	591	363	954
Mean (SD)	6.9 (9.9)	8.6 (18.3)	7.5 (13.6)	7.9 (11.9)	7.4 (15.1)	7.7 (13.2)
Median (min, max)	4.9 (0.6, 119.2)	5.4 (1.2, 264.8)	5.2 (0.6, 264.8)	5.3 (1.2, 150.3)	5.0 (1.2, 261.5)	5.2 (1.2, 261.5)
Normalized brain volume at baseline (cm3)
N	591	323	914	586	356	942
Mean (SD)	1443.2 (80.8)	1435.9 (78.0)	1440.6 (79.9)	1443.8 (76.2)	1441.9 (78.4)	1443.1 (77.0)
Median (min, max)	1447.4 (1184.4, 1709.4)	1440.6 (1220.0, 1675.3)	1444.7 (1184.4, 1709.4)	1448.4 (1193.3, 1671.0)	1442.6 (1217.1, 1635.0)	1445.5 (1193.3, 1671.0)
Race
Caucasian, n (%)	555 (93.1)	268 (81.0)	823 (88.8)	541 (91.4)	294 (81.0)	835 (87.4)
Others, n (%)	41 (6.9)	63 (19.0)	104 (11.2)	51 (8.6)	69 (19.0)	120 (12.6)
Randomized treatment group
Ofatumumab 20 mg, n (%)	301 (50.5)	164 (49.5)	465 (50.2)	294 (49.7)	187 (51.5)	481 (50.4)
Teriflunomide 14 mg, n (%)	295 (49.5)	167 (50.5)	462 (49.8)	298 (50.3)	176 (48.5)	474 (49.6)
MS subtype
RRMS, n (%)	566 (95.0)	306 (92.4)	872 (94.1)	562 (94.9)	340 (93.7)	902 (94.5)
SPMS, n (%)	30 (5.0)	25 (7.6)	55 (5.9)	30 (5.1)	23 (6.3)	53 (5.5)

BMI, body mass index; DMT, disease-modifying therapy; EDSS, Expanded Disability Status Scale; Gd, gadolinium; IVD, in vitro diagnostic; MS, multiple sclerosis; NfL, neurofilament light chain; RRMS, relapsing-remitting multiple sclerosis; SD, standard deviation; SPMS, secondary progressive multiple sclerosis.

^aPatients in the full analysis set in each study who had valid NfL values at baseline.

^bBased on N = 581 with available scans.

^cBased on N = 324 with available scans.

^dBased on N = 905 with available scans.

^eBased on N = 581 with available scans.

^fBased on N = 359 with available scans.

^gBased on N = 940 with available scans.

Identification and validation of the NfL threshold

The optimal cutoff value for the Atellica^® IM NfL assay to dichotomize pwRMS into higher-risk versus lower-risk groups for future neT2 lesion formation was identified as 12.9 pg/mL, based on a marked peak in maximum likelihood (Figure 2).

Figure 2.

Log likelihood across potential NfL cutoff points^a.

Dashed line indicates the selected baseline NfL threshold: 12.9 pg/mL. NfL, neurofilament light chain.

^aData from ASCLEPIOS I.

In ASCLEPIOS I, the adjusted annualized mean rate of neT2 lesions for patients with baseline NfL measurements above (n = 176) versus below (n = 406) the threshold was 4.70 versus 1.96, respectively (relative risk (RR) = 2.40; 95% confidence interval (CI) = 1.72–3.35; p < 0.001), corresponding to a relative increase in lesion formation of 140.4% (Figure 3(a)). Thus, the formal success criteria for threshold identification were met in ASCLEPIOS I. Supplemental Figure 2 shows how the number of neT2 lesions increased as a continuous function of baseline NfL quartiles in both arms of ASCLEPIOS I.

Figure 3.

Identification and independent validation of the optimal NfL threshold.

High NfL, baseline NfL ⩾ 12.9 pg/mL. Low NfL, baseline NfL < 12.9 pg/mL. The number of neT2 lesions was analyzed using a negative binomial model with adjustments for NfL category; the natural log of the time from the screening scan (in years) was used as the offset. n, number of participants included in the analysis; ne, new or enlarging; NfL, neurofilament light chain.

In ASCLEPIOS II, the adjusted annualized mean rate of neT2 lesions for patients with baseline NfL measurements above (n = 197) versus below (n = 373) the threshold was 4.45 versus 2.55, respectively (RR = 1.75; 95% CI = 1.27–2.40; p < 0.001), corresponding to a relative increase in lesion formation of 74.6% (Figure 3(b)). Thus, the formal success criteria were also met in the independent validation data set.

An NfL concentration of 12.9 pg/mL was therefore considered the optimized and clinically validated threshold for the Atellica^® IM assay to dichotomize pwRMS into higher-risk versus lower-risk groups for neT2 lesion formation over 2 years.

Prognostic value of the NfL threshold for individual patient’s risk of persistent disease activity

Overall, 330 pwRMS had an annualized neT2 lesion rate of >3, equating to a prevalence of 23.3% (95% CI = 21.1%–25.5%) for the entire study population (i.e. risk without NfL testing (pretest risk)). With testing, the probability of developing > 3 lesions per year over 2 years in patients with baseline NfL levels above the threshold was 28.8% (higher than the pretest prevalence), whereas the risk was 18.0% in patients with NfL levels below the threshold (less than the pretest prevalence). A significant absolute risk difference of > 10% was therefore observed on an individual patient level between those prognosticated as being at higher-risk versus lower-risk of neT2 lesions (Table 3). More relevant to clinical practice than the overall test performance across pooled treatment data is its prognostic performance after the influence of various therapies (see Supplemental Table 2 for test performance metrics in different therapeutic contexts).

Table 3.

Prognostic value of the NfL threshold on the individual patient’s risk of persistent disease activity (defined by >3 lesions per year over 2 years) ^a .

NfL pg/mL	N b	Annualized neT2 rate		Likelihood ratio (95% CI)	Absolute risk (95% CI) c	Annualized neT2 rate(95% CI)	Annualized relapse rate (95% CI)
		> 3	⩽ 3
⩾ 12.9	694	200	494	1.34 (1.19, 1.48)	28.8% (200/694) (25.4%, 32.2%)	3.85 (3.35, 4.42)	0.33 (0.29, 0.38)
< 12.9	724	130	594	0.72 (0.62, 0.83)	18.0% (130/724) (15.2%, 20.8%)	2.06 (1.80, 2.36)	0.24 (0.21, 0.28)
Total	1418	330	1088		Prevalence 23.3% (330/1418) (21.1%, 25.5%)

High NfL, baseline NfL ⩾ 12.9 pg/mL. Low NfL, baseline NfL < 12.9 pg/mL. Across the pooled dataset: sensitivity 60.6%, specificity 54.6%, positive predictive value 28.8%, and negative predictive value 82.0%. Performance metrics in a specific therapeutic context are provided in Supplemental Table 2. Risk of an individual developing > 3 neT2 lesions per year over 2 years was calculated as a function of the NfL test result and provided with the following additional summary measures: absolute risk (post-test risk), likelihood ratio between the low- and high-NfL groups, annualized rate of neT2 lesions, and annualized relapse rate. CI, confidence interval; N, number of participants included in the analysis; ne, new or enlarging; NfL, neurofilament light chain.

^aPooled data from ASCLEPIOS II, FREEDOMS, and TRANSFORMS.

^bApplies to neT2 endpoint.

^cAbsolute risk of an individual developing > 3 lesions per year over 2 years (high vs. low NfL).

Among patients with a high-NfL test result who remained untreated after the test, 61.7% developed persistently high lesion activity—defined by more than three lesions per year—over the subsequent 2 years (positive predictive value (PPV) on placebo). When such patients with a high-NfL test result instead received high-efficacy treatment, the risk of developing high lesion activity dropped to 14.3% (PPV on ofatumumab). Conversely, after a low NfL test result, 69.1% of untreated patients did not develop persistently high lesion activity, compared to 97.2% of patients receiving effective therapy (negative predictive value on placebo or ofatumumab, respectively). Thus, while the NfL test helps to distinguish risk levels for disease activity over the mid-term, the choice of treatment plays a significant role in patient outcomes.

Generalizability of the NfL threshold and clinical scenarios

The prognostic value of the NfL threshold was shown across relevant subgroups of pwRMS, defined by age, gender, race, body mass index (BMI), disability level, previous use of a DMT, MS subtype at study entry, and randomized treatment received (Supplemental Figures 3 and 4).

For pwRMS who had previously received a DMT—such as dimethyl fumarate, glatiramer acetate, any form of interferon, natalizumab, or other therapies—and with baseline NfL concentrations above versus below the threshold, a higher annual number of neT2 lesions developed, regardless of which DMT was last used. No significant interaction was found between the prognostic effect of NfL and treatment history (p = 0.89; see Supplemental Figure 5), supporting the prognostic value of NfL irrespective of prior medication. Similar findings were obtained for both previously treated and treatment-naive patients, with a trend in the same direction for clinical relapse rates (Supplemental Figure 6).

The prognostic value of the NfL threshold for neT2 lesion formation was also observed in pwRMS who initiated different treatments across a range of efficacy classes (placebo, interferon-β-1a, teriflunomide, fingolimod, and ofatumumab; Figure 4). A trend in the same direction was observed for clinical relapse rates (Supplemental Figure 7).

Figure 4.

Prognostic value of the NfL threshold on the annualized rate of neT2 lesions across treatment groups^a.

High NfL, baseline NfL ⩾ 12.9 pg/mL. Low NfL, baseline NfL < 12.9 pg/mL. The number of neT2 lesions was analyzed using a negative binomial model with adjustments for NfL category, treatment, and interactions of NfL category by treatment; the natural log of the time from the baseline scan (in years) was used as the offset. The usual limitations apply when comparing treatment arms, as these are from different clinical trials. n, number of participants included in the analysis; ne, new or enlarging; NfL, neurofilament light chain.

^aPooled data from ASCLEPIOS II, FREEDOMS, and TRANSFORMS.

In placebo-treated patients (FREEDOMS data), approximating what the natural course of the disease would be if the absolute risk of neT2 lesions was not modulated by treatment, the absolute difference was > 7 additional lesions per year in pwRMS classified as being at higher-risk versus lower-risk (95% CI = 4.04–11.16; p < 0.001; Figure 5). In these patients, baseline NfL levels above the threshold were accompanied by a significantly increased annualized relapse rate (p = 0.039).

Figure 5.

Prognostic value of the NfL threshold in the natural history of the disease^a.

High NfL, baseline NfL ⩾ 12.9 pg/mL. Low NfL, baseline NfL < 12.9 pg/mL. ARR, annualized relapse rate; CI, confidence interval; HR, hazard ratio; ne, new or enlarging; NfL, neurofilament light chain.

^aData from FREEDOMS.

^bThe number of neT2 lesions was analyzed using a negative binomial model with adjustments for NfL category, treatment, and interactions of NfL category by treatment; the natural log of the time from the baseline scan (in years) was used as the offset (n, number of participants included in the analysis).

^cARR was analyzed using a negative binomial model with log-link to the number of relapses adjusted for baseline NfL category, treatment, and interaction of NfL category by treatment; the natural log of the time in study was used as the offset to annualize the relapse rate (n, total number of relapses included in the analysis; Y, number of patient-years in study).

The prognostic value of the NfL threshold for neT2 lesion formation was also seen in pwRMS receiving active treatment (teriflunomide or ofatumumab) when the NfL test was performed (Supplemental Figure 8). A high on-treatment NfL level was associated with a significantly higher relapse rate in the teriflunomide arm only. In the ofatumumab arm, relapse rates were consistently low (annualized relapse rate (ARR) < 0.1), irrespective of the on-treatment NfL test result.

Supplemental Figure 9 compares the classification of NfL levels above or below the optimized fixed threshold for this assay versus the alternative of an age-adjusted threshold based on Z-scores (using the 95th percentile level for the NfL range in healthy adults, as measured by the Atellica^® IM NfL assay) ²⁴ in samples from pwRMS aged 18–55 years. Compared with the fixed threshold (Table 3), the prognostic value of the age-adjusted NfL threshold showed minimal differences in the likelihood ratios and absolute risk of > 3 neT2 lesions and in the annualized neT2 rate (Supplemental Table 3).

Discussion

The CE-marked Atellica^® IM NfL assay is clinically validated for prognostic use in pwRMS and is intended to be used as a tool to supplement the current standard of care. ²⁵ A strong correlation (R² ⩾ 0.95) between the Atellica^® IM NfL assay and the Quanterix Simoa^® assay, presently the most frequently used NfL assay although currently not Food and Drug Administration (FDA)-authorized or CE-marked, has been observed ²⁹ and will be discussed in a separate manuscript describing the analytical validation of this assay. Other NfL assays have now become available on ELLA^® (ProteinSimple) and Lumipulse^® (Fujirebio) platforms. While these specific NfL assays are not CE-marked or FDA-authorized, the Lumipulse^® instruments themselves are approved for clinical diagnostics, whereas the ELLA^® platform is not. An NfL threshold of 12.9 pg/mL was validated as the optimal cutoff value for the Atellica^® IM assay to risk stratify pwRMS for the prognostication of radiological disease activity, as defined by neT2 lesions, over the next 2 years. This single threshold also had prognostic value at the individual patient level, with those categorized as having high-NfL levels at baseline being at a higher risk of experiencing persistently high disease activity (> 3 lesions per year over 2 years) versus those categorized as having low NfL levels (28.8% vs. 18.0%). The generalizability of the prognostic threshold was shown across relevant patient subgroups, including those defined by age and BMI. The prognostic effect was confirmed to be independent of prior treatment history and was maintained regardless of whether the test was applied off-treatment (i.e. at baseline) or while on active treatment, and it was observed even in those receiving high-efficacy treatment (HET; i.e. ofatumumab). Findings for clinical relapse activity were generally in the same direction as radiological disease activity.

The NfL test result provides objective information to support physician–patient discussions on benefit–risk considerations and inform clinical decision-making in RMS. After accounting for potentially confounding comorbidities, elevated NfL levels in pwRMS likely reflect neuroaxonal damage secondary to inflammation, including neuronal injury that may occur outside the field of view of a brain scan (e.g. spinal cord damage). ¹⁵ NfL is elevated from around the time of lesion formation and for approximately 3 months, which provides a larger window into recent disease history than standard radiological assessments based on Gd-enhancing lesions.^17,31 Indeed, additional tools for evaluating neuronal damage are particularly valuable due to recent recommendations to limit the use of Gd-based contrast agents to evaluate inflammatory disease activity. ³² Incorporation of NfL measurements into routine practice would provide a relatively inexpensive and noninvasive indication of MS disease activity that can be repeated at regular intervals. ³³ This contrasts with MRI scans, which are not always available or tolerated by patients and are limited by long scanning times and high cost. ³³

Our findings highlight several clinical scenarios wherein routine assessment of NfL using the Atellica^® IM NfL assay is likely to provide value in the clinical management of pwRMS:

• Treatment-naive patients: pwRMS who were treatment naive and categorized as having baseline NfL levels above versus below the threshold experienced significantly more neT2 lesions, and this was accompanied by a consistent trend toward higher relapse rates. The prognostic utility of NfL may be particularly important in early MS (i.e. at presentation) when disease activity levels are the highest. ¹ Evidence suggests that addressing this disease activity through the implementation of HET as first-line treatment delays disability worsening and improves long-term clinical outcomes.^2,3,7,34

• Patients initiating/switching to a new DMT: The prognostic value of the NfL threshold for lesion formation was observed irrespective of the treatment initiated. Patients categorized as having baseline NfL levels above versus below the threshold developed more neT2 lesions regardless of therapy, although, as expected, the absolute lesion count reflected the efficacy class of the specific treatment.^35,36 There was also a consistent trend toward higher relapse rates in pwRMS categorized as having NfL levels above the threshold within each therapeutic context except ofatumumab. For ofatumumab, relapse rates were below one in 10 years, irrespective of NfL levels.

• Patients on active treatment: The prognostic value of the NfL threshold was evaluated at Month 12 while patients were on active treatment (teriflunomide or ofatumumab). Regardless of treatment arm, NfL levels while on treatment were prognostic for the risk of future (breakthrough) lesion formation in pwRMS who continued the same treatment. Specifically, pwRMS who had baseline NfL levels above versus below the threshold experienced significantly more lesions when continuing treatment with the same DMT. In guidelines on the use of NfL in clinical practice, it is recommended that a change in therapy or escalation to HET be considered if NfL levels do not decrease within 6 months of initiating treatment. ¹⁶

Generalizability of the optimized prognostic threshold for the Atellica^® IM NfL assay was successfully shown in the intended use population (pwRMS aged 18–55 years) based on subgroup analyses, including those defined by age.

Age has been discussed as the main confounding factor influencing NfL levels, particularly in people aged > 50 years, ¹⁶ and a methodology has been proposed whereby the NfL concentration from an individual with RMS is compared to an age-dependent Z-score derived from the NfL distribution in a healthy reference population. ²⁰ While validating such an approach is challenging for an IVD, and fitting such an analysis to clinical laboratory workflows would be difficult, as a supplementary analysis, we compared the performance of the proposed Z-scoring methodology with our optimized single threshold method in their ability to predict subsequent disease activity. We evaluated the impact of age on NfL measurements using the Atellica^® assay with a cohort of healthy individuals (n = 684), in which the average NfL concentration showed a significant but quantitatively minimal variation of less than 5 pg/mL due to aging. ²⁶ This aligns well with what is reported by Benkert et al. ²² The average aging-related change of <5 pg/mL in healthy humans between the age of 18 and 55 is relatively small when compared to the variation in NfL due to disease activity in RMS patients, which spans from 2.2 to 300 pg/mL in the target population. ¹⁵ When comparing our optimized fixed threshold for prognosticating neT2 lesion formation over 2 years with an age-dependent Z-scoring approach, the classification of NfL levels as “above” or “below” the threshold remained unaffected for the vast majority (92.6%) of samples independent of the approach taken; differences in classification were seen only in samples with NfL concentrations close (within ±3 pg/mL) to the threshold (Supplemental Figure 9). In line with this observation, the assay performance in terms of prognostic value remained similar with a fixed threshold compared to an age-dependent cutoff (Supplemental Table 3). These results support the view that, to simplify prognostication of inflammatory disease activity in routine clinical care, adjusting for age using Z-scores is unnecessary, since age is not a significant confounder in the target population. ¹⁵ Z-scoring offers little performance gain but adds complexity, as thresholds rely on demographic factors and external reference populations. A validated fixed threshold is therefore more practical for prognostic IVD use in clinical practice.

Our analysis established the validity and the generalizability of a single optimized cut-off for routine clinical practice with the Atellica^® IM NfL assay for the prognostication of the risk of future MS disease activity over a time period of 2 years. Clinicians will need to interpret results on an individualized basis, in conjunction with existing clinical and imaging tools, accounting for potential confounders and individual benefit–risk considerations. For example, comorbidities such as stroke or diabetes, although expected to be infrequent in young pwRMS, can influence NfL findings and need to be considered when interpreting results. Supplemental Table 4 presents mitigation strategies for the main confounding factors. Supplementary Materials show the limitations associated with the analyses.

In conclusion, the Atellica^® IM NfL assay, which can be run on a globally available routine clinical laboratory platform, has been analytically and clinically validated for prognostic use in pwRMS in clinical practice. It provides the first validated test for the prognostication of disease activity in MS patients. Based on our validated threshold, this NfL assay may aid physicians in stratifying patients based on their likelihood of disease activity from the earliest stages of MS, helping to optimize benefit–risk considerations, clinical decision-making, and early implementation of HET.