NfL predicts relapse-free progression in a longitudinal multiple sclerosis cohort study

Abstract

Background: Easily accessible biomarkers enabling the identification of those patients with multiple sclerosis (MS) who will accumulate irreversible disability in the long term are essential to guide early therapeutic decisions. We here examine the utility of serum neurofilament light chain (sNfL) for forecasting relapse-free disability progression and conversion to secondary progressive MS (SPMS) in the prospective Neurofilamentandlongtermoutcome inMS (NaloMS) cohort.

Methods: The predictive ability of sNfL at Baseline and sNfL follow-up (FU)/ Baseline (BL) ratio with regard to disability progression was assessed within a development cohort (NaloMS, n=196 patients with relapsing-remitting MS (RRMS) or clinically isolated syndrome) and validated with an external independent cohort (Düsseldorf, Essen, n=204). Both relapse-free EDSS-progression (RFP: inflammatory-independent EDSS-increase 12 months prior to FU) and SPMS-transition (minimum EDSS-score of 3.0) were investigated.

Findings: During the study period, 17% (n=34) of NaloMS patients suffered from RFP and 14% (n=27) converted to SPMS at FU (validation cohort RFP n=42, SPMS-conversion n=24). sNfL at BL was increased in patients with RFP (10.8 pg/ml (interquartile range (IQR) 7.7-15.0) vs. 7.2 pg/ml (4.5-12.5), p<0.017). In a multivariable logistic regression model, increased sNfL levels at BL (Odds Ratio (OR) 1.02, 95% confidence interval (CI) 1.01-1.04, p=0.012) remained an independent risk factor for RFP and predicted individual RFP risk with an accuracy of 82% (NaloMS) and 83% (validation cohort) as revealed by support vector machine. In addition, the sNfL FU/BL ratio was increased in SPMS-converters (1.16 (0.89-1.70) vs. 0.96 (0.75-1.23), p=0.011). This was confirmed by a multivariable logistic regression model, as sNfL FU/BL ratio remained in the model (OR 1.476, 95%CI 1.078-2,019, p=0.015) and individual sNfL FU/BL ratios showed a predictive accuracy of 72% in NaloMS (63% in the validation cohort) as revealed by machine learning.

Interpretation: sNfL levels at baseline predict relapse-free disability progression in a prospective longitudinal cohort study 6 years later. While prediction was confirmed in an independent cohort, sNfL further discriminates patients with SPMS at follow-up and supports early identification of patients at risk for later SPMS conversion.

Funding: This work was supported by the German Research Council (CRC-TR-128), Else Kröner Fresenius Foundation and Hertie-Stiftung.

Keywords: Disease progression; Multiple sclerosis; Neurofilament light chain; SPMS transition.

Fig. 1

Scheduled visits and study profile a) Scheduled visits within NaloMS. b) Models were developed using the NaloMS cohort (Mainz) and validated with an independent cohort (Düsseldorf, Essen). Abbreviations: RFP, Relapse-free EDSS-progression; SPMS, secondary progressive multiple sclerosis; mo, months; y, years; EDSS, Expanded disability status scale; sNfL, serum neurofilament light chain; MRI, magnetic resonance imaging. Created using Biorender.com

Fig. 2

sNfL levels at baseline predict relapse-free disability progression in a prospective longitudinal study (NaloMS cohort) a) Baseline NfL was increased in patients with relapse free EDSS-progression (RFP) (median (IQR) 10.8 (7.7-15.0) vs. 7.2 (4.5-12.5), p=0.017); data is displayed as a violin plot, left side: density of all data points, right side: median and IQR. b) Machine learning by support vector machine assessed the predictive accuracy of sNfL at Baseline for RFP under consideration of covariates (EDSS Baseline, Number of Gadolinium-enhancing lesions at Baseline, T2-hyperintense lesions at Baseline, Age at Baseline, Disease Duration and Relapses within the last 12 months prior to FU; grey bar). In addition, the predictive accuracy for RFP of a combination of Age + T2-hyperintense lesion number (green bar, without covariate adjustment) and a combination of Age + T2-hyperintense lesion number + sNfL at Baseline (blue bar) was similarly analyzed. c,d) Area under the receiver operating characteristic curve (ROC-AUC) for a combination of Age+T2-hyperintense lesions and additional inclusion of sNfL at Baseline in NaloMS (development cohort; c) and validation cohort (Düsseldorf, Essen; d). c) AUC increased from 0.714 (95%CI 0.645-0.776, green line) to 0.755 (95%CI 0.688-0.813, blue line) after additional consideration of sNfL within the NaloMS cohort (development). d) This finding could be confirmed within the validation cohort (Düsseldorf, Essen): AUC increased from 0.613 (95%CI 0.543-0.680) to 0.715 (95%CI 0.648-0.776) after inclusion of sNfL. e) Receiver operating characteristic curve in the combined NaloMS and validation cohort (Düsseldorf, Essen) with regard to prediction of RFP by sNfL at Baseline according to quartiles based on patient age at sampling (<26 years (grey line), 26-31 years (light green line), 32-42 years (dark green line), >42 years (black line)). f) Kaplan-Meier survival analysis for occurrence of RFP in patients with sNfL at baseline ≥7.3 pg/ml (green line) or <7.3 pg/ml (blue line), logrank test p=0.0135. Patients with sNfL ≥7.3 pg/ml at baseline suffer a 190% increased risk of experiencing RFP at follow-up (Hazard Ratio 2.90, 95%CI 1.19-7.03, p=0.009).

Fig. 3

Baseline sNfL level discriminates patients with SPMS conversion at follow-up a) Patients converting to secondary progressive multiple sclerosis (SPMS) showed similar NfL levels at baseline (conversion 9.2 pg/ml (interquartile range (IQR) 4.5-13.7) vs no conversion 7.9 pg/ml (IQR 4.7-12.9), p=0.657) and an increase in sNfL levels at follow up (conversion 10.4 pg/ml (IQR 6.9-17.6) vs. no conversion 6.9 pg/ml (IQR 5.0-9.2), p<0.001). b) NfL increase at follow-up compared to baseline NfL levels occurred in 70.4% of patients suffering from SPMS-conversion (19 of 27 patients) and in 47.3% of patients without SPMS-conversion (80 out of 169 patients, Chi²-test, p=0.026) supporting the use of sNfL level at FU in case clinical SPMS conversion is suspected. c) Machine learning by support vector machine assessed the accuracy of sNfL FU/BL ratio for identification of patients with SPMS transition. The following covariates were considered: EDSS at BL, Number of gadolinium-enhancing lesions at BL, Number of T2-hyperintense lesions at BL, Age at BL, Disease Duration and Relapses within the last 12 months prior to FU; grey bar). In addition, the accuracy for identification of SPMS-transition of a combination of EDSS at BL + T2-hyperintense lesion number at BL (green bar, without covariate adjustment) and a combination of EDSS at BL + T2-hyperintense lesion number at BL + sNfL FU/BL ratio (blue bar) are also shown. d) A ratio of sNfL at FU divided by sNfL at BL (sNfL FU/BL ratio) was used to draw the Area under the receiver operating curve (ROC-AUC) with regard to discrimination between patients with and without conversion into SPMS at follow-up (AUC 0.651, 95%CI 0.580-0.717, p=0.007). A cut-off above 19% NfL increase at follow-up compared to Baseline levels was determined by Youden's index. e,f) sNfL values at BL and FU in patients with inflammation (Inflam.; new T2-hyperintense lesions, gadolinium-enhancing lesions, clinical relapse) and progression (Progr.; EDSS-progression, MRI signs of atrophy) in an age- and gadolinium-matched subgroup of NaloMS. e) sNfL at BL: Bonferroni-correction: No inflammation no progression, 5.3 (3.0-8.3) pg/ml; inflammation only, 7.1 (4.4-12.2); progression only, 13.8 (5.1-29.0); inflammation and progression, 12.3 (5.2-18.5). p stable vs. inflammation and progression =0.004, stable vs. progression only p=0.004. f) sNfL at FU: Bonferroni-correction: No inflammation no progression, 4.8 (3.0-6.1); inflammation only, 7.1 (4.5-9.6); progression only, 6.0 (3.9-10.1); inflammation and progression, 7.9 (4.5-11.3). p for inflammation and progression vs. stable =0.001, inflammation only vs. stable p=0.039.