Serum neurofilament light chain as a biomarker of inflammatory activity in relapsing-remitting multiple sclerosis: a real-world longitudinal cohort study

Abstract

Abstract Objectives Serum neurofilament light chain (sNfL) reflects neuroaxonal injury in relapsing-remitting multiple sclerosis (RRMS), but its clinical utility is limited by high interindividual variability and confounders such as renal function. Population-based thresholds may therefore be suboptimal for longitudinal monitoring. The reference change value (RCV), an intraindividual approach based on biological variation, remains insufficiently validated in real-world settings. Methods In this retrospective cohort study, 967 sNfL determinations from 295 RRMS patients were analyzed (LUMIPULSE G600II). Measurements were matched to clinical relapse and MRI activity within ±90 days. Associations with log-sNfL were assessed using linear mixed-effects models. RCV and population-based threshold (PBT) were compared as factors associated with inflammatory activity using cluster-robust logistic regression and ROC analysis. Correlation with new MRI lesion count was also evaluated. Results The intraclass correlation coefficient was 0.536. RCV elevation showed a highly significant association with active MRI (OR 4.93, 95 % CI 2.49–9.77) and clinical relapse (OR 6.48, 95 % CI 2.75–15.25), whereas PBT elevation did not reach statistical significance for either outcome (active MRI: OR 1.52, p=0.290; clinical relapse: OR 2.38, p=0.073). Clinical relapse and MRI activity were independently associated with mean sNfL increases of +72.1 % (95 % CI +53.7 %, +92.7 %) and +24.5 % (95 % CI +13.7 %, +36.3 %), respectively. Renal impairment was a major confounder, with sNfL elevations of +16.0 % in eGFR G2 and +67.8 % in G3–G5. Negative predictive values exceeded 91 % across all markers and outcomes. sNfL correlated with new MRI lesion count (Spearman ρ=0.498, p<0.0001), with a Youden-optimal threshold of ≥4 lesions. Conclusions sNfL measured on the LUMIPULSE platform is a valid real-world biomarker of inflammatory activity in RRMS. Its high interindividual variability makes RCV-based intraindividual interpretation substantially more informative than population-based thresholding for longitudinal monitoring. The consistently high negative predictive value supports its use as a rule-out tool for active neuroinflammation and provides a rational basis for risk-adapted MRI surveillance in clinically stable patients.