Serum Neurofilament Light Chain in Multiple Sclerosis: Superiority of Age- and BMI-Corrected Z Scores/Percentiles Over Absolute Cutoff Values for Prediction of Treatment Response

Objective: Prognostication of disease course and prediction of treatment response in multiple sclerosis is an unmet need. We compared the performance of serum neurofilament light chain Z scores (age- and BMI-adjusted) with absolute concentrations for the prediction of response to disease-modifying therapy. Methods: Observational cohort study including the first serum sample of participants after the start of fingolimod therapy. We estimated hazard ratios for future relapses comparing participants with high (upper quartile) versus lower neurofilament light chain levels, based on either absolute concentration or Z score cutoffs. We compared the prognostic/predictive performance of these two measures for the occurrence of new/enlarging T2w lesions in longitudinal MRI. Results: We included 447 participants (median [IQR] age, 41.3 [32.1–49.2] years; 65.1% female); median follow-up 8.3 years [6.0–10.3]. Participants with a high neurofilament light chain Z score (Z ≥ 1.2/88.5 percentile) were more likely to experience future relapses (HR: 1.80, 95% CI 1.27–2.54, p < 0.001) compared to those below this threshold while this dichotomy could not be demonstrated with absolute concentration cutoffs (≥ vs. < 10.8 pg/mL; HR: 0.94, 95% CI 0.64–1.38, p = 0.75). Furthermore, patients with upper quartile Z scores were associated with a higher incidence of new/enlarging T2w lesions compared with those below this threshold (OR: 1.88, 95% CI 1.31–2.70, p < 0.001); again, absolute concentration cutoffs failed to identify this risk (OR: 1.20, 95% CI 0.82–1.77, p = 0.34). These findings were confirmed when patients having started alternative oral treatments were also included (n = 713). Interpretation: Serum neurofilament light chain Z scores consistently outperformed absolute concentration cutoffs for prognostication of clinical/radiological disease activity and may facilitate individual prediction of treatment response.