Early neurofilament light and glial fibrillary acidic protein levels improve predictive models of multiple sclerosis outcomes

Early risk-stratification in multiple sclerosis (MS) may impact treatment decisions. Current predictive models have identified that clinical and imaging characteristics of aggressive disease are associated with worse long-term outcomes. Serum biomarkers, neurofilament (sNfL) and glial fibrillary acidic protein (sGFAP), reflect subclinical disease activity through separate pathological processes and may contribute to predictive models of clinical and MRI outcomes.We conducted a retrospective analysis of the Comprehensive Longitudinal Investigation of Multiple Sclerosis at the Brigham and Women's Hospital (CLIMB study), where patients with multiple sclerosis are seen every 6 months and undergo Expanded Disability Status Scale (EDSS) assessment, have annual brain MRI scans where volumetric analysis is conducted to calculate T2-lesion volume (T2LV) and brain parenchymal fraction (BPF), and donate a yearly blood sample for subsequent analysis. We included patients with newly diagnosed relapsing-remitting MS and serum samples obtained at baseline visit and 1-year follow-up (both within 3 years of onset), and were assessed at 10-year follow-up. We measured sNfL and sGFAP by single molecule array at baseline visit and at 1-year follow-up. A predictive clinical model was developed using age, sex, Expanded Disability Status Scale (EDSS), pyramidal signs, relapse rate, and spinal cord lesions at first visit. The main outcome was odds of developing of secondary progressive (SP)MS at year 10. Secondary outcomes included 10-year EDSS, brain T2LV and BPF. We compared the goodness-of-fit of the predictive clinical model with and without sNfL and sGFAP at baseline and 1-year follow-up, for each outcome by area under the receiver operating characteristic curve (AUC) or R-squared.A total 144 patients with median MS onset at age 37.4 years (interquartile range: 29.4-45.4), 64% female, were included. SPMS developed in 25 (17.4%) patients. The AUC for the predictive clinical model without biomarker data was 0.73, which improved to 0.77 when both sNfL and sGFAP were included in the model (P = 0.021). In this model, higher baseline sGFAP associated with developing SPMS (OR=3.3 [95%CI:1.1,10.6], P = 0.04). Adding 1-year follow-up biomarker levels further improved the model fit (AUC = 0.79) but this change was not statistically significant (P = 0.15). Adding baseline biomarker data also improved the R-squared of clinical models for 10-year EDSS from 0.24 to 0.28 (P = 0.032), while additional 1-year follow-up levels did not. Baseline sGFAP was associated with 10-year EDSS (ß=0.58 [95%CI:0.00,1.16], P = 0.05). For MRI outcomes, baseline biomarker levels improved R-squared for T2LV from 0.12 to 0.27 (P<0.001), and BPF from 0.15 to 0.20 (P = 0.042). Adding 1-year follow-up biomarker data further improved T2LV to 0.33 (P = 0.0065) and BPF to 0.23 (P = 0.048). Baseline sNfL was associated with T2LV (ß=0.34 [95%CI:0.21,0.48], P<0.001) and 1-year follow-up sNfL with BPF (ß=-2.53% [95%CI:-4.18,-0.89], P = 0.003).Early biomarker levels modestly improve predictive models containing clinical and MRI variables. Worse clinical outcomes, SPMS and EDSS, are associated with higher sGFAP levels and worse MRI outcomes, T2LV and BPF, are associated with higher sNfL levels. Prospective study implementing these predictive models into clinical practice are needed to determine if early biomarker levels meaningfully impact clinical practice.