ABSTRACT NUMBER: ESOC2026A2083 EARLY PLASMA NFL PREDICTS COGNITIVE OUTCOMES 5 YEARS AFTER STROKE Authors: Jule Filler, Gilamo Atari Kazemi, Marco Duering, Michael Goertler, Silke Wunderlich, Matthias Endres, Inga Zerr, Gabor Petzold, Martin Dichgans, Steffen Tiedt Published: 2026 ABSTRACT AbstractBackground and aimsCognitive impairment is common after stroke, yet early risk stratification remains challenging. Blood-based biomarkers of neuroaxonal and astroglial injury, such as neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP), may help identify patients at high risk for cognitive impairment.MethodsIn a prospective multicentre cohort of dementia-free patients with ischaemic or haemorrhagic stroke (n = 558; median NIHSS 3 [IQR 1–5]; DEMDAS), plasma samples were collected at a median of 3 days post-stroke (IQR 2-5). NfL and GFAP were measured using single-molecule detection technology. Neuropsychological assessments were conducted at 6, 12, 36, and 60 months. The primary outcome was post-stroke cognitive impairment (PSCI; ≥1 domain <-1.5 SD) within five years. Secondary outcomes included global cognitive performance and domain-specific outcomes. Associations with longitudinal outcomes were assessed using mixed-effects models, adjusted for relevant confounders.ResultsHigher NfL was associated with greater white matter hyperintensity volume and infarct volume. NfL, but not GFAP, predicted PSCI over 5 years (OR = 1.57, 1.28–1.91, P < 0.0001; Figure, Panel A), lower global cognitive scores (Figure, Panel B), and poorer attention and executive function. Machine-learning models consistently ranked NfL among the top predictors of PSCI (Figure, Panel C), outperforming GFAP and conventional stroke-related factors such as infarct volume and NIHSS.ConclusionsPlasma NfL measured shortly after stroke robustly predicts long-term cognitive impairment, particularly in executive domains. It likely captures brain injury not fully explained by clinical or imaging markers and may improve early risk identification, targeted monitoring, and post-stroke trial enrichment.Conflict of interestThe authors declare no conflicts of interest.Figure 1 - belongs to Conclusions