ABSTRACT NUMBER: ESOC2026OT199 EARLY PREDICTION OF POST-STROKE COGNITIVE IMPAIRMENT USING AN INTERPRETABLE MACHINE-LEARNING MODEL: AN ONGOING PROSPECTIVE COHORT STUDY

Abstract Background and aims Post-stroke cognitive impairment (PSCI) affects a substantial proportion of stroke survivors and is associated with poorer functional recovery and reduced quality of life. Despite its clinical relevance, early cognitive screening and structured risk stratification are not routinely integrated into stroke care pathways. Identifying patients at high risk for PSCI early after stroke may enable personalised follow-up and targeted rehabilitation strategies. Methods To develop and prospectively validate an interpretable machine-learning model for predicting PSCI at 3 months after stroke based on early clinical variables. Results This is an ongoing single-centre prospective observational cohort study enrolling consecutive adult patients with acute ischaemic or haemorrhagic stroke within 7 days of symptom onset. Baseline variables include demographics, vascular risk factors, stroke severity (NIHSS), lesion characteristics, acute treatment variables, and functional status (mRS). Cognitive assessment using the Montreal Cognitive Assessment (MoCA) is performed on days 5–7 and repeated at 3 months. Predictive modelling will include logistic regression and gradient boosting with internal validation and assessment of discrimination and calibration. Conclusions Primary outcome: PSCI at 3 months defined as MoCA <26 (education-adjusted). Secondary outcomes include association between early cognitive screening and 3-month functional outcome (mRS). Conflict of interest