DISCO: A Delta-NIHSS-Based Machine Learning Model for Predicting Recurrence, Disability, and Mortality Following Acute Ischemic Stroke

BackgroundAn accurate, robust, clinically accessible, and explainable predictive model for post-stroke composite outcomes could identify high-risk patients for targeted interventions. However, such a model is currently lacking. This study leverages artificial intelligence to develop and validate a predictive model for post-stroke outcomes at three months and over five years, leveraging comprehensive data from the third China National Stroke Registry (CNSR-III), one of Chinas largest nationwide multi-center ischemic stroke registries with five-year follow-up and the CHANCE-2 trial, a genotype-guided dual anti-platelet therapy trial in China. MethodsWe evaluated 309 hospitalization variables, including baseline characteristics, medical history, hospitalization data, biomarkers, geographical factors, NIHSS/mRS scores, and stroke polygenic risk scores (PRS), using an extreme gradient boosting tree model. Feature importance was assessed via Shapley values. Primary outcomes were three-month stroke recurrence (5.6%), disability (mRS > 2, 13.75%), and mortality (1.18%). Secondary outcomes were assessed at six additional time points over five years. A nested cross-validation scheme was employed for feature selection and internal validation in 80% of patients (n=11,313) from CNSR-III cohort. External validation of the model was performed in the remaining 20% patients (n=2,627) from CNSR-III cohort and in CHANCE-2 trail (n=5,158). ResultsGlobal and domain-specific delta-NIHSS(admission-discharge) emerged as the strongest predictor of stroke recurrence, disability and mortality. The Delta-NIHSS-Based Predictor for Post-Stroke Composite Outcomes (DISCO) model, integrating 16 delta-NIHSS(admission-to-discharge) and 8 clinically accessible variables, achieved AUCs > 0.8 for recurrence and disability and > 0.9 for mortality at three months. The highest-risk 1% patients exhibited a >10-fold relative risk (RR) for recurrence, >30-fold RR for disability, and >100-fold RR for mortality at three months. The DISCO model is clinically accessible at http://www.discosysu.cn. ConclusionThe DISCO model, incorporating 24 readily obtainable clinical variables, demonstrates high accuracy, robustness, clinical accessibility, and explainability in predicting post-stroke outcomes. The predictive strength of delta-NIHSS(admission-discharge) provides mechanistic insights into stroke outcomes and informs future acute stroke treatment and rehabilitation strategies.