Related Experiment Video
Updated: Jan 7, 2026
09:10
Determining the Functional Status of the Corticospinal Tract Within One Week of Stroke
Published on: February 22, 2020
9.1K
Artifical intelligence-powered delta-NIHSS-based model for predicting recurrence, disability and mortality after
Shiyao Cheng1,2,3, Yuandan Wei2, Huaguang Zheng1
1China National Clinical Research Center for Neurological Diseases, and the Department of Neurology, Beijing Tiantan Hospital, Capital Medical University, Beijing, 100070, China.
Eclinicalmedicine
|January 1, 2026
Summary
A new AI model, DISCO, accurately predicts post-stroke outcomes like recurrence, disability, and mortality using changes in NIHSS scores. This tool helps identify high-risk patients for timely interventions, improving stroke care.
Area of Science:
- Artificial Intelligence in Medicine
- Clinical Prediction Modeling
- Stroke Research
Background:
- Stroke is a major global cause of death and disability.
- Effective identification of high-risk patients is crucial for targeted interventions.
- Existing predictive models may lack clinical accessibility or explainability.
Purpose of the Study:
- To develop and validate a clinically accessible and explainable AI-powered predictive model for post-stroke composite outcomes.
- To identify patients at high risk for adverse events following a stroke.
- To utilize comprehensive data for robust prognostic modeling.
Main Methods:
- Utilized an extreme gradient boosting tree model with 309 variables from the China National Stroke Registry (CNSR-III).
- Developed and validated the model using internal and external cohorts, including the CHANCE-2 trial and CRCS-5 cohort.
- Assessed feature importance using Shapley values, focusing on delta-NIHSS (admission-discharge).
Main Results:
- The DISCO model, integrating 16 delta-NIHSS measures and other clinical variables, achieved high AUCs (e.g., 0.852 for 3-month disability in validation cohort 2).
- Delta-NIHSS (admission-discharge) was the strongest predictor for stroke recurrence, disability, and mortality.
- The top 1% highest-risk patients showed significantly elevated relative risks for adverse outcomes at 3 months.
Conclusions:
- The DISCO model demonstrates high accuracy, robustness, and explainability in predicting post-stroke outcomes.
- The predictive power of delta-NIHSS offers mechanistic insights and may guide future stroke treatment and rehabilitation.
- Further validation across diverse populations and mechanistic studies are warranted.
