Machine learning-based prediction model for long-term mortality after ischemic stroke
Hee-Soo Kim1, Seung-Bo Lee1, Changi Kim2
1Department of Medical Informatics, Keimyung University School of Medicine, Daegu, South Korea.
Scientific Reports
|June 21, 2026
Summary
Machine learning models predict long-term stroke survival more accurately than traditional scores. A Gradient Boosting Cox model identified key factors like age and NIH Stroke Scale score for improved patient outcome predictions.
Area of Science:
- Neurology
- Medical Informatics
- Biostatistics
Background:
- Stroke is a leading cause of mortality, necessitating accurate long-term survival predictions.
- Existing prediction methods are often short-term focused and data-intensive.
- There is a need for convenient, time-aware models for post-stroke mortality prediction.
Purpose of the Study:
- To develop and validate machine learning (ML) and deep learning (DL) models for predicting long-term mortality in acute stroke patients.
- To compare the performance of ML/DL models against traditional risk scores.
- To identify key predictive features for practical application.
Main Methods:
- Development of various ML and DL models using clinical data from 3,411 patients (developmental cohort).
- External validation using data from 502 patients (secondary cardiovascular center).
- Comparison of the best-performing model against the PREMISE score.
Main Results:
- The Gradient Boosting Cox Proportional Hazards model achieved the highest performance (C-index 0.785 internal, 0.845 external).
- The ML model significantly outperformed the conventional PREMISE score in the external dataset (C-index 0.845 vs. 0.783).
- Key predictors included age, National Institutes of Health Stroke Scale score, and hemoglobin levels.
Conclusions:
- A validated ML-based model offers superior accuracy for post-stroke survival prediction compared to existing scores.
- This model can improve individual patient prognostication and aid in medical resource allocation.
- The identified key features provide practical insights for clinical management.