Integrative Multidimensional Machine Learning Models for Stroke Prognosis: Age-Stratified and History Engineered
Gawon Lee1,2,3, Sunyoung Kwon1, Seung-Ho Shin4
1Division of DataScience, Hallym University, Chuncheon 24252, Republic of Korea.
Diagnostics (Basel, Switzerland)
|May 13, 2026
Summary
Machine learning models accurately predict stroke patient mortality by integrating vital signs, lab results, and medical history. Pulse rate is a key predictor, especially in younger patients, enabling personalized stroke care.
Area of Science:
- Neurology
- Data Science
- Biostatistics
Background:
- Stroke is a major cause of death and disability globally.
- Accurate prognosis is crucial for effective stroke management.
- Previous models often neglected patient history and age-specific risks.
Purpose of the Study:
- Develop and validate machine learning models for stroke mortality prediction.
- Incorporate diverse patient data including vitals, labs, demographics, and history.
- Explore subgroup-specific predictive factors.
Main Methods:
- Retrospective analysis of 1780 stroke patients (2018-2023).
- Utilized Random Forest models with original and binarized clinical data.
- Assessed model performance using AUC and variable importance (Mean Decrease Gini, SHAP).
Main Results:
- Highest accuracy (AUC 0.995) achieved for patients under 60 using binarized data.
- Pulse rate identified as the most significant predictor across models.
- Platelet count and diastolic blood pressure also important; higher pulse rate linked to increased mortality risk.
Conclusions:
- Integrating clinical, demographic, and historical data improves stroke mortality prediction accuracy and interpretability.
- Stratified modeling and monitoring pulse rate are vital for precision stroke care.