Machine Learning Applications Within the Earlier Medicine Framework for Stroke: A Scoping Review

Area of Science:

• Neurology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• The Earlier Medicine framework emphasizes proactive and personalized stroke care across primary, secondary, and tertiary prevention levels.
• Machine learning (ML) offers potential for enhancing stroke research and clinical decision-making.
• A comprehensive understanding of ML applications in stroke is needed.

Purpose of the Study:

• To conduct a scoping review of machine learning applications in stroke research.
• To map the use of ML across primary, secondary, and tertiary stroke prevention.
• To identify common predictors, ML models, and challenges in the field.

Main Methods:

• Systematic literature search conducted in PubMed and Scopus databases.
• Adherence to PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines.
• Inclusion of 105 eligible studies focusing on ML in stroke research.

Main Results:

• The majority of studies concentrated on tertiary prevention (e.g., predicting mortality, complications, functional outcomes).
• Fewer studies addressed primary (risk prediction) or secondary (acute phase detection) prevention.
• Commonly used predictors included age, NIHSS score, glucose, stroke volume, BMI, hypertension, and diabetes.
• Traditional ML models (logistic regression, random forest, SVM) are dominant, with a rise in Ensemble/Hybrid and Deep Learning models since 2021.

Conclusions:

• Machine learning shows significant promise for improving stroke prediction and patient outcomes.
• Challenges include data heterogeneity, lack of model transparency, and the need for robust external validation.
• Further research is required to overcome these limitations and facilitate clinical translation.