Predicting Risk of Stroke From Lab Tests Using Machine Learning Algorithms: Development and Evaluation of Prediction...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Public Health

Background:

• Stroke is a leading cause of death and disability, imposing significant health and financial burdens.
• Health-related behaviors are key stroke risk factors, driving the need for effective prevention strategies.
• Existing stroke prediction models often use lifestyle or imaging data, but not laboratory test results.

Purpose of the Study:

• To develop and evaluate machine learning models for stroke prediction using laboratory test data.
• To investigate the efficacy of different data selection and machine learning techniques for this purpose.

Main Methods:

• Utilized National Health and Nutrition Examination Survey datasets.
• Compared three data selection methods: no resampling, imputation, and resampling.
• Evaluated four machine learning classifiers using six performance metrics.

Main Results:

• Machine learning models can effectively predict stroke from lab test data.
• The data resampling approach significantly improved model performance.
• The random forest algorithm achieved the highest performance, with an accuracy of 0.96 and AUC of 0.97.

Conclusions:

• A highly accurate and user-friendly predictive model for stroke was developed using lab test data.
• This approach offers a novel method for stroke risk assessment.
• Future research will focus on developing models for specific stroke types.