Functional Outcome Prediction in Ischemic Stroke: A Comparison of Machine Learning Algorithms and Regression Models
Shakiru A Alaka1, Bijoy K Menon1,2,3, Anita Brobbey1
1Department of Community Health Sciences, O'Brien Institute for Public Health, University of Calgary, Calgary, AB, Canada.
Frontiers in Neurology
|September 28, 2020
Summary
Machine learning and logistic regression models show similar accuracy in predicting functional outcomes for stroke patients after endovascular treatment. These models help assess stroke-related functional impairment risk.
Area of Science:
- Neurology
- Medical Informatics
- Biostatistics
Background:
- Stroke-related functional risk scores are crucial for predicting patient outcomes post-stroke.
- Accurate prediction aids in treatment planning and patient management.
Purpose of the Study:
- To evaluate the predictive accuracy of machine learning (ML) algorithms versus traditional regression models.
- To predict functional outcomes in acute ischemic stroke patients undergoing endovascular treatment.
Main Methods:
- Utilized data from 614 ischemic stroke patients in the PROVE-IT study.
- Compared various ML models (RF, CART, SVM, etc.) and logistic regression.
- Validated models internally and externally using the INTERRSeCT cohort, assessing AUC, MCC, and Brier scores.
Main Results:
- 40.5% of patients experienced 90-day functional impairment (modified Rankin Scale > 2).
- Both ML and logistic regression models demonstrated comparable predictive accuracy.
- Internal validation showed AUCs of 0.65-0.72, external validation showed 0.66-0.71.
Conclusions:
- Machine learning algorithms and logistic regression offer similar predictive performance for stroke-related functional impairment.
- These findings support the use of both approaches for risk stratification in stroke patients.
