Predicting Ischemic Stroke Patients to Transfer for Endovascular Thrombectomy Using Machine Learning: A Case Study.

Area of Science:

• Neurology
• Data Science
• Health Services Research

Background:

• Endovascular thrombectomy (EVT) is a critical treatment for ischemic stroke with large vessel occlusion.
• Patients often require transfer from local hospitals to specialized urban centers for EVT.
• Current patient selection for transfer can lead to a significant number of futile transfers, impacting resource allocation and patient outcomes.

Purpose of the Study:

• To evaluate the potential of machine learning (ML) models in improving the accuracy of patient selection for EVT transfer.
• To assess if ML can serve as a decision support tool to minimize futile transfers.

Main Methods:

• Retrospective analysis of ischemic stroke patient data from Nova Scotia, Canada (2018-2022).
• Application of four supervised binary classification ML algorithms: logistic regression, decision tree, random forest, and support vector machine, plus an ensemble method.
• Model performance evaluated using accuracy, futile transfer rate, and false negative rate, with k-nearest neighbors for missing data imputation and five-fold cross-validation.

Main Results:

• Analysis included 93 patients after exclusions from an initial cohort of 5156.
• The decision tree and random forest models demonstrated higher accuracy (79% and 74%, respectively).
• The random forest model achieved a 0% futile transfer rate with a 5.37% false negative rate, while the decision tree had an 18.9% futile transfer rate and a 4.3% false negative rate.

Conclusions:

• Machine learning models hold potential for reducing futile transfer rates in the context of EVT for stroke.
• Further research with larger, diverse datasets is necessary to validate these findings and enable broader clinical implementation.