Predicting the likelihood of readmission in patients with ischemic stroke: An explainable machine learning approach...

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Ischemic stroke impacts millions globally, with high readmission rates necessitating improved patient management.
• Predicting and preventing 90-day readmissions is crucial for enhancing stroke survivor quality of life.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting 90-day readmission in ischemic stroke patients.
• To identify key predictors of stroke readmission using electronic health record data.

Main Methods:

• Retrospective analysis of 1,136 ischemic stroke patients' electronic medical records.
• Utilized six machine learning and three deep learning models with synthetic minority over-sampling technique.
• Employed SHapley Additive exPlanations (SHAP) for feature importance interpretation.

Main Results:

• 17.2% of patients were readmitted within 90 days; males showed higher readmission rates.
• The LightGBM model achieved an area under the curve of 0.94 for predicting readmission.
• Key predictors included renal and metabolic variables: creatinine, blood urea nitrogen, calcium, sodium, and potassium.

Conclusions:

• Machine learning models effectively predict 90-day stroke readmission using Common Data Model (CDM) data.
• Findings support personalized post-discharge care strategies for stroke patients.
• The study provides a foundation for future multicenter research on stroke readmission prediction.