Claims-Based Machine Learning Classifier of Modified Rankin Scale in Acute Ischemic Stroke.

Area of Science:

• Neurology
• Health Informatics
• Public Health

Background:

• Acute ischemic stroke severity is crucial for patient outcomes and research.
• The modified Rankin Scale (mRS) is a standard measure of stroke disability.
• Medicare claims data offers a large dataset for studying stroke populations.

Purpose of the Study:

• To develop and validate a classifier predicting acute ischemic stroke severity using Medicare claims.
• To utilize the modified Rankin Scale (mRS) at discharge as the outcome measure.
• To enhance stroke outcomes research and national surveillance capabilities.

Main Methods:

• A multistate study linked Paul Coverdell National Acute Stroke Program data with Medicare claims.
• Lasso-penalized logistic regression was used to develop a binary classifier for mRS outcomes.
• Performance was assessed using area under the receiver operator characteristic curve (AUROC), precision-recall, sensitivity, and specificity.

Main Results:

• The study analyzed data from 68,636 Medicare beneficiaries aged 65+ hospitalized for acute ischemic stroke.
• The classifier achieved an AUROC of 0.86 and a precision-recall area under the curve of 0.90.
• Sensitivity was 0.81 and specificity was 0.73, indicating strong predictive performance.

Conclusions:

• A claims-based classifier demonstrated excellent performance for modified Rankin Scale classification in Medicare beneficiaries with acute ischemic stroke.
• The developed classifier shows potential for improving stroke outcomes research.
• This tool can support the development of national surveillance for acute ischemic stroke.