Predicting 90-day prognosis for patients with stroke: a machine learning approach

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Stroke is a leading global cause of death and disability.
• Accurate prognosis prediction is crucial for effective stroke patient management.
• Current predictive methods may not fully capture post-discharge outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting 90-day stroke patient prognosis.
• To identify key factors influencing stroke recovery and functional outcomes.
• To enhance personalized care planning and preventive strategies for stroke survivors.

Main Methods:

• Utilized a large national multiethnic stroke registry (7,452 patients).
• Applied feature selection techniques (correlation, permutation importance).
• Evaluated six ML classifiers, including Random Forest (RF), for prediction.

Main Results:

• The RF model achieved high predictive accuracy (0.823) and discrimination (AUC 0.893).
• Key predictors identified: stroke type, hospital-acquired infections, admission location, and length of stay.
• Demonstrated the potential of ML in forecasting post-stroke functional status.

Conclusions:

• The developed RF model shows significant promise for predicting stroke prognosis.
• Enables tailored care plans and proactive interventions for stroke patients.
• Further prospective validation is recommended for clinical implementation.