Machine learning-based prediction of clinical outcomes after first-ever ischemic stroke

Area of Science:

• Neurology
• Medical Informatics
• Biostatistics

Background:

• Accurate prediction of clinical outcomes post-acute stroke is crucial for optimizing patient care and treatment strategies.
• First-ever ischemic stroke patients require precise prognostication for functional recovery, cognitive status, depression, and mortality.
• Advanced machine learning (ML) techniques offer a promising approach to enhance predictive accuracy in stroke patient management.

Purpose of the Study:

• To systematically compare ML models for predicting functional recovery, cognitive function, depression, and mortality in ischemic stroke patients.
• To identify the leading prognostic factors influencing these clinical outcomes.
• To validate the predictive capabilities of ML in the context of first-ever ischemic stroke.

Main Methods:

• Utilized 43 baseline features from 307 first-ever ischemic stroke patients (PROSpective Cohort with Incident Stroke Berlin study).
• Employed Support Vector Machine (linear and RBF kernels) and Gradient Boosting Classifier models with nested cross-validation.
• Identified key prognostic features using Shapley additive explanations (SHAP).

Main Results:

• ML models demonstrated significant predictive performance for modified Rankin Scale (mRS), Barthel Index (BI), Mini-Mental State Examination (MMSE), TICS-M, and CES-D at various time points.
• National Institutes of Health Stroke Scale (NIHSS) emerged as the top predictor for functional recovery outcomes.
• Educational attainment was identified as a significant predictor for cognitive function and depression.

Conclusions:

• Machine learning models can effectively predict clinical outcomes after first-ever ischemic stroke.
• Identified key prognostic factors, including NIHSS and education, that significantly contribute to outcome prediction.
• This ML-driven approach aids in optimizing stroke treatment and patient care planning.