Developing machine learning models to predict multi-class functional outcomes and death three months after stroke in...

Area of Science:

• Neurology
• Data Science
• Public Health

Background:

• Stroke is a leading cause of death and disability globally, imposing significant societal costs.
• Accurate stroke outcome prediction is crucial for optimizing healthcare resource allocation and improving patient care.
• Existing prediction methods may not fully leverage the potential of routinely collected data.

Purpose of the Study:

• To develop and evaluate supervised machine learning models for predicting stroke outcomes (functional dependence and death).
• To compare the performance and explainability of artificial neural networks (ANN), eXtreme Gradient Boosting (XGBoost), and support vector machines against traditional multinomial logistic regression (mLR).
• To utilize routinely-collected data from the Swedish Stroke Registry (Riksstroke) for prognostic modeling.

Main Methods:

• A prognostic study using data from 102,135 adult patients in the Swedish Stroke Registry (2015-2020).
• Predictive factors included demographics, pre-stroke status, risk factors, medications, acute care, stroke type, and severity.
• Models evaluated: Support Vector Machines, ANN, XGBoost, and mLR, with predictions explained using SHAP values.

Main Results:

• All models achieved similar overall accuracy (69%-70%).
• ANN and XGBoost significantly outperformed mLR in predicting functional dependence, evidenced by higher F1-scores (0.603 and 0.577, respectively, vs. 0.544 for mLR).
• Predictive factors identified were consistent across models and aligned with clinical knowledge.

Conclusions:

• ANN and XGBoost models offer a modest improvement in predicting stroke outcomes and provide better explainability compared to mLR.
• These advanced machine learning approaches, using routinely collected data, can enhance the prediction of functional dependence after stroke.
• Improved prediction accuracy is vital for effective planning and organization of acute stroke care and rehabilitation services.