RGX Ensemble Model for Advanced Prediction of Mortality Outcomes in Stroke Patients

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Prediction Models

Background:

• Stroke outcome prediction is challenging due to a lack of models quantifying clinical variable impact on survival.
• Existing methods struggle to handle missing data in clinical datasets, hindering accurate prognosis.

Purpose of the Study:

• To develop and validate a comprehensive model (RGX) for predicting stroke patient mortality.
• To improve personalized treatment planning by identifying key prognostic indicators.
• To enhance the accuracy of stroke mortality prediction using machine learning and explainable AI.

Main Methods:

• Developed and evaluated multiple machine learning models for stroke patient mortality prediction.
• Integrated Shapley Additive Explanations (SHAP) to elucidate the influence of risk factors on predictions.
• Assessed model performance using metrics like area under the curve, accuracy, and specificity.

Main Results:

• The RGX model achieved 92.18% accuracy on complete datasets, surpassing state-of-the-art models by 11.38%.
• RGX maintained strong predictive performance (84.62% accuracy) even with significant missing data.
• SHAP analysis provided insights into the contribution of various patient indicators to mortality prediction.

Conclusions:

• The RGX ensemble model offers a highly accurate tool for clinicians in predicting stroke patient survival.
• This approach advances the understanding of factors influencing stroke prognosis, supporting precision medicine.
• The model's robustness with missing data makes it valuable for real-world clinical applications.