Developing an Explainable Prognostic Model for Acute Ischemic Stroke: Combining Clinical and Inflammatory Biomarkers...

Area of Science:

• Neurology
• Biomarkers
• Machine Learning

Background:

• Accurate prognosis prediction for acute cerebral infarction (ACI) is vital for patient management.
• Existing models often lack integration of clinical and biological indicators for precise predictions.

Purpose of the Study:

• To develop and validate a predictive model for ACI prognosis.
• Integrate clinical assessments and inflammatory biomarkers for improved accuracy and interpretability.

Main Methods:

• Retrospective cohort study of 1,017 ACI patients.
• Developed and evaluated 12 machine learning models using a training/validation split.
• Employed Shapley Additive exPlanations (SHAP) for feature interpretability.

Main Results:

• Logistic Regression model identified NIHSS at 24h, NIHSS change, D-dimer, neutrophil count, lymphocyte percentage at 24h, and length of stay as key predictors.
• NIHSS at 24h proved a critical early prognostic indicator.
• Inflammatory markers significantly improved predictive performance.

Conclusions:

• A robust, interpretable model for ACI prognosis was developed by integrating clinical and inflammatory biomarkers.
• The model highlights the prognostic value of early assessment and inflammatory markers for personalized treatment.
• Future research should focus on multi-center validation and incorporating additional variables.