Evaluating Predictive Performance of Machine Learning Algorithms That Integrate Routine Clinical Variables With...

Area of Science:

• Neurology
• Medical Informatics
• Biostatistics

Background:

• Stroke recurrence poses a significant challenge in patient management.
• Traditional prediction models for stroke recurrence often lack sufficient accuracy.
• Integrating diverse data types is crucial for enhancing predictive capabilities.

Purpose of the Study:

• To compare the performance of various machine learning (ML) algorithms in predicting stroke recurrence risk.
• To evaluate ML models that combine routine clinical variables with imaging-derived features.
• To identify the optimal ML model for stroke recurrence prediction.

Main Methods:

• A retrospective cohort study of 350 ischemic stroke patients.
• Collected routine clinical data (age, gender, hypertension, diabetes) and imaging features (infarct size, location).
• Applied logistic regression, random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost) for model development and evaluation using AUC, sensitivity, specificity, and accuracy.

Main Results:

• The XGBoost model achieved the highest predictive performance with an Area Under the Curve (AUC) of 0.86.
• Random forest (0.82), SVM (0.78), and logistic regression (0.75) models followed.
• Key predictors identified were infarct size, history of hypertension, and fasting blood glucose levels.

Conclusions:

• Machine learning algorithms effectively predict stroke recurrence risk by integrating clinical and imaging data.
• The XGBoost model exhibits superior predictive performance compared to other evaluated ML algorithms.
• These findings support the development of individualized clinical decision-making for stroke patients.