Machine Learning Models for Predicting Stroke-Associated Pneumonia: A Systematic Review and Meta-Analysis

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Clinical Prediction Models

Background:

• Stroke-associated pneumonia (SAP) is a significant post-stroke complication.
• Machine learning (ML) models are increasingly developed for SAP prediction.

Purpose of the Study:

• To systematically evaluate the predictive performance of ML, deep learning (DL), and neural network (NN) models for SAP.
• To provide pooled performance metrics for these predictive models.

Main Methods:

• Systematic literature search of PubMed, Embase, Scopus, and Web of Science.
• Meta-analysis of 27 studies using R to calculate pooled AUC, accuracy, sensitivity, specificity, and DOR.
• Analysis of model types (ML, DL, NN) and input data (clinical, imaging, etc.).

Main Results:

• Pooled AUC of 0.84 and pooled accuracy of 0.80 indicate strong predictive performance.
• Pooled sensitivity was 0.73 and specificity was 0.85.
• ML models, primarily using clinical data, showed promising results without significant differences between ischemic and hemorrhagic stroke subgroups.

Conclusions:

• ML-based models demonstrate significant potential for early SAP risk identification in clinical practice.
• Further external validation and integration into clinical workflows are necessary for widespread adoption.