Clinical performance of machine-learning algorithms to predict intraoperative hypotension: a meta-analysis

Area of Science:

• Anesthesiology and Critical Care Medicine
• Artificial Intelligence in Healthcare
• Medical Informatics

Background:

• Intraoperative hypotension is a frequent adverse event during surgery, linked to increased postoperative complications.
• Predictive tools for hypotension can improve patient outcomes and surgical safety.
• Machine learning (ML) offers advanced capabilities for real-time prediction and detection of intraoperative hypotension.

Purpose of the Study:

• To evaluate the clinical performance and diagnostic accuracy of ML algorithms for predicting intraoperative hypotension.
• To synthesize evidence from observational studies on the effectiveness of ML in hypotension prediction.
• To identify randomized controlled trials (RCTs) investigating AI-guided intraoperative management.

Main Methods:

• A systematic literature search was performed across major databases (PubMed, Scopus, Web of Science, Google Scholar) up to August 2024.
• Two independent reviewers screened studies, focusing on observational accuracy studies and RCTs.
• Quantitative synthesis of diagnostic/prognostic accuracy metrics (sensitivity, specificity, AUROC) from included observational studies.

Main Results:

• Five observational studies with 42,509 participants were included.
• ML algorithms utilized electrocardiography and blood pressure data for prediction.
• Pooled analysis demonstrated acceptable performance: specificity 0.73, sensitivity 0.75, and AUROC 0.83.

Conclusions:

• ML algorithms exhibit acceptable accuracy and offer advantages in predicting intraoperative hypotension.
• The integration of ML tools in surgical settings holds promise for mitigating complications.
• Further research, including RCTs, is warranted to fully establish the clinical utility of AI in intraoperative management.