Diagnostic performance of machine-learning algorithms for sepsis prediction: An updated meta-analysis

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Diagnostics

Background:

• Early sepsis identification significantly enhances patient prognosis.
• Sepsis remains a critical global health challenge requiring improved diagnostic tools.
• Timely intervention is key to reducing sepsis-related mortality and morbidity.

Purpose of the Study:

• To systematically evaluate the diagnostic efficacy of machine-learning algorithms for sepsis prediction.
• To synthesize evidence from existing studies on the accuracy of AI in sepsis detection.
• To provide a comprehensive overview of machine learning's role in early sepsis identification.

Main Methods:

• A systematic literature search was performed across PubMed, Embase, and Cochrane databases up to December 2023.
• Studies predicting sepsis using machine learning were included, with data extracted on sensitivity, specificity, and AUC.
• A meta-analysis was conducted on 21 studies involving over 4 million patient records.

Main Results:

• The overall pooled sensitivity was 0.82, specificity 0.91, and AUC 0.94, indicating high diagnostic accuracy.
• Subgroup analysis showed strong performance in emergency departments (AUC 0.94) and ICUs (AUC 0.93).
• The findings highlight the robust performance of machine learning models across different clinical settings.

Conclusions:

• Machine-learning algorithms demonstrate excellent diagnostic accuracy for sepsis prediction.
• These algorithms show significant potential for clinical application in early sepsis detection.
• The findings support the integration of AI tools into clinical workflows for improved sepsis management.