Early detection of sepsis using machine learning algorithms: a systematic review and network meta-analysis

Area of Science:

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

Background:

• Machine learning (ML) is increasingly used in medicine, with significant interest in its application for sepsis prediction.
• Sepsis prediction is critical due to the 'golden hour' timeframe for intervention.
• This study evaluates factors affecting ML model efficacy in sepsis prediction for clinical optimization.

Purpose of the Study:

• To assess the efficacy of machine learning models for real-time sepsis prediction in adult patients.
• To compare the performance of ML models against traditional scoring systems.
• To identify factors influencing the accuracy of ML-based sepsis prediction.

Main Methods:

• A systematic literature search was conducted across multiple databases (Medline, PubMed, Google Scholar, CENTRAL) up to October 2023.
• A network meta-analysis (NMA) using the CINeMA approach compared ML models and traditional scoring systems.
• Meta-regression was employed to identify factors impacting model performance, with Area Under the Curve (AUC) as the primary outcome.

Main Results:

• Analysis of 73 articles with 457,932 patients and 256 models revealed a pooled AUC of 0.825 for ML models.
• ML models significantly outperformed traditional scoring systems in sepsis prediction.
• Neural Network and Decision Tree models exhibited the highest AUC metrics, with model type, dataset type, and prediction window identified as significant factors.

Conclusions:

• Machine learning models, particularly Neural Networks and Decision Trees, demonstrate superior performance in sepsis prediction.
• Model type and dataset characteristics are crucial for prediction accuracy.
• Standardized reporting and validation are essential for ML healthcare applications, with a call for broader clinical implementation.