Machine learning algorithms in sepsis.

Area of Science:

• Medical Informatics
• Clinical Diagnostics
• Computational Biology

Background:

• Sepsis presents a significant global health burden with high mortality and morbidity.
• Early sepsis detection is challenging due to variable clinical signs.
• Machine learning (ML) integration into laboratory medicine offers potential for improved sepsis identification and outcome prediction.

Purpose of the Study:

• To comprehensively review current research on ML applications in laboratory diagnostics for sepsis.
• To assess the strengths and limitations of existing ML approaches for sepsis.
• To identify areas for improvement in ML model development and validation for clinical use.

Main Methods:

• Extensive literature search of PubMed and Scopus databases (keywords: Sepsis, Machine Learning, Laboratory) until January 2023.
• Two independent investigators screened and evaluated 135 articles, selecting 39 for inclusion.
• Studies were analyzed based on design, intent (diagnostic/prognostic), clinical setting, data, ML methods, and validation.

Main Results:

• The majority of included studies (30/39) focused on ML for sepsis diagnosis, with fewer (8/39) for prognosis.
• ML algorithms are being developed across diverse journals, indicating interdisciplinary interest.
• Significant variation exists in study designs, feature definitions, and validation methodologies.

Conclusions:

• ML holds considerable promise for enhancing early sepsis diagnosis through laboratory data.
• There is a critical need for standardized validation protocols and feature definitions for ML models in sepsis.
• Standardization is essential to ensure the reliability and clinical applicability of ML tools for sepsis management.