Artificial intelligence based predictive models for early sepsis detection in intensive care units: a scoping review

Area of Science:

• Critical care medicine
• Medical informatics
• Artificial intelligence in healthcare

Background:

• Early sepsis detection in intensive care units (ICUs) is a significant clinical challenge.
• AI-based predictive models offer potential for early sepsis identification.
• Current AI models for sepsis detection exhibit heterogeneous clinical readiness and methodological rigor.

Purpose of the Study:

• To systematically review and critically analyze existing evidence on AI-based predictive models for early sepsis detection in ICUs.
• To synthesize findings on the development, validation, and performance of these AI models.

Main Methods:

• A comprehensive scoping review was performed adhering to the PRISMA ScR framework.
• Systematic searches of multiple databases identified studies on AI models for sepsis detection in adult ICUs.
• Data extraction focused on study design, data sources, model types, prediction timelines, validation methods, and performance metrics.

Main Results:

• Thirty-seven studies were included in the review.
• The majority of AI models utilized retrospective electronic health record data and machine learning techniques.
• External validation was limited, performance metrics varied significantly, and few studies addressed clinical implementation or interpretability.

Conclusions:

• AI models demonstrate potential for improving early sepsis detection in ICUs.
• Significant gaps exist in external validation, clinical integration, and real-world applicability of current AI models.
• Future research must emphasize methodological transparency and evaluation focused on practical implementation.