Artificial Intelligence-Based Models for Prediction of Mortality in ICU Patients: A Scoping Review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Real-time clinical decision support systems can improve healthcare professionals' ability to predict patient mortality and recovery.
• Accurate patient condition assessment aids in informed resource allocation within healthcare systems.

Purpose of the Study:

• To conduct a scoping review analyzing Artificial Intelligence (AI) algorithms for death prediction in Intensive Care Unit (ICU) patient populations.
• To evaluate the performance and accuracy of various AI-driven mortality prediction models.

Main Methods:

• Systematic database searches were conducted in MEDLINE, Embase, and PubMed up to July 2022.
• Keywords included mortality, survival, ICU, and terminal care.
• Identified AI models' variables, characteristics, and performance (Area Under the Curve - AUC) were summarized and compared.

Main Results:

• An initial search yielded 8271 articles, narrowed down to 16 studies focusing on AI-based mortality prediction models in ICUs.
• AI models demonstrated higher accuracy in death prediction compared to traditional tools, with some achieving AUCs up to 92.9%.
• Model performance varied based on patient populations and specific medical conditions, with overall mortality rates ranging from 5% to over 60%.

Conclusions:

• AI-based models show variable performance in mortality prediction across different patient groups.
• Customizing AI models for specific patient populations and medical contexts is recommended to improve accuracy.
• Incorporating additional variables, such as genetic information, can further enhance the predictive power of these models.