Systematic review of Artificial Intelligence-based methods for glycemic control and risk prediction in intensive care units

Comprehensive Summary

The integration of artificial intelligence (AI) and machine learning (ML) within the ICU is gaining prominence, particularly regarding effective glucose monitoring. Maintaining patients within a euglycemic range (normal glucose levels) has consistently been shown to improve clinical outcomes. This study examines the rising prevalence and current performance of AI and ML in ICU glucose management and clinical decision-making. Evidence suggests that patients experience fewer instances of hyper- or hypoglycemia when utilizing new continuous glucose monitoring (CGM) and AI-guided administration. Advancements in technology have also improved the ability to anticipate glycemic fluctuations, facilitating more precise management. However, the review identifies several limitations in existing data, including a lack of data homogeneity, inflated quality metrics, non-generalizable datasets, and irregular sampling frequencies. The authors conclude that future prospective clinical validation is essential to enhance data quality and overall generalizability.

Outcomes and Implications

The application of AI and ML in glucose control is ushering in an era of improved patient outcomes in the ICU. Highly trained care teams who understand the nuances of tight glycemic control, paired with advanced monitoring technology, are positioned to further improve patient results. While AI-based monitoring has demonstrated real-world efficacy, rigorous quality oversight remains necessary. It is imperative that the medical community prioritizes the collection and distribution of organized, generalizable data to optimize the effectiveness of these models and improve future patient outcomes.