Machine Learning Models for Inpatient Glucose Prediction.

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Diabetes Management

Background:

• Hospital glucose management is complex due to dynamic patient factors like medication, renal function, and diet.
• Suboptimal glycemic control can lead to adverse patient outcomes.
• Machine learning (ML) offers potential solutions for predicting glucose trends.

Purpose of the Study:

• To review the clinical evidence supporting the use of ML-based models for predicting hospitalized patients' glucose trajectories.
• To assess the current state and future directions of ML in inpatient glycemic management.

Main Methods:

• Review of published literature on ML algorithms applied to glucose prediction in hospitalized patients.
• Analysis of studies utilizing data from continuous glucose monitors and electronic health records (EHRs).
• Evaluation of trends in predictive accuracy with increasing data and algorithmic complexity.

Main Results:

• ML models have demonstrated increasing predictive accuracy for glycemic outcomes in hospital settings.
• Validated ML algorithms exist for predicting hypoglycemia and hyperglycemia.
• Current models require prospective clinical validation and integration into EHR systems.

Conclusions:

• ML-based models show significant potential for improving inpatient glucose management and preventing adverse glycemic events.
• Further prospective studies are essential to evaluate the clinical effectiveness and safety of ML decision support tools.
• Integration of ML models into EHR systems is a critical next step for clinical implementation.