Continuous glucose monitoring using machine learning models and IoT device data: A meta-analysis

Area of Science:

• Diabetes Management
• Artificial Intelligence in Healthcare
• Internet of Things (IoT)

Background:

• Machine learning (ML) offers various approaches for diabetes blood glucose management.
• Selecting appropriate ML algorithms is crucial for effective diabetes care.
• Integrating data from IoT devices enhances real-time blood glucose monitoring models.

Purpose of the Study:

• To evaluate the effectiveness of ML models in predicting blood glucose levels.
• To assess the impact of IoT device data integration on prediction accuracy.
• To compare the performance of different ML algorithms for blood glucose prediction.

Main Methods:

• Systematic literature search of electronic databases (2019-2023).
• Inclusion of studies with ML model derivation and performance metrics.
• Quality assessment using the Quality Assessment of Diagnostic Accuracy Studies tool.
• Comparison of ML models for blood glucose (BG) prediction across various prediction horizons (PHs).

Main Results:

• Ten studies were analyzed across 15, 30, 45, and 60-minute prediction horizons.
• Mean absolute Root Mean Square Error (RMSE) values ranged from 15.02 to 35.89 mg/dL.
• Random Forest algorithm demonstrated superior performance compared to other ML models.

Conclusions:

• Significant heterogeneity was observed across study subgroups.
• Increasing prediction horizons led to higher RMSE values for blood glucose prediction.
• Random Forest consistently showed the highest relative performance among evaluated ML models.