Use of continuous glucose monitoring systems in pediatric patients in the perioperative environment: Challenges and machine learning opportunities

Table 6 Machine learning applications commonly used in continuous glucose monitors

AI technique	Predictive modeling	Pattern recognition	Event prediction
Methodology	(1) Supervised learning: Use of labeled data in training models to predict future glucose levels (Hemoglobin A1c); (2) Case based reasoning: Adapts solutions based on previous data history	Unsupervised learning: Use of unlabeled data in training models used to identify patterns and relationships in the data	Supervised learning: Uses datasets with specific outcomes (hypo and hyperglycemia) for event prediction
Algorithms used	(1) Linear regression, decision trees, random forests, neural networks; (2) Similarity measures, nearest neighbor	k-means clustering, isolation forests	support vector machines, logistic regression and deep learning (require more resources for efficiency)
Diabetic applications	Strength: Provides early alerts for high and low glucose readings. Weakness: Requires copious data points for accuracy	Strength: Detects glucose patterns and atypical readings. Weakness: Clustering can require interpretation, can result in false + especially when readings are varied	Strength: Prompts critical alerts in hypoglycemia. Weakness: May produce false positive alarms
Examples in diabetic care	Predictive modeling, self-management tools, retinal screening	Decision support, prediction models, self-management tools, retinal screening	Predictive modeling, patient self-management tools, retinal screening

AI: Artificial intelligence.