CGM Metric Target Modeling for Multi-Metric Glycemic Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current blood glucose management systems lack universally agreed-upon international targets for glycemic control metrics, particularly for Type 1 diabetes, leading to trade-offs in achieving specific glycemic targets that affect other metrics, and there is a need to determine corresponding targets for continuous glucose monitoring (CGM) metrics to guide automated insulin delivery systems effectively.
Innovation Solution
Techniques involve training classifiers to classify CGM data samples, using receiver operating characteristic (ROC) curves and dimensionality reduction methods like principal component analysis (PCA) to determine appropriate targets for CGM metrics, maximizing sensitivity and specificity, and providing personalized or population-based targets for glycemic control systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a target value for one CGM metric is set to guide automated insulin delivery, then glycemic control for that specific metric is improved, but other CGM metrics may not be optimized due to lack of corresponding targets
Solution Approach 1:
The system establishes a framework where a single target value input for one CGM metric automatically generates corresponding target values for multiple other CGM metrics through classifier models. This multi-functional approach allows the automated insulin delivery system to simultaneously optimize multiple glycemic control metrics (such as TIR, TB70, TB54, TA180, TA250) based on a single user input, resolving the contradiction between specialized control accuracy and multi-metric adaptability
2Ease of operation
If automated insulin delivery systems use single-metric targets, then control simplicity is maintained, but trade-offs between multiple glycemic metrics cannot be balanced
Solution Approach 1:
The system enables self-service operation where the automated insulin delivery system automatically calculates and provides corresponding target values for multiple CGM metrics based on a single user-input target value. The classifier models perform the complex multi-metric optimization calculations autonomously, maintaining ease of operation for the user while ensuring reliable comprehensive glycemic control through systematic target value generation across all metrics
3Adaptability or versatility
If corresponding targets for multiple CGM metrics are determined using classifier models, then multi-metric optimization is achieved, but system complexity increases
Solution Approach 1:
The system applies preliminary action by pre-training classifier models (such as random forest, support vector machine, neural network) with ROC curve analysis to establish the relationships between different CGM metrics before actual use. This pre-computation of target value correspondences stores the complex analytical results in accessible formats, allowing the system to generate multi-metric targets efficiently during operation without performing complex ROC curve analysis in real-time, thus reducing operational complexity while maintaining adaptability
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed herein are techniques for blood glucose management. In one example, a processor-implemented method includes receiving an input of a target value of a first continuous glucose monitoring (CGM) metric, estimating a target value of at least a second CGM metric that corresponds to the target value of the first CGM metric, and providing the estimated target value of at least the second CGM metric to a user. In some examples, the processor-implemented method also includes determining that the estimated target value of at least the second CGM metric meets a predetermined criterion, and configuring an insulin delivery system based on the target value of the first CGM metric.