Glucose Prediction Funnel Using Multi-Horizon Alert Logic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing glucose prediction models for hypoglycemic and hyperglycemic events lack individualization and effective alarm-raising strategies, leading to inconsistent accuracy and efficacy in predicting and managing blood glucose levels.
Innovation Solution
A machine-learning-based approach using a linear prediction model and a rule-based alarm-raising strategy that considers multiple prediction horizons, incorporating clinical impact and individual variability, to improve hypoglycemic and hyperglycemic event prediction and alert systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single prediction horizon is used in glucose prediction models, then the model complexity is reduced, but the prediction accuracy and clinical effectiveness deteriorate due to inability to capture individual variability and multiple time scales
Solution Approach 1:
The patent segments the prediction task into multiple prediction horizons (e.g., short-term, medium-term, long-term forecasts) rather than using a single prediction model. This allows the system to capture different temporal patterns and individual variability at multiple time scales, improving prediction accuracy without requiring an overly complex single-model solution
Solution Approach 2:
The patent adds the dimension of multiple prediction horizons to the prediction framework, transforming a single-timepoint prediction into a multi-horizon prediction system. This dimensional expansion enables the model to account for different temporal dynamics and individual patient variability across time scales
2Device complexity
If standard alarm-raising strategies are used without considering clinical impact, then the alarm system is simpler to implement, but the reliability of alerts deteriorates due to false positives and missed critical events
Solution Approach 1:
The patent changes the parameters used for alarm triggering by incorporating clinically-relevant features and multiple prediction horizons into the alarm-raising logic. Instead of using simple threshold-based alarms, the system evaluates predictions across different time horizons and clinical contexts, improving alert reliability while maintaining reasonable system complexity
Solution Approach 2:
The patent implements feedback mechanisms where prediction results across multiple horizons inform alarm decisions, and alarm outcomes can refine future predictions. This feedback loop allows the system to learn from clinical outcomes and improve both prediction accuracy and alarm reliability over time
3Ease of manufacture
If individualization is not incorporated into glucose prediction models, then the model is easier to implement, but the adaptability to different patients and conditions deteriorates
Solution Approach 1:
The patent creates a universal prediction framework that can adapt to individual patients through multiple prediction horizons and clinically-relevant features. The multi-horizon approach serves multiple functions: capturing short-term dynamics, predicting longer-term trends, and adapting to individual patient variability, all within a unified system that maintains implementation feasibility
Data Source
AI summary
Certain aspects of the present disclosure relate to methods and systems for providing decision support around glucose management for patients with diabetes. Time-varying inputs including blood glucose, meal intake information, and amount of infused insulin are processed using a machine learning model to obtain predicted glucose levels for a plurality of prediction horizons and uncertainties for the predictions. A confidence interval is generated for each prediction and the confidence intervals are compared to hypo- and hyperglycemic thresholds. If a confidence interval is entirely below or entirely above the hypo- and hyperglycemic thresholds, respectively, then a decision support output is provided.