Nighttime Hypoglycemia Risk Prediction Using Glucose Event History
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to provide a robust and reliable prediction of hypoglycemic events, particularly at night, due to challenges in modeling and aggregating past glucose values effectively.
Innovation Solution
A computer-implemented method using machine learning input parameters, including recent glucose measurements, hypoglycemic event history, and insulin intake, to predict the risk of hypoglycemic events during a prediction time interval, utilizing a trained model to generate actionable outputs for users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models use historical glucose monitoring data to predict hypoglycemic events, then prediction reliability is improved, but model complexity and difficulty of implementation increase
Solution Approach 1:
The prediction model is segmented into distinct input parameters (recent glucose measurements, hypoglycemic event share value, and other historical data) that can be independently calculated and processed. This segmentation simplifies the overall model complexity while maintaining prediction reliability by allowing each parameter to be computed separately using straightforward algorithms.
Solution Approach 2:
The patent introduces intermediate calculated parameters (such as the hypoglycemic event share value derived from historical data) that act as mediators between raw historical glucose data and the final prediction output. These intermediaries simplify the relationship between input data and predictions, making the model easier to implement while preserving reliability.
2Measurement precision
If the prediction model incorporates multiple input parameters including recent glucose measurements and historical event data, then prediction accuracy is improved, but data processing requirements increase
Solution Approach 1:
The patent extracts only the most relevant features from historical glucose monitoring data, specifically focusing on recent glucose measurements (within 10 minutes) and the hypoglycemic event share value. This selective extraction maintains prediction accuracy by capturing critical information while significantly reducing the overall data processing requirements compared to using complete historical datasets.
3Object-affected harmful factors
If the system provides predictions before the prediction time interval to allow preventive measures, then user safety is improved, but prediction time requirements increase
Solution Approach 1:
The system performs preliminary calculations of the hypoglycemic event share value and other prediction parameters in advance, using historical data from previous prediction time intervals. This preliminary action enables the model to generate accurate predictions quickly when needed, improving user safety by providing advance warning while minimizing the actual prediction time required at the moment of need.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method for predicting a risk whether a hypoglycemic event will occur during a prediction time interval is disclosed. The method comprises: providing user values for a plurality of machine learning input parameters, the user values determined from glucose monitoring data with glucose measurement values detected for a user before a cut-off time, the plurality of machine learning input parameters comprising: at least one first glucose measurement value determined from glucose measurements values measured within at most 10 minutes before the cut-off time, and a hypoglycemic event share value indicative of a share of previous prediction time intervals with detected hypoglycemic events out of a total number of previous prediction time intervals before the cut-off time; predicting, based on the user values and using a machine learning model, a risk whether a hypoglycemic event will occur during a prediction time interval associated with assumed sleep of the user, the machine learning model generated based on the machine learning input parameters and using historical glucose monitoring data of a user population; and generating an output based on the predicted risk. Further, a data processing system, a computer program, a system, and a remote control are disclosed.