Blood Glucose Probability Prediction Using Kernel Density Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional spot monitoring blood glucose measurements provide limited insight into glucose dynamics and risks of adverse events due to sparse data, leading to a loss of resolution and inadequate understanding of problematic time periods for diabetes management.
Innovation Solution
A method and system using kernel density estimation and Bayes' rule to determine the probability of blood glucose values being in adverse ranges, applying different kernel bandwidths for hypoglycemia and hyperglycemia, enabling accurate prediction of adverse events and informing therapy adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If spot monitoring blood glucose measurements are used, then the measurement method is simple and easy to perform, but the data is sparse and provides limited insight into glucose dynamics
Solution Approach 1:
The patent introduces kernel density estimation as an intermediary statistical method that bridges sparse spot monitoring measurements and continuous glucose dynamics. The KDE algorithm processes the limited spot measurements and generates a continuous probability density function that represents glucose levels throughout the day, effectively mediating between the simple measurement approach and comprehensive glucose information
Solution Approach 2:
The patent transforms the parameter representation from discrete spot measurements to continuous probability density functions. By changing how glucose data is represented (from individual time-point values to continuous distributions), the system extracts more information from the same sparse measurements, enabling identification of problematic time periods without requiring more frequent measurements
2Reliability
If measurements are grouped into large time slots to obtain feasible statistics, then statistical analysis becomes possible, but resolution is lost
Solution Approach 1:
The patent replaces static time slot grouping with dynamic kernel density estimation that adapts to the actual distribution of glucose measurements. Instead of forcing measurements into fixed hourly or quarter-daily bins, the KDE method dynamically creates continuous probability distributions that reflect the true temporal patterns, maintaining both statistical reliability and temporal precision
3Measurement precision
If different kernel bandwidths are applied for hypoglycemia and hyperglycemia, then prediction accuracy for adverse events is improved, but computational complexity increases
Solution Approach 1:
The patent applies different kernel bandwidths for different glucose ranges (hypoglycemia vs. hyperglycemia), making the analysis locally optimized for each condition. This local quality approach recognizes that different glucose states require different levels of smoothing in the density estimation, improving prediction accuracy for adverse events while keeping the overall system relatively simple through targeted differentiation
Data Source
AI summary
A method of determining a probability that a patient's blood glucose value is in an adverse range at a prediction time. Spot monitoring blood glucose measurement data is provided and includes blood glucose measurement values and associated measurement times. The values include first and second sets assigned to first and second adverse ranges, respectively. A kernel density estimation and Bayes' rule are used to determine the probability of the blood glucose value of the patient being in the first and second adverse blood glucose ranges at the prediction time. In the kernel density estimation, a first kernel bandwidth is applied for all or some of the first blood glucose measurement values and a second kernel bandwidth different from the first kernel bandwidth is applied for all or some the second blood glucose measurement values. Output data is provided indicative of the prediction time and the probability at the prediction time.