AMD Dose Scheduling With Near-Term Weighted Glucose Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated medicament delivery (AMD) systems face inaccuracies in predicting future blood glucose levels due to individual user variations, leading to sub-optimal medicament delivery recommendations, especially when analyte values vary rapidly or exhibit trends, and limited prediction horizons result in insufficient responses.
Innovation Solution
A computer-implemented method that generates multiple medicament dose sets, calculates predicted analyte values, applies a cost function weighing earlier doses more heavily, and selects the optimal dose set for delivery, using varying weight schemes based on prediction accuracy and user variability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If population-average models are used for prediction, then the system can provide automated medicament delivery recommendations, but prediction accuracy deteriorates over the prediction horizon due to individual user variations
Solution Approach 1:
The system dynamically adjusts the prediction horizon parameter based on prediction accuracy. When prediction accuracy deteriorates (as indicated by increasing error metrics or user variability), the system shortens the prediction horizon to focus on more reliable near-term predictions, thereby maintaining recommendation quality despite using population-average models
Solution Approach 2:
The prediction horizon is made dynamic rather than fixed. The system continuously evaluates prediction reliability and adjusts the horizon length in real-time, extending it when predictions are reliable and contracting it when individual variations cause accuracy to deteriorate, allowing the system to adapt to changing conditions while maintaining automation
2Device complexity
If a fixed prediction horizon is used, then the system operates simply, but it fails to respond adequately when analyte values vary rapidly or exhibit trends
Solution Approach 1:
The prediction horizon transforms from a static fixed value to a dynamic parameter that adjusts based on analyte behavior. When rapid variations or trends are detected in analyte values, the system automatically modifies the horizon length to ensure adequate response time, balancing operational simplicity with reliable adaptation to changing physiological conditions
3Device complexity
If equal weighting is applied to all predicted values, then the cost function is simple to compute, but later predictions closer to the horizon end contribute equally despite lower accuracy
Solution Approach 1:
The cost function applies different weights to different portions of the prediction horizon, creating local quality variations in how predictions are valued. Earlier, more reliable predictions receive higher weights while later, less reliable predictions receive lower weights, ensuring that the optimization process prioritizes accuracy in the most trustworthy time ranges without requiring complex computational structures
Data Source
Figure 1
Figure 2A~2B
Figure 3A~3C
AI summary
Exemplary embodiments relate to automated medicament delivery (AMD) devices. Such devices may measure the level of an analyte and deliver medicament with the intent of maintaining the analyte at a target level or in a target range. The described methods and apparatuses apply a model to evaluate an effect of proposed future medicament doses on predicted analyte levels. The proposed future medicament doses and predicted analyte levels may be supplied to a cost function to select which of the proposed future medicament dose schedules results in optimal control of the predicted analyte levels. The cost function may apply a weighting scheme that weighs predictions in the near- and/or medium-term future more than longer-term predictions.