Automated Insulin Delivery Systems

Predicting meal and exercise events for a user with diabetes using their historical glucose and insulin data to improve insulin delivery. The method involves calculating residuals between actual and predicted glucose values over time. Positive residual rates indicate a meal event and negative rates indicate exercise. This allows the insulin pump to respond appropriately to meals and exercise without manual input.

Source

The systems, devices, and methods described herein for providing medication dose guidance, particularly insulin dose guidance, for diabetes management. The dose guidance is based on analyzing a user's historical and real-time physiological data, like glucose levels, diet, activity, etc., to determine optimal insulin doses. The guidance can also account for factors like hypoglycemic risk, glucose dysregulation, and low alarm frequency. The goal is to provide personalized and safe insulin dose recommendations without requiring users to manually analyze their data.

Source

Determining optimal basal insulin infusion rates for diabetes management using a CGM system that accounts for sensor reliability. The method involves calculating an adjustment to the basal rate based on a current risk metric indicating hypo/hyper risk and a control-to-range algorithm with an aggressiveness parameter. This allows customizing the basal rate response to the sensor data quality. If sensor reliability is low, the aggressiveness parameter reduces the basal rate adjustment to prevent over-correction. Conversely, if sensor reliability is high, the parameter allows more aggressive adjustments for better glucose control.

Source

Method to accelerate rate of adaptivity of automated insulin delivery (AID) systems for people with diabetes who experience significant variations in insulin needs. The method involves calculating a modification coefficient using a ratio of current insulin variability to historical variability. This coefficient is used to generate a faster adaptive rate for calculating daily insulin requirements. This allows the AID system to more quickly respond to user's changing insulin needs, reducing time in hyper/hypoglycemia when they experience sudden spikes or drops in sensitivity.

Source

Robust method for optimizing insulin dosing to achieve target glucose levels in diabetes patients. It involves using historical autonomous glucose measurements and insulin infusions to calculate glycemic risk and insulin sensitivity factors for a patient. These factors are then used to generate a personalized basal rate titration plan and fasting glucose profile model. The plan guides insulin adjustments during future time periods to minimize risks and maintain target fasting glucose.

Source

Self-learning insulin dose supervision system for diabetes patients that uses machine learning to calculate personalized insulin doses based on a patient's blood sugar levels and meal parameters. The system acquires patient data, learns the relationship between blood sugar, carbs, insulin, and post-meal levels, and generates a dynamic medication adjustment model for that patient. It then calculates the pre-meal insulin dose using the model and alerts the patient if the calculated dose exceeds the doctor's limits or changes too much. This allows personalized insulin adjustment and monitoring beyond what a doctor can provide.

Source

Automatically determining the user's risk of hypoglycemia or hyperglycemia based on factors calculated using data available in an insulin pump or diabetes management system. The system analyzes factors like current glucose, insulin delivery history, and other datasets to calculate a risk. It provides real-time alerts to the user when the risk is indicated, along with recommendations for mitigating the risk. The risk calculation can also be adjusted based on factors like exercise to account for how activities affect glucose levels. Machine learning can further improve the risk calculation by weighting factors based on accuracy.

Source

A glucose level control system for diabetes management that can adapt insulin dosing based on long-acting insulin administration. The system determines a long-acting insulin dose recommendation based on fast-acting insulin during a first therapy period. During a second therapy period after long-acting insulin administration, it modifies fast-acting insulin dosing based on the recommendation. This enables optimization of overall insulin therapy when both fast-acting and long-acting insulins are used.

Source

A system and method for intelligent insulin dosing using a pen-based insulin delivery device. The system estimates insulin sensitivity adaptively to improve accuracy of insulin dose recommendations. It uses a Gaussian process model trained on patient data to predict blood sugar levels considering insulin sensitivity. A classifier based on physiological scenes like exercise, emotion, and ketosis relief further refines the prediction. This adaptive insulin sensitivity estimation significantly improves blood sugar prediction for better dosing decisions.

Source

Integrated diabetes management system that connects glucose monitoring devices, insulin pens, and other devices to a central reader and displays a unified view of glucose levels, insulin doses, and other data. The system aims to provide better diabetes management by allowing users to view and analyze glucose and insulin data in a more actionable and correlated way. It also enables automated data transfer from devices like insulin pens to the central reader. The system can display metrics like average glucose, low glucose events, insulin amounts, and carbohydrate intake over time. It can also provide alerts for missed doses or high glucose levels. The centralized view and analytics aim to provide more insights and enable better diabetes management compared to viewing device-specific logs separately.

Source

Customizing a glucose prediction model in an insulin delivery device to accurately predict glucose levels for users with insulin sensitivity or insulin insensitivity. The customization is done based on the user's past glucose readings using regression analysis to determine weightings for those readings. These weightings are then used to predict future glucose levels for the user. This customized prediction model provides more accurate glucose predictions for users with varying insulin sensitivity, improving insulin dosing decisions.

Source

CGM-based automated basal insulin titration for Type 2 Diabetes patients uses historical CGM data, basal insulin doses, hypoglycemia reports, and past recommendations to generate adjusted insulin doses that minimize hypoglycemia risk and avoid overdosing. The method involves estimating personalized dose-response models from CGM metrics, regularizing fits in the early days, incorporating glucose variability, and ensuring safe dose changes. By leveraging CGM's comprehensive view of glucose levels, this approach aims to improve insulin titration compared to fingerstick-based manual titration, reducing the risk of overdosing and minimizing hypoglycemia.

Source

An automated insulin dosing system for diabetes patients using subcutaneous insulin pumps operates via a central server that receives blood glucose data from glucose meters. The server calculates optimal insulin dosages based on aggregated glucose trends and transmits them to the patient's insulin pump. This system enables personalized insulin dosing based on historical glucose patterns rather than relying solely on current measurements.

Source

Managing diabetes using personalized machine learning algorithms and patient data to recommend and automatically administer insulin and carbohydrates. The system collects continuous glucose monitor (CGM) data, processes it, and uses machine learning to predict optimal treatments. It provides real-time updates to patients and caregivers and allows remote intervention. The algorithms are personalized based on factors like activities, foods, and weight. The system also facilitates communication between patients, caregivers, and providers.

Source

Apparatus to optimize insulin dosage regimen for diabetes patients between clinic visits. The device analyzes patient blood glucose data and adjusts insulin dosages to maintain future levels within a range. It uses algorithms to determine how much to vary long-acting, meal, and correction insulin based on past glucose trends. The device stores the initial insulin regimen set by the doctor, and then continually evaluates if adjustments are needed based on new glucose readings. It can also suggest immediate insulin corrections for high glucose readings.

Source

Allowing users of insulin pumps to customize insulin delivery parameters in a way that is easier and more intuitive compared to traditional methods. It presents a user interface with input devices for subjective parameters like "insulin need" that are interpreted by the pump's automatic insulin delivery algorithm. When the user inputs a subjective insulin need, it modifies a coefficient value used in the algorithm's calculations. This allows users to adjust a key parameter without having to understand the complex factors used by the algorithm. The pump then collects physiological data and determines insulin dosage based on the modified coefficient value and the collected data. This provides personalized insulin delivery that adapts to the user's specific needs and conditions.

Source

Self-adaptive insulin infusion method and system for closed-loop artificial pancreas that can complete insulin bolus infusion even if blood glucose detection signal is lost. The method involves calculating a blood glucose estimate based on the last detected value and derivatives. When conditions trigger, a corrected bolus is calculated and split into fast and slow infusions. This ensures insulin delivery during meals despite sensor loss.

Source

Automated blood glucose control system that uses a predictive algorithm with adaptive parameters to optimize insulin delivery based on subject-specific factors. The system calculates insulin doses using a pharmacokinetic (PK) model with control parameters that can change over time for each subject. These parameters, like absorption time constants, are adapted based on factors like insulin type, blood glucose level, subject characteristics, etc. The system uses sensors and user input to modify the PK model parameters to better match the subject's insulin response. This allows personalized insulin delivery that accounts for subject-specific factors to more accurately control blood glucose levels.

Source

Automated insulin delivery system that anticipates and responds to glycemic disturbances like meals and exercise using historical data. The system generates disturbance profiles from patient history that represent deviations from target glucose levels. Current patient data is compared to the profiles and probabilities are calculated to assess likelihood of matching. An anticipated disturbance profile is determined based on the probabilities. Insulin dosage is then calculated based on the anticipated disturbance profile. This allows the system to proactively adjust insulin delivery to mitigate glycemic excursions.

Source

Predicting blood glucose levels in people with Type 1 diabetes using a deep learning model trained on historical data. The model takes in inputs like meal intake, insulin doses, and baseline glucose levels to predict future glucose levels. This enables personalized insulin dosing schedules based on predicted glucose levels. The model uses a convolutional neural network (CNN) for feature extraction and a long short-term memory (LSTM) network for temporal dynamics. The goal is to accurately forecast glucose levels multiple steps ahead to better manage insulin administration for people with Type 1 diabetes.

Source