Reducing hyperglycemia using online disorder suppression and prediction system
A system using disorder curves and prediction models to determine insulin dosage addresses the absorption delay issue in conventional systems, effectively managing blood glucose levels and reducing hyperglycemia and hypoglycemia risks in type 1 diabetes patients.
Patent Information
- Application Number
- CN202280076197.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2021-09-15
- Filing Date
- 2022-09-15
- Publication Date
- 2024-12-03
AI Technical Summary
Conventional automated insulin dosing systems struggle to effectively prevent hyperglycemia after meals and exercise in type 1 diabetes patients due to the inherent absorption delay of subcutaneous insulin analogs, requiring feed-forward control that is not adequately addressed.
A system that generates disorder curves based on historical patient data, applies a prediction model to assess the likelihood of blood glucose deviations, and determines insulin dosage to manage insulin levels, incorporating a processor to output signals for insulin management devices.
The system improves blood glucose control by reducing postprandial hyperglycemia and hypoglycemia risks, enhancing the effectiveness of insulin delivery systems in managing blood glucose fluctuations.
Smart Images

Figure CN119072754A_ABST
Abstract
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This patent application is related to and claims priority to U.S. Provisional Application No. 63 / 244,368, filed on September 15, 2021, the entire contents of which are incorporated herein by reference. Technical Field
[0002] Embodiments relate to systems and methods for informing, determining, or controlling insulin dosage based on disturbance profiles associated with disturbed patterns of blood glucose levels caused by physical activity, mealtime activity, and the like. Background Art
[0003] For people with type 1 diabetes (T1D), preventing hyperglycemia after meals, exercise, etc. remains a persistent problem, even with the use of conventional automated insulin dosing (AID) systems that variably deliver insulin based on internal closed-loop algorithms. For people with T1D, the unnecessary rise in blood glucose (BG) after meals, exercise, etc. is primarily a result of the dynamics associated with subcutaneous insulin injections, carbohydrate ingestion, exercise, etc. Currently available subcutaneous insulin analogs have slower time constants than consumed carbohydrates, so feed-forward control is required to prevent hyperglycemia. Anticipatory and automated drug injection in an advanced control system framework has been shown to compensate for the inherent absorption delay of current insulin analogs. However, such systems can be improved. Summary of the invention
[0004] Exemplary embodiments relate to a system for notifying, determining, or controlling insulin dosage. The system may include a processor. The system may include a memory including instructions that cause the processor to perform functions associated with embodiments of the methods disclosed herein. The instructions may cause the processor to generate a plurality of disorder curves, each disorder curve being a data representation of historical patient data based on deviations from a threshold blood glucose level. The instructions may cause the processor to receive current patient data. The current patient data and the historical patient data include any data indicating patient characteristics. The instructions may cause the processor to apply a prediction model. The processor is configured by the instructions to compare the current patient data with at least one disorder curve. The processor is configured by the instructions to assess the likelihood that one of the plurality of disorder curves is an expected disorder curve based on a probability analysis, the expected disorder curve being a disorder curve that matches the current patient data based on the probability analysis. The instructions may cause the processor to determine an insulin dosage based on the expected disorder curve. The instructions may cause the processor to output a signal representing the insulin dosage to a device configured to monitor, influence, and / or manage a patient's insulin level.
[0005] Exemplary embodiments relate to a computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform functions associated with embodiments of the methods disclosed herein. The instructions may cause the processor to generate a plurality of disorder curves, each disorder curve being a data representation of historical patient data based on deviations from a threshold blood glucose level. The instructions may cause the processor to receive current patient data. The current patient data and the historical patient data include any data indicating patient characteristics. The instructions may cause the processor to apply a prediction model. The processor is configured by the instructions to compare the current patient data with at least one disorder curve. The processor is configured by the instructions to assess the possibility that one of the disorder curves in the plurality of disorder curves is an expected disorder curve based on a probability analysis, the expected disorder curve being a disorder curve that matches the current patient data determined based on the probability analysis. The instructions may cause the processor to determine an insulin dose based on the expected disorder curve. The instructions may cause the processor to output a signal representing the insulin dose to a device configured to monitor, influence and / or manage the patient's insulin level.
[0006] Exemplary embodiments relate to a method for notifying, determining or controlling insulin dosage. The method includes generating a plurality of disorder curves, each disorder curve being a data representation of historical patient data related to deviations from a threshold blood glucose level. The method includes receiving current patient data. The current patient data and the historical patient data include any data indicating patient characteristics. The method includes applying a predictive model, wherein the current patient data is compared with at least one disorder curve, and based on a probability analysis, assessing the likelihood that one of the plurality of disorder curves is an expected disorder curve, the expected disorder curve being a disorder curve that matches the current patient data based on the probability analysis. The method includes determining an insulin dosage based on the expected disorder curve. The method includes outputting a signal representing the insulin dosage to a device configured to monitor, influence and / or manage a patient's insulin level. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] Other features and advantages of the present disclosure will become more apparent when the following detailed description is read in conjunction with the accompanying drawings, in which like elements are represented by like numerals, wherein:
[0008] Figure 1 An exemplary system architecture for implementing embodiments of the method is shown.
[0009] Figure 2 An exemplary schematic diagram of the tested control strategy is shown.
[0010] Figure 3 Shown are average continuous glucose monitoring (CGM) data for one subject.
[0011] Figure 4 Insulin data for one subject is shown.
[0012] Figure 5 An estimate of the disorder characteristic d is shown.
[0013] Figure 6 The effect of a weighting function applied to the curve values is shown.
[0014] Figure 7 The disorder curve of a representative subject is shown.
[0015] Figure 8 A timeline of outpatient study visits is shown.
[0016] Fig. 9 The changes in CGM of the control group (dark grey) and the experimental group (light grey) are shown.
[0017] Fig.10Shown are box plot representations of the post-lunch CGM slope trends observed in the first 2 hours (left) and second 2 hours (right) after lunch values in the study visit.
[0018] Fig.11 A block diagram illustrating an example of a machine upon which one or more aspects of embodiments of the invention may be implemented. DETAILED DESCRIPTION
[0019] refer to Figure 1 and Fig.11 , embodiments may relate to a system 1000 for notifying, determining, or controlling insulin dosage. System 1000 may include a processor 1102 configured to build and / or implement a prediction model. The system may include a memory 1104, a memory 1106, a memory 1122, which has instructions for causing the processor 1102 to perform any method steps disclosed herein. Processor 1102 may be any of the processors 1102 disclosed herein. Processor 1102 may be part of or communicate with machine 1100 (logic device, one or more components, circuits (e.g., modules) or structures). Processor 1102 may be hardware (e.g., processor, integrated circuit, central processing unit, microprocessor, core processor, computer device, etc.), firmware, software, etc., and the processor is configured to perform operations by executing instructions contained in algorithms, data processing program logic, artificial intelligence programming, automatic reasoning programming, etc. It should be noted that the processor 1102 used herein includes any one or combination of a graphics processing unit (GPU), a field programmable gate array (FPGA), a central processing unit (CPU), etc. The processor 1102 may include one or more processing modules. A processing module may be a software or firmware operating module configured to implement any method steps disclosed herein. The processing module may be embodied as software and stored in a memory, which is operably associated with the processor 1102. The processing module may be embodied as a web application, a desktop application, a console application, etc. Exemplary embodiments of the processor 1102 and the machine 1100 will be discussed later.
[0020] The processor 1102 may include or be associated with a computer or machine readable medium 1122. As discussed in more detail below, the computer or machine readable medium 1122 may include a memory. Any memory discussed herein may be a computer readable memory configured to store data. The memory may include volatile or non-volatile, temporary or non-temporary memory, and may be embodied as internal memory, active memory, cloud memory, etc. Embodiments of the memory may include processor modules and other circuits to allow data to be input and output in the memory, which may include transmitting data to and from other components of the communication system. Such transmission may be through hardwire or wireless transmission. The communication system may include a transceiver that may be used in conjunction with a switch, a receiver, a transmitter, a router, a gateway, a waveguide, etc., to facilitate communication via a communication method or protocol to control and coordinate signal transmission and processing to any other component or combination of components of the communication system. Transmission may be performed via a communication link. The communication link may be electronically based, optically based, photoelectrically based, quantum based, etc.
[0021] The computer or machine readable medium 1122 may be configured to store thereon one or more instructions 1124. The instructions 1124 may be in the form of an algorithm, program logic, etc. that causes the processor 1102 to build and / or implement a predictive model.
[0022] Processor 1102 can communicate with other processors of other devices 1004 (e.g., blood sugar status monitoring device, glucose management system, insulin suggestion system, insulin delivery device, glucose monitoring device, activity tracking device, meal tracking device, etc.). Any of these other devices 1004 may include any exemplary processor 1102 disclosed herein. Any processor may have a transceiver or other communication device / circuit to facilitate the transmission and reception of wireless signals. Any processor may include an application programming interface (Application Programming Interface, API) as a software intermediary, which allows two applications to talk to each other. The use of API can allow the software of the processor 1102 of system 1000 to communicate with the software of the processor of other devices 1004.
[0023] In some embodiments, the processor 1102 is configured to build a prediction model. In some embodiments, the processor 1102 is configured to implement a prediction model (e.g., the prediction model has been built via another processor and stored in a memory associated with or accessible to the processor 1102 (e.g., memory 1104, memory 1106, memory 1122, data storage 1002, etc.). In some embodiments, the processor 1102 is configured to build and implement a prediction model.
[0024] As will be explained in more detail later, embodiments of the prediction model can be applied to disorder curves to assess the likelihood that any one or combination of these disorder curves is an expected disorder curve, which is a disorder curve that closely matches the current patient data in order to predict a disorder in the patient's blood glucose level. In some embodiments, the processor 1102 is configured to generate a disorder curve. In some embodiments, the processor 1102 is configured to implement a prediction model that has been constructed using a disorder curve via another processor and stored in a memory (e.g., memory 1104, memory 1106, memory 1122, data memory 1002, etc.) associated with the processor 1102 or accessible by the processor 1102. In some embodiments, the processor 1102 is configured to generate a disorder curve and use the disorder curve that has been constructed. The disorder curve is based on historical patient data, but can also be updated by current patient data.
[0025] In an exemplary embodiment, instructions 1124 may cause processor 1102 to generate at least one disturbance curve. Each disturbance curve may be a data representation based on historical patient data related to deviations from a threshold blood glucose level. For example, historical patient data may be analyzed to determine that on a particular day, at a particular time of day, etc., a patient did something that produced a "disturbance" that caused the patient's "normal" or average blood glucose level to deviate. For example, people (especially those with (T1D)) tend to follow a schedule or at least exhibit habitual behaviors—for example, they eat at a particular time, exercise at a particular day / time, etc. Meals and physical activity can cause fluctuations in blood glucose levels, but these fluctuations may be part of the "normal" because of the habitual nature of this occurrence. However, for example, an analysis may be performed to determine that a patient eats early 30% of the time on Fridays. As another example, while a patient may exercise on Saturday mornings, this physical activity may occur 80% of the time at 9 a.m., 15% of the time at 10 a.m., and 5% of the time without exercise. Based on this historical data analysis (which will be explained in detail later), a disturbance curve may be generated to capture these disturbances. One or more thresholds may be set to determine what constitutes a disturbance. Thresholds may be set based on the patient's average blood glucose level, a range of blood glucose levels (e.g., a band set by standard deviations about the mean), an expected blood glucose level, a desired blood glucose level, and the like. These may be overall averages / ranges / expected / desired levels for a period of historical data, averages / ranges / expected / desired levels for a given day, averages / ranges / expected / desired levels for a given time of day, and the like. One or more thresholds may be set such that a deviation from a threshold blood glucose level of a disordered curve is the presence of a blood glucose level caused by physical activity, meal activity, and the like, thereby increasing the risk of hyperglycemia and / or hypoglycemia. Physical activity, meal activity, and the like that increases the risk of hyperglycemia and / or hypoglycemia may be identified by state-of-the-art fault detection or classification techniques, such as logistic regression or artificial neural networks. As non-limiting examples, physical activity, meal activity, and the like may include physical activity, meal activity, and the like exhibited by a patient that deviates from a threshold value, which is an average behavior associated with physical activity, meal activity, and the like.
[0026] Although the exemplary embodiments discuss analysis using blood glucose levels or blood glucose measurements, it should be understood that other levels / measurements (eg, interstitial glucose) may be used / analyzed in addition to or instead of blood glucose.
[0027] Clustering techniques (e.g., k-means with Hamming distance) can be used to determine which dates, times, and activities should be grouped together to form a disorder curve. An exemplary implementation of clustering will be discussed in detail later. It is expected that more than one disorder curve will be generated. Although clustering techniques can be used to determine which dates, times, and activities should be grouped together to form a disorder curve, it should be understood that the disorder curve can be bound to a day (e.g., only for Fridays), bound to a certain activity (e.g., only for meals), bound to a certain activity that occurred earlier, etc. As a non-limiting example, historical patient data can include data related to more than one day, and clustering can be used to group dates with similar deviations from threshold blood glucose levels. For example, clustering techniques can cluster dates to determine that, for example, every Friday at 4:00 p.m., there is a specific disorder due to a specific type of activity.
[0028] Each disorder curve may be a data representation (eg, a graph, trend line, polynomial function, array of data, etc.) of historical data that mathematically models the disorder. The data representation of any one or combination of the disorder curves may be stored in a memory.
[0029] Instruction 1124 may cause processor 1102 to receive current patient data. Current patient data and / or historical patient data may include any data indicating patient characteristics (e.g., data providing information about the patient's metabolic state, blood sugar state, meal activity, physical activity, active insulin, etc.). Examples of current patient data and / or historical patient data may be glucose measurements, meal intake data, physical activity measurements, insulin injection data, etc. Historical patient data may be data that has affected or contributed to a disorder, which may include baseline data. Current patient data may be data that: 1) is / can / may affect or contribute to a disorder; 2) is / can / may affect or contribute to a disorder; and / or 3) is / can / expected to / affect or contribute to a disorder. For example, current patient data may be data received in the past 24 hours or less.
[0030] Instructions 1124 may cause the processor 1102 to apply a predictive model. The details of the predictive model will be explained later, but the predictive model may be used to determine the likelihood that the patient is entering / experiencing or is about to enter / experience a disorder based on the received current patient data. For example, the processor 1102 may be configured by instructions 1124 to compare the current patient data with at least one disorder curve. This comparison may occur continuously, periodically, at a predetermined time, etc. For example, the current patient data may be generated and / or received continuously, periodically, at a predetermined time, etc., wherein the processor 1102 compares immediately when the data is generated and / or received. As another example, the current patient data may be stored in a memory after being generated, wherein the processor 1102 extracts the data from the memory continuously, periodically, at a predetermined time, etc. and compares it. As a non-limiting example discussed in more detail herein, the processor 1102 is configured by instructions 1124 to compare the current patient data with at least one disorder curve every five minutes.
[0031] Processor 1102 may be configured via instruction 1124 to evaluate the likelihood that one of the plurality of disordered curves is an expected disordered curve. This evaluation may be performed via a probabilistic analysis. The expected disordered curve is a disordered curve that matches the current patient data based on the probabilistic analysis (the matching process may use any state-of-the-art technique to perform a distance measurement between the two sets of data, such as the minimum distance L between the disordered curve and the data that is linearly transformed). 2 , also known as linear regression). There can be a match with more than one disorder curve, for example, there can be more than one expected disorder curve. One or more thresholds can be set to determine what constitutes a match. The threshold used to determine a match between the current patient data and one disorder curve can be the same or different than the threshold used to determine a match between the current patient data and another disorder curve. In addition, the current patient data can be segmented (e.g., into physical activity data, meal data, incremental data for the past hour, etc.), where one segment of the current patient data can be compared using the same or different threshold as another segment of the current patient data.
[0032] Processor 1102 can be configured by instruction 1124 to determine insulin dosage based on expected disorder curve. Insulin dosage can be determined by cost function, objective function, loss function, etc. In some embodiments, the determination can be based on the best match with one or more disorder curves. In some embodiments, the determination can be based on a weighting factor assigned to one or more disorder curves, one or more weighting factors assigned to one or more disorder curves, etc. For example, in a non-limiting example, the current patient data can be compared with the disorder curve, and the weighting factor can be assigned to the disorder curve based on the comparison. The disorder curve can be assigned a weighting factor based on the possibility of the expected disorder curve being assessed, for example, the disorder curve is assigned a weighting factor based on the degree of matching with the current patient data. Any one or combination of the disorder curves can be assigned a weighting factor. Any one or combination of weighting factors can be updated continuously, periodically, at a predetermined time, etc. based on the current patient data, for example, the weighting factor can be updated based on the time when the current patient data is received. In some embodiments, each disorder curve is assigned a weighting factor—e.g., if disorder curve-1 has a very poor match with the current patient data, then disorder curve-1 may be assigned a weighting factor of 0.0 or close to 0.0, if disorder curve-2 has a moderate match with the current patient data, then disorder curve-2 may be assigned a weighting factor of 0.5, and if a disorder curve has a very good match with the current patient data, then the disorder curve may be assigned a weighting factor of 1.0 or close to 1.0. A cost function, objective function, loss function, etc. may then be applied to all disorder curves with weighting factors (or any number or combination of disorder curves with weighting factors) to determine the best (e.g., by optimization) insulin dose.
[0033] Other data processing, such as smoothing, weighting, etc., may be performed on any one or combination of the disordered curves. For example, any one or combination of the disordered curves may be smoothed and a weighting function may be applied to emphasize or weaken certain disordered curves. For example, it may be desirable to weaken a portion of the disordered curve associated with the patient's sleep time (e.g., from 10:00 p.m. to 6:00 a.m.). Doing so may cause the probability of the disordered curve to return to its prior value at 6:00 a.m. each day.
[0034] There may be a situation where the current patient data does not match any disorder curve. In this case, the processor 1102 can create or at least start to create a new disorder curve. The new disorder curve can be stored in the memory together with the multiple disorder curves. In addition, or in an alternative, the processor 1102 can update or modify the existing disorder curve.
[0035] Any one or combination of the turbulence curves and / or the prediction model may be stored in a memory associated with the processor 1102. Alternatively, any one or combination of the turbulence curves and / or the prediction model may be stored in a memory external to the processor 1102, wherein the processor 1102 may be configured to communicate with the external memory.
[0036] Processor 1102 can be configured by instruction 1124 to output a signal representing insulin dosage to device 1004, which is configured to monitor, affect and / or manage the insulin level of the patient. In addition, or alternatively, instruction 1124 can make processor 1102 consider additional data, such as certain attributes of current patient data, previous insulin dosage data, total daily insulin (total daily insulin, TDI) data, active insulin (insulin on board, IOB) data, etc., and use this / these data to modify the determined insulin dosage. Processor 1102 can be configured by instruction 1124 to output a signal representing the modified insulin dosage to device 1004. Instruction 1124 can make processor 1102 output the modified insulin dosage signal to replace insulin dosage, output the modified insulin dosage and the modified insulin dosage, output the modified insulin dosage with the indicator of the modified insulin dosage recommended, etc.
[0037] The current patient data can indicate that the patient has entered hyperglycemia or hypoglycemia state or has the high risk of entering hyperglycemia or hypoglycemia state.Therefore, processor 1102 can be configured to determine whether there is any current deviation with threshold blood sugar level that increases hyperglycemia and / or hypoglycemia risk by instruction 1124.Using this current deviation assessment, processor 1102 can be configured to determine the insulin dosage to be applied based on current deviation rather than based on expected disorder curve by instruction 1124.Alternatively, processor 1102 can be configured to modify the insulin dosage based on expected disorder curve by instruction 1124, to consider current deviation.Instruction 1124 can make processor 1102 output this modified insulin dosage signal to replace described insulin dosage, output this modified insulin dosage and described insulin dosage, output have the described insulin dosage of the indicator of recommending this modified insulin dosage etc.
[0038] The historical patient data and the current patient data may be generated by the device 1004. The device 1004 may communicate with the processor 1102 and / or with a memory that is in communication with or otherwise accessible to the processor 1102, for example, the device 1004 may transmit the patient data directly to the processor 1102, the device 1004 may transmit the patient data to the memory from which the processor 1102 receives the data, or the device 1004 may send the patient data to a combination of the two. The device 1004 may measure / obtain / collect / generate patient data continuously, periodically, at predetermined times, in accordance with user requirements of the device 1004, user requirements of the processor 1102, algorithm provisions of the device 1004, algorithm provisions of the processor 1102, etc. The data transmission from the device 1004 or the memory to the memory or the processor 1102 may be a push operation, a pull operation, or a combination thereof. The device 1004 may be a glucose monitoring device, an activity tracking device, a meal tracking device, etc., any of which has a processor configured to perform the function of measuring / obtaining / collecting / generating patient data. Any one or combination of these devices 1004 can be in communication with the processor 1102, or in communication with a memory associated with the processor 1102 or in communication with the processor 1102. Thus, the processor 1102 can be configured to receive historical patient data and / or current patient data from the glucose monitoring device, the activity tracking device, and / or the meal tracking device. Optionally, the glucose monitoring device, the activity tracking device, and / or the meal tracking device can be configured to transmit the historical patient data and / or current patient data to the data store 1002, wherein the processor 1102 is configured to receive the historical patient data and / or current patient data from the data store 1002.
[0039] The output of the processor 1102 can be configured to be displayed 1110 on a display screen (e.g., via a user interface): as a command signal to control an insulin device, as a communication signal to generate an alarm or notification, as an input to a monitoring device, etc. Therefore, the output can be configured to be transmitted to a device 1004, which includes any one or a combination of an insulin delivery device, a blood glucose status monitoring device, a blood glucose management system, an insulin suggestion system, etc. In addition, or alternatively, the output can be transmitted to a memory that communicates with any one of these devices 1004. The transmission from the processor 1102 or memory to the memory or device 1004 can be performed continuously, periodically, or at a predetermined time according to the user requirements of the device 1004, the user requirements of the processor 1102, the algorithm regulations of the device 1004, the algorithm regulations of the processor 1102, etc. The data transmission from the memory or processor 1102 to the device 1004 or memory can be a push operation, a pull operation, or a combination thereof. Any one or a combination of these devices may include a processor configured to perform functions such as dose delivery, monitoring, display, management, etc. Any of these devices 1004 may be part of or configured to operate in an open-loop system, a closed-loop system, a dosing activation system, an artificial pancreas system, etc. For example, processor 1102 may be a control module of an automated insulin dosing system.
[0040] Thus, instruction 1124 may cause processor 1102 to: 1) configure the output signal as notification information that causes device 1004 to recommend administration of said insulin dose and / or modified insulin dose; 2) configure the output signal as a command signal that causes device 1004 to administer said insulin dose and / or modified insulin dose; and / or 3) configure the output signal as a data point to allow device 1004 to graphically display said insulin dose and / or modified insulin dose.
[0041] Exemplary embodiments relate to a computer-readable medium 1122 having stored thereon instructions 1124 which, when executed by a processor, cause the processor 1102 to execute an algorithm. The instructions 1124 may cause the processor 1102 to generate a plurality of disorder curves, each disorder curve being a data representation of historical patient data related to a threshold blood glucose level deviation. The instructions 1124 may cause the processor 1102 to receive current patient data. The current patient data and / or historical patient data may include any data indicating patient characteristics. The instructions 1124 may cause the processor 1102 to apply a prediction model. The instructions 1124 may cause the processor 1102 to compare the current patient data with at least one disorder curve. The instructions 1124 may cause the processor 1102 to evaluate the possibility that one disorder curve in a plurality of disorder curves is an expected disorder curve based on a probability analysis. The expected disorder curve may be a disorder curve that matches the current patient data determined based on a probability analysis. The instructions 1124 may cause the processor 1102 to determine an insulin dose based on the expected disorder curve. The instructions 1124 may cause the processor 1102 to output a signal representing the insulin dose to a device 1004 configured to monitor, influence, and / or manage a patient's insulin levels.
[0042] Exemplary embodiments relate to methods for notifying, determining or controlling insulin dosage. The method may include generating a plurality of disorder curves, each disorder curve being a data representation based on historical patient data related to a threshold blood glucose level deviation. The method may include receiving current patient data. The current patient data and / or historical patient data may include any data indicating patient characteristics. The method may include applying a prediction model. The current patient data may be compared with at least one disorder curve. Probabilistic analysis may be used to assess the likelihood that one of the plurality of disorder curves is an expected disorder curve, the expected disorder curve being a disorder curve that matches the current patient data based on the probability analysis. The method may include determining the insulin dosage based on the expected disorder curve. The method may include outputting a signal representing the insulin dosage to a device configured to monitor, influence and / or manage the patient's insulin level.
[0043] Example
[0044] The following discussion relates to exemplary implementations and test results related to embodiments of the present disclosure.It should be understood that the following examples are non-limiting examples of how to make, use, and test the systems, devices, and methods disclosed herein.
[0045] Example 1
[0046] Postprandial hyperglycemia is a recurring challenge for patients with type 1 diabetes, and even the most advanced automated systems currently require manual entry of carbohydrate amounts. To achieve fully automated system development, an exemplary embodiment provides a control system that can automatically deliver insulin doses or recommended doses, initiate dosing, anticipate meal behaviors, anticipate physical activity behaviors, etc. to improve postprandial closed-loop or open-loop control.
[0047] A model predictive control (MPC) system is enhanced with an automated dosing system that reacts to early glucose rises and / or a multistage MPC (MS-MPC) framework to predict historical patterns. Priming is achieved by detecting large glycemic disturbances (such as meals) and delivering a portion of the patient's TDI modulated by the likelihood of the disturbance (bolus priming system, BPS). In the anticipation module, clustering is used to group days with similar behavior and glycemic disturbance curves are generated from historical data; the probability of each cluster is then evaluated at each controller step and the probability is informed to the MS-MPC framework to predict each curve. Four configurations were tested: simulated MPC, MPC+BPS, MS-MPC, and MS-MPC+BPS to compare the effect of each controller module.
[0048] As will be explained in more detail, the results showed that MS-MPC+BPS had the largest range of postprandial time at 60.73±25.39%, but each module improved compared to MPC (51.79±26.12%): 56.95±25.83 for MPC+BPS and 54.83±26.00% for MS-MPC. All controllers maintained hypoglycemic exposure (time below 70 mg / dL <0.5%), with the improvement coming primarily from a reduction in postprandial time within this range (MS-MPC+BPS: 39.10±25.32%, MPC+BPS: 42.99±25.81%, MS-MPC: 45.09±25.96%, MPC: 48.18±26.09%). The dosing activation system and the anticipatory perturbation profile improved blood glucose (BG) control and were most effective when used in combination.
[0049] Fully closed loop AID
[0050] Several AID systems that do not require user input of meal sizes have been tested. In 2014, Harvey et al. evaluated a fully closed loop (FCL) AID system in 12 adult participants with T1D in an inpatient clinical trial. Overall, the system produced glycemic control with 80% of blood glucose readings within the range of 70-180 mg / dL (time in range (TIR)), but resulted in elevated postprandial glucose values after unannounced meals. Forlenza et al. presented the performance of FCL with pre-ordered postprandial excursions in a 72-hour hotel-based study, with a TIR of 63.6 ± 9.2%. However, the time above range (>180 mg / dL, TAR) was 60.9 ± 23.3% within 4 hours after unannounced meals. A recent study by Dovc et al. used two insulin preparations, FIASP and aspart, in a 27-hour inpatient study of their FCL AID system, with TIRs of 53.3% and 57.9% for aspart and FIASP insulin, respectively. Their approach relies on a meal detection algorithm, but faces the same difficulty with unannounced meals. According to recent early outpatient results, multi-hormone therapy has better results. In 2021, Haidar et al. conducted an open-label crossover trial of 30 participants, and the results showed that an AID system using simple meal notifications and empagliflozin was noninferior to a mixed system using meal notifications (mean blood glucose was 153±25.2 mg / dL and 153±27.0 mg / dL, respectively). In another 2021 pilot study, Majdpour et al. demonstrated that the insulin-pramlintide FCL AID was noninferior to the mixed AID (with TIRs of 81% and 83%, respectively).
[0051] Cameron’s 2012 manuscript describes an implementation of an automated insulin dosing system that uses multiple blood glucose prediction models, each informed by a different disturbance. When the system determines a disturbance is more likely to occur, the predictions of that model are given additional weight. Additionally, information about the likely timing of meal disturbances based on normal meal times, last meal times, and sleep schedules is included. Using this approach, Cameron et al. were able to reduce the two-hour prediction error by 45% without meal detection and 18% with meal detection. The three-hour prediction error was reduced by 60% without meal detection and 30% with meal detection.
[0052] Cameron et al. expanded on these findings in a 2014 clinical trial. In an inpatient study, 10 patients used a so-called multiple model probabilistic controller (MMPC) in which they had five unannounced meals. For the six patients using the final version of the controller, the average continuous glucose monitoring (CGM) TIR was 78%. There was only one case of hypoglycemia caused by the controller during the hospital stay.
[0053] In an inpatient clinical study, MMPC was evaluated in 10 patients with an overall mean TIR of 142 mg / dL and an average nighttime TIR of 125 mg / dL. In a hotel-based study, another version of the algorithm was tested in 15 subjects with an overall mean glucose value of 152 mg / dL and an average nighttime glucose value of 139 mg / dL.
[0054] Using behavioral patterns to predict blood sugar disturbances
[0055] Several simulations and clinical experiments have shown that the anticipatory curves integrated into the AID system can reduce the adverse effects of glucose disturbances. Simulations showed that multistage model predictive control (MS-MPC) could reduce hypoglycemia by predicting the effects of moderate exercise through disturbance curves, a result that was later confirmed in a randomized crossover clinical trial with 15 adult participants: participants had fewer hypoglycemic events when using the MS-MPC system compared with well-adjusted standard MPC (9 vs. 33, respectively), and the percentage of time with blood glucose <70 mg / dL (time below range, TBR) was reduced by 1.3%. The overall reduction in hypoglycemia did not result in a significant increase in TAR.
[0056] Computer experiments also demonstrated the ability of the MS-MPC framework to predict meal-induced glycemic disturbances. In this work, disturbance curves were generated based on a representative real subject, and then closed-loop experiments were performed using a 100-adult cohort from the University of Virginia (UVA) / Padua simulator using the generated disturbance curves. The results showed that when the hybrid closed-loop method was used, TIR increased by an average of 1.6%, and when the FCL method was used, TIR increased by an average of 16.4%.
[0057] One of the goals of this experiment was to build on past experience and further investigate the impact of automatic dosing and upset prediction using the MS-MPC framework. To test the effect of each component of the control system, a 2 × 2 experimental design was implemented (each module was turned on or off), resulting in four treatment strategies: Model MPC, MS-MPC, MPC+BPS, and MS-MPC+BPS.
[0058] method
[0059] Control strategy
[0060] Figure 2 An exemplary modular control strategy is described. Using the traditional modular architecture (eg, data management, safety system modules, auto-correction, state estimation) as a guide, four different module combinations are considered to suppress large positive disturbances.
[0061] Core Controller
[0062] MPC encompasses a general control paradigm rather than a single control strategy; it integrates predictions from an explicit mathematical model of the user's glucose-dependent metabolic state into a real-time optimization problem that aims to find the best insulin injection subject to possible constraints on the model variables. In this setting, the MPC component commands optimized basal insulin injections through a pump every five minutes to exclude internal metabolic disturbances. MS-MPC is a robust control strategy that considers N parallel MPC controllers, each perturbed by a specific disturbance realization. In this case, the disturbance realization consists of N different subject-specific disturbance profiles representing typical meal behavior. The final control action is reached after the N controllers reach consensus through unexpected constraints.
[0063] Generation of subject-specific disorder curves
[0064] In order to construct a manageable set of disorder curves for MS-MPC prediction, the current set of d(t) features is reduced to a few representative curves. To do this, significant disorder occurrences are clustered and then representative disorder inputs are constructed for each cluster. This requires (i) constructing disorder features, (ii) detecting disorder events, (iii) clustering the daily patterns of detected events, and (iv) constructing features that represent each cluster. This process is detailed below.
[0065] Historical Disorder Estimates
[0066] A Kalman Filter (KF) with the same internal model as the controller was used to estimate the disturbance signature d(t) for each day of the history (e.g., based only on CGM and insulin, not meal logs). In this case, the operating point for model linearization was defined based on the subject’s average CGM value for the day, the basal rate, and the steady-state solution of the model equations. KF tuning was performed by modifying the covariance matrices of the measured (R) and modeled (Q) noise. The diagonal elements of the R and Q matrices were tested to be in the range [1 × 10 -4 1×10 4 ] to determine values that adequately represent the system inputs, in addition to a smoother estimate of the d(t) signal, to reflect the meal-related glucose excursions observed in the CGM. Therefore, the values of Q and R are fixed in such a way that d has a suitable behavior to represent food intake during high glucose excursions. Figure 3 CGM data for one subject is shown; Figure 4 Insulin data for one subject is shown; Figure 5 An estimate of the disorder characteristic d is shown.
[0067] Kalman filter (KF) design
[0068] The KF is designed based on a linearized version of the controller model given by the following equation:
[0069] The model parameters were determined from the data collected for each subject, and the operating point was determined based on the mean of the daily CGM, the subject-specific basal rate, and the steady-state solution of the model equations. After linearization, the discretization was performed by combining the discretized system (w k ) in the dynamics and the noise in the discrete-time measurements (v k ), the model can be written as: x k+1 =f(x k ,u k ,d k )+w k (6) y k =g(x k )+v k (7)
[0070] f(·) and g(·) are linear functions, x k=[G,X,I sc1 ,I sc2 ,I] T ,u k = J,y k = G are the state vector, input vector and output vector respectively. k and v k , assuming a normal distribution with zero median and covariance matrices Q and R. Then, the KF was designed according to: is the observer error, K k is the observer gain calculated as a function of Q and R by solving the Riccati equation.
[0071] Turbulence Detection
[0072] Since there is no meal record in the fully closed-loop system, meal-like disturbances are retrospectively reconstructed based on CGM and insulin data. In order to filter out any small deviations and focus on disturbances that are worth predicting, an automatic disturbance detector is used, which uses features of the estimated disturbance values d and continuous glucose measurements cgm that characterize the historical data collected every day. These features are selected to represent changes in glucose and disturbance values that are unique to blood sugar disturbances (such as meals). Table 1 lists the calculated feature descriptions for every five-minute sample. Table 1 - Characteristics of retrospective disorder detection algorithms feature describe <![CDATA[f1]]> <![CDATA[The intercept term of the second-order polynomial conforms to c1 = cgm(t - 6), …, cgm(t + 6)]]> <![CDATA[f2]]> <![CDATA[The slope term of the first-order polynomial conforms to c2 = cgm(t), …, cgm(t + 12)]]> <![CDATA[f3]]> <![CDATA[The curvature term of the second-order polynomial conforms to c1 = cgm(t - 6), …, cgm(t + 6)]]> <![CDATA[f4]]> <![CDATA[f2·f3]]> <![CDATA[f5]]> <![CDATA[The intercept term of the second-order polynomial conforms to d1 = d(t - 6), …, d(t + 6)]]> <![CDATA[f6]]> <![CDATA[The slope term of the first-order polynomial conforms to d2 = d(t), …, d(t+6)]]> <![CDATA[f7]]> <![CDATA[The curvature term of the second-order polynomial conforms to d1 = d(t - 6), …, d(t + 6)]]> <![CDATA[f8]]> <![CDATA[f6·f7]]> <![CDATA[f9]]> <![CDATA[The maximum d value d within the next hour max = max d2]]>
[0073] After generating features at every five-minute interval, logistic regression is used to determine the probability of a large disorder π detect (t) (see Table 2). The test time is determined by finding the local minimum of the CGM value when the test vector is equal to 1 for at least 60 minutes apart (or the first index if there is no local minimum). If the test times differ by less than one hour, the first test time is used. Table 1 - Logistic regression parameter values
[0074] Chaos curve generation
[0075] After detection of a major glucose disturbance, daily indicator signals were defined to group similar days into clusters (equal to 1 within two hours of detection of the disturbance and equal to 0 otherwise). Using k-means with a Hamming distance metric, these signals were clustered with k = 1, …, 5. The number of clusters k for each individual was based on which produced the highest Calinski-Harabasz score, maximizing cluster separation and cohesion. After grouping the data for several days, the clustered disturbance signal d for each day within that five-minute interval was used. m,j The average value is used to determine the curve trajectory ω of each cluster i at every five-minute interval on day j. n days,i is the number of days grouped into cluster i. These curves are then smoothed using a one-hour centered moving average. Multiply the above values by Figure 6 The weighting function shown, to weaken the curve at night, allows the curve probabilities to return to their previous values at the beginning of each day.
[0076] The previous probability value π for each cluster prior,i The proportion of days that are estimated to be data assigned to this given cluster. Where n days,i is the number of days of data in cluster i, n days,total represents the total number of days considered. These previous probability values are used as a starting point so that the initial weights of each curve in MS-MPC are related to historical data.
[0077] Chaos curve generation
[0078] Figure 7 An example of a curve for one subject is shown. Here it can be seen that this person's disordered curve rises after the typical times for breakfast (6:00-8:00 AM), lunch (12:00-2:00 PM), and dinner (6:00-8:00 PM). The associated prior probabilities indicate that this person ate earlier on 31% of the days, and on 46% of the days in the data used, the subject ate later. On 23% of the days, they had less discernible meal patterns determined from the detected disorder.
[0079] Curvilinear Probability Estimation of Online MS-MPC Weights
[0080] The probability value for each curve is updated in real time based on the current disorder measure, which is found using the same technique applied to the retrospective data. i(t) is estimated using the method developed by Patek et al. This method allows the probability of each curve to move dynamically following the turmoil the user is currently experiencing.
[0081] A further "night mode" was added, which allows the curve probability estimates to reset back to their previous values in real time (so each day is not affected by the previous day). From 11:00 pm to 1:00 am, the probability is reset from its value before night mode started to π i,night Decrease linearly to the previous value,
[0082] For the interval n after night mode starts, π i (t) = l d,i (n) (13)
[0083] To ensure that the currently observed disorder is better than the prediction of the expected disorder, all curve probability values are scaled based on an exponent ρ0 that is proportional to the current disorder estimate. The calculation of ρ0 is detailed below.
[0084] The probability of the current disorder estimate is calculated as follows: in:
[0085] The final adjusted probability π adjusted,i (t) is obtained by multiplying the curve probability by 1-ρ0. π adjusted,i (t) = π i (t)·(1-ρ0) (10)
[0086] Experimental design
[0087] Data collected from the unsupervised home portion of a large-scale pivotal trial (NCT03563313) conducted at the University of Virginia was used to evaluate an embodiment of the method. The data came from 124 adult and adolescent participants with T1D for more than 6 months, during which they used an automated insulin delivery (AID) system with meal notifications. Data from 100 clinical subjects were randomly selected and paired with a computer subject in the UVA / Padova T1D simulator platform. 80% of the data was used to create meal curve clusters. Seven days with at least one meal recorded were randomly selected from the remaining months of data collection. Then, this week's meal records, in proportion to body weight, were used as a meal plan for the simulation experiment, including the variability of insulin sensitivity on the day. In this embodiment of the simulation platform, subject-specific (e.g., class-specific) time-varying curves that affect endogenous glucose production and insulin-dependent glucose utilization are used to simulate the diurnal changes in insulin sensitivity and dawn phenomenon. The use of real meal records enables more realistic behavior in the simulation and the presence of final meal patterns. This experimental configuration was then tested with the following four configurations of the control system: MPC, MS-MPC, MPC+BPS, and MS-MPC+BPS. Treatments were compared overall and within 4 hours postprandially using the relevant metrics described by the AID system evaluation criteria of Maahs et al. Because the assumptions tested were not particularly informative in the simulated environment, no statistical significance was reported.
[0088] result
[0089] The simulation results showed that the use of the anticipatory curve in MS-MPC and BPS reduced BG values overall. The primary outcome (TIR throughout the simulation) improved from 72±17.7% for MPC alone to 73.4±17.4% for anticipatory and 75.5±17.1% for priming. The combination of priming and anticipatory had the greatest effect, with a TIR of 77.2±16.7%, +5.2% higher than MPC alone. On average, when BPS was dosed 27.64±29.70 minutes after the actual meal time, 4% of patients had a TDI delivered at 64.11% of the meal, 7% had a TDI delivered at 36.72% of the meal, and 10% had a TDI delivered at 22.18% of the meal. No BPS dose was given after 35.89% of the meals. The overall false positive rate was 0.69%.
[0090] The mean blood glucose of the MS-MPC+BPS group was the lowest (155.14±31.88mg / dL). The mean blood glucose values of other controller configurations of MPC, MS-MPC and MPC+BPS were higher, 165.49±33.49mg / dL, 161.61±33.60mg / dL and 159.65±33.20mg / dL, respectively. The trend of lower BG values was similarly shown in the time below the range (e.g. <50mg / dL, <60mg / dL and <70mg / dL), in the time within the strict range (e.g. 70mg / dL-140mg / dL) and above the range (e.g. >180mg / dL, >250mg / dL and >300mg / dL). The BPS and expected curves also reduced the standard deviation of the BG values and enabled the system to deliver more insulin overall. The overall results of the experiment are shown in Table 3. Table 3 - Average and standard deviation values of artificial pancreas evaluation indicators in the simulation experiment Controller MPC MS-MPC MPC+BPS MS-MPC+BPS <50mg / dL(%) 0.03±0.21 0.07±0.40 0.02±0.15 0.14±0.54 <60mg / dL(%) 0.06±0.41 0.23±0.85 0.10±0.45 0.33±1.01 <h2 style=";text-align:left;direction:ltr"><70mg / dL(%)<h2 style=";text-align:left;direction:ltr"> 0.14±0.64 0.47±1.37 0.24±0.81 0.65±1.58 70-140 mg / dL (%) 49.16±17.94 50.35±19.25 51.92±17.61 53.81±18.89 70-180 mg / dL (%) 72.02±17.67 73.37±17.38 75.50±17.07 77.17±16.73 >180 mg / dL (%) 27.85±17.47 26.16±17.25 24.27±16.84 22.18±16.45 >250mg / dL (%) 9.92±11.51 9.31±11.25 7.66±10.71 7.02±10.28 >300 mg / dL (%) 4.81±8.29 4.52±8.03 3.74±7.86 3.43±7.36 Average value (mg / dL) 165.49±33.49 161.61±33.60 159.65±33.20 155.14±31.88 Standard Deviation (mg / dL) 53.90±33.49 53.62±32.72 49.51±34.32 48.91±32.82 CV(%) 30.46±12.09 31.12±11.89 28.85±12.01 29.44±11.88 Total insulin per day (u) 36.72±17.33 37.65±17.57 38.09±18.04 39.20±18.34
[0091] The effects of the expected curve and BPS were more pronounced within four hours after a meal. During this period, the TIR of MS-MPC+BPS was 60.73±25.39%. The average TIR of MPC+BPS, MS-MPC, and MPC configurations was reduced by 3.78%, 5.9%, and 8.94%, respectively. The average postprandial blood glucose in the MS-MPC+BPS group was also 4.51mg / dL-10.35mg / dL lower than that of the other controller groups. During this period, the TBR of all controller groups was less than 1%. Table 4 lists the results for the postprandial period. Table 4 - Mean values and standard deviations of artificial pancreas evaluation indicators in the 4-hour postprandial simulation experiment Controller MPC MS-MPC MPC+BPS MS-MPC+BPS <50mg / dL(%) 0.00±0.02 0.00±0.02 0.00±0.01 0.03±0.15 <60mg / dL(%) 0.01±0.08 0.03±0.16 0.01±0.06 0.08±0.33 <h2 style=";text-align:left;direction:ltr"><70mg / dL(%)<h2 style=";text-align:left;direction:ltr"> 0.03±0.17 0.08±0.33 0.06±0.32 0.17±0.53 70-140 mg / dL (%) 23.02±18.78 27.79±22.09 25.67±18.92 31.24±22.28 70-180 mg / dL (%) 51.79±26.12 54.83±26.00 56.95±25.83 60.73±25.39 >180 mg / dL (%) 48.18±26.09 45.09±25.96 42.99±25.81 39.10±25.32 >250mg / dL (%) 17.87±19.85 16.61±19.28 13.77±18.46 12.46±17.91 >300 mg / dL (%) 8.68±14.73 8.13±14.34 6.69±13.77 6.19±13.16 Average value (mg / dL) 197.20±54.44 191.11±54.17 188.59±54.36 181.72±52.27 Standard Deviation (mg / dL) 53.87±30.73 54.59±30.37 49.67±32.03 50.00±30.96 CV(%) 25.69±9.09 26.97±8.95 24.61±8.96 25.83±8.76 Insulin delivery (u) 3.52±1.92 3.61±1.98 3.67±2.07 3.79±2.13
[0092] discuss
[0093] The simulation included real meal behaviors in T1DM and showed that MS-MPC+BPS had the lowest mean blood glucose values, followed by MPC+BPS, MS-MPC, and MPC, both overall and after meals. This relationship was maintained based on the percentage of time BG was in the normoglycemic range (e.g., 70mg / dL-140mg / dL and 70mg / dL-180mg / dL) and the hypoglycemic range (e.g., <50mg / dL, <60mg / dL, and <70mg / dL). The opposite was true when blood glucose was in the hyperglycemic range (e.g., >180mg / dL, >250mg / dL, and >300mg / dL). This suggests that both the prospective curve and BPS have the effect of reducing overall blood glucose values. Overall, MPC+BPS had a mean blood glucose of 5.84mg / dL lower than MPC and a TIR of 3.38% higher than MPC. MS-MPC increased TIR by 1.35% compared to MPC without the curve. The difference in TIR between MPC+BPS and MS-MPC+BPS was 1.67%, indicating a synergistic effect between BPS and the expected disorder curve.
[0094] The overall volume of hypoglycemia did not change significantly across the four configurations. Comparison of MPC and MS-MPC+BPS suggests that these modules may be responsible for increasing the volume of TBR by less than 5 minutes while increasing TIR by more than 5% overall, which is clinically relevant.
[0095] The most insulin was used in the condition where both BPS and the curve were active (MS-MPC+BPS). Less insulin was used when only BPS was used (MPC+BPS), even less when only the curve was used (MS-MPC), and the least insulin was used when standard MPC was used. Interestingly, the MS-MPC condition had the highest mean CV, followed by MPC, MS-MPC+BPS, and MPC+BPS.
[0096] Compared to the standard MPC, the postprandial TIR increased by nearly 10% when using the prospective curve and BPS. In addition, there was an increase in the 70mg / dL-140mg / dL range, but this amount was slightly smaller. This difference was due to a decrease in TAR and a decrease in mean BG of approximately 16mg / dL. The combination of the MS-MPC structure and BPS produced the greatest effect by reducing BG values without increasing hypoglycemia during the postprandial period.
[0097] in conclusion
[0098] The experiment showed that in a simulated setting, both BPS and the derangement profile had a positive impact on the amount of time BG values were in the normoglycemic range, while also reducing hyperglycemia. The modules were most influential in the four hours after a meal. The two modules independently improved TIR and did even better in combination. The reduction in mean BG values due to BPS and the derangement profile did increase hypoglycemia, but only to an extent that was not clinically significant. BPS and the derangement profile appear to affect glucose variability, although the reasons for this are unclear.
[0099] Example 2
[0100] Physical activity is recommended for all and has been shown to reduce the risk of cardiovascular disease, cancer, hypertension, obesity, depression, and osteoporosis. In addition to general health benefits, physical activity has been associated with increased lifespan and a reduced risk of diabetes-related complications such as retinopathy and neuropathy. Given the role of physical activity in improving health and preventing these complications, it is concerning that more than 60% of people with T1D remain sedentary.
[0101] Despite the well-documented physical activity guidelines for glycemic management and advances in glycemic control technology, physical activity remains one of the major barriers to optimizing glycemic control in T1D. Current guidelines provide strategies based on the intensity and duration of physical activity and the patient’s metabolic status (e.g., glucose and insulin levels) during structured physical activity bouts. The performance of these strategies is limited by the accuracy of the patient’s assessment of these factors. The increasing availability of readily available physical activity trackers offers a more objective alternative to self-assessment. Recent studies have shown that using physical activity signals from these trackers in a closed-loop control system is effective in detecting physical activity and improving glycemic control.
[0102] In addition to structured physical activity, even short walks accumulated throughout the day have a significant effect on glycemic control. Therefore, glycemic control in patients with T1D may improve when all daily physical activity is considered in the design and adjustment of insulin therapy.
[0103] Figure 8 The timeline of outpatient study visits is shown. In this study, prandial insulin therapy was adjusted for physical activity beyond the individual's usual range. The motivation was to promote glucose management related to physical activity in a patient-specific manner and based on objectively measured physical activity. Results are presented from a pilot study designed to evaluate the safety and feasibility of this approach—using step count as a universally available measure of physical activity—to adjust prandial insulin to improve protection against hypoglycemia after the physical activity of walking. To this end, the performance of this approach with respect to postprandial glucose control was compared with standard therapy in a randomized, crossover clinical trial.
[0104] Research Design and Methods
[0105] Participants
[0106] Fifteen insulin pump users with T1D participated in a randomized crossover study. Inclusion criteria were diabetes duration of at least one year, insulin pump use for at least six months, meal calculation using defined parameters and insulin dosing corrections, and willingness to maintain a consistent physical activity regimen during data collection. Exclusion criteria were pregnancy within 6 months prior to recruitment, diabetic ketoacidosis or severe hypoglycemia, use of non-insulin medications designed to lower glucose (eg, GLP-1 agonists, metformin), current use of a well-defined insulin dosing adjustment method to compensate for significant physical activity, inability to exceed 30 minutes of physical activity per day, or current participation in another interventional clinical trial. The Institutional Review Board at the University of Virginia approved the study protocol and consent form. Informed consent was obtained from each participant.
[0107] Program Plan
[0108] The study consisted of a free-living data collection period followed by two one-day outpatient supervised visits at the University of Virginia Diabetes Technology Center using a randomized crossover design (clinicaltrials.gov: NCT03394352). During the data collection period, CGM, insulin, meal, and physical activity records were collected over 20 days while participants were free-living. CGM data were acquired via a Dexcom G6 (Dexcom, San Diego, CA) and physical activity data were collected via a Fitbit Charge 2 (Fitbit, San Francisco, CA) wristband. Subjects were asked to enter any consumed carbohydrates into the dosing calculator of their insulin pump or the CGM mobile app.
[0109] These two outpatient supervision visits have the same timeline (see Figure 8) and were separated by at least 2 days to allow for washout. These visits began at 7:30 AM after the participant arrived at the study unit. Upon arrival, each participant's basal insulin rate was set to their regular basal insulin curve, and no temporary basal insulin rate was used during the visit. Around 8:00 AM, a mixed meal with 24 g of carbohydrate was provided for breakfast. In both visits, the breakfast dose was calculated based on the participant's regular treatment parameters, with a target glucose of 110 mg / dl. Until 10:45 AM, participants had free time during the first visit with unrestricted activity. In their second visit, they were asked to repeat physical activity behaviors similar to the previous visit. At 10:45 AM, physical activity began. Participants walked briskly at a steady pace on a running track for a designated 30- or 45-minute period. The duration for each participant was determined in a way that exceeded their habitual accumulated physical activity during lunch. At noon, the carbohydrate content of the lunch provided matched each participant's daily lunch. The participants' habitual accumulated physical activity at lunch and the daily lunch carbohydrate content were derived from their data collected during the free-living data collection phase. Each participant ate the same breakfast and lunch at both visits to control for meal effects. Participants selected their own foods without any ingredient restrictions, except for the amount of carbohydrates.
[0110] In the control visit, the lunch insulin dosage is calculated according to standard therapy (ST). In the experimental visit, these dosages are adjusted according to the dosage method based on physical activity. The adjustment is limited to 50% of the standard meal dosage. This is the safety layer selected empirically in this pilot study. After lunch, the physical activity of the participants is required to be limited to the lowest possible level until the end of the two study visits in order to observe the performance of postprandial blood glucose control with minimal distortion. Hypoglycemic events are treated with a certain amount of carbohydrates determined by the research physician and are always recorded. Both study visits end after the completion of the post-lunch blood glucose excursion, approximately 4.5 to 5 hours after lunch. At the end of each study visit, the CGM, insulin pump and physical activity tracker data of the participants are downloaded and used for the analysis disclosed herein.
[0111] Algorithm Description
[0112] The standard therapeutic dosing amount for pre-meal insulin dosing calculations in patients with T1D has three components: the amount needed to compensate for carbohydrates consumed at the meal, the amount needed to correct any current elevated glucose levels at the time of the meal, and active insulin on board (IOB), which is the active insulin from the previous injection. The resulting dosing calculation formula is as follows: Among them, CHO Intake is the amount of carbohydrates consumed in the meal, CR is the ratio of carbohydrates to insulin (g / U), BG target is the target BG value, CF is the BG correction factor (mg / dL / U) (or a measure of insulin sensitivity), BG is the mealtime self-monitored BG value or CGM reading, and IOB is the active insulin from the previous basal and correction insulin injections.
[0113] The IOB component is an important factor in explaining the lasting effects of previously administered insulin on blood glucose. The similarity between the effects of physical activity and insulin is that physical activity also has a prolonged effect on blood glucose. Therefore, a physical activity-based insulin dosing method was designed that adds a physical activity component inspired by the IOB concept to the standard therapeutic dosing recipe. This component corresponds to the amount of insulin required to compensate for the estimated glycemic perturbation associated with previously performed physical activity. The modulation relies on a wearable activity tracker that calculates the cumulative physical activity by the weighted sum of the historical step count over the past 12 hours. The resulting metric, AOB, characterizes the cumulative physical activity that is still actively affecting glucose uptake. Once the AOB is obtained at mealtime dosing, the physical activity-based dosing amount can be calculated as follows:
[0114] AOB d,m is the AOB calculated for meal m consumed on day d, AOB profile,m is a curve that records the participants’ daily cumulative physical activity around a selected standard meal (e.g., breakfast, lunch, or dinner). Note that AOB profile,m As a reference to the patient's usual cumulative physical activity at the selected mealtime, an average treatment is designed for that mealtime. Deviations from usual physical activity are expected to result in changes in insulin requirements, and dosing based on physical activity is designed to adjust insulin doses accordingly. A patient-specific dosing correction parameter, the activity factor (AF), converts the expected blood glucose changes produced by physical activity deviations into insulin units with similar impact. Therefore, the AF and the above formula can be used to modify the disorder curve, enhance the probability analysis of the prediction model and / or be modified by the expected curve itself. For example, the AF can be used as a multiplier for the disorder curve to adjust the difference between the observed activity amount and the activity amount in the expected curve (converting the activity amount to insulin amount). Another example of use can be AOB d,m -AOB profile,m The quantity is used to determine the distance between the data and the curve (rather than the direct activity tracker measurement). Finally, PA-based dosing can use various curves (indexed by i below) that are matched by the above matching process with weights pi (derived from the matching distance) notification. In this case, the dosage can be calculated as:
[0115] The direction of the resulting dosing adjustment depends on the AOB d,m Is it greater than or less than AOB? profile,m .AOB profile,m The magnitude of is a function of the amount of deviation from the curve and the value of AF. Following the steps provided in the Supplementary File and detailed in Ozaslan et al., optimized meal dosing parameters CR and AF for physical activity-based dosing calculations were obtained for each participant.
[0116] For the calculation of lunchtime dosing, optimized CRs were used at both clinic visits, whereas AF was used only at the experimental visit to correct for prior physical activity.
[0117] result
[0118] The primary outcome measure was the duration of hypoglycemia during the postprandial window after lunch. Secondary outcomes were time spent in the target range (70 mg / dL-180 mg / dL), low blood glucose index (LBGI), and high blood glucose index (HBGI). Paired Wilcoxon rank tests were used to compare the percentage of time spent in the target range and the HBGI index because of their non-Gaussian distribution. Paired t-tests were used to compare the percentage of time spent in the target range and the HBGI index. To assess the potential of both dosing methods to alleviate hypoglycemia, the rate of change of glucose in the first 2 hours after the meal was reported. This time window corresponds to the phase with the greatest rate of glucose uptake caused by physical activity. Linear mixed-effects analysis was used for this assessment, with covariates being the categorical study visit variables and the continuous lunchtime glucose variable. In the model, treatment for hypoglycemia from 1 hour before lunch to the end of the postprandial window was controlled. Participant effects were considered random effects.
[0119] result
[0120] Fifteen subjects (6 males and 9 females) completed the study. Table 5 lists the demographic characteristics of the participants, expressed as mean ± SD. Table 5 - Demographic characteristics of study participants
[0121] For four participants, CR optimization for lunch was not feasible (e.g., lack of mealtime refills between 11:00 AM and 2:00 PM on most days). Therefore, the original lunchtime CR for these participants during both study visits was used. For six participants, optimization failed to find the best AF within the optimization boundaries. In these cases, 2200 cumulative steps were used as an approximate mean AF. In cases where CR optimization was feasible, it resulted in a 2% ± 18% change in the participant's original CR, but there was no statistically significant difference (original: 10.6 ± 2.7 g / U, optimized: 11 ± 3.8 g / U, P = .64). The mean of the best AFs calculated from the nine patients who could be optimized was 2926 ± 1267 cumulative steps / U.
[0122] During the study visits, the reduction in medication dosing associated with physical activity averaged 28% ± 15% of total medication dosing. Between study visits, participants had similar glucose values at lunchtime (control: 126 ± 41 mg / dL, study: 127 ± 35 mg / dL, P = .94). There were no significant differences between visits for AOB at lunchtime (control: cumulative steps 5438 ± 1048, study: cumulative steps 5258 ± 969, P = .32). Nine episodes of hypoglycemia occurred within the hour before lunch (5 in the study group and 4 in the control group) and were treated with rescue carbohydrates. Post-lunch hypoglycemia was quite rare during both visits; only 3 of 30 observations had glucose levels less than 70 mg / dL (1 in the study visit and 2 in the control visit) and were treated with 25.7 ± 5.1 g of rescue carbohydrates. Thus, there was no statistically significant difference in the percentage of time spent in hypoglycemia between visits (median [IQR], control group: 0 [0]%, experimental group: 0 [0]% P = 1). The percentage of time spent in the target range was lower than the percentage of time under physical activity-based dosing (control group: 77% ± 27%, experimental group: 59% ± 31%, P = .03). However, no significant differences were observed for the LBGI (median [IQR], control group: 0 [0.03], experimental group: 0 [0.02], P = .62) or HBGI (control group: 0.82 ± 0.8, experimental group: 0.78 ± 0.6 in the post-lunch period, P = .83).
[0123] Overall, afternoon physical activity measured by total steps during the study visit was significantly lower than participants' daily physical activity extracted from the data collection period (afternoon physical activity, control group vs. data collection group, experimental group vs. data collection group, control group: 1095 ± 451 steps, experimental group: 1288 ± 702 steps, data collection group: 2555 ± 1618 steps, P < .01). There was no significant difference between the total number of steps taken by participants in the afternoon portion of the control and experimental groups (P = .18).
[0124] like Fig. 9 As shown, on average, standard-of-care dosing produced late post-dinner glucose levels below mean morning glucose, whereas physical activity-based dosing produced glucose levels above this reference value; although not statistically significant (control group ΔCGM 参考 :-16±74mg / dL, experimental group ΔCGM 参考 : 10 ± 68 mg / dL, P = .09). Differences in the direction and magnitude of glucose changes in the early and late postlunch glucose excursions were assessed by CGM slope in a linear mixed-effects regression analysis, and the results are shown in Table 6. Although the study visit was a significant factor associated with CGM trends in the first two hours after the dosing injection (P = .032), there was no significant difference in CGM trends between visits in the late postlunch period (second and fourth hours after lunch) (P = .71).
[0125] Fig. 9 The changes in CGM for the control group (dark grey) and the experimental group (light grey) are shown. The reference CGM value for each visit and participant was their average morning CGM, which was calculated based on the CGM readings between 7:30 and 10:45 am of the relevant visit. All values were then obtained by subtracting the subject- and visit-specific reference CGM values from the CGM readings observed between 7:30 am and 16:30 pm during the study visit. Blood glucose values varied more in the afternoon than in the morning because the study physician managed glucose levels in the morning to a safe glucose range for physical activity. In addition, the meal range for lunch was wider than that for breakfast.
[0126] In the control visits, the CGM slope in the first two hours was negative in 40% of the observations, decreasing to 7% in the experimental visits (see Fig.10 ). Fig.10 Shown are box plot representations of the post-lunch CGM slope trends observed in the first 2 hours (left) and second 2 hours (right) after lunch values in the study visit. Table 6 - Regression analysis results of the post-lunch CGM slope trend in the first 2 hours after lunch (left) and the second 2 hours after lunch (right)
[0127] discuss
[0128] One goal of this pilot feasibility study was to evaluate whether step-based physical activity information could be used to enhance hypoglycemia protection with meal dosing after brisk walking. Negative CGM slopes in the early postprandial phase support the need to formally incorporate prior physical activity into insulin dosing calculations. While a physical activity-based dosing approach was able to successfully compensate for the drop in blood glucose levels, the results suggest that it may be too conservative with respect to hypoglycemia risk because it produced a mean blood glucose trend that was higher than the participant-specific reference blood glucose mean calculated in the experimental arm of this study. In other words, the results showed that standard meal therapy after walking activity, with blood glucose levels below the reference level, represented an increased risk of hypoglycemia, although not as much as expected. Despite full dosing in the control group, a considerably lower hypoglycemia rate was observed, as well as a longer time spent in the target range, compared with the experimental group. More specifically, only three cases of post-lunch hypoglycemia occurred during the trial. Therefore, participants did benefit from the walking exercise in terms of time spent in the target range. These findings suggest that additional factors should be considered when adjusting insulin dosing for physical activity. Factors that attenuated the activity-induced increase in blood glucose uptake in this study may have included the high protein and fat content of the breakfast, which was designed to maintain minimal IOB during walking exercise. In addition, the afternoon portion of the study incorporated minimal physical activity, which was significantly lower than the participants' usual afternoon physical activity, and treatment parameters were prescribed and optimized based on usual afternoon physical activity. These circumstances, combined with the unused rescue carbohydrates at lunch, may have contributed to the higher-than-expected blood glucose values.
[0129] The variable “experimental visit” in the regression model is clear, and the coefficient associated with it shows that there is a marginal difference in slope in the experimental visits compared to the control visits.
[0130] There are three aspects of the approach taken in this paper that warrant further investigation. First, the physical activity participated in is accounted for by convolution with a reference signal. Different reference signal values may affect the outcome of dosing decisions. It can be assumed, based on mathematical intuition, that if the reference signal is close enough to the decremental nature of the effect of physical activity on blood glucose, the corresponding adjustment decision will be effective for the intended purpose. Second, there is limited literature on the long-term effects of physical activity on blood glucose in patients with T1D. For example, the reference study showed a biphasic change in glucose uptake in adolescents after they exercised late in the day. However, Davey et al. showed that when exercise was performed at noon, the change in glucose uptake was quite stable from 1 to 11 hours after exercise. Similar studies in adults at different times of the day and with different physical activity patterns could provide more accurate signals for this approach. A third aspect is that the range of effect of physical activity is longer (12 hours) than that of insulin injection (5 hours). Since the AOB predicts a further drop in blood glucose within hours after insulin administration, this difference may lead to excessive reductions in insulin doses during physical activity. Furthermore, based on the study by McMahon et al., which looked at blood glucose results up to five hours after the end of physical activity, it is expected that about half of the total blood glucose effect of the physical activity performed would occur in the hours after our observation window. One possible way to reconcile this discrepancy would be to perform a physical activity correction to compensate only for the changes expected to occur in the duration of insulin action.
[0131] Of note, during the first 2 h of the post-lunch phase, standard dosing produced a trend toward greater reductions in blood glucose compared with dosing based on physical activity. These results are consistent with current knowledge of increased glucose uptake within 2 h after moderate-intensity physical activity.
[0132] In conclusion, this pilot study provides early safety and feasibility data for a novel approach to quantify physical activity by a commonly available measure (step count) and to use this measure to adjust prandial insulin doses. This approach can be adapted to different physical activity indices, physical activity accumulation methods, and changes in individual baseline physical activity behavior through the parameters AOB and AF. The results suggest that individualized treatment adjustments by quantifying physical activity based on step counts are safe, feasible, and have the potential to mitigate the BG decline caused by early postprandial physical activity in patients with T1D.
[0133] Example 2 demonstrates that glucose excursions can occur during and after physical activity, leading to barriers to optimal insulin dosing in patients with type 1 diabetes (T1D). A pilot clinical trial evaluated the safety and feasibility of physical activity-based mealtime insulin dosing recommendations that adjust mealtime dosing according to prior physical activity based on step data collected via an off-the-shelf physical activity tracker. Fifteen adults with T1D, each using a continuous glucose monitor (CGM) and an insulin pump with carbohydrate counting, completed two randomized crossover daily visits. Participants performed a 30- to 45-minute brisk walk before lunch, and lunchtime insulin dosing was calculated according to their standard of care (ST) or physical activity-based dosing method. Post-lunch glucose excursions were assessed using CGM readings. There was no significant difference between the two visits during post-lunch hypoglycemia (median [IQR] standard: 0 [0]% vs physical activity-based: 0 [0]%, P = NS). Standard-of-care dosing spent more time in the 70 mg / dL to 180 mg / dL target range (mean ± SD: 77% ± 27% vs physical activity-based: 59% ± 31%, P = .03). However, it was associated with a steeper negative slope in the early postprandial period (P = .032). Therefore, using step count to adjust prandial insulin after walking has been shown to be safe and feasible in a cohort of 15 T1D subjects. Physical activity-based insulin dosing of meals taken shortly after walking has the potential to mitigate the glucose drop associated with physical activity in the early postprandial period.
[0134] Examples 1 and 2 demonstrate that a personalized multistage model predictive control (MS MPC) algorithm for glycemic stabilization and improved postprandial glucose control in patients with type 1 diabetes (T1D) with consistent meal patterns is achievable. The multistage concept can utilize different meal patterns as perturbed implementations into the glucose-insulin system, and the best possible control input can then be evaluated in all possible scenarios. Perturbed implementations in the form of glucose occurrence rate tracking can be estimated by using meal records (time and carbohydrate amount) as inputs to a personalized oral model. Meal features can then be clustered using a k-medoids algorithm to obtain meal patterns. Both hybrid closed-loop (HCL) and fully closed-loop (FCL) MS-MPC methods were tested on a full computer adult cohort using the FDA-approved UVA / Padova metabolic simulator and compared with their respective control treatments (hybrid and fully automated MPC, respectively). The results confirmed that at 70mg / dl-180mg / dl, the postprandial time and total time percentage were improved by 85.2±15.5, 89.6±12.2 and 94.1±6.3, 95.7±5.0 respectively when using the HCL method; and by 37.8±15.7, 63.4±16.6 and 65.8±12.7, 82.2±9.2 respectively when using the FCL method.
[0135] In healthy conditions, the pancreas regulates the production of insulin, which causes glucose to be excreted from the plasma, and glucagon, which stimulates the liver to release glucose into the blood. In T1D, this precise feedback process is dysregulated. The pancreas does not produce insulin, resulting in high blood glucose (BG), also known as hyperglycemia. To prevent this, exogenous insulin must be administered lifelong to maintain blood glucose levels within a safe range.
[0136] Treatment guidelines recommend that patients with T1D calculate mealtime insulin doses based on carbohydrate counting (CC) and insulin therapy parameters (American Diabetes Association (2016), Tascini et al., 2018). In addition, to minimize the effect of ingested carbohydrates on glucose levels, insulin should ideally be injected 15–20 minutes before meals (Slattery et al., 2018). Results related to accuracy in CC in patients with T1D vary between studies, but there is a consensus that errors in CC are common in real life. Deeb et al. reported that children and adolescents could only make accurate carbohydrate estimates for 67% of meals (Deeb et al., 2017). Another study showed that adults with T1D could only estimate the carbohydrate content of foods to within 5 grams of the true amount 44% of the time (Meade and Rushton, 2016). A survey by Lancaster et al. reported that adolescents with T1D perceived CC as a barrier to achieving glycemic control (Lancaster et al., 2010).
[0137] Postprandial glucose management remains a challenge for conventional artificial pancreas systems (APS). In hybrid closed-loop (HCL) approaches, inaccurate CC and parameters such as carbohydrate ratio (CR) and insulin-to-carbohydrate correction factor (ICF) have a major impact on glucose control as under / overdosage of insulin may occur ( Bally et al., 2017 , Kovachev et al., 2017 ). Although promising results have been reported for fully closed-loop (FCL) systems, balancing the aggressiveness of the controller to mitigate meal-related glucose excursions without increasing the risk of late hypoglycemia remains an open challenge for conventional systems ( Dassau et al., 2013 ; Turksoy et al., 2016 ; Bally et al., 2017 ; Colmegna et al., 2018 ; Sánchez- et al., 2018).
[0138] Pre-meal prediction can improve post-meal control of APS in HCL and FCL systems. For example, Hughes et al. explored the delay between subcutaneous blood glucose measurement and insulin effect by using randomized meal profiles to predict meals (Hughes et al., 2011). The in silico results in this paper showed that the HCL approach using meal anticipation and insulin dosing resulted in attenuated post-meal blood glucose excursions compared with an open-loop control using meal dosing.
[0139] Fig.11 A block diagram illustrating an example of a machine upon which one or more aspects of embodiments of the invention may be implemented.
[0140] refer to Fig.11 , one aspect of embodiments of the present invention includes, but is not limited to, systems, methods, and computer-readable media, which provide: a model for notifying, determining, or controlling insulin dosage, which illustrates a block diagram of an example machine 1100 on which one or more embodiments (e.g., the methods discussed) can be implemented (e.g., run).
[0141] Examples of machine 1100 may include logic, one or more components, circuits (e.g., modules), or mechanisms. A circuit is a tangible entity configured to perform certain operations. In one example, the circuits may be arranged in a specified manner (e.g., internally or relative to external entities such as other circuits). In one example, one or more computer systems (e.g., stand-alone, client, or server computer systems) or one or more hardware processors (processors) may be configured by software (e.g., instructions, application portions, or applications) to operate as circuits to perform certain operations described herein. In one example, the software may reside (1) on a non-transitory machine-readable medium (e.g., non-transitory, non-volatile memory), or (2) in a transmission signal. In one example, when the software is executed by the underlying hardware of the circuit, it causes the circuit to perform certain operations.
[0142] In one example, the circuit can be implemented mechanically or electronically. For example, the circuit may include dedicated circuits or logic that are specifically configured to perform one or more of the techniques described above, such as including a dedicated processor, a field programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). In one example, the circuit may include programmable logic (e.g., a circuit contained within a general-purpose processor or other programmable processor) that can be temporarily configured (e.g., by software) to perform certain operations. It should be understood that the decision to implement the circuit mechanically (e.g., in a dedicated and permanently configured circuit) or in a temporarily configured circuit (e.g., configured by software) can be driven by cost and time considerations.
[0143] Thus, the term "circuit" is understood to include a tangible entity that may be physically constructed, permanently configured (e.g., hardwired) or temporarily (e.g., temporarily) configured (e.g., programmed) to operate in a particular manner or perform a particular operation. In one example, given a plurality of temporarily configured circuits, each circuit need not be configured or instantiated at any one instance in time.
[0144] For example, where the circuit includes a general purpose processor configured by software, the general purpose processor may be configured as corresponding different circuits at different times. The software may configure the processor accordingly, for example, to constitute a particular circuit at one time instance and to constitute a different circuit at a different time instance.
[0145] In one example, a circuit can provide information to other circuits and receive information from other circuits. In this example, a circuit can be considered to be communicatively coupled to one or more other circuits. When multiple such circuits exist simultaneously, communication can be achieved by signal transmission connecting these circuits (e.g., through appropriate circuits and buses). In embodiments where multiple circuits are configured or instantiated at different times, communication between these circuits can be achieved, for example, by storing and retrieving information in a memory structure accessible to multiple circuits. For example, a circuit can perform an operation and store the output of the operation in a storage device communicatively coupled thereto. Then, another circuit can access the memory device at a later time to retrieve and process the stored output. In one example, a circuit can be configured to initiate or receive communication with an input device or an output device, and can operate on a resource (e.g., a collection of information).
[0146] The various operations of the method examples described herein may be performed at least in part by one or more processors that are temporarily configured (e.g., by software) or permanently configured to perform the relevant operations. Whether temporarily configured or permanently configured, such processors may constitute a processor-implemented circuit that operates to perform one or more operations or functions. In one example, the circuits mentioned herein may include processor-implemented circuits.
[0147] Similarly, the methods described herein may be implemented at least in part by a processor. For example, at least some of the operations of the method may be performed by one or more processors or circuits implemented by the processor. The performance of certain operations may be distributed among one or more processors, not only residing in a single machine, but also deployed on multiple machines. In one example, one or more processors may be located at a single location (e.g., in a home environment, an office environment, or as a server farm), while in other examples, the processors may be distributed at multiple locations.
[0148] One or more processors may also operate in a "cloud computing" environment or as "software as a service" (SaaS) to support performance of related operations. For example, at least some operations may be performed by a group of computers (as an example of a machine including a processor), which may be accessed via a network (e.g., the Internet) and via one or more appropriate interfaces (e.g., an Application Program Interface (API)).
[0149] Example embodiments (e.g., devices, systems, or methods) may be implemented in digital electronic circuitry, computer hardware, firmware, software, or any combination thereof. Example embodiments may be implemented using a computer program product (e.g., a computer program embodied in an information carrier or machine-readable medium for execution by, or to control the operation of, a data processing device such as a programmable processor, a computer, or multiple computers).
[0150] A computer program may be written in any form of programming language, including compiled or interpreted languages, and may be deployed in any form, including as a stand-alone program or as a software module, subroutine or other unit suitable for use in a computing environment. A computer program may be deployed to be executed on one computer or on multiple computers at one site, or distributed across multiple sites and interconnected by a communication network.
[0151] In one example, the operations may be performed by one or more programmable processors executing a computer program to perform functions by operating on input data and generating output. Examples of method operations may also be performed by special purpose logic circuitry (e.g., a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC)), and example devices may be implemented as special purpose logic circuitry.
[0152] A computing system may include a client and a server. The client and the server are usually remote from each other and usually interact through a communication network. The relationship between the client and the server is due to the computer programs running on their respective computers and the client-server relationship between each other. In the embodiment of the deployment of a programmable computing system, it should be understood that both hardware and software architectures need to be considered. Specifically, it should be understood that it can be a design choice to choose whether to implement certain functions in permanently configured hardware (e.g., ASIC), temporarily configured hardware (e.g., a combination of software and a programmable processor), or a combination of permanently and temporarily configured hardware. The hardware (e.g., machine 1100) and software architecture that can be deployed in an example embodiment are listed below.
[0153] In one example, the machine 1100 may operate as a standalone device, or the machine 1100 may be connected (e.g., using a network) to other machines.
[0154] In a network deployment, the machine 1100 can operate in the capacity of a server or a client machine in a server-client network environment. In an example, the machine 1100 can act as a peer machine in a peer-to-peer (or other distributed) network environment. The machine 1100 can be a personal computer (PC), a tablet computer, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a network appliance, a network router, a switch or a bridge, or any machine capable of executing instructions (sequential or otherwise) for the specified machine 1100 to take (e.g., perform) operations. In addition, although only a single machine 1100 is shown, the term "machine" should also be understood to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methods discussed herein.
[0155] The example machine (e.g., computer system) 1100 may include a processor 1102 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), or both), a main memory 1104, and a static memory 1106, some or all of which may communicate with each other via a bus 1108. The machine 1100 may also include a display unit 1110, an alphanumeric input device 111 (e.g., a keyboard), and a user interface (UI) navigation device 1114 (e.g., a mouse). In an example, the display unit 1110, the input device 1117, and the UI navigation device 1114 may be a touch screen display. The machine 1100 may also include a storage device (e.g., a drive unit) 1116, a signal generating device 1118 (e.g., a speaker), a network interface device 1120, and one or more sensors 1121, such as a global positioning system (GPS) sensor, a compass, an accelerometer, or other sensors.
[0156] The storage device 1116 may include a machine-readable medium 1122 having stored thereon one or more sets of data structures or instructions 1124 (e.g., software) that embody or are utilized by any one or more of the methods or functions described herein. The instructions 1124 may also reside, in whole or in part, in the main memory 1104, in the static memory 1106, or in the processor 1102 during execution of the instructions 1124 by the machine 1100. In an example, one or any combination of the processor 1102, the main memory 1104, the static memory 1106, or the storage device 1116 may constitute a machine-readable medium. Although the machine-readable medium 1122 is illustrated as a single medium, the term "machine-readable medium" may include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) configured to store one or more instructions 1124. The term "machine-readable medium" may also be understood to include any tangible medium that is capable of storing, encoding or carrying instructions executed by a machine and causing the machine to perform any one or more methods of the present disclosure, or capable of storing, encoding or carrying data structures used by or associated with these instructions. Therefore, the term "machine-readable medium" may be understood to include, but is not limited to, solid-state memory, optical and magnetic media. Specific examples of machine-readable media may include non-volatile memory, for example, semiconductor memory devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices; disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.
[0157] Instructions 1124 may also be sent or received over the communications network 1126 using a transmission medium via the network interface device 1120 using any of a variety of transmission protocols (eg, Frame Relay, IP, TCP, UDP, HTTP, etc.).
[0158] Example communication networks may include a local area network (LAN), a wide area network (WAN), a packet data network (e.g., the Internet), a mobile phone network (e.g., a cellular network), a Plain Old Telephone (POTS) network, and a wireless data network (e.g., a wireless network called The IEEE 802.11 family of standards, known as The term "transmission medium" shall be understood to include any intangible medium that is capable of storing, encoding or carrying instructions for machine execution, and includes digital or analog communication signals or other intangible media that facilitates such software communication.
[0159] It should be understood that the embodiments disclosed herein can be modified to meet a specific set of design criteria. For example, any component or feature can be of any suitable number or type to meet a specific goal. Therefore, although certain exemplary embodiments of the systems, devices, and methods of making and using the systems and devices disclosed herein have been discussed and described, it should be clearly understood that the present invention is not limited thereto, but can be implemented and practiced differently within the scope of the following claims.
[0160] It should be understood that some components, features, and / or configurations may be described in conjunction with only one particular embodiment, but these same components, features, and / or configurations may be applied or used with many other embodiments and should be considered applicable to other embodiments unless otherwise specified or unless such components, features, and / or configurations are technically impossible to use with other embodiments. Therefore, the components, features, and / or configurations of the various embodiments may be combined together in any manner, and such combinations are specifically contemplated and disclosed by this statement.
[0161] Those skilled in the art will appreciate that the present invention can be embodied in other specific forms without departing from its spirit or essential features. Therefore, the embodiments disclosed at present are considered to be illustrative rather than restrictive in all aspects. The scope of the present invention is indicated by the appended claims rather than the foregoing description, and all changes and equivalents thereof within its meaning and scope are intended to be included therein. In addition, the range of disclosed values discloses each numerical value within the range, including endpoints.
[0162] The following references are incorporated herein by reference in their entirety. 1.V.Gingras,N.Taleb,A.Roy-Fleming,L.Legault,and R.Rabasa-Lhoret,“Thechallenges of achieving postprandial glucose control using closed-loopsystems in patients with type 1diabetes,”Diabetes,Obes.Metab.,vol.20,no.2,pp.245-256,Feb.2018,doi:10.1111 / dom.13052. 2.J.P.Corbett,P.Colmegna,J.Garcia-Tirado,and M.D.Breton,“AnticipatingMeals with Behavioral Profiles in an Artificial Pancreas System-An InformedMultistage Model Predictive Control Approach,”IFAC-PapersOnLine,vol.53,no.2,pp.16305-16310,Jan.2020,doi:10.1016 / j.ifacol.2020.12.652. 3.R.A.Harvey et al.,“Clinical evaluation of an automated artificialpancreas using zone-model predictive control and health monitoring system,”Diabetes Technol.Ther.,vol.16,no.6,pp.348-357,Jun.2014,doi:10.1089 / dia.2013.0231. 4.G.P.Forlenza et al.,“Fully Closed-Loop Multiple Model ProbabilisticPredictive Controller Artificial Pancreas Performance in Adolescents andAdults in a Supervised Hotel Setting,”Diabetes Technol.Ther.,vol.20,no.5,pp.335-343,May 2018,doi:10.1089 / dia.2017.0424. 5.K.Dovc et al.,“Faster Compared with Standard Insulin Aspart duringDay-And-Night Fully Closed-Loop Insulin Therapy in Type 1Diabetes:A Double-Blind Randomized Crossover Trial,”Diabetes Care,vol.43,no.1,pp.29-36,Jan.2020,doi:10.2337 / dc19-0895. 6.H.Blauw,A.J.Onvlee,M.Klaassen,A.C.van Bon,and J.H.DeVries,“FullyClosed Loop Glucose Control With a Bihormonal Artificial Pancreas in AdultsWith Type 1Diabetes:An Outpatient,Randomized,Crossover Trial,”Diabetes Care,vol.44,no.3,pp.836-838,Mar.2021,doi:10.2337 / dc20-2106. 7.A.Haidar et al.,“Reducing the need for carbohydrate counting intype 1 diabetes using closed-loop automated insulin delivery(artificialpancreas)and empagliflozin:A randomized,controlled,non-inferiority,crossoverpilot trial,”Diabetes,Obes.Metab.,2021,doi:10.1111 / dom.14335. 8.D.Majdpour et al.,“Fully Automated Artificial Pancreas for Adultswith Type 1 Diabetes using Multiple Hormones:Exploratory Experiments,”Can.J.Diabetes,Feb.2021,doi:10.1016 / j.jcjd.2021.02.002. 9.J.Garcia-Tirado,J.P.Corbett,D.Boiroux,J.B. and M.D.Breton,“Closed-loop control with unannounced exercise for adults with type 1diabetes using the Ensemble Model Predictive Control,”J.Process Control,vol.80,pp.202-210,Aug.2019,doi:10.1016 / j.jprocont.2019.05.017. 10.J.F.Garcia-Tirado,P.Colmegna,J.P.Corbett,B.Ozaslan,and M.D.Breton,“Ensemble Model Predictive Control Strategies Can Reduce ExerciseHypoglycemia in Type 1 Diabetes:In Silico Studies,”in 2019 American ControlConference(ACC),2019,vol.2019-July,pp.4752-4758,doi:10.23919 / acc.2019.8814728. 11.J.Garcia-Tirado,P.Colmegna,J.P.Corbett,B.Ozaslan,and M.D.Breton,“In Silico Analysis of an Exercise-Safe Artificial Pancreas With MultistageModel Predictive Control and Insulin Safety System,”J.Diabetes Sci.Technol.,vol.13,no.6,2019,doi:10.1177 / 1932296819879084. 12.J.Garcia-Tirado et al.,“Anticipation of Historical ExercisePatterns by a Novel Artificial Pancreas System Reduces Hypoglycemia Duringand After Moderate-Intensity Physical Activity in People with Type 1Diabetes,”Diabetes Technol.Ther.,vol.23,no.4,Dec.2020,doi:10.1089 / dia.2020.0516. 13.B.Ozaslan et al.,“Safety and Feasibility Evaluation of Step CountInformed Meal Boluses in Type 1 Diabetes:A Pilot Study,”J.DiabetesSci.Technol.,p.193229682199791,Apr.2021,doi:10.1177 / 1932296821997917. 14.C.Ellingsen et al.,“Safety constraints in an artificial pancreaticβcell:An implementation of model predictive control with insulin on board,”J.Diabetes Sci.Technol.,vol.3,no.3,pp.536-544,2009,doi:10.1177 / 193229680900300319. 15.R.W.Hamming,“Error Detecting and Error Correcting Codes,”BellSyst.Tech.J.,vol.29,no.2,pp.147-160,Apr.1950,doi:10.1002 / j.1538-7305.1950.tb00463.x. 16.S.P.Lloyd,“Least Squares Quantization in PCM,”1982.Accessed:Aug.15,2019.[Online].Available: https: / / sites.cs.ucsb.edu / ~veronika / MAE / kmeans_LLoyd_Least_Squares_Quantization_in_PCM.pdf. 17.T. and J.Harabasz,“A Dendrite Method Foe Cluster Analysis,”Commun.Stat.,vol.3,no.1,pp.1-27,1974,doi:10.1080 / 03610927408827101. 18.S.D.Patek,“Open-loop feedback control under multiple disturbancefunction hypotheses,”2010,doi:10.1109 / CDC.2010.5717476. 19.S.A.Brown et al.,“Six-Month Randomized,Multicenter Trial ofClosed-Loop Control in Type 1 Diabetes,”N.Engl.J.Med.,vol.381,no.18,pp.1707-1717,Oct.2019,doi:10.1056 / NEJMoa1907863. 20.R.Visentin et al.,“The UVA / Padova Type 1 Diabetes Simulator GoesFrom Single Meal to Single Day,”J.Diabetes Sci.Technol.,vol.0,no.0,pp.1-9,2018,doi:10.1177 / 1932296818757747. 21.D.M.Maahs et al.,“Outcome measures for artificial pancreasclinical trials:A consensus report,”Diabetes Care,vol.39,no.7,pp.1175-1179,Jul.2016,doi:10.2337 / dc15-2716.22.J.W.White,A.Rassweiler,J.F.Samhouri,A.C.Stier,and C.White,“Ecologists should not use statistical significancetests to interpret simulation model results,”Oikos,vol.123,no.4,pp.385-388,Apr.2014,doi:10.1111 / j.1600-0706.2013.01073.x. 23.Warburton DER,Nicol CW,Bredin SSD,“Health benefits of physicalactivity:the evidence,”Can Med Assoc J.2006;174(6):801-809. 24.Paffenbarger RS,Hyde R,Wing AL,Hsieh C,“Physical activity,all-cause mortality,and longevity of college alumni,”N Engl J Med.1986;314(10):605-613. 25.Balducci S,Iacobellis G,Parisi L,et al.,“Exercise training canmodify the natural history of diabetic peripheral neuropathy,”J DiabetesComplications 2006;20(4):216-223. 26.Bohn B,Herbst A,Pfeifer M,et al.,“Impact of physical activity onglycemic control and prevalence of cardiovascular risk factors in adults withtype 1 diabetes:a cross-sectional multicenter study of 18,028 patients,”Diabetes Care.2015;38(8):1536-1543. 27.Brazeau A-S,Rabasa-Lhoret R,Strychar I,Mircescu H,“Barriers tophysical activity among patients with type 1 diabetes,”Diabetes Care.2008;31(11):2108-2109. 28.Riddell MC,Gallen IW,Smart CE,et al.,“Exercise management in type1 diabetes:a consensus statement,”Lancet Diabetes Endocrinol.2017;5(5):377-390. 29.Colberg SR,Sigal RJ,Yardley JE,et al.,“Physical activity / exerciseand diabetes:a position statement of the American Diabetes Association,”Diabetes Care.2016;39(11):2065-2079. 30.Cinar A,“Multivariable adaptive artificial pancreas system in type1 diabetes,”Curr Diab Rep.2017;17(10):88. 31.Riddell MC,Zaharieva DP,Yavelberg L,Cinar A,Jamnik VK,“Exerciseand the development of the artificial pancreas:one of the more difficultseries of hurdles,”J Diabetes Sci Technol.2015;9(6):1217-1226. 32.DeBoer MD, DR,Topchyan K,Kovatchev BP,Francis GL,BretonMD,“Heart rate informed artificial pancreas system enhances glycemic controlduring exercise in adolescents with T1D,”Pediatr Diabetes.2017;18(7):540-546. 33.Breton MD,Brown SA,Karvetski CH,et al.,“Adding heart rate signalto a control-to-range artificial pancreas system improves the protectionagainst hypoglycemia during exercise in type 1 diabetes,”Diabetes TechnolTher.2014;16(8):506-511. 34.Turksoy K,Bayrak ES,Quinn L,Littlejohn E,Cinar A,“Multivariableadaptive closed-loop control of an artificial pancreas without meal andactivity announcement,”Diabetes Technol Ther.2013;15(5):386-400. 35.Garcia-Tirado J,Brown SA,Laichuthai N,et al.,“Anticipation ofhistorical exercise patterns by a novel artificial pancreas system reduceshypoglycemia during and after moderate-intensity physical activity in peoplewith type 1 diabetes,”Diabetes Technol Ther.Published online December 1,2020.doi:10.1089 / dia.2020.0516 36.Dunstan DW,Kingwell BA,Larsen R,et al.,“Breaking up prolongedsitting reduces postprandial glucose and insulin responses,”DiabetesCare.2012;35(5):976-983. 37.Bailey DP,Locke CD.“Breaking up prolonged sitting with light-intensity walking improves postprandial glycemia,but breaking up sitting withstanding does not”.J Sci Med Sport.2015;18(3):294-298. 38.Peddie MC,Bone JL,Rehrer NJ,Skeaff CM,Gray AR,Perry TL,“Breakingprolonged sitting reduces postprandial glycemia in healthy,normal-weightadults:a randomized crossover trial,”Am J Clin Nutr.2013;98(2):358-366. 39.Manohar C,Levine JA,Nandy DK,et al.,“The effect of walking onpostprandial glycemic excursion in patients with type 1 diabetes and healthypeople,”Diabetes Care.2012;35(12):2493-2499. 40.Ozaslan B,Patek SD,Breton MD,“Impact of daily physical activity asmeasured by commonly available wearables on meal time glucose control in type1 diabetes,”Diabetes Technol Ther.2020;22(10):742-748. 41.Cappon G,Marturano F,Vettoretti M,Facchinetti A,Sparacino G,“Insilico assessment of literature insulin bolus calculation methods accountingfor glucose rate of change,”J Diabetes Sci Technol.2019;13(1):103-110. 42.Swan KL,Dziura JD,Steil GM,et al.,“Effect of age of infusion siteand type of rapid-acting analog on pharmacodynamic parameters of insulinboluses in youth with type 1 diabetes receiving insulin pump therapy,”Diabetes Care.2009;32(2):240-244. 43.McMahon SK,Ferreira LD,Ratnam N,et al.,“Glucose requirements tomaintain euglycemia after moderate-intensity afternoon exercise inadolescents with type 1 diabetes are increased in a biphasic manner,”J ClinEndocrinol Metab.2007;92(3):963-968. 44.Ozaslan B,Fabris C,Patek SD,Breton M,“Automatically accounting forphysical activity in insulin dosing for type 1 diabetes,”Comput MethodsPrograms Biomed.Published online September 21,2020.doi:10.1016 / j.cmpb.2020.105757 45.Davey RJ,Howe W,Paramalingam N,et al.,“The effect of mid-daymoderate-intensity exercise on postexercise hypoglycemia risk in individualswith type 1 diabetes,”J Clin Endocrinol Metab.2013;98(7):2908-2914.
Claims
1. A system for notifying, determining or controlling insulin dosage, the system include: processor; as well as a memory comprising instructions to cause the processor to: generating a plurality of disorder curves, each disorder curve being based on a data representation of historical patient data associated with a deviation from a threshold blood glucose level; receiving current patient data; wherein the current patient data and the historical patient data include any data indicating patient characteristics; Apply a predictive model where: The processor is configured by the instructions to compare the current patient data to at least one disturbance curve; and The processor is configured by the instructions to assess, based on a probability analysis, a likelihood that a turbulent curve among the plurality of turbulent curves is an expected turbulent curve, the expected turbulent curve being a turbulent curve determined based on the probability analysis to match the current patient data; determining an insulin dose based on the expected disturbance profile; and A signal representative of the insulin dose is output to a device configured to monitor, influence and / or manage a patient's insulin levels.
2. The system according to claim 1, in: The plurality of turbulence curves and / or the prediction model are stored in the memory; or The plurality of turbulence curves and / or the prediction model are stored in a memory external to the processor, and the processor is configured to communicate with the external memory.
3. The system according to claim 1, in: The processor is configured by the instructions to compare the current patient data to at least one disturbance curve continuously, periodically, or at a predetermined time.
4. The system according to claim 3, in: The processor is configured by the instructions to compare the current patient data to at least one disturbance curve every five minutes.
5. The system according to claim 1, in: The historical patient data includes glucose measurements, meal intake data, physical activity measurements, and / or insulin injection data; and The current patient data includes glucose measurements, meal intake data, physical activity measurements, and / or insulin injection data.
6. The system according to claim 5, in: A deviation from the threshold blood glucose level of the disordered curve is the occurrence of a blood glucose level caused by physical activity and / or meal activity that increases the risk of hyperglycemia and / or hypoglycemia.
7. The system according to claim 6, in: The physical activity and / or dining activity includes physical activity and / or dining activity exhibited by the patient that deviates from average behavior associated with physical activity and / or dining activity.
8. The system according to claim 7, in: The physical activities and / or mealtime activities that increase the risk of hyperglycemia and / or hypoglycemia were identified by logistic regression.
9. The system according to claim 8, in: The historical patient data includes data related to more than one day; Clustering was used to group days with similar deviations from threshold blood glucose levels.
10. The system according to claim 1, in, The instructions cause the processor to: At least one turmoil curve is updated and / or a new turmoil curve is created based on current patient data.
11. The system according to claim 1, in, The instructions cause the processor to: A modified insulin dosage is determined based on the current patient data, previous insulin dosage data, and the patient's total daily insulin TDI.
12. The system according to claim 11, in: The modified insulin dose is also based on the active insulin IOB dose.
13. The system according to claim 11, in, The instructions cause the processor to: A signal representative of the modified insulin dose is output.
14. The system according to claim 11, in, The instructions cause the processor to: determining a current deviation from said threshold blood glucose level that increases the risk of hyperglycemia and / or hypoglycemia; and determining a dose of insulin to be administered based on the current deviation rather than on an expected disturbance profile; or The insulin dose based on the expected disturbance profile is modified to account for the current deviation.
15. The system of claim 1, in combination with: A glucose monitoring device, an activity tracking device, and / or a meal tracking device, any one or a combination of the foregoing devices are in communication with the processor and are configured to collect or generate patient data.
16. The combination according to claim 15, in: The processor is configured to receive the historical patient data and / or the current patient data from the glucose monitoring device, the activity tracking device, and / or the meal tracking device; and / or The glucose monitoring device, the activity tracking device and / or the meal tracking device are configured to transmit the historical patient data and / or the current patient data to a data store, wherein the processor is configured to receive the historical patient data and / or the current patient data from the data store.
17. The system of claim 1, in combination with: An insulin delivery device, a blood glucose status monitoring device, a glucose management system and / or an insulin recommendation system, any one or a combination of the aforementioned devices communicates with the processor and is configured to receive output from the processor.
18. The system according to claim 1, in, The instructions cause the processor to: configuring the output signal to be a notification message, the notification message causing the device to recommend administration of the insulin dose; configuring the output signal to be a command signal causing the device to administer the insulin dose; and / or The output signals are configured as data points to allow the device to graphically display the insulin dosage.
19. The system according to claim 13, in, The instructions cause the processor to: configuring the output signal as a notification message causing the device to recommend administration of the modified insulin dose; configuring the output signal to be a command signal, the command signal causing the device to administer the modified insulin dose; and / or The output signals are configured as data points to allow the device to graphically display the modified insulin dose.
20. The system according to claim 1, in, The processor is a control module of the automatic insulin delivery system.
21. A computer readable medium having instructions stored thereon, which, when executed by a processor, cause the processor to: generating a plurality of disorder curves, each disorder curve being based on a data representation of historical patient data associated with a deviation from a threshold blood glucose level; receiving current patient data; in, The current patient data and the historical patient data include any data indicative of patient characteristics; Apply a predictive model where: The instructions cause the processor to compare the current patient data to at least one disturbance curve; and The instructions cause the processor to evaluate, based on a probability analysis, a likelihood that a turbulent curve among the plurality of turbulent curves is an expected turbulent curve, the expected turbulent curve being a turbulent curve determined based on the probability analysis to match the current patient data; determining an insulin dose based on the expected disturbance profile; and A signal representative of the insulin dose is output to a device configured to monitor, influence and / or manage a patient's insulin levels.
22. A method for notifying, determining or controlling insulin dosage, the method include: generating a plurality of disorder curves, each disorder curve being based on a data representation of historical patient data associated with a deviation from a threshold blood glucose level; receiving current patient data; wherein the current patient data and the historical patient data include any data indicating patient characteristics; Apply a predictive model where: comparing the current patient data to at least one disturbance curve; and assessing, based on a probability analysis, a likelihood that a turbulent curve among the plurality of turbulent curves is an expected turbulent curve, the expected turbulent curve being a turbulent curve determined based on the probability analysis to match the current patient data; determining an insulin dose based on the expected disturbance profile; and A signal representative of the insulin dose is output to a device configured to monitor, influence and / or manage a patient's insulin levels.
