Switching and Customizing Glucose Prediction Models in Drug Delivery Devices

Abstract

Exemplary embodiments may provide for switching of a glucose prediction model in response to certain conditions. For example, the glucose prediction model may be switched in response to a crash state of the detected glucose value. Exemplary embodiments may dynamically customize parameters such as the coefficient values of the glucose prediction model for the user. Exemplary embodiments may customize the glucose prediction model based on the user's glucose level history and the history of insulin delivery to the user. Exemplary embodiments may determine a set of parameters that provides an improved fit of the parameters to the user's glucose level and insulin delivery history. The improved fit parameters may be used to adapt the parameter set to the most recent run.

Description

Technical Field

[0001] Related Applications This application claims the benefit of U.S. Provisional Patent Application No. 63 / 343,739, filed May 19, 2022, the entire content of which is incorporated herein by reference.

Background Art

[0002] Some conventional automated insulin delivery (AID) systems rely on a glucose prediction model that predicts a user's future glucose value. These glucose prediction models can predict a user's future glucose value based at least in part on the user's most recent glucose value and insulin delivery to the user that can still affect the user's glucose value. The predicted future glucose value of the user can be used when controlling basal insulin delivery to the user. Thus, it is desirable for the predicted future glucose value to be accurate in order for the AID system to deliver an appropriate insulin dose to the user in basal insulin delivery.

[0003] One complexity of some conventional glucose prediction models is that the glucose prediction model reacts slowly to a rapid decrease in the user's glucose value. As a result, the predicted future glucose value becomes very high. Thus, an AID system incorporating such a conventional glucose prediction model may continue to administer insulin to the user even though the user's glucose value is rapidly decreasing. Excessive insulin delivery exacerbates the rapid drop in the user's glucose value.

Summary of the Invention

Problems to be Solved by the Invention

[0004] Another problem with conventional glucose prediction models used in conventional AID systems is that the conventional glucose prediction models may not be very suitable for many users. Generally, conventional glucose prediction models are formulated by evaluating the average characteristics of a group. Unfortunately, as a result, for many users, the accuracy of the glucose prediction model is low. This may be largely due to the fact that there are significant differences in insulin sensitivity among users. Conventional glucose prediction models tend to be less reactive in order to avoid under-dosing or over-dosing insulin to users.

Means for Solving the Problems

[0005] According to one aspect of the present invention, a drug delivery device for delivering insulin to a user includes a non-transitory computer-readable storage medium storing computer program instructions and a processor configured to execute the computer program instructions. By executing the computer program instructions, the processor predicts a first future glucose value of the user using a first model that predicts future glucose values based on the user's glucose value history and drug delivery to the user, and detects a crash state of the user's glucose value. Also, by executing the computer program instructions, the processor switches to a second model to predict the user's next future glucose value in response to detecting the crash state of the glucose value, and the second model is more conservative than the first model such that when the user's glucose value is decreasing, the second model predicts the user's future glucose value lower than the first model.

[0006] The processor may detect the crash state of the glucose value by determining the difference between consecutive pairs of the user's most recent glucose value measurements extending from the measurement of the oldest glucose value in the sequence to the measurement of the newest glucose value in the sequence and comparing each of the differences to a threshold value. The processor may detect the crash state of the glucose value by comparing the differences to each other to determine whether the difference between consecutive pairs of the user's glucose value measurements becomes more negative as the difference spreads from the pair of the oldest glucose value measurement to the pair of the newest glucose value measurement. The processor may detect the crash state of the user's glucose value when each of the differences is negative and less than the threshold value and the difference between consecutive pairs of the user's glucose value measurements becomes more negative as the difference spreads from the pair of the oldest glucose value measurement to the pair of the newest glucose value measurement.

[0007] Detecting the crash state of the glucose value may involve comparing the user's insulin on board (IOB) to a threshold value. The processor may detect the crash state of the glucose value by looking for a rapidly decreasing measurement of the user's glucose value. In response to detecting the crash state of the glucose value, the processor may decrease the upper limit of the amount of insulin delivered to the user over a time period by a drug delivery device. The computer program instructions may, when executed, cause the processor to determine a basal insulin delivery amount for the user based on the predicted next future glucose value of the user predicted using a second model. The drug delivery device may have a cannula and / or a needle for delivering a drug to the user, and the computer program instructions may, when executed, cause the processor to initiate delivery of the basal insulin delivery dose to the user via the needle and / or the cannula.

[0008] According to another aspect of the invention, a drug delivery device for delivering insulin to a user comprises a non-transitory computer-readable storage medium storing computer program instructions and a processor configured to execute the computer program instructions. By executing the computer program instructions, the processor may access the user's insulin delivery dosage and glucose value history. Based on the history, the processor may determine an improved fit of the parameters for the glucose prediction model from the history to the current parameters of the glucose prediction model, and a glucose prediction model using the improved fit parameters predicts future glucose values more accurately than a glucose prediction model using the current parameters when compared to the prediction of the glucose values of the user's glucose value history. The glucose prediction model determines the user's future glucose values predicted from the past glucose values and past drug delivery dosages of the history. By executing the computer program instructions, it may involve adapting the parameters based on the improved fit of the parameters to generate the adapted parameters, and the processor may use the glucose prediction model with the adapted parameters to predict at least one future glucose value of the user.

[0009] When determining an improved fit of the parameters of the glucose prediction model, the glucose prediction model may determine an optimal fit of the parameters to the glucose prediction model. The improved fit may be the best fit of the parameters of the glucose prediction model. The processor may use a genetic algorithm to determine the improved fit of the parameters. The drug delivery device may have an operating cycle, may receive a measured value of at least one glucose value for each operating cycle, and may deliver a basal insulin dose to the user for each operating cycle. The processor may perform an inspection over a plurality of operating cycles of the drug delivery device when determining an improvement in the fitness of the parameters. A maximum deviation of the parameters from the population mean of the parameter candidates may be set, and at least one of the parameters of the improved fit of the parameters may be subject to the maximum deviation. The parameters of the improved fit of the parameters may need to conform to stability constraints. Glucose values of the glucose value history affected by external disturbances may be excluded from consideration when determining the improved fit of the parameters.

[0010] According to another aspect of the invention, a drug delivery device for delivering insulin to a user has a non-transitory computer-readable storage medium storing computer program instructions and a processor configured to execute the computer program instructions. By executing the computer program instructions, the processor may use a current model to predict a user's analyte level response to the delivery of a drug dose. Also, by executing the computer program instructions, the processor may exchange the current model or parameters for another model or parameters or update the current model based on new analyte level data in response to detecting a state. Further, by executing the computer program instructions, the processor may use another model or the updated current model to change the delivery of a drug dose to the user.

[0011] The detected state may be rapidly changing analyte level data. The change may be to stop drug delivery or to change the parameters of the model for drug dosage calculation.

Brief Description of the Drawings

[0012]

Figure 1

[0013]

Figure 2

[0014]

Figure 3

[0015]

Figure 4

[0016]

Figure 5

[0017]

Figure 6

[0018]

Figure 7

[0019]

Figure 8

[0020]

Figure 9

[0021]

Figure 10

[0022]

Figure 11

[0023]

Figure 12

[0024]

Figure 13

[0025]

Figure 14

[0026]

Figure 15

[0027]

Figure 16

[0028] Exemplary embodiments may provide for switching of a glucose prediction model in response to certain conditions. For example, the glucose prediction model may be switched in response to a crash state of a detected glucose value. Conventional glucose prediction models tend to maintain the same fixed model parameters even when there is a period during which the decrease in the glucose value is accelerating, and thus deliver insulin also during the period after the user's glucose value has crashed. Continued delivery of insulin under such conditions may only exacerbate the problems caused by the crash state of the glucose value. Exemplary embodiments may avoid such problems during continued insulin delivery by switching to a more aggressive model when ending or reducing insulin delivery in response to detection of a crash state of the user's glucose value. The glucose prediction model of the exemplary embodiments may be configured to more accurately predict the user's glucose value when the blood glucose value is decreasing more rapidly than a conventional glucose prediction model. Thus, under such conditions, insulin delivery may be decreased or stopped more rapidly.

[0029] In addition, in an exemplary embodiment, insulin delivery constraints may be changed when detecting a crash state of glucose values. For example, in an exemplary embodiment, when detecting a crash state of glucose values, the one-time upper limit constraint that restricts the basal insulin delivery rate by the drug delivery device may be reduced. Further, in an exemplary embodiment, when detecting a crash state of glucose values, the overall constraint that restricts how much basal insulin can be delivered over a predetermined period such as several hours may be reduced.

[0030] In an exemplary embodiment, parameters such as coefficient values of a glucose prediction model may be customized for a user. In an exemplary embodiment, the glucose prediction model may be customized based on the user's glucose value history and the history of insulin delivery to the user. In an exemplary embodiment, a set of parameters that provides an improved fit of the parameters to the user's glucose value history for the most recent glucose value execution may be determined. The improved fit may be relatively optimal for the most recent glucose value in some exemplary embodiments. Strategies such as genetic algorithms and other parameter fitting strategies may be used in an exemplary embodiment. The improved fit parameters are used when adapting the current parameters to the most recent glucose value.

[0031] Although the discussion focuses on the case where the drug delivery device is an insulin delivery device and the analyte to be measured is the user's glucose value, it should be understood that the systems, methods, and storage media described herein may also be used for the delivery of drugs other than insulin or insulin alone and may depend on analyte levels other than glucose such as the user's ketone level. Exemplary drugs that may be used include insulin, glucagon-like peptide-1 receptor agonists (GLP-1), glucose-dependent insulinotropic polypeptide (GIP), or other hormones and / or combinations of pharmaceuticals such as two or more of insulin, GLP-1, and GIP, or other homologous hormones or weight loss drugs.

[0032] Figure 1 depicts an exemplary drug delivery system 100 suitable for delivering drugs to user 108 according to an exemplary embodiment. The drug delivery system 100 may include a drug delivery device 102. The drug delivery device 102 may be wearable and attached to the body of user 108 or carried by user 108. The drug delivery device 102 may be directly coupled to the user (e.g., directly attached to a body part and / or skin of the user via an adhesive or the like) or may be carried by the user with the drug delivery device 102 connected to an injection site where the drug is injected using a needle and / or cannula (e.g., may be placed in a belt or pocket). The surface of the drug delivery device 102 may include an adhesive to facilitate attachment to user 108.

[0033] The drug delivery device 102 may include a processor 110. The processor 110 may be, for example, a microprocessor, logic circuit, field programmable gate array (FPGA), application specific integrated circuit (ASIC), or microcontroller. The processor 110 may hold dates and times as well as perform other functions (e.g., calculations, etc.). The processor 110 may be operable to execute a control application 116 encoded by computer program instructions stored in a storage device 114, and this control application 116 enables the processor 110 to direct the operation of the drug delivery device 102. The control application 116 may be a single program, multiple programs, modules, libraries, etc. The processor 110 may execute computer program instructions stored in the storage device 114 for a user interface 117 that may include one or more display screens presented on a display 127. The display 127 may display information to user 108 and, in some cases, may receive input from user 108, such as when the display 109 is a touch screen.

[0034] The control application 116 may control the delivery of drugs to the user 108 according to a control approach as described herein. The control application may use a glucose prediction model as described below to predict the future glucose values of the user 108. The storage device 114 may hold the user's history 111 such as the history of basal delivery, the history of bolus delivery, and / or other histories such as the meal event history, the exercise event history, the glucose value history, the other analyte level history, etc. Further, the processor 110 may be operable to receive data or information. The storage device 114 may include both a primary storage device and a secondary storage device. The storage device 114 may include a random access memory (RAM), a read-only memory (ROM), an optical storage device, a magnetic storage device, a removable storage medium, a solid-state storage device, etc.

[0035] The drug delivery device 102 may have a tray or cradle and / or one or more housings that house various components including a pump 113, a power source (not shown), and a tank 112 that stores the drug to be delivered to the user 108. A fluid path to the user 108 may be provided, and the drug delivery device 102 may discharge the drug from the tank 112 to deliver the drug to the user 108 via the fluid path using the pump 113. The fluid path may include, for example, a tube that connects the drug delivery device 102 to the user 108 (e.g., a tube that connects a cannula to the tank 112), and may include a conduit to a separate injection site. The drug delivery device 102 may have an operating cycle, for example, every 5 minutes, that calculates and delivers the basal dose of the drug as needed. These procedures are repeated for each cycle.

[0036] For example, there may be one or more communication links with one or more devices physically separated from the drug delivery device 102, including the management device 104 of the user 108 and / or the caregiver of the user 108, the (one or more) sensors 106, the smartwatch 130, the fitness monitor 132, and / or other various wearable devices 134. The communication link may include any wired communication link or wireless communication link that operates according to any known communication protocol or standard, such as Bluetooth®, Wi-Fi®, short-range wireless communication standards, cellular standards, or any other wireless protocol.

[0037] The drug delivery device 102 can communicate with the network 122 via a wired communication link or a wireless communication link. The network 122 may include a local area network (LAN), a wide area network (WAN), a cellular network, a Wi-Fi® network, a short-range wireless communication network, or a combination thereof. A computer device 126 may communicate with the network, and the computer device may communicate with the drug delivery device 102 or the management device 104.

[0038] The drug delivery system 100 may have one or more sensors 106 that detect the level of one or more analytes. The (one or more) sensors 106 may be coupled to the user 108, for example, by an adhesive or the like, and provide information or data regarding one or more medical conditions, physical attributes, or analyte levels of the user 108. The (one or more) sensors 106 may be physically separate from or an integrated component of the drug delivery device 102. The (one or more) sensors 106 may include, for example, glucose monitors such as continuous glucose monitors (CGMs) and / or non-invasive glucose monitors. The (one or more) sensors 106 may include ketone sensors, other analyte sensors, heart rate monitors, respiratory rate monitors, motion sensors, temperature sensors, sweat sensors, blood pressure sensors, alcohol sensors, and the like. Some sensors 106 may detect the characteristics of the components of the drug delivery device 102. For example, the sensors 106 of the drug delivery device may include voltage sensors, current sensors, temperature sensors, and the like.

[0039] The drug delivery system 100 may or may not have a management device 104. In some embodiments, since the drug delivery device 102 can manage itself, a management device is not required. The management device 104 may be a dedicated device such as a dedicated personal diabetes manager (PDM) device. The management device 104 may be a programmed general-purpose device such as any portable electronic device including a dedicated controller such as a processor, a microcontroller, etc. The management device 104 may be used to program or adjust the operation of the drug delivery device 102 and / or the sensor(s) 106. The management device 104 may be any portable electronic device including, for example, a dedicated device, a smartphone, a smartwatch or a tablet. In the example depicted, the management device 104 may have a processor 119 and a storage device 118. The processor 119 may execute a process of managing the user's glucose value and controlling the delivery of drugs to the user 108. The drug delivery device 102 may supply data from the sensor 106 and other data to the management device 104. The data may be stored in the storage device 118. The processor 119 may be operable to execute program code stored in the storage device 118. For example, the storage device 118 may be operable to store one or more control applications 120 for execution by the processor 119. The storage device 118 may be operable to store historical information such as drug delivery information, analyte level information, user input information, output information or other historical information. The control application 120 may play a role in controlling the drug delivery device 102 such as the control of automatic drug delivery (ADD) (or, for example, automatic insulin delivery (AID)) of drugs to the user 108. The storage device 118 may store the control application 120, a history 121 as described above for the drug delivery device 102 and other data and / or programs.

[0040] To display information, a display 140 such as a touch screen may be provided. The display 140 may display a user interface (UI) 123. The display 140 may be used to receive input as it is when it is a touch screen. The management device 104 may further have input elements 125 such as a keyboard, buttons, knobs, etc. to receive input from the user 108.

[0041] The management device 104 may communicate with a network 124 such as a LAN or a WAN or a combination of those networks via a wired communication link or a wireless communication link. The management device 104 may communicate with one or more servers or cloud services 128 via the network 124. In some embodiments, data such as sensor values may be directly transmitted from the drug delivery device 102 to one or more cloud services / servers 128 for storage and processing or transmitted from the management device 104 to one or more cloud services / servers 128.

[0042] Other devices such as smartwatch 130, fitness monitor 132, and wearable device 134 may be part of the drug delivery system 100. These devices 130, 132, and 134 may communicate with the drug delivery device 102 and / or the management device 104 for receiving information and / or issuing commands to the drug delivery device 102. These devices 130, 132, and 134 may execute computer program instructions to perform part of the control functions executed by the processor 110 or the processor 119 via, for example, the control applications 116 and 120. These devices 130, 132, and 134 may have a display for displaying information. The display can display a user interface for providing inputs by the user 108 such as a request to change or pause the dosage or a request to deliver a drug bolus, start or confirmation, or can display a user interface for displaying an output such as a change in the dosage (e.g., basal delivery amount) determined by the processor 110 or the management device 104. These devices 130, 132, and 134 may have a wireless communication connection with the sensor 106 for directly receiving analyte measurement data. Other delivery devices 105 such as a drug delivery pen (e.g., insulin pen) may be configured (e.g., when determining the IOB) or provided for delivering drugs to the user 108.

[0043] The functions described later for the exemplary embodiments may be under the control of or executed by the control application 116 of the drug delivery device 102 or the control application 120 of the management device 104. In some embodiments, the functions may be under the control of or executed by a cloud service or server 128, a computer device 126, or other enumerated devices, and the other enumerated devices include the smartwatch 130, the fitness monitor 132, or another wearable device 134.

[0044] In closed-loop mode, control applications 116, 120 continuously determine the drug delivery amount for user 108 based on a feedback loop. For example, in the case of an insulin delivery device, the purpose of the closed-loop mode is to bring the user's glucose value to a target glucose value, or to keep the user's glucose value within a range of glucose values.

[0045] In some embodiments, drug delivery device 102 need not deliver a single drug alone. Instead, drug delivery device 102 may deliver one drug, such as insulin, to lower the glucose value of user 108 and also deliver another drug, such as glucagon, to raise the glucose value of user 108. As described above, drug delivery device 102 may deliver other drugs, such as glucagon-like peptide (GLP)-1 receptor agonist drugs, to lower glucose or delay gastric emptying to delay the post-meal glucose spike, instead of insulin. In other embodiments, drug delivery device 102 may deliver other drugs that replace pramlintide or insulin. In other embodiments, drug delivery device 102 may deliver concentrated insulin. In some embodiments, the drug or drugs delivered by the drug delivery device may be a combination of two or more of the drugs specified above. In a preferred embodiment, the drug delivery device delivers insulin. Thus, throughout this application, reference is made to insulin and insulin delivery devices, but those skilled in the art will understand that drugs other than insulin can be delivered instead of or in addition to insulin.

[0046] Exemplary embodiments may use an analyte prediction model to predict a user's future analyte levels. For example, the predicted analyte level may be the user's future glucose value, and in some exemplary embodiments, the drug may be insulin. The glucose prediction model may predict the user's future glucose value as follows. [Number] Here, [Number] is the predicted glucose value of cycle k expressed as a relative delta with respect to the set value, [Number] are the predicted glucose values of cycles k-1, k-2, k-3 expressed as relative deltas with respect to the set value, b 1 , b 2 , b 3 are respectively pre-determined prediction coefficients, K is a weight coefficient, I Δ (k-3) is the insulin dosage at cycle k-3, expressed as the difference with respect to the standard basal dosage.

[0047] Figure 2 shows a flowchart 200 of exemplary steps that may be performed in an exemplary embodiment when using the glucose prediction model. At 202, the glucose prediction model may be used by control application 116 or 120 to determine one or more predicted future glucose values of the user from the glucose value and insulin delivery dosage information of history 111 and / or 121. The glucose prediction model may predict the glucose value for only a single future operating cycle of drug delivery device 102 or for multiple future operating cycles. At 204, the next basal insulin dosage may be determined considering the predicted (one or more) glucose values and the target glucose value (e.g., the set value). A cost function having a glucose cost component and an insulin cost component may be used when determining the basal insulin delivery dosage. The insulin dosage candidate with the lowest cost may be selected as the next basal insulin delivery dosage by control application 116 or 120. Suitable cost functions are as follows. [Number] Here, J(t) is the cost of the insulin dose for cycle t, target(i) is the user's target glucose value for cycle i, CGM(i) is the user's predicted glucose value for cycle i, M is the last cycle of the future period for which the glucose cost is determined, Q is the weight coefficient of the glucose cost component, and I p (i) is the predicted insulin dose in the i-th cycle, and I b (i) is the predicted ideal basal dose in the i-th cycle (this may be based on dividing the value obtained by dividing the user's total daily insulin (TDI) value by 24 by 2). N is the cycle number of the last cycle for which the insulin cost is determined, and R is the weight coefficient of the insulin cost. [Number] is the glucose cost component of the cost function, [Number] is the insulin cost component of the cost function. As can be seen from the above formula, the predicted future glucose value enters the glucose cost component of the cost function. Finally, at 206, delivery of the determined basal insulin dose may be initiated for user 108 by drug delivery device 102.

[0048] As described above, the exemplary embodiment may switch the glucose prediction model in response to several conditions. One such condition is the crash state of the glucose value. The exemplary embodiment may change from one glucose prediction model to another in response to the detection of one or more states. For example, the same glucose prediction function (see Equation 1) may be used by the model, but the coefficients (b 1 , b 2 and b 3) may be changed. Changing the coefficients is referred to as changing the model in this specification. FIG. 3 shows a flowchart 300 of exemplary steps that may be performed in an exemplary embodiment to effect a switch in the glucose prediction model. First, at 302, a first glucose prediction model may be used to predict future glucose values. This first model may be well-suited for most operating conditions. However, the first model may not be suitable for certain conditions. One such condition is a crash state of the glucose value where the user's glucose value drops acceleratively over a period of time. At 304, a check is made as to whether there is a crash state of the glucose value. If not, there is no need to change the model and the process ends. If there is a crash state of the glucose value, at 306, the glucose prediction model is changed to a second model suitable for dealing with the crash state of the glucose value. The coefficients used in the second model may indicate or bring about a greater drop in the predicted glucose value more quickly than the first model during the crash state of the glucose value, and thus may be able to react to the crash state of the glucose value more quickly than the first model. The first model may be more aggressive in that it is more likely to deliver insulin more easily and in larger amounts than the second model. Thus, the second model may be characterized as being less aggressive or more conservative.

[0049] In an exemplary embodiment, the second glucose prediction model may continue to be used until the crash state of the glucose value is alleviated. FIG. 4 shows a flowchart 400 of exemplary steps that may be performed in an exemplary embodiment regarding the switchback from the second model to the first model. At 402, a check may be made as to whether the crash state of the glucose value still exists. This may involve checking whether the condition(s) examined (as described below) to identify whether the crash state of the glucose value is still being met are still being met. If not, the switchback does not occur. If the examined state indicates the crash state of the glucose value, at 404, a switchback to the first glucose prediction model may be performed to use the first glucose prediction model.

[0050] The detection of the crash state of the glucose value may be determined in various ways. FIG. 5 depicts a flowchart 500 of exemplary steps that may be performed in one of these ways to detect the crash state of the glucose value. In this example, to detect the crash state of the glucose value, the difference between successive predicted glucose values needs to indicate that the user's glucose value is decreasing acceleratingly over a period of time. Thus, at 502, the difference between the measured values of successive predicted glucose values of the user may be determined. The sequence may include the measured value of the user's most recent predicted glucose value. In some examples, the three most recent predicted glucose values from history 111 and / or 121 may be used. In that case, the difference may be calculated as follows. [Number] Here, G(k) is the predicted glucose value at cycle k, G(k - 1) is the user's predicted glucose value at cycle k - 1, G(k - 2) is the user's predicted glucose value at cycle k - 2, G(k - 3) is the user's predicted glucose value at cycle k - 3, d 1 , d2 , d 3 is the difference between consecutive glucose values.

[0051] At 504, it may be checked whether all of the differences are zero or less. If the difference is not negative, there is no possibility that the glucose value is in a crash state. Therefore, if not all of the differences for a sequence of consecutive predicted glucose values are zero or less, at 506, a crash state of the glucose value is not detected. If all of the differences are zero or less, a crash state of the glucose value is detected. Additionally or alternatively, at 508, it may be further checked whether the rate of decrease of the predicted glucose value is accelerating by checking whether all of the differences other than the last difference are more negative than the previous difference. This is expressed as checking whether d n is greater than or equal to d n+1 for all but the last difference in the sequence. If not, at 506, a crash state of the glucose value is not detected. If so, at 512, it may be considered that a crash state of the glucose value has been detected. In some embodiments, the process may include any checks related to IOB at 510 (shown in phantom form to indicate that it is optional). Specifically, it may be checked whether the user's IOB is greater than 3% of the total daily insulin amount (i.e., the average total of the basal insulin delivery and bolus insulin delivery per day). The basis for such a check at 510 is that when the user has little IOB (e.g., less than 3% of the TDI), the user has much less risk of hypoglycemia when more insulin is delivered to the user. Therefore, it is not very important to rapidly reduce insulin delivery in response to a crash state of the glucose value, and there is less need to switch the glucose prediction model. If the user has an IOB greater than 3% TDI, at 512, it may be considered that a crash state of the glucose value has been detected. If not, at 506, a crash state of the glucose value is not detected. Values other than 3% may be used.

[0052] It should be understood that the crash state of the glucose value may be determined by other methods.

[0053] The response of the system to detecting the crash state of the glucose value is not limited to simply changing the coefficients of the glucose value prediction model. Additional measures may be taken. For example, in response to detecting the crash state of the glucose value, specific constraints may be changed. FIG. 6 shows a flowchart 600 of exemplary steps that may be performed in an exemplary embodiment regarding changing the insulin delivery constraint. At 602, the crash state of the glucose value may be detected. In response thereto, at 604, the upper limit of the amount of insulin delivered per injection may be reduced. For example, the maximum basal insulin delivery dose may be reduced to be set to a value such as twice the standard basal delivery dose. Additionally or alternatively, at 606, the overall constraint may be reduced. The overall constraint may limit the amount of insulin that may be delivered by the drug delivery device 102 over a predetermined period, for example, a period of several hours. For example, the constraint may be reduced to nine times the normal basal hourly delivery amount.

[0054] As described above, the parameters of the glucose prediction model may be customized for the user and updated taking into account the most recent glucose values. FIG. 7 shows a flowchart 700 of steps that may be performed in an exemplary embodiment when customizing the parameters of the glucose prediction model for user 108. First, at 702, the current parameters are used by the glucose prediction model. At 704, a trigger to update the parameters arrives, for example, a new operating cycle, a new delivery cycle, a new time, a new day, a new week, the use or application of a new drug delivery device in place of an old drug delivery device or the arrival of another period. The trigger is customizable. "Execution" is used herein because the glucose values received between the last update and the last trigger point refer to a predetermined period or cycle. More generally, an event or other trigger that prompts an update of the parameters may arrive. At 706, parameters may be determined that provide improved fit to the glucose values being executed. In some embodiments, parameters that provide the best fit to the user (or the received glucose values) may be determined. At 708, the parameters with improved fit are used to adapt the current parameters of the glucose prediction model taking into account the glucose values or levels of execution, as described below. At 710, the adapted parameters are applied when predicting future glucose values using the glucose prediction model.

[0055] FIG. 8 shows a flowchart 800 illustrating exemplary steps that may be performed in an exemplary approach for generating an improved set of fitting parameters according to an exemplary embodiment. This approach uses a genetic algorithm that generates generations of parameter sets and mutates the generations (such as by perturbing the values from the previous generation) until an improved set of fitting parameters (such as an optimal set of parameters) is found. At 802, a new generation of parameter sets is generated. An initial parameter set is randomly selected, and subsequent generations are generated by mutating the best parameter set of the current generation. For each generation, at 804, a residual is calculated for each of the parameter sets of the generation. The residual may be calculated using a fitness function. In this example, the fitness function may calculate the difference between the predicted parameters in each of the parameter sets with respect to the actual past glucose values. These differences may be aggregated for each of the parameter sets to determine an aggregate value of the residuals of the parameter sets.

[0056] Figure 9 shows an exemplary table 900 of values for generating parameter sets. Table 900 includes a column 902 of past glucose values and a column 904 of past insulin delivery amounts for cycles used in the glucose prediction model. Columns 906, 908, 910, 912, 914, 916, and 918 are associated with predicted glucose values respectively generated by parameter sets labeled P1, P2,..., P10. Rows 920, 922, 924, 926, and 928 hold information regarding predictions generated for the user's actual past glucose values and insulin delivery dosage for cycle k - 3 as used in the glucose prediction model (see Equation 1). Thus, for row 920, columns 906, 908, 910, 912, 914, 916, and 918 hold the predicted glucose values generated when each of the parameter sets is used with the predicted glucose model for a cycle where the actual glucose value was 100 mg / dL and the insulin delivery amount was 0.5 units. Row 930 holds the aggregated residuals for each of the parameter sets. As can be seen, parameter set P5 has the smallest residual. Thus, the parameter set of P5 is the most fitting for this generation.

[0057] Referring to FIG. 8, at 806, a parameter set with the smallest residual is selected. At 808, the residual of the selected parameter set is compared with a threshold to determine whether the selected parameter set can be accepted as an improved fitting parameter set. If not, at 802, a new generation is generated by changing the selected parameter set. In that case, at 810, the selected parameter set is used as the improved fitting parameter set.

[0058] Adaptation does not necessarily require fitting the parameters of a single cycle. Instead, it may attempt to find the parameters over multiple cycles. FIG. 10 shows a flowchart 1000 of exemplary steps that may be performed in an exemplary embodiment to make predictions over multiple cycles. In such an example, at 1002, a parameter set is applied to a glucose prediction model to predict a user's future glucose values over multiple cycles. For example, glucose values may be predicted for the next 12 cycles. At 1004, the predicted glucose values are compared with the corresponding actual glucose values for each cycle to calculate the residual for each cycle. At 1006, the residuals are aggregated (i.e., summed) over the cycles. These aggregated residuals are used when selecting an improved set of fitting parameters as described above.

[0059] Parameters that may be created during the change phase for creating the next generation may be constrained. FIG. 11 shows a flowchart 1100 of exemplary steps that may be performed in an exemplary embodiment to constrain such parameters. At 1102, a maximum deviation of a parameter from a population mean (e.g., the average value of the parameters during execution) may be established. For example, the parameter does not deviate more than a threshold from the average value, e.g., does not deviate more than 25% from the average value. At 1104, when a parameter set is generated through a process of finding improved parameter fitting, the parameter may not deviate more than a threshold from the average value of the parameters.

[0060] Stability constraints may be applied to the parameters. FIG. 12 shows a flowchart of exemplary steps that may be performed in an exemplary embodiment regarding stability constraints. At 1202, stability constraints are established. As a result, at 1204, the parameters of the glucose prediction model are restricted such that the poles of the model are located within the unit circle. In other words, Equation 1-b 1,new z -1 -b 2,new z -2 -b 3,new z-3 The solution for =0 must not exceed from -1 to 1.

[0061] To avoid data caused by external disturbances, glucose values caused by external disturbances such as diet intake or exercise may be removed from the running data to be processed to find an improved set of adaptation parameters. FIG. 13 shows a flowchart 1300 of exemplary steps that may be performed in an exemplary embodiment to remove the effects of such external disturbances. At 1302, it may be determined that there is an external disturbance. For example, the user may have indicated that they have eaten a meal or are exercising. Alternatively, the drug delivery system of FIG. 1 may include logic in software to process glucose values and make a determination that there has been an interference. For example, the glucose value may increase rapidly as to indicate a meal intake or decrease as to indicate exercise. At 1304, the glucose values affected by the external disturbance may be removed from the execution being processed when updating the parameter set to an improved set of adaptation parameters. For example, the post-meal glucose values may be removed by omitting the glucose values for a predetermined period such as a 5-hour period after the intake of a meal of a size exceeding a threshold such as a meal having at least 40 grams of carbohydrates. Similarly, the glucose values during the exercise period may be omitted.

[0062] Another approach to limiting the impact of external disturbances is depicted in flowchart 1400 of FIG. 14. FIG. 14 shows a flowchart 1400 of exemplary steps that may be performed in an exemplary embodiment to remove glucose values affected by external disturbances. At 1402, a threshold value may be established. The threshold value may represent how much the residual can vary from the threshold value. The threshold value may be a percentage of the actual glucose value. For example, the threshold value may establish that a residual indicating that the actual glucose value changes by more than 20% from the predicted glucose value is discarded or not considered. Other percentages or absolute values (e.g., above 20 mg / dL) may be used. Alternatively, the threshold value may be based on the average residual value and may represent the maximum allowable deviation of the residual value from the average residual value (e.g., 1 standard deviation). At 1404, the magnitude of the residual between the predicted glucose value and the predicted glucose value may be compared to the threshold value. If the residual exceeds the threshold value, at 1408, the actual glucose value may be removed from the glucose value history. Otherwise, at 1406, the glucose value may be retained in the glucose value history.

[0063] Once improved fitting parameters are determined based on the execution of newly processed glucose values, the glucose value prediction model may be refitted to further reflect the improved parameters. FIG. 15 shows a flowchart 1500 of exemplary steps that may be performed to refit the parameters. In a specific example, each coefficient b 1 , b 2 and b 3 or other parameters may be refitted to reflect the change in the improved fitting parameter values relative to the previous parameter values, and this may be done for each "execution" or each cycle as described above with reference to FIG. 7. Continuing with the example of the b coefficients, at 1502, the value of b n-1.new may be determined. b n-1.new is the value of the coefficient at the n-1 cycle of the improved fitting parameter set as described above. At 1504, b n-1.newFor example, weights may be applied to the number of days of execution, N days.new In 1506, a weight may be applied to the coefficient weights used to predict the glucose value for cycle n-1 using the current parameter set. A suitable weight may be 0.2 times the number of days in the run. Thus, using these example numbers, if there is only one day of new data, the new b coefficient (b n,final ) when determining the previous b coefficient (b n-1,final ) is weighted 80%. Other weightings may be used. At 1508, these weighted values ​​are summed. At 1510, the new coefficient value b n.final Set equal to the sum, which can be expressed as:

number

[0064] Adaptation of parameter values ​​may be used when using multiple models as described above. It is assumed that there are at least two glucose prediction models and which glucose prediction model is used depends on the current conditions. FIG. 16 shows a flow chart 1600 of exemplary steps that may be performed in an exemplary embodiment to continuously adapt parameters using a Kalman filter. The Kalman filter may use inputs and statistical noise to generate estimates by estimating a joint probability distribution over the parameters for each cycle. At 1602, the user's most recent glucose value and most recently delivered basal insulin dose may be received. In response to these inputs, at 1604, parameters of the glucose prediction model may be updated, for example, using a Kalman filter. At 1606, the updated parameters may be applied to the glucose prediction model to estimate one or more glucose values ​​of the user.

[0065] Although exemplary embodiments have been described in this specification, various changes in form and detail can be made without departing from the intended scope of the appended claims.

Claims

1. A drug delivery system for delivering a drug to a user, comprising: a non-transitory computer-readable storage medium storing computer program instructions; a processor, predicting a first future glucose value of the user using a first model that predicts a future glucose value based on a history of the user's glucose values and drug delivery to the user; detecting a crash state of the user's glucose value; in response to detecting the crash state of the glucose value, switching to a second model to predict the user's next future glucose value, wherein the second model has different parameter values than the first model such that when the user's glucose value is decreasing, the second model predicts a lower future glucose value of the user than the first model; a processor configured to execute computer program instructions that cause the processor to perform the above; A drug delivery system comprising the above.

2. The drug delivery system according to claim 1, wherein the processor determines a difference between consecutive pairs of measurements of the user's most recent glucose value extending from a measurement of the oldest glucose value in the sequence to a measurement of the newest glucose value in the sequence and compares each of the differences to a threshold value to detect the crash state of the glucose value.

3. The drug delivery system according to claim 2, wherein the processor compares the differences to each other to determine whether the difference between consecutive pairs of measurements of the user's glucose value becomes more negative as the difference spreads from a pair of measurements of the oldest glucose value to a pair of measurements of the newest glucose value to detect the crash state of the glucose value.

4. The drug delivery system according to claim 3, wherein the processor detects the crash state of the user's glucose value when each of the differences is negative and less than the threshold value and the difference between consecutive pairs of measurements of the user's glucose value becomes more negative as the difference spreads from a pair of measurements of the oldest glucose value to a pair of measurements of the newest glucose value.

5. The processor detecting a crash state of the glucose value also involves comparing the user's insulin on board (IOB) with a threshold, the drug delivery system according to claim 4.

6. The processor detects the crash state of the glucose value by looking for a measured value of the user's glucose value that is decreasing rapidly, the drug delivery system according to claim 1.

7. In response to detecting the crash state of the glucose value, the processor reduces the upper limit of the amount of insulin delivered to the user over a time period by a drug delivery device, the drug delivery system according to claim 1.

8. The computer program instructions, when executed, cause the processor to determine a basal insulin delivery dosage for the user based on the predicted next future glucose value of the user predicted using the second model, the drug delivery system according to claim 1.

9. Further comprising a cannula and / or a needle for delivering a drug to the user, and the computer program instructions, when executed, cause the processor to initiate delivery of a basal insulin delivery dosage to the user via the needle and / or the cannula, the drug delivery system according to claim 8.

10. A drug delivery system for delivering a drug to a user, A non-transitory computer-readable storage medium storing computer program instructions, A processor, Accessing the history of the user's drug delivery dosage and glucose value, Based on the history, determining an improved fit of the parameters for the glucose prediction model from the history to the current parameters of the glucose prediction model, wherein the glucose prediction model using the improved fit parameters predicts future glucose values more accurately than the glucose prediction model using the current parameters to predict future glucose values when compared to the prediction of the glucose values in the history of the user's glucose value, and the glucose prediction model determines the user's future glucose values predicted from the past glucose values and past drug delivery dosages in the history, Adapting the parameters based on the improved fit of the parameters to generate the adapted parameters. To predict at least one future glucose value of a user, using the glucose prediction model having the adapted parameters, A processor configured to execute computer program instructions that cause the processor to execute, A drug delivery system comprising.

11. The drug delivery system according to claim 10, wherein when determining an improved adaptation of the parameters of the glucose prediction model, the glucose prediction model determines an optimal adaptation of the parameters of the glucose prediction model.

12. The drug delivery system according to claim 11, wherein the improved adaptation is an improved adaptation of the parameters of the glucose prediction model.

13. The drug delivery system according to claim 10, wherein the processor uses a genetic algorithm to determine an improved adaptation of the parameters.

14. The drug delivery device has an operating cycle, receives a measured glucose value for each operating cycle, delivers a basal insulin dose to the user for each operating cycle, and the processor that determines an improved adaptation of the parameters performs inspections over a plurality of operating cycles of the drug delivery device. The drug delivery system according to claim 10.

15. A maximum deviation of the parameters with respect to the population mean of the parameter candidates is set, and at least one of the parameters of the improved adaptation of the parameters is subject to the maximum deviation. The drug delivery system according to claim 10.

16. The drug delivery system according to claim 10, wherein the parameters of the improved adaptation of the parameters need to conform to stability constraints.

17. The glucose values of the history of glucose values affected by disturbances are excluded from consideration when determining the improved adaptation of the parameters. The drug delivery system according to claim 10.

18. A drug delivery system for delivering a drug to a user, A non-transitory computer-readable storage medium storing computer program instructions, A processor, Using the current model to predict the user's analyte level response to the delivery of a drug dose, In response to detecting a state, replacing the current model with another model or updating the current model based on new analyte level data, To change the delivery of the drug dosage to the user, use the other model or the updated current model, A processor configured to execute computer program instructions that cause the processor to execute, A drug delivery system comprising.

19. The drug delivery system according to claim 18, wherein the detected state is rapidly changing analyte level data.

20. The drug delivery system according to claim 18, wherein the change is to stop the delivery of the drug.

Images