Method for controlling blood glucose and artificial pancreas system for carrying out the method
Patent Information
- Application Number
- JP2026512699
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-08-25
- Filing Date
- 2024-08-23
- Publication Date
- 2026-08-27
Smart Images
Figure 2026529166000146 
Figure 2026529166000147 
Figure 2026529166000148
Abstract
Description
[Technical Field]
[0001] The present invention relates to the technical field of glucose control. Specifically, the present invention relates to a regulatory control action by administering insulin, and an antagonistic control action by administering glucagon and / or rescue carbohydrates to more effectively control glucose levels (concentrations).
[0002] The object of the present invention is to provide a method for controlling glucose, which allows for the management of arbitrary meal and exercise notifications by a flexibly configured dual hormone control system.
[0003] A second object of the present invention is to provide a flexibly configured dual-hormone artificial pancreas for the coordinated administration of insulin, glucagon, and / or relief carbohydrates. [Background technology]
[0004] Type 1 diabetes (T1DM) is an autoimmune disorder that destroys pancreatic beta cells, resulting in an inability to secrete insulin. This hormone plays a crucial role in glucose homeostasis because it is involved in the decrease of plasma glucose concentration.
[0005] As a result, T1DM patients tend to have high levels of blood glucose (sugar) over the long term (hyperglycemia), leading to serious long-term health problems such as cardiovascular disease, renal impairment, retinal damage, and neuropathy. This is why exogenous insulin administration is necessary.
[0006] The artificial pancreas (AP) system emerged as a technical treatment for T1DM, improving blood glucose control compared to conventional methods. However, daytime control remains a challenging issue because dietary intake and exercise cause substantial fluctuations in glucose levels. Current hybrid AP systems are based on a dietary bolus using carbohydrates and are configured to compensate for (offset) the rise in glucose caused by meals. In open-loop treatment methods, patients must proactively address the responsibility of estimating their carbohydrate intake and informing the system of this (meal notification).
[0007] Therefore, some patients may not recognize the usefulness of AP and may discontinue its use. Thus, the AP system must be able to perform at an acceptable level without imposing the responsibility of meal notifications. However, a complete system without meal notifications (fully automated) may not be suitable for all patients. Some patients with extensive experience in carbohydrate estimation may, at least in some cases, prefer to assume the responsibility of meal notifications to enhance postprandial control.
[0008] However, meal notifications as a feedforward action may interfere with feedback actions in APs designed to function fully automatically without such notifications, potentially leading to insulin overdose and resulting hypoglycemia (abnormally low glucose levels). Therefore, AP systems must incorporate a mechanism that allows users to notify about meals without increasing the risk of hypoglycemia.
[0009] Exercise is another major challenge for the performance of AP systems. Very strenuous exercise events can lead to hyperglycemia (abnormally high glucose levels), but light to moderate aerobic exercise, the most common physical activity practiced by non-athletes, lowers glucose levels. Current hybrid systems require users to plan their exercise so that insulin infusions are reduced in the hours leading up to the exercise.
[0010] To eliminate exercise-induced hypoglycemia, the AP system must have an antagonistic control action that raises glucose levels. Carbohydrate supplementation is the most practical option for treating exercise-related mild hypoglycemia. However, individuals concerned about weight gain may prefer subcutaneous administration of glucagon as a non-caloric treatment for hypoglycemia. A dual-hormone AP system with automated insulin and glucagon administration effectively reduced the duration of exercise-induced hypoglycemia in clinical trials.
[0011] For individuals with low levels of physical activity and a low tendency towards hypoglycemia, the additional cost of a dual-hormone AP system for timely glucagon administration may not be justifiable. Self-administered low-dose glucagon pens may be a more suitable option for such patients to safely and effectively alleviate mild hypoglycemia.
[0012] Based on the above analysis, the following three characteristic features were identified: 1) To effectively supplement meals without providing any mealtime notifications. 2) If the patient wishes, enable the patient to provide meal notifications without impairing the performance of the feedback controller, and 3) In the case of exercise events, etc., for which no notification is given, provide antagonistic control actions to manage and alleviate hypoglycemia. This is recommended for AP systems.
[0013] Many methods have been proposed to eliminate meal notifications, but none of them have a mechanism to reduce the interaction with feedback actions that patients might provide if they do.
[0014] Regarding feedforward effects resulting from unplanned exercise, most glucagon-insulin systems lack a carbohydrate suggesting module, and vice versa.
[0015] Several strategies that may work using glucagon and carbohydrates to alleviate hypoglycemia require dietary notification. However, as described above, a coordinated dual-hormone AP system that does not provide meal notifications, but allows patients to provide meal notifications if they so desire, and enables the determination of antagonistic regulatory actions between glucagon and carbohydrates, has not yet been identified in the art. [Overview of the project]
[0016] The present invention provides a method for controlling blood glucose for a flexibly configured dual-hormone artificial pancreas. This method allows for the handling of optional dietary and exercise notifications by using coordinated control actions.
[0017] This system coordinates (harmonizes) regulatory control actions (insulin) and antagonistic regulatory control actions (glucagon and / or carbohydrates). As described above, the method of the present invention manages meals for which notifications are not given; however, if the user so desires, the user may provide meal notifications as a result of a non-interactive (non-interfering) feedback-feedforward scheme that reduces the mutual interference between feedback control and feedforward control.
[0018] In addition to insulin, the system provides antagonistic control actions to address hypoglycemia induced by, for example, unannounced exercise events. Users can choose whether to have these antagonistic control actions implemented as automated glucagon infusion via a pump, as suggestions regarding glucagon administered by the patient using a pen, or as suggestions regarding carbohydrate intake.
[0019] Therefore, the method of the present invention can manage meals (with or without notification, depending on user preference) and exercise events without notification, providing parameters within the range recommended in clinical guidelines in an in silico trial. The most significant change in performance corresponds to the patient deciding whether or not to provide meal notifications. As expected, when meal notifications are provided, the controller improves the percentage of time spent in a hyperglycemic state compared to when operated without meal notifications, without a significant drop in hypoglycemia-related parameters.
[0020] Furthermore, in in silico trials, when using glucagon or carbohydrates as antagonistic control actions, it must be considered that there is no meaningful difference in performance parameters between the two. This is because, by design and theory, they provide the same control effort as glucagon administration, which has the advantage of being a non-caloric supply and being automatically administered using a pump.
[0021] The present invention enables the provision of nearly optimal postprandial glucose control, regardless of whether or not meal or exercise notifications are given. "Optimal" is defined as a control method that minimizes the postprandial blood glucose peak while simultaneously preventing glucose levels from falling below a defined threshold, in other words, a predefined hypoglycemic limit (e.g., 70 mg / dl), under nominal conditions.
[0022] The feedback controller provides feedback control actions that achieve near-optimal postprandial glucose control when no meal notification is given. Therefore, optional meal notifications are managed by a non-interactive feedback-feedforward scheme that prevents insulin infusion after a preprandial bolus administered based on an optional (if necessary) notified meal, thereby preventing insulin overdose and maintaining performance.
[0023] The proposed solution defines a novel control strategy designed to function regardless of whether or not there is a meal or exercise notification, achieving near-optimal postprandial glucose control under nominal conditions. This strategy includes a pre-meal bolus under insulin administration control, designed to achieve near-optimal postprandial glucose even without meal notification, in order to achieve optimized coordination between feedback control and feedforward action. At the same time, it allows for the calculation of whether or not an antagonistic control action is necessary, for example, due to an exercise event.
[0024] The optimal open-loop policy involves administering an insulin bolus before mealtime, in other words, an instantaneous feedforward control action. However, it should be noted that this action cannot be generated by feedback, as feedback control cannot be implemented until the output is affected by a disturbance. Therefore, the best performance achievable through feedback is obtained by administering an optimally sized bolus at the point when the output is affected by a disturbance.
[0025] The expected postprandial variation (y) of CGM caused by the intake of food notified by the patient and the bolus administered in connection therewith. *The expected trajectory is calculated based on the meal and insulin model. This expected trajectory is used to calculate a correction term for the control action provided by the feedback controller, preventing mutual interference between the feedback action and the feedforward action that could lead to insulin overdose. As a result, the corrected feedback control action following the notified meal controls (adjusts) only the deviation from the expected trajectory during the postprandial period caused by the meal bolus.
[0026] Therefore, the method of the present invention is designed to achieve near-optimal blood glucose control with or without meal and / or exercise notifications, and is configured as a near-optimal feedback controller designed to function without meal notifications, and as a coordinated strategy between feedback and feedforward actions to adapt to meal boluses from optionally notified meals. At the same time, the method of the present invention allows for the calculation of antagonistic control actions if necessary.
[0027] This method involves measuring a plasma glucose signal (G(t)) by a continuous glucose monitor (CGM), which is represented by its Laplace transform (G(s)). Hereinafter, each signal can be represented by its time domain (t) or Laplace domain (s) associated by the Laplace transform. An increase in the plasma glucose measurement (y) is introduced, which is y(t)=G(t)-G b (In the formula, G b This is the baseline glucose injection volume (u b (This is the baseline glucose value resulting from the application of the ) It is calculated as follows.
[0028] Therefore, the increase in insulin infusion is, u(t)=u T (t)-u b (where (u(t)) is the total insulin infusion amount) is defined by
[0029] Furthermore, the increase in plasma glucose (y) is considered while taking into account the carbohydrate intake (d), insulin infusion amount (u), glucagon administration amount (v), and rescue carbohydrate administration amount (w), y(s)=G u (s)·u(s)+G v (s)·v(s)+G w (s)·w(s)+G d (s)·d(s) (where u(t)≧-u b , v(t)≧0, and w(t)≧0) is defined by
[0030] The carbohydrate administration amount (w) can be further processed by quantification means (quantifiers) to provide an amount that can be easily administered manually by the patient. Also, when easy manual administration by pen is suggested to the patient, the glucagon administration amount can also be quantified. As an alternative, in some cases, a quantified glucagon signal is automatically administered by a pump instead of administering a continuous glucagon signal.
[0031] In the context of the present invention, G u , G v , G w , and G d are time-invariant linear transfer functions that respectively associate the intake amounts of insulin, glucagon, rescue carbohydrates, and carbohydrates from meals with glucose, and can be defined by a plurality of models known in the art, provided that G u (s) / G v (s), and G u (s) / G w (s) must be realizable, that is, the degree of the numerator must be less than or equal to the degree of the denominator.
[0032] [[ID=__50]] Preferably, the models G u , G v , Gw , and G d The structure is as follows:
[0033]
number
[0034] Defined by, in the formula,
[0035]
number
[0036] teeth,
[0037]
number
[0038] (In the formula, l u ,l v ,l w , and l u This is a parameter that represents the delay,
[0039]
number
[0040] τ 1u , τ 2u , τ 1v , τ 2v , τ 1w , τ 2w , τ 1d , and τ 2d (where is the parameter obtained so far, and j is the index representing the patient.) This is the transfer function of the form.
[0041] Carbohydrate intake is
[0042]
number
[0043] (In the formula, M * (This is the estimated carbohydrate content reported by the patient.) It is defined as follows. Therefore, the expected increase in postprandial plasma glucose (y * )teeth, y * (s) = G u (s)·u * (s) + G d (s)·d * (s) (In the formula, u * (s) is an insulin bolus administered in connection with the patient's manual notification of meals. It is defined as follows.
[0044] Preferably, insulin bolus u * (s) uses the feedforward controller C(s): u * (s) = C(s)·d * (s) It is calculated as follows.
[0045] Therefore, the corrected increase in plasma glucose
[0046]
number
[0047] , corrected insulin infusion volume
[0048]
number
[0049] , and adjusted carbohydrate intake
[0050]
number
[0051] teeth,
[0052]
number
[0053] It is defined as follows. Corrected insulin infusion volume
[0054]
number
[0055] Using this, the virtual control action μ(s) is:
[0056]
number
[0057] It is defined as, and as a result,
[0058]
number
[0059] This is the result. This newly acquired virtual control action is then used for the control action (μ r ) and antagonistic control action (μ cr ) is divided into:
[0060]
number
[0061] This formula describes antagonistic control actions,
[0062]
number
[0063] (In the formula,
[0064]
number
[0065] Like this, μ cr,v This represents the antagonistic control effect introduced (executed) by glucagon injection, and μ cr,w This is further divided as representing the antagonistic regulatory effect introduced by carbohydrate suggestions.
[0066] Controlling actions and counter-controlling actions are,
[0067]
number
[0068] (In the formula, σ(t) is a function (switching signal) that defines the change between the controllative action and the counter-controllative action, for example, σ(t) = μ0(t) or μ0(t) filtering such as a moving average filter with a specific sample window, γ is an adjustable coefficient, and k cr This is the controller gain obtained so far, and μ0 is, μ0(s) = μ dob (s) It is defined as, however, μ dob This is a nominal value pre-filter F r As a 2-degree-of-freedom (2-DOF) feedback controller having (s),
[0069]
number
[0070] μ dob (s) = K(s)·(F r ·r(s)-y(s))+K(s)·y * (s) (In the formula, K(s) is the central linear controller designed to stabilize the system.) This is calculated as follows:
[0071]
number
[0072] is a disturbance
[0073]
number
[0074] This can be attenuated and, in a preferred embodiment, calculated as a control based on a disturbance observer. Preferably, the central controller K(s) is
[0075]
number
[0076] Defined as, filter F(s) and filter F r (s) can be a low-pass filter. More specifically, these filters are
[0077]
number
[0078] (In the formula, k, k r , α, and α r These are the parameters calculated so far, and r(s) is G b (This is the nominal value or target of the increase in glucose relative to [the target value].) It is defined as follows.
[0079] This method involves antagonistic control actions between glucagon administration and rescue carbohydrate administration. μ cr,v (s) = (1 - θ cr)·μ cr (s) μ cr,w (s) = θ cr ·μ cr (s) (In the formula, θ cr (This is an adjustable parameter between 0 and 1.) The process further includes the step of separating (distributing) into different forms.
[0080] In some embodiments, θ cr (t) may change over time under certain conditions in which one antagonistic control action is preferable to the other. For example, when insulin uptake is high (e.g., after a large insulin bolus), or when the cumulative dose of glucagon is excessively high, it is preferable to administer a relief carbohydrate rather than glucagon to avoid side effects such as nausea, or according to the user's preference.
[0081] Therefore, the applicable control action is:
[0082]
number
[0083] It is calculated as follows. The method of the present invention enhances safety when applying the obtained control action.
[0084]
number
[0085] (In the equation, σ(t) is a function that defines the change between the controllative action and the counter-controllative action, as explained above.) The procedure may further include steps to introduce modifiers for controllative and counter-controllative actions.
[0086] Furthermore, the method of the present invention may further include the step of introducing a correction for μ0 in order to prevent glucagon administration as a feedforward action. In the above case, μ dob teeth, μ fb (s) = K(s)·(F r (s)·r(s)-y(s)) μ ff (s) = K(s)·y * (s) As such, feedback action (μ fb ) and feedforward action (μ ff ) is divided into two parts.
[0087] Therefore, μ0 is
[0088]
number
[0089] It is calculated as follows. As explained above, the control action for rescue carbohydrates is quantified at the level (q w ) can be discretized to administer a specific dose. In this case, the discretized control action
[0090]
number
[0091] teeth,
[0092]
number
[0093] (In the formula,
[0094]
number
[0095] represents the nearest integer operator, q w This is the minimum cumulative carbohydrate threshold for activating the suggestion. q w w and
[0096]
number
[0097] (In the formula,
[0098]
number
[0099] is a specific time axis t H This is the predicted glucose value in [location].
[0100]
number
[0101] and G(k) thr These are thresholds for predicted and measured glucose, respectively, and enable HypoFlag. It can be calculated by ). The prediction for G(k) is
[0102]
number
[0103] It is calculated as follows, where δ G (k) is the discrete, filtered glucose derivative (derivative value), or in other words,
[0104]
number
[0105] (where τ δ >τ s is the time constant determined so far) is the discretization of, and
[0106]
Number
[0107] (where BW is the user's body weight (kg), τ s >0 is the sampling period, and τ cl is the clearance) is.
[0108] Preferably, the threshold value [[ID=
[0115] (where q v is the minimum threshold of cumulative glucagon to activate the suggestion, q v <q v and
[0116]
Number
[0117] ]> and δ G (k) is the discretized glucose derivative value after filtering, that is, <00^00754>
[0118]
Number
[0119] (where τ δ is the time constant determined so far) is the discretization of
[0120]
Number
[0121] is calculated by Preferably, when a specific dose is used as the control action, the expected increase in postprandial plasma glucose y * (s) is
[0122]
Number
[0123] (where η v and η w [[ID=7^]]are adjustable gains) can be calculated by adding a term corresponding to the specific dose of the antagonistic control action.
[0124] Preferably, η v and η w This is when the antagonistic control action (v,w) is quantified at the level (q v ,q w The increase in the desired maximum output is caused when it is equal to ).
[0125]
number
[0126] Designed to achieve,
[0127]
number
[0128] (In the formula, L -1 {·} represents the inverse Laplace transform operator.
[0129]
number
[0130] This is the quantification level (q v ,q w (This is the desired output change caused by...) It is obtained as follows. Preferably, the expected increase in postprandial plasma glucose is in a saturated state.
[0131]
number
[0132] This could be the case, but as a result
[0133]
number
[0134] (In the formula,
[0135]
number
[0136] These are the upper and lower thresholds that have been determined (measured) so far. It is expressed as follows. Optionally, the adjustable gain k mentioned above is given by k = 0.65 × 6 × (1 / CR j ) is defined as.
[0137] As an alternative, the adjustable gain k is:
[0138]
number
[0139] (In the formula, L -1 is the inverse Laplace transform, and y min and
[0140]
number
[0141] This is an adjustable parameter, however
[0142]
number
[0143] represents the expected amount of food in the worst-case scenario, and y min That meal
[0144]
number
[0145] (This is the acceptable lower limit for the increase in the postprandial glucose response in response to [the given condition].) is defined as. Optionally,
[0146] [Number]
[0147] is defined, for each meal notification made by the patient, as the amount of the meal (M * ) and y min is defined as the acceptable lower limit for the increase in the post - meal glucose response in response to that meal M * , and the calculated value of k is held for an adjustable time window from the time the meal is notified.
[0148] In some embodiments, the insulin bolus (u * ) administered is calculated as a bolus defined by u * (s) = C(s)M * (s) (where C(s) is * a feed - forward controller that can be defined as
[0149] [Number]
[0150] ). As an alternative, the feed - forward controller C(s) can
[0151] [Number]
[0152] conform to a super - bolus strategy defined as, provided that Tσ > 0 is the period during which the pump is stopped. The present invention is also an artificial pancreas system incorporating a method for controlling blood glucose as defined above, Cooperative control action (u T ) in accordance with the delivery of automated insulin infusions and / or coordinated control actions
[0153]
number
[0154] Accordingly, a pump for delivering automated glucagon infusion, A continuous glucose monitor for detecting plasma glucose signals (G(t)), A first computing unit configured to carry out the steps of the method of the present invention, Regarding systems for preparing for such situations.
[0155] Alternatively, the system may include a second pump for delivering automated glucagon infusions. In alternative embodiments, the artificial pancreas system includes a display for notifying the patient of recommended control actions regarding the administration of glucagon and / or relief carbohydrates. In those cases, the system may also include a pen-type device for the patient to deliver the recommended dose of glucagon.
[0156] The present invention also relates to a computer program configured to perform steps of a defined method using a computing unit of an artificial pancreas system, and a computer-readable storage medium containing the said computer program.
[0157] To complement the prepared specification and to aid in a better understanding of the features of the invention based on its preferred practical embodiments, a series of drawings are attached as an integral part of the specification. These drawings depict the following illustrative and non-limiting features. [Brief explanation of the drawing]
[0158] [Figure 1]It is a diagram showing a schematic view of a control strategy according to the method of the present invention. [Figure 2] It is a diagram showing a graph of the percentage of the time during which glucose is within a specific range. [Figure 3] It is a diagram showing control actions calculated for the configuration of FIG. 2 according to the method of the present invention.
Embodiments for Carrying Out the Invention
[0159] In order to obtain parameters, a patient-oriented control model based on an identification method is presented, and as a result, personalized gains are obtained based on easily available relevant clinical parameters of the patient.
[0160] The interaction between glucose and insulin is a complex system and can be represented by a high-order non-linear model. However, when aiming at control design, a simple model using some personalized parameters is often used.
[0161] The proposed AP has been verified using 10 adult cohorts from the academic version of the UVa / Padova simulator in a challenging scenario including meal and exercise events.
[0162] FIG. 1 represents the proposed fully autonomous cooperative artificial pancreas system. The core of the system includes four basic components: (A) asymmetric cooperative control, (B) carbohydrate and glucagon recommendation module, (C) patient action adaptation module, and (D) central controller and prefilter. The asymmetric cooperative control distributes the output of the patient action adaptation module to the insulin infusion (amount) u(t) and the antagonistic control actions that lower and raise the glucose level, respectively. As the antagonistic control actions, the user can set the automatic glucagon infusion amount (v(t)), for example, the quantified dose of glucagon administered using a pen
[0163]
number
[0164] Suggestions regarding, or carbohydrate intake derived from the carbohydrate and glucagon recommendation module.
[0165]
number
[0166] The controller can be configured to provide suggestions regarding meals. The central controller is designed to operate without meal notifications, but the system can also manage notified meals through a patient action adaptation module. This module adapts the output of the central controller when the user manually administers quantified antagonistic actions, such as carbohydrate intake or glucagon injection.
[0167] In the system design, the following simplified relationship is assumed: the increase in glucose y (mg / dl) relative to the baseline state, and the baseline injection volume u. b The relationship between the increase in insulin infusion rate u (pmol / kg / min), glucagon infusion rate v (mg / kg / min), and rescue carbohydrate administration rate w (mg / kg / min).
[0168] y(s)=G u (s)·u(s)+G v (s)·v(s)+G w (s)·w(s)+G d (s)·d(s) (1) In the equation, d(s) corresponds to the oral carbohydrate intake rate (mg / kg / min) of size M, modeled using a pulse signal; in other words, d(s) = M. Time delay G d (s), G u (s), G v (s), and G wThe linear transfer function with (s) is:
[0169]
number
[0170] (In the formula, the superscript 'j' represents the gain that has been adapted for the patient.) This is described by: The control input has a saturation limit u(t)≧-u for all t. b And it is subject to the constraints v(t)≧0 and w(t)≧0. The dynamics in equation (2) are assumed to be the following three approximations (estimates) in order to simplify model identification: 1) The two compartments represent the absorption of the input (u, d, v, or w). 2) One compartment simulates the effect of this input on glucose, and 3) All models have the same delay equal to 15 minutes. It should be noted that, when the goal is to control diabetes, personalized models with similar gains are conventional (traditional).
[0171] The above model was identified for 10 hypothetical adults included in the UVa / Padova distribution version. The parameters of the control-oriented patient model corresponding to the hypothetical adults identified in the UVa / Padova simulator are shown in Table I.
[0172] [Table 1]
[0173] The proposed control strategy should allow patients to provide meal notifications, if desired, and consequently enable insulin bolus administration. However, doing so without notifying the feedback controller could lead to insulin overdose. To avoid this situation, the central controller receives meal notifications through a non-interactive feedback-feedforward scheme implemented in the patient action adaptation module. The expected output variability is: y * (s) = G d (s)·d * (s) + G u (s)·u * (s) (3) It is defined as follows. However, this is,
[0174]
number
[0175] (In the formula, d * (s) = M * is the carbohydrate content (g) of the notified meal, CIR is the ratio of carbohydrates to insulin (g / U), BW is body weight (kg), v∈(0,1) is the decay coefficient, and the term (6000BW) -1 Insulin bolus u is calculated by converting insulin units from U to pmol / kg. * (s) = B * Based on this, subtracting (1) and (3) gives,
[0176]
number
[0177] (In the formula,
[0178]
number
[0179] (is) This is obtained. Thus, a controller can be designed based on a model (4) that already incorporates feedforward actions performed by the patient. Note that in the ideal case where the patient provides accurate information about the amount of food consumed, resulting in the appropriate amount of insulin bolus administration, the feedback controller will not be activated (it will be disabled) as long as the output follows the expected pathway, and therefore, undesirable interaction between feedback actions and feedforward actions is avoided.
[0180] Having more manipulated inputs than controlled outputs, as in the MISO (multiple input single output) model (4), can be used to improve performance under different operating conditions and manage input saturation, but it can also introduce some difficulties in the design process. To simplify this challenge and to take into account the nature of different control actions, model (4) is...
[0181]
number
[0182] It is convenient to rewrite it as follows. However,
[0183]
number
[0184] It is considered a virtual control action that includes both controllative and counter-controllative actions.
[0185]
number
[0186] Defined by: Here, the last two terms represent the antagonistic regulatory effects introduced by the amount of glucagon infusion (v) or carbohydrate intake (w), which are
[0187]
number
[0188] Provided by [company name]. The antagonistic control mode should be more aggressive to compensate for any decrease in glucose levels that could have serious consequences for the patient. Let μ0(s) be the result of the patient action adaptation module. The final control effort μ for distribution is:
[0189]
number
[0190] (In the formula, μ0 is defined as follows) It is defined as follows:
[0191]
number
[0192] However, in antagonistic control mode, the controller gain is k cr It increases by ≥1. Note that this definition is imposed by equation (6) for all t, except when saturation is in effect. μ(t) = μ r (t)+μ cr (t) Please note that this constraint must be followed. Furthermore, in accordance with equation (8), and in order to ensure coordination of the two available antagonistic control actions, the following distribution is necessary. μ cr,v (s) = (1 - θ cr )·μ cr (s), μ cr,w (s) = θ cr ·μcr (s) (12) The following is selected. However, θ cr ∈{0,1} is time-varyingly adjustable, allowing for the selection of either glucagon or a rescue carbohydrate.
[0193] The above cooperative logic incorporates two mechanisms to minimize the consumption of antagonistic control actions. In one, switching is performed on the filtered signal σ. This signal is calculated by applying a moving average filter with a 3-sample window to the signal μ0. This avoids the undesirable activation of the antagonistic control mode due to noise in the control action. Furthermore, the threshold for activating the antagonistic control mode is slightly shifted by an adjustable coefficient γ. This modification creates a small dead zone—γ·u—where no antagonistic control action is performed. b <σ(s)<-u b This implies the existence of μ. The min and max functions are used solely to guarantee (ensure) the feasibility of the resulting control action. Finally, the μ defined above. r (s), μ cr,v (s), and μ cr,w Using (s), the control signal that guarantees closed-loop coordination is:
[0194]
number
[0195] Therefore, it can be calculated from (13) to (15). A positive insulin pulse, or in other words, an insulin bolus, is the optimal insulin infusion that limits the postprandial peak while simultaneously minimizing glucose deficiency. From the equivalent model (5), the following central controller: μ dob (s) = K(s)·(F r (s)·r(s)-y(s))+K(s)·y *(s) (16) This can be derived. However,
[0196]
number
[0197] The adjustable parameters are k, k r , α, α r This is an extremely suitable filter with >0. The control action (16) is, μ dob (s) = μ fb (s) + μ ff (s) μ fb (s) = K(s)·(F r (s)·r(s)-y(s)) μ ff (s) = K(s)·y * (s) (17) This can be rewritten as follows, which shows that the output of the central controller is the contribution of two terms, one of which is the tracking error μ fb This term is obtained from μ, and the other is μ ff This is a correction term resulting from the feedforward action provided as follows. Note that in some cases, for example, in the post-meal period after a pre-meal bolus, y * It should be noted that overestimating this can lead to counter-controlling actions, which are undesirable. To avoid this, the contributions of both are considered.
[0198]
number
[0199] It will be integrated like this. The suggestion for the relief carbohydrate must be provided as a specific quantified dose, not as a continuous rate. Therefore, the continuous signal w must be converted into a quantified pulse. τ sIf >0 is the sampling period, the recommended carbohydrate intake at sampling time k is:
[0200]
number
[0201] It is expressed as,
[0202]
number
[0203] It is given by the formula,
[0204]
number
[0205] represents the nearest integer operator, and q w This is a quantification level, q w w This is the minimum cumulative carbohydrate threshold for activating the suggestion. Signal z w (k) is,
[0206]
number
[0207] It is an auxiliary variable defined by . In the formula, τ σ teeth,
[0208]
number
[0209] The clearance τ described by cl >τ s This is the time constant determined so far, which depends on the rate of cumulative carbohydrate accumulation. HypoFlag is an alert indicating the risk of hypoglycemia. This is due to the following conditions
[0210]
number
[0211] This corresponds to the following. In the formula, G(k) is the actual reading from the continuous glucose monitor (CGM),
[0212]
number
[0213] teeth,
[0214]
number
[0215] (In the formula, δ G (k) is,
[0216]
number
[0217] This is the discretization of the time constant τ. δ (This is the differential value of glucose filtered using a low-pass filter.) This is the predicted glucose level after one hour, calculated using [the specified method / tool].
[0218] In the above quantification scheme, z w (k) q w As long as it reaches q w The recommended carbohydrate intake, which is a multiple of (mg), is activated. Next, the suggested carbohydrates are used as a cumulative signal.
[0219]
number
[0220] It is subtracted from and the excess is compensated for. To mitigate delays in the accumulation process, q w is q w It is set to less than . As a result of the delayed quantification, if the continuous equivalent quantity w(k) is close to 0, the controller may activate the suggestion too late. The condition HypoFlag in equation (19) is provided to reduce the risk of such delayed suggestions. Furthermore,
[0221]
number
[0222] This includes a forgetting factor, which is set to 0 if the condition HypoFlag is not met, even if it is large enough to enable the suggestion. Unlike rescue carbohydrate intake, glucagon infusion rates can be administered automatically via a pump (other than an insulin pump or dual-chamber pump). The system may be able to continuously administer glucagon while simultaneously providing quantified carbohydrate suggestions. However, some patients may prefer manual injection of low-dose glucagon, which does not require an additional (or more complex) pump, as a non-caloric alternative to recover from mild hypoglycemia. In glucagon quantification strategies, z w ,
[0223]
number
[0224] , q w , and q w Then, using the corresponding method, z v ,
[0225]
number
[0226] , q v , and q v By substituting this, an equation equivalent to the carbohydrate quantification scheme is introduced. The quantified antagonistic control action may differ from that calculated by the central controller. To notify the central controller of this situation, a non-interactive feedback-feedforward scheme is used to calculate the expected output variation in equation (3).
[0227]
number
[0228] (In the formula, η v and η w This defines the desired increase in maximum output that results when the antagonistic control action is equivalent to the corresponding quantification level. This is corrected by redefining it as follows. Therefore, these parameters are
[0229]
number
[0230] (In the formula, L -1 {·} represents the inverse Laplace transform operator.
[0231]
number
[0232] is, q v or q w (Represents the desired output change caused by the change) It can be derived from this. Furthermore, in order to prevent the output from accidentally causing hyperglycemia or hypoglycemia, signal y * teeth
[0233]
number
[0234] It becomes saturated like this. It should be noted that the expected output fluctuations modified in equation (24) are practical considerations for implementation purposes, and the design of the central controller is still based on the factorized model equation (5) derived from (4).
[0235] [Table 2]
[0236] Table II contains the controller parameters. The gain k of the central controller is given by the following formula:
[0237]
number
[0238] (In the formula, η is a safety factor set to 0.7) Accordingly, it was individualized for each target j. α, α in Table II r The values of , and v were derived through simulation. The antagonistic control gain was adjusted k crSince the coefficient γ was initially heuristic, an optimization-based strategy was used to individualize these parameters. However, because optimization-based adjustments are not applicable to clinical trials, population values were also evaluated by adopting the median of the optimal value for each parameter. In the simulations, no significant difference was found between using optimal adjustments and using population values. Therefore, for the sake of simplification, population parameters were selected in this study. For quantification-related parameters, the quantification level for carbohydrates was set to the conventional dose in commercially available gels, in other words, 15 g. The glucagon suggestion was quantified to 0.08 mg because this quantification level has achieved safe and effective control in recent clinical trials. The remaining quantification-related parameters (q) are shown in Table II. v , q w , and τ cl The thresholds defining the conditions, q, and HypoFlag, were adjusted through simulation. v and q w It can be noted, based on Table II, that all of these were set to 1 / 3 of the corresponding quantification level.
[0239] The results obtained were simulated using 10 virtual adults included in an extended version of the UVa / Padova simulator for the six proposed controller configurations (a hybrid system using glucagon, a hybrid system using carbohydrate suggestions, a system using glucagon without meal notification, and a system using carbohydrate suggestions without meal notification – where glucagon was assumed to be administered as a suggestion or continuous infusion).
[0240] The scenario included three meals a day for 14 days, randomized in terms of carbohydrate content and timing: 37.0 [30.0, 45.0] g (median [25th percentile, 75th percentile]) at 8:05 [7:45, 8:15]h, 53.0 [36.0, 58.0] g at 13:05 [12:50, 13:15]h, and 75.5 [53.0, 90.0] g at 19:50 [19:35, 20:10]h. In addition, eight aerobic exercise sessions were randomly scheduled every other day from day 1 at 19:17 [14:40, 22:10], with a duration of 55.0 [50.0, 60.0] minutes and an intensity of 48.5 [47.5, 53.0]% of maximum oxygen uptake. The effect of exercise on glucose was demonstrated by increasing insulin sensitivity. Finally, the controller configuration was tested for noise (built-in sensor model dexcom25) and the following sources of variation introduced in the academic version of the simulator: the nominal values of the dietary absorption rate and the carbohydrate bioavailability parameters of the dietary absorption model varied with a uniform distribution of ±30% and ±10% per meal, respectively; the nominal values of the parameters describing insulin pharmacokinetics were corrected according to a uniform distribution of ±30% per meal; the circadian variability of insulin sensitivity was represented as a 24-hour periodic sinusoidal change with random amplitude and random phase following a uniform distribution of ±30%; and error estimation of the amount of carbohydrates in the meal was introduced. Note that in the case of controller configurations with carbohydrate and glucagon suggestions, users should be aware that they should follow the recommendations (i.e., carbohydrate intake or glucagon administration) simultaneously with the suggestions. More ideally, this premise is considered in other studies in the literature. The performance of the controller configuration was evaluated through the percentage of time spent within, below, or above the range, as well as through daily consumption of insulin, glucagon, and carbohydrates.
[0241] Figure 2 shows a graph of the percentage of time (%) that glucose levels are within a specific range. The lower and upper boxes represent the 25th and 75th percentiles, respectively, and the thick black horizontal line corresponds to the median. The cross represents the mean.
[0242] The ends of the upper and lower whiskers represent the maximum and minimum values, respectively, that do not exceed 1.5 times the interquartile range (up to 1.5 times). The black dots outside the whiskers represent outliers. Notified meals and non-notified meals will be considered. Furthermore, glucagon administered continuously by a pump (glucagon) and glucagon suggestions quantified by a pen (glucagon suggestions) will be evaluated. Carbohydrates will always be quantified, and suggestions regarding manual intake (carbohydrate suggestions) will be made to the patient.
[0243] As shown in Figure 2, all controller configurations were capable of managing both unannounced meal and exercise events, with mean values falling within their respective ranges. Episodes of hypoglycemia were rare in all configurations. Only subjects 3 and 6 had episodes with CGM < 54 mg / dl, but the time spent within the aforementioned range was less than 0.70%. Subject 6 had episodes with CGM < 70, but these were clinically acceptable. Even in the patient with the longest duration of hypoglycemia (subject 3), the time spent below 70 mg / dl was only 1.6%. Reduction in the duration of hypoglycemia is achieved without the application of excessive antagonistic control actions.
[0244] Figure 3 shows the control actions calculated for the configuration shown in Figure 2 according to the method of the present invention. As shown in Figure 3, the required median daily dose for the glucagon configuration is less than 1 mg (the threshold at which subjects frequently experience nausea). The required daily dose for the carbohydrate configuration is similar to that of other controllers in the literature evaluated under comparable synthetic scenarios.
Claims
1. A method for controlling glucose in a flexibly configured dual-hormone artificial pancreas capable of managing optional dietary and exercise notifications through coordinated control actions, The steps include measuring the plasma glucose signal (G(t)) using a continuous glucose monitor (CGM), and y(t)=G(t)-G ｂ (In the formula, G ｂ (This is the baseline glucose value.) The steps include calculating the increase in plasma glucose measurements and y(s)=G ｕ (s)・u(s)+G ｖ (s)・v(s)+G ｗ (s)・w(s)+G ｄ (s)・d(s) (where u(t) ≥ -u ｂ , v(t) ≥ 0, and w(t) ≥ 0, and G ｕ , G ｖ , G ｗ , G ｄ are time-invariant linear transfer functions that relate the carbohydrate intake (d), the increment in insulin infusion (u), the glucagon dose (v), and the rescue carbohydrate dose (w) to the increment in plasma glucose (y), and G ｕ (s) / G ｖ (s) and G ｕ (s) / G ｗ (s) are subject to the restriction that the degree of the numerator does not exceed the degree of the denominator) As such, carbohydrate intake (d), baseline injection amount u ｂ The steps include defining a model for the increase in plasma glucose (y) that takes into account the increase in insulin infusion (u), glucagon dose (v), and rescue carbohydrate dose (w), and Carbohydrate intake [Math 1] (In the formula, M ＊ (This is the estimated carbohydrate amount reported by the patient.) The step of defining it as, Expected increase in postprandial plasma glucose y ＊ (s) y ＊ (s)=G ｕ (s)・u ＊ (s)+G ｄ (s)・d ＊ (s) (In the formula, u ＊ (s) is an insulin bolus administered in connection with the patient's manual notification of meals. The step of defining it as, Corrected increase in plasma glucose [Math 2] , corrected insulin infusion volume [Math 3] , and adjusted carbohydrate intake [Math 4] of, [Math 5] The step of defining it as, The virtual control action μ(s) [Math 6] Defined as such, and as a result, [Number 7] The steps include, Control action μ(s) [Number 8] Controlling control action μ ｒ and antagonistic control action μ ｃｒ The steps are to divide it into and Counter-controlling actions [Number 9] (In the formula, μ ｃｒ，ｖ μ represents the antagonistic regulatory effect introduced by glucagon injection, ｃｒ，ｗ This represents the antagonistic regulatory effect introduced by carbohydrate intake. [Number 10] The steps are to distribute as follows: Controlling actions and counter-controlling actions [Math 11] (wherein σ(t) is a function that defines the change between the controllative action and the countercontrollative action, γ is an adjustable coefficient, and k ｃｒ This is the controller gain obtained so far, μ ０ teeth, m ０ (s) = m ｄｏｂ (s) Defined as μ ｄｏｂ This is a nominal value prefilter F ｒ As a two-degree-of-freedom (2-DOF) feedback controller having (s), [Math 12] μ ｄｏｂ (s)=K(s)・(F ｒ (s)・r(s)-y(s))+K(s)・y ＊ (s) (In the formula, K(s) is the system, i.e.) [Number 13] It is a central linear controller designed to stabilize disturbances [Number 14] (to attenuate) The steps to calculate as follows, Counter-controlling actions, m ｃｒ，ｖ (s) = (1-θ) ｃｒ )・m ｃｒ (s) m ｃｒ，ｗ (s) = θ ｃｒ ・m ｃｒ (s) (In the formula, θ ｃｒ (∈[0,1] is an adjustable parameter; y) The steps involve calculating in the following form: Control Action [Number 15] The steps to calculate as and A method that includes this.
2. Model G ｕ G ｖ G ｗ G ｄ but, [Number 16] It is obtained as, in the formula, [Number 17] teeth, [Number 18] (In the formula, l ｕ , l ｖ , l ｗ , and l ｕ This is a parameter that represents the delay, [Number 19] τ １ｕ , τ ２ｕ , τ １ｖ , τ ２ｖ , τ １ｗ , τ ２ｗ , τ １ｄ , and τ ２ｄ (These are the parameters obtained so far, and j is the index representing the patient.) The method according to claim 1, wherein the transfer function is of the form of the transfer function. 【Request Item 3】 【Number 20】 (In the equation, σ is a function that defines the change between the controllative action and the counter-controllative action, and γ is an adjustable coefficient.) The method according to claim 1 or 2, further comprising the step of introducing a modifier for controllative and counter-controllative actions in a specified form.
4. σ(t) is μ ０ (t) is equal to μ using a specific sample window. ０ The method according to any one of claims 1 to 3, wherein the moving average filter is of (t).
5. μ ０ In contrast, [Math 21] The steps include introducing corrections and Feedback action (μ ｆｂ ) and feed forward action (μ ｆｆ )of, μ ｆｂ (s)=K(s)・(F ｒ (s)・r(s)-y(s)) μ ｆｆ (s)=K(s)・y ＊ (s) (In the formula, r(s) is G ｂ (This is the nominal value of the increase in glucose relative to [the value of glucose].) The method according to any one of claims 1 to 4, further comprising the step defined as:
6. Controller K(s) [Number 22] Defined as, Filter F(s) and F ｒ (s) is, [Number 23] (In the formula, k, k ｒ , α, and α ｒ These are the parameters calculated so far, and r is G ｂ (This is the nominal value of the increase in glucose relative to [the value of glucose].) The method according to any one of claims 1 to 5, as defined as
7. θ ｃｒ The method according to any one of claims 1 to 6, wherein one antagonistic control action is preferred over the other, and is varied over time.
8. Relief carbohydrates, [Number 24] (In the formula, q ｗ is the minimum cumulative carbohydrate threshold for activating the suggestion, and q ｗ <q ｗ and [Number 25] (In the formula, [Number 26] This is the defined target period t Ｈ This is a prediction for G(k) in the context of, [Number 27] and G(k) ｔｈｒ (These are predetermined thresholds for predicted and measured glucose, respectively.) and [Number 28] (In the formula, BW is the user's weight (kg), and τ ｓ >0 is the sampling period, and τ ｃｌ (This is clearance.) (is) The quantification level (q) calculated by ｗ The method according to any one of claims 1 to 5, administered as a specific dose of )
9. The prediction for G(k) is [Number 29] It is calculated as δ Ｇ (k) is the differential value of glucose after filtering: [Number 30] (In the formula, τ σ >τ ｓ (This is the time constant determined so far.) The method according to claim 8, which is a discretization of the method.
10. Glucagon, [Number 31] (In the formula, q ｖ is the minimum cumulative glucagon threshold for enabling the suggestion, and q ｖ <q ｖ and [Number 32] (In the formula, [Number 33] is a prediction for G(k) during the defined target period t Ｈ and is [Number 34] and G(k) ｔｈｒ (These are predetermined thresholds for predicted and measured glucose, respectively.) and [Number 35] (In the formula, BW is the user's weight (kg), and τ ｓ >0 is the sampling period, and τ ｃｌ (This is clearance.) The quantification level (q) calculated by ｖ The method according to any one of claims 1 to 9, administered as a specific dose of )
11. The prediction for G(k) is [Number 36] It is calculated as δ Ｇ (k) is the differential value of glucose after filtering: [Number 37] (where τ σ > τ ｓ is the time constant determined so far) The method according to claim 10, which is a discretization of the method. [Request Item 12] [Number 38] The concentration is 60 mg / dl, and G(k) ｔｈｒ The concentration was 54 mg / dl, and the target period t Ｈ The method according to any one of claims 8 to 11, wherein the time is 60 minutes.
13. Expected increase in postprandial plasma glucose y ＊ (s) is, [Number 39] (In the formula, η ｖ and η ｗ (This is an adjustable gain.) The method according to any one of claims 8 to 12, calculated by adding a term corresponding to a specific dose of an antagonistic control action, as shown above.
14. Adjustable gain η ｖ and η ｗ However, the antagonistic control action (v, w) is at the quantifiable level (q ｖ ,q ｗ When equal to the desired maximum output, [Number 40] This is determined as the gain required to achieve this, [Number 41] (In the formula, L －１ {・} represents the inverse Laplace transform operator. [Number 42] This is the quantification level (q ｖ ,q ｗ (This is a change in Italian output caused by...) The method according to claim 13, obtained as follows.
15. Expected increase in postprandial plasma glucose y ＊ (s) is, [Number 43] The method according to any one of claims 1 to 14, wherein the product is saturated as follows. [Request Item 16] [Number 44] is 180 and [Number 45] The method according to claim 15, wherein is 70.
17. The adjustable gain k of controller K(s) is given by k = 0.65 × 6 × (1 / CR ｊ The method according to any one of claims 6 to 16, as defined as ).
18. The adjustable gain k is [Number 46] (In the formula, L －１ is the inverse Laplace transform, and y ｍｉｎ and [Number 47] This is an adjustable parameter, [Number 48] represents the expected amount of food in the worst-case scenario, and y ｍｉｎ That meal [Number 49] This is the acceptable lower limit for the increase in the postprandial glucose response. The method according to any one of claims 6 to 16, as defined as: [Request Item 19] [Number 50] However, each time an optional meal notification is given, the amount of food notified by the patient (M ＊ Defined as, and y ｍｉｎ However, that meal M ＊ The method according to claim 18, wherein the calculated value of k, defined as an acceptable lower limit for the increase in the postprandial glucose response to , is maintained for an adjustable time window from the time the meal was notified.
20. Insulin bolus (u) administered as a result of optional meal notification ＊ ) is the amount of food (M) notified by the patient. ＊ ) from, u ＊ (s) = C(s)M ＊ (s) (wherein C(s) is [Number 51] The method according to any one of claims 1 to 19, which is calculated as a bolus defined by a feedforward controller provided by
21. The insulin bolus administered (u ＊ ) is the amount of food (M) notified by the patient. ＊ ) from, u ＊ (s) = C(s)M ＊ (s) (wherein C(s) is [Number 52] A feedforward controller provided by T σ The method according to any one of claims 1 to 19, calculated as a superbolus defined by (where >0 is the period during which the pump is stopped).
22. An artificial pancreas system for carrying out the method according to any one of claims 1 to 21, with the aim of controlling blood glucose, Cooperative control action (u Ｔ ) in order to deliver insulin and / or coordinated control actions [Number 53] Accordingly, a pump (3) for delivering the automatic infusion of glucagon, A continuous glucose monitor (2) for measuring plasma glucose signal (G(t)), A first computing unit (1) configured to carry out the steps described in any one of claims 1 to 21 and An artificial pancreas system equipped with [a specific feature / feature].
23. The artificial pancreas system according to claim 22, further comprising a second pump for delivering an automated infusion of glucagon.
24. The artificial pancreas system according to claim 22 or 23, further comprising a display for notifying of recommended control actions regarding the administration of glucagon and / or relief carbohydrates.
25. An artificial pancreas system according to any one of claims 22 to 24, further comprising a pen-type device for the patient to deliver a recommended glucagon control action.
26. A computer program configured to perform steps of the method according to any one of claims 1 to 21 using a defined computing unit of the artificial pancreas system according to any one of claims 22 to 25.
27. A computer-readable storage medium containing the computer program described in claim 26.