Method for validating a control algorithm

Abstract

A method for validating a control algorithm, the method being implemented by a computer and comprising the steps of receiving a plurality of states (50); receiving a plurality of reference actions (52); receiving a plurality of reference outputs (54); processing the plurality of states (56); calculating at least one action difference (58) consisting of calculating at least one difference between at least one control action and at least one reference action; evaluating the at least one action difference (60) by calculating at least one evaluation score; and validating the control algorithm (62), the control algorithm being validated if at least one evaluation score satisfies an evaluation score criterion. Figure for the abstract: Fig. 1

Description

Title of the invention: Method for validating a control algorithm. FIELD OF THE INVENTION

[0001] The present invention relates to a method and device for validating a control algorithm using states comprising at least one glucose measurement value.

[0002] BACKGROUND OF THE INVENTION

[0003] In the field of healthcare, and more specifically in the field of diabetes treatment, it is known to use control algorithms to control a patient's blood glucose, also called blood sugar, within a safe range also known as normoglycemia or euglycemia. The control algorithms send actions, also called control parameters, to an insulin pump, for example, to inject a quantity of insulin based on a measured glucose value.

[0004] Recently, so-called "closed-loop" systems have been developed, in which a processor is programmed to evaluate the insulin injection volume using a control algorithm, based on patient-related and / or time-series data, such as past and / or present blood glucose measurements, and to control the insulin injection based on this evaluation. Furthermore, the processor can be programmed to evaluate the insulin injection volume under certain special circumstances, in particular, meals and / or physical activity. The amount can be injected into the patient, subject to their approval. Such systems are also called "hybrid closed-loop" systems because of the requirement for the patient to declare some of these special circumstances.

[0005] An incorrect amount of insulin to be injected may lead to blood glucose concentrations in unacceptable intervals, where the patient may be in hypoglycemia and / or hyperglycemia.

[0006] It is therefore useful to validate a control algorithm before using it in order to increase patient safety. In general, control algorithms are validated during clinical trials.

[0007] One of the main disadvantages of this procedure is that it can be potentially dangerous for the patient, who must therefore be monitored.

[0008] Another disadvantage is that validating a control algorithm using a clinical trial takes a lot of time.

[0009] The invention thus aims to address at least partially the technical problems presented above.

[0010] BRIEF SUMMARY OF THE INVENTION

[0011] Thus, the invention relates to a method for validating a control algorithm, the method being implemented by a computer and comprising the steps of: - receive a plurality of states, each state comprising at least one glucose measurement value, the glucose measurement value being representative of a user's measured glucose level associated with an associated time; - receive a plurality of reference actions from a reference algorithm, each reference action being associated with a state and corresponding to a control parameter determined by the reference algorithm on the basis of said associated state; - receive a plurality of reference outputs, each reference output corresponding to at least one glucose measurement value and being associated with at least one state; - to process the plurality of states, each state being processed using the control algorithm to generate a control action associated with said processed state and corresponding to a control parameter determined by the control algorithm on the basis of said associated state; - calculate at least one difference in action, the calculation of at least one difference in action consisting of calculating at least one difference between at least one control action and at least one reference action; - evaluate at least one difference in action, the evaluation of at least one difference in action consisting of calculating at least one evaluation score of at least one difference in action on the basis of at least one reference output and a predetermined target; - validate the control algorithm, the control algorithm being validated if at least the evaluation grade satisfies one evaluation grade criterion.

[0012] Such a configuration allows for the accurate validation of a control algorithm without resorting to a simulator, which is susceptible to bias, or a time-consuming clinical trial. Indeed, such a method requires only real data to obtain states, reference controls, and reference outputs to quickly validate a bias-free control algorithm. Virtual patient simulators are difficult to design, often contain modeling biases—which limit variability compared to real life and potentially lead to unrealistic simulations—and are associated with potentially high computational costs. important (such as solving ordinary differential equations by numerical integration).

[0013] According to the present invention, a glucose measurement value is a blood glucose measurement value or a value indicating a blood glucose measurement value, such as an interstitial glucose measurement value.

[0014] According to the present invention, the predetermined target corresponds to at least one blood glucose value of an associated user.

[0015] According to the present invention, a computer is an electronic device that can receive, store, process and produce data according to predefined instructions or algorithms, such as a smartphone, a server or a desktop computer for example.

[0016] According to one embodiment, the plurality of glucose measurement values ​​is received from a continuous glucose monitoring (CGM) sensor, for example.

[0017] According to one embodiment, a control parameter is a control parameter configured to control a subcutaneous insulin delivery device configured to deliver exogenous insulin into a user's subcutaneous tissue in response to a control parameter, in particular continuous infusion insulin and / or bolus insulin, for example.

[0018] According to one embodiment, a state represents the state of the user at an associated time.

[0019] According to the present invention, each state of the plurality of states also includes at least one past glucose measurement value representative of a user's measured glucose level associated with a specific time. Such a configuration allows the control algorithm and the reference algorithm to generate more precise actions by estimating a trend, for example, and thus improve the accuracy of the validation.

[0020] According to the present invention, each state of the plurality of states also includes at least one past insulin injection representative of a user's measured glucose level at a specific time. Such a configuration allows the control algorithm and the reference algorithm to generate more precise actions by estimating the lasting effect of past insulin administrations, for example, and thus improve the accuracy of the validation.

[0021] According to one embodiment, each state of the plurality of states also includes at least one prior carbohydrate intake representative of a user's measured glucose level associated with a specific time. Such a configuration allows testing of the control algorithm and the reference algorithm under different conditions, such as meal management or rest management, and thus improves the accuracy of the validation.

[0022] In one embodiment, a state is a reference output associated with a previous state. In other words, since the control algorithm is validated by its ability to send precise control parameters that keep a user's blood glucose close to the target, a state containing at least one glucose measurement value is a reference output of a previous state. Indeed, the glucose measurement value associated with time t+1 is the result / reference output of the reference action performed at time t based on a state having at least one glucose measurement value associated with time t.

[0023] According to one embodiment, the reference actions and the control actions are positive numbers and are proportional to a quantity of insulin to be injected into a user, for example.

[0024] According to one embodiment, if the control algorithm is validated, the method also includes a step of sending at least one control parameter to a subcutaneous insulin delivery device configured to administer exogenous insulin into a user's subcutaneous tissue in response to a control parameter, in particular continuous infusion insulin and / or bolus insulin, for example.

[0025] According to one embodiment, the reference actions and the control actions can be of any type, such as rescue carbohydrates, glucagon, or insulin, provided that both the reference and control actions are of the same type. Such a configuration makes it easier to compare the reference and control actions and thus to validate the control algorithm more accurately. Furthermore, since rescue carbohydrates, glucagon, or insulin have a monotonic effect on blood glucose levels, this allows for more precise validation of a control algorithm.

[0026] According to the present invention, rescue carbohydrates are a recommendation for carbohydrate intake or a carbohydrate-related control parameter.

[0027] According to one embodiment, the evaluation score criteria can be of any type, such as a predetermined threshold for example.

[0028] According to one embodiment, the evaluation score criterion corresponds to a threshold guaranteeing that the control algorithm is at least as effective as the reference algorithm.

[0029] According to one embodiment, the evaluation score is calculated using a plurality of action differences.

[0030] Such a configuration makes it possible to take into account a plurality of situations and therefore to evaluate more precisely how the control algorithm would have reacted in particular circumstances represented by the states compared to the reference algorithm.

[0031] According to one embodiment, the evaluation score is calculated as follows:

[0032] [Math. 1]

[0033] evaluation grade = (A + B) / (A + B + C + D);

[0034] in which:

[0035] A corresponds to the proportions of i in [0,N] such that D(i) > 0 and o(i, p +h) > predetermined objective;

[0036] B corresponds to the proportions of i in [0,N] such that D(i) < 0 and o(i, p +h) < predetermined objective;

[0037] C corresponds to the proportions of i in [0,N] such that D(i) > 0 and o(i, p +h) < predetermined objective;

[0038] D corresponds to the proportions of i in [0,N] such that D(i) < 0 and o(i, p +h) > predetermined objective;

[0039] i corresponds to a sequence of states and associated actions;

[0040] N corresponds to a sequence of states and associated actions;

[0041] D(i) corresponds to the sum of the differences in action between the indices 0 and p;

[0042] p corresponds to a first constant;

[0043] h corresponds to a second constant;

[0044] o(i, p+h) corresponds to a reference output associated with a sequence of states and actions associated i to a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

[0045] According to the present invention, the term "corresponds to" means "is".

[0046] Such a configuration makes it possible to create an objective and unbiased evaluation score, because the cases falling into category A are cases for which the final glucose measurement value was higher than the predetermined target and the control algorithm would have generated a corresponding control action of a larger quantity of insulin to be injected: we can assume that the control algorithm handled this particular situation better, and in the same way, mutatis mutandis, for other cases B, C and D.

[0047] According to the present invention, p is a first constant corresponding to a number of states, reference actions associated with said states, and reference outputs also associated with said states, sufficiently small such that changes in insulin administration during the number p of states, reference actions associated with said states, and reference outputs also associated with said states have not yet had a significant impact on the reference output o(i, p). Such a configuration makes it possible to accurately calculate an evaluation score, since insulin has not yet had an impact on the reference output, and said reference output therefore cannot be considered different from a control output. A control output corresponding to at least one glucose measurement value if The associated control action was sent to a subcutaneous insulin delivery device. In other words, having a relatively small p-value means that the user's state is not significantly altered between 0 and p under the reference actions, so it can be reasonably assumed that it has not changed. Therefore, the sequences of control actions generated by the control algorithm between 0 and p represent the actions that would have been taken in real-life situations on this user.

[0048] According to the present invention, h is a second constant corresponding to a number of states, reference actions associated with said states, and reference outputs also associated with said states, sufficiently large such that changes in insulin administration during the number p of states, reference actions associated with said states, and reference outputs also associated with said states, have a total impact on the reference output o(i, p+h). Such a configuration makes it possible to accurately calculate an evaluation score, since the effect of the control measures taken during the number p of states is fully active at a state in which the time associated with the blood glucose measurement value associated with the state is equal to p+h.

[0049] According to one embodiment:

[0050] A corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p +h) > predetermined target + high margin;

[0051] B corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p +h) < predetermined objective - low margin;

[0052] C corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p +h) < predetermined objective + high margin;

[0053] D corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p +h) > predetermined target - low margin;

[0054] Such a configuration makes it possible to artificially reduce the evaluation score and therefore to validate only an effective control algorithm.

[0055] Having a high margin that differs from the low margin allows for the artificial reduction of the evaluation score based on the expected effectiveness of the control algorithm. For example, a larger margin prevents the validation of a control algorithm that provides only minor improvements to the control output, in favor of a control algorithm that provides more significant improvements. This avoids making decisions about a control algorithm based solely on small differences in glucose measurement values ​​due to the variability of continuous glucose monitoring sensors. Having a lower low margin than the high margin allows the margins to reflect the smallest distance between dangerously low glucose values ​​leading to hypoglycemia and dangerously high glucose values ​​leading to to hyperglycemia. Therefore, such a configuration only allows for the validation of an effective control algorithm.

[0056] According to one embodiment, the upper margin and the lower margin are variables.

[0057] According to the present invention, the upper margin and the lower margin vary as a function of o(i, p+h), the reference output associated with a sequence of states and actions associated with i to a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

[0058] Such a configuration allows the control algorithm to be validated more safely, because it improves the safety of the process by reducing the variance of the evaluation score, which allows a more precise calculation of an evaluation score and only validating an effective control algorithm.

[0059] According to one embodiment, p is equal to 0.

[0060] Such a configuration makes it possible to calculate an evaluation score based on the difference in performance between a single control action and a reference action. Therefore, the evaluation score will accurately determine the control algorithm's ability to outperform the reference algorithm in various cases, such as meal bolus estimation. According to the present invention, meal bolus estimation refers to the generation of a control action in response to carbohydrate intake, such as a meal.

[0061] According to one embodiment, the step of calculating at least one difference of action consists of calculating at least one difference between a sum of several control actions and a sum of several reference actions.

[0062] Such a configuration allows more actions to be taken into account and therefore allows for a more precise evaluation of at least one difference of action at the stage of evaluating at least one difference of action.

[0063] According to one embodiment, the step of calculating at least one action difference consists of calculating an action difference such that:

[0064] [Math. 2]

[0065] D(i) = (Icont - Iref) / Irap

[0066] D(i) corresponds to a difference in action of a state i;

[0067] Icont corresponds to a sum of a plurality of control actions;

[0068] Iref corresponds to a sum of a plurality of reference shares;

[0069] Irap corresponds to a normalization constant.

[0070] Such a configuration makes it possible to take into account certain particular situations such as D(i) being greater than zero and o(i, p+h) being greater than the predetermined target, but with a control parameter strong enough to modify o(i, p+h) so that it is less than the predetermined target if the control parameter is received by the subcutaneous insulin delivery device instead of the reference parameter for example. Consequently, such a configuration allows for a more precise calculation of an action difference and therefore a more precise calculation of an evaluation score, and for validating only an effective control algorithm.

[0071] According to one embodiment, Icont = Sj=ip : i F(s(i,j)), with s(i,j) corresponding to the jth state of the ith sequence and F(s(i,j)) corresponding to the jth control action of the ith sequence; the control action resulting from the processing of state s(i,j). Such a configuration allows for taking into account more actions and therefore allows for a more precise evaluation of at least one difference of action at the at least one difference of action evaluation step.

[0072] According to one embodiment, Iref = 2j=ip : ia(i,j), with s(i,j) corresponding to the jth state of the ith sequence and a(i,j) corresponding to the jth reference action of the ith sequence; the reference action resulting from the processing of state s(i,j). Such a configuration makes it possible to take into account more actions and therefore to evaluate at least one difference of action more precisely at the step of evaluating at least one difference of action.

[0073] According to an embodiment in which each state includes a baseline reference rate, Irap varies according to the value of the baseline reference rate. Such a configuration makes it possible to calculate a differential action relative to a baseline reference rate and thus to calculate a personalized and meaningful differential action for the associated user. Such a configuration also makes it possible to handle values ​​close to zero in the denominator.

[0074] According to the present invention, the basal reference level is an insulin level, in units per hour or as a fraction of the TDD for example, associated with a user.

[0075] According to an embodiment in which each state includes a basal reference rate and a total daily dose (TDD), if the basal reference rate is greater than one quarter of the TDD and if Iref is greater than zero, Irap = Iref.

[0076] According to an embodiment in which each state includes a basal reference rate and a total daily dose (TDD), if the basal reference rate is greater than one quarter of the TDD and if Iref is equal to zero, Irap = Icont.

[0077] According to an embodiment in which each state includes a baseline reference level and a total daily dose (TDD), if the baseline reference level is less than or equal to one quarter of the TDD, Irap = (½ TDD) / 24.

[0078] According to the present invention, the TDD is a total daily dose of insulin associated with a user and theoretically required by the associated user to maintain glucose values ​​in normoglycemia.

[0079] According to one embodiment, the step of evaluating at least one action difference consists of calculating an action difference evaluation score based on a difference between a reference output and the predetermined target.

[0080] Such a configuration makes it possible to improve the accuracy of the validation of the control algorithm by comparing the differences in action with a reference output and a predefined target. This improvement is achieved using an evaluation score based on the difference between the reference output and the predetermined target, which provides a more precise measure of the effectiveness of the control algorithm. The comparison with a reference output ensures that the process can effectively identify deviations, leading to a more accurate validation process.

[0081] Such a configuration makes it easier to identify performance gaps between the validated control algorithm and a reference algorithm by evaluating differences in action. This ease of identification is achieved by comparing the predetermined target with the reference output, which highlights the gaps in the algorithm's effectiveness. By highlighting these gaps, the method effectively identifies areas for improvement and facilitates the optimization of the control algorithm being validated.

[0082] According to one embodiment, D(g) represents the difference between a reference output and the predetermined target.

[0083] According to one embodiment, the step of evaluating at least one action difference consists of calculating an evaluation score for the action difference based on D(i) and D(g). Such a configuration allows for a more precise calculation of an evaluation score, as said evaluation score depends both on the result through the reference output, taking into account the expectation through the predetermined target, and on the control action, taking into account the reference action. Indeed, if the reference action results in a reference output that is only slightly lower or higher than the predetermined target, a control parameter that is too far from the reference control would not merit a high evaluation score.

[0084] According to one embodiment, the plurality of control actions and the plurality of reference actions are associated with the same plurality of states and at least one glucose measurement value of the same plurality of states is associated with time in at least one continuous time interval.

[0085] According to the present invention, a continuous time interval is continuous with respect to the associated time of at least one glucose measurement value. In other words, all states comprising at least one glucose measurement value associated with a time within the continuous time interval are included in the same plurality of states. A past glucose measurement value, while having an associated time, is not considered a glucose measurement value in this particular case, since it refers to a glucose measurement performed in the past.

[0086] According to one embodiment, the plurality of control actions and the plurality of reference actions are associated with the same plurality of states and at least one value glucose measurement of the same plurality of states being associated with time in a plurality of continuous time intervals linked to at least one particular period of a day.

[0087] According to the present invention, a particular period of the day could be of any type, such as a postprandial period, a meal period, a period of physical activity or a period of sleep, for example.

[0088] Such a configuration allows the method to validate the control algorithm specifically for a period of the day. Consequently, a user could use several control algorithms, each control algorithm being validated for each particular period of the day. Such a configuration allows for a significant improvement in the regulation of a user's blood glucose.

[0089] According to one embodiment, the step of calculating at least one difference in action consists of calculating at least one difference between at least one control action and at least one reference action as follows:

[0090] [Math. 3]

[0091] D(i) = E(k = j...p)(F(s(i,j)) - a(i,j))

[0092] s(i,j) corresponds to the jth state of the ith sequence;

[0093] a(i,j) corresponds to the jth reference action of the ith sequence; the reference action resulting from the processing of the state s(i,j); and

[0094] F(s(i,j)) corresponds to the jth control action of the ith sequence; the action of control resulting from the processing of the state s(i,j).

[0095] According to one embodiment, the validation step of the control algorithm consists of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if at least one of a first worst-case score and a second worst-case score satisfies a worst-case score criterion, the first worst-case score is calculated as follows:

[0096] first worst-case score = mean(i such that D(i) > 0 and o(i, p+h) < objective); and in which

[0097] The second worst-case score is calculated as follows:

[0098] second worst case score = mean(i such that D(i) < 0 and o(i, p+h) > objective).

[0099] i corresponds to a number of states;

[0100] D(i) corresponds to a sum of the differences in action between the indices 0 and p;

[0101] o(i, p+h) corresponds to a reference output associated with a sequence of states and of actions associated with a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

[0102] Such a configuration allows the process to validate only a control algorithm that has an acceptable first worst case score or a second worst case score and therefore further improves the safety of the process.

[0103] According to one embodiment, the validation step of the control algorithm consists of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if the first worst-case score satisfies the worst-case score criterion. Such a configuration allows the process to validate only a control algorithm that has an acceptable first worst-case score, corresponding to a glucose value lower than the target, such as hypoglycemia, for example, and thus further improves the safety of the process.

[0104] According to one embodiment, the validation step of the control algorithm consists of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if both the first worst-case score and the second worst-case score satisfy the worst-case score criterion. Such a configuration allows the process to validate only a control algorithm that has an acceptable first and second worst-case score, and thus further improves the safety of the process.

[0105] According to one embodiment, the first result, the first worst-case score, and the second worst-case score can be of any type, such as a predetermined threshold, for example.

[0106] According to one embodiment, the first worst-case score and the second worst-case score correspond to thresholds ensuring that the control algorithm never generates, during the plurality of states processing step, a control parameter worse than the reference parameter, while said reference parameter causes the reference outputs to fall outside a safety range. The safety range can be of any type, such as a range corresponding to normoglycemia, for example.

[0107] The invention also relates to a closed-loop automated blood glucose monitoring system for controlled insulin administration to a user, comprising:

[0108] a continuous glucose monitoring sensor configured to provide a plurality of glucose measurement values ​​representative of a user-measured glucose level at a plurality of associated measurement times;

[0109] a subcutaneous insulin delivery device configured to deliver exogenous insulin into a user's subcutaneous tissue in response to a control parameter;

[0110] a controller programmed to receive glucose measurement values ​​and provide a control parameter to the subcutaneous insulin delivery device;

[0111] the controller being programmed to determine the control parameter using a validated control algorithm according to the embodiments described above.

[0112] The different incompatible aspects defined above can be combined. Brief description of the drawings

[0113] Embodiments of the invention will be described below with reference to the drawings, briefly described below: - Figure 1 illustrates a method for validating a control algorithm according to one embodiment of the invention; and - [Fig.2] shows an automated closed-loop blood glucose control system according to one embodiment of the invention. DETAILED DESCRIPTION OF THE INVENTION

[0114] Figure 1 represents a method for validating a control algorithm, the method being implemented by a computer and comprising a step of receiving a plurality of states 50, each state comprising at least one glucose measurement value, the glucose measurement value being representative of a user's measured glucose level associated with an associated time. According to the present invention, a glucose measurement value is a blood glucose measurement value or an indicative value of a blood glucose measurement value, such as an interstitial glucose measurement value. According to the present invention, a computer is an electronic device that can receive, store, process, and produce data according to predefined instructions or algorithms, such as a smartphone, a server, or a desktop computer, for example.The plurality of glucose measurement values ​​is received from a continuous glucose monitoring (CGM) sensor, for example.

[0115] A state represents the user's state at an associated time. Each state in the plurality of states also includes at least one past glucose measurement value representative of a glucose level measured by a user at an associated time. Such a configuration allows the control algorithm and the reference algorithm to generate more precise actions by estimating a trend, for example, and thus improve the accuracy of the validation. Each state in the plurality of states also includes at least one past insulin administration by a user at an associated time. Such a configuration allows the control algorithm and the reference algorithm to generate more precise actions by estimating the lasting effect of past insulin administrations, for example, and thus improve the accuracy of the validation.Each state in the plurality of states also includes at least one previous carbohydrate intake representative of a user's measured glucose level associated with a specific time. Such a configuration allows testing the control algorithm and the reference algorithm under different conditions, such as meal management or rest management, and thus improves the accuracy of the validation. Therefore, a state is also a reference output associated with a previous state. In other words, like the algorithm of... Control is validated by its ability to send precise control parameters that maintain a user's blood glucose close to the target. A state containing at least one glucose measurement value is a reference output of a previous state. Indeed, the glucose measurement value associated with time t+1 is the result / reference output of a reference action undertaken at time t based on a state to which at least one glucose measurement value at time t is associated.

[0116] The method includes a step of receiving a plurality of reference actions 52 from a reference algorithm, each reference action being associated with a state such that a particular action is associated with the particular state representative of the consequences of said action, and corresponding to a control parameter determined by the reference algorithm on the basis of said associated state. A control parameter is a control parameter configured to control a subcutaneous insulin delivery device configured to deliver exogenous insulin into a user's subcutaneous tissue in response to a control parameter, in particular continuous infusion insulin and / or bolus insulin.

[0117] The method includes a step of receiving a plurality of reference outputs 54, each reference output corresponding to at least one glucose measurement value and being associated with at least one state.

[0118] The method includes a step of processing the plurality of states 56, each state being processed using the control algorithm to generate a control action associated with the processed state and corresponding to a control parameter determined by the control algorithm based on the associated state. The reference actions and the control actions are positive numbers and are proportional to a quantity of insulin to be injected into a user, for example. The reference actions and the control actions can be of any type, such as rescue carbohydrates, glucagon, or insulin, provided that both the reference actions and the control actions are of the same type. Such a configuration makes it easier to compare the reference actions and the control actions and thus to validate the control algorithm more accurately.Furthermore, since rescue carbohydrates, glucagon, or insulin have a monotonic impact on blood glucose, this allows for more precise validation of a control algorithm. For example, a monotonic impact occurs when more insulin is injected into a user; the user's blood glucose decreases unless carbohydrates are ingested. According to the present invention, rescue carbohydrates are a carbohydrate intake recommendation or a carbohydrate-related control parameter.

[0119] The method includes a step of calculating at least one action difference 58, the calculation of at least one action difference 58 consisting of calculating at least one difference between at least one control action and at least one reference action. Furthermore, the step of calculating at least one action difference 58 involves calculating at least one difference between a sum of several control actions and a sum of several reference actions. Such a configuration allows for the consideration of more actions and therefore enables a more precise evaluation of the at least one action difference in the at least one action difference evaluation step.

[0120] More specifically, the step of calculating at least one action difference 58 consists of calculating an action difference as:

[0121] [Math. 4]

[0122] D(i) = (Icont - Iref) / Irap

[0123] D(i) corresponds to a difference in action of a state i;

[0124] Icont corresponds to a sum of a plurality of control actions;

[0125] Iref corresponds to a sum of a plurality of reference shares;

[0126] Irap corresponds to a normalization constant. Irap is used to convert the absolute amounts of insulin are converted into "relative amounts" compared to the user's baseline insulin requirements, so that each user contributes as much as any other user to the fine-tuning.

[0127] Such a configuration allows for the consideration of certain specific situations, such as D(i) being greater than zero and o(i, p+h) being greater than the predetermined target, but with a control parameter strong enough to modify o(i, p+h) so that it is less than the predetermined target if the control parameter is received by the subcutaneous insulin delivery device instead of the reference parameter, for example. Therefore, such a configuration allows for a more precise calculation of the difference in action, and thus for a more precise calculation of an evaluation score and for validating only an effective control algorithm.

[0128] Iref = Sj=ip : ia(i,j), with s(i,j) corresponding to the jth state of the ith sequence and a(i,j) corresponding to the jth reference action of the ith sequence; the reference action resulting from the processing of state s(i,j). Such a configuration allows for taking into account more actions and therefore allows for a more precise evaluation of at least one difference of action at the at least one difference of action evaluation step.

[0129] Icont = Sj=ip : i F(s(i,j)), with s(i,j) corresponding to the jth state of the ith sequence and F(s(i,j)) corresponding to the jth control action of the ith sequence; the control action resulting from the processing of state s(i,j). Such a configuration allows for more actions to be taken into account and therefore allows for a more precise evaluation of at least one difference of action at the at least one difference of action evaluation step. Iref = Sj=ip : ia(i,j), with s(i,j) corresponding to the jth state of the ith sequence and a(i,j) corresponding to the jth reference action of the ith sequence; the action of reference resulting from the processing of the state s(i,j). Such a configuration allows for more actions to be taken into account and therefore allows for a more precise evaluation of at least one difference of action at the step of evaluating at least one difference of action.

[0130] Each state includes a basal reference rate and a total daily dose (TDD), and the Irap varies according to the value of the reference rate. Such a configuration allows for the calculation of a differential action relative to a basal reference rate and thus for the calculation of a personalized and meaningful differential action for the associated user. This configuration provides a more user-independent, standardized metric. The basal reference rate is an insulin delivery rate, in units per hour or as a fraction of the TDD, for example, associated with a user. If the basal reference rate is greater than one-quarter of the TDD and Iref is greater than zero, Irap = Iref. If the basal reference rate is greater than one-quarter of the TDD and Iref is zero, Irap = Icont. If the basal reference rate is less than or equal to one-quarter of the TDD, Irap = (½ TDD) / 24.According to the present invention, the TDD is a total daily insulin dose associated with a user and theoretically required by the associated user to maintain normoglycemic glucose levels. Such a configuration also allows for addressing values ​​close to zero in the denominator.

[0131] According to another embodiment, the step of calculating at least one difference in action 58 consists of calculating at least one difference between at least one control action and at least one reference action as follows:

[0132] [Math. 5]

[0133] D(i) = E(k = j...p)(F(s(i,j)) - a(i,j))

[0134] s(i,j) corresponds to the jth state of the ith sequence;

[0135] a(i,j) corresponds to the jth reference action of the ith sequence; the reference action resulting from the processing of the state s(i,j); and

[0136] F(s(i,j)) corresponds to the jth control action of the ith sequence; the action of control resulting from the processing of the state s(i,j).

[0137] The method includes a step of evaluating at least one action difference 60, the evaluation of at least one action difference 60 consisting of calculating at least one evaluation score of at least one action difference based on at least one reference output and a predetermined target. According to the present invention, the predetermined target corresponds to at least one blood glucose value of an associated user. The step of evaluating at least one action difference 60 consists of calculating an evaluation score of the action difference based on a difference between a reference output and the predetermined target. Such a configuration makes it possible to improve the accuracy of the validation of the control algorithm by comparing the action differences with a reference output and a predefined target. This Improvement is achieved using an evaluation score based on the difference between the reference output and the predetermined target, providing a more precise measure of the control algorithm's effectiveness. Comparison with a reference output ensures that the process can effectively identify deviations, leading to a more accurate validation process. This setup also facilitates the identification of effectiveness gaps between the validated control algorithm and a reference algorithm through the evaluation of differences in action. This ease of identification is achieved by comparing the predetermined target with the reference output, highlighting deviations in the algorithm's effectiveness. By highlighting these deviations, the process effectively identifies areas for improvement and facilitates the optimization of the control algorithm during validation.

[0138] The step of evaluating at least one action difference 60 consists of calculating an evaluation score for the action difference based on D(i) and D(g), where D(g) represents the difference between a reference output and the predetermined target. Such a configuration allows for a more precise calculation of an evaluation score, as said evaluation score depends both on the result through the reference output, taking into account the expectation through the predetermined target, and on the control action, taking into account the reference action. Indeed, if the reference action results in a reference output that is only slightly lower or higher than the predetermined target, a control parameter that is too far from the reference control would not merit a high evaluation score.

[0139] The method includes a validation step for the control algorithm 62, the control algorithm being validated if at least the evaluation score satisfies an evaluation score criterion. The evaluation score criterion corresponds to a threshold ensuring that the control algorithm is at least as effective as the reference algorithm. Such a configuration makes it possible to accurately validate a control algorithm without resorting to a simulator, which is susceptible to bias, or a time-consuming clinical trial. Indeed, such a method requires only real data to obtain the states, the reference control, and the reference output to quickly validate a bias-free control algorithm.Indeed, virtual patient simulators are difficult to design, often include modeling biases – which limits variability compared to real life and potentially leads to unrealistic simulations – and are associated with potentially significant computational costs (such as solving ordinary differential equations by numerical integration).

[0140] The evaluation score is calculated using a plurality of action differences. Such a configuration makes it possible to take into account a plurality of situations and therefore to more precisely assess how the control algorithm would have reacted in particular circumstances represented by the states compared to the reference algorithm.

[0141] According to a certain embodiment, the evaluation grade is calculated as follows:

[0142] [Math. 6]

[0143] evaluation grade = (A + B) / (A + B + C + D);

[0144] in which:

[0145] A corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p +h) > predetermined objective;

[0146] B corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p +h) < predetermined objective;

[0147] C corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p +h) < predetermined objective;

[0148] D corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p +h) > predetermined objective;

[0149] i corresponds to a sequence of states and associated actions;

[0150] N corresponds to a sequence of states and associated actions;

[0151] D(i) corresponds to the sum of the differences in action between the indices 0 and p;

[0152] p corresponds to a first constant;

[0153] h corresponds to a second constant;

[0154] o(i, p+h) corresponds to a reference output associated with a sequence of states and actions associated with i to a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

[0155] Such a configuration makes it possible to create an objective and unbiased evaluation score, because the cases falling into category A are cases for which the final glucose measurement value was higher than the predetermined target and the control algorithm would have generated a corresponding control action of a larger quantity of insulin to be injected: we can assume that the control algorithm handled this particular situation better, and in the same way, mutatis mutandis, for other cases B, C and D.

[0156] p is a first constant corresponding to a number of states, reference actions associated with said states, and reference outputs also associated with said states, sufficiently small such that changes in insulin administration during the number p of states, reference actions associated with said states, and reference outputs also associated with said states have not yet had a significant impact on the reference output o(i, p). Such a configuration allows for the accurate calculation of an evaluation score, since insulin has not yet had an impact on the reference output, and said reference output therefore cannot be considered as distinct from a control output. A control output corresponds to at least one glucose measurement value if the associated control action was sent to a subcutaneous insulin delivery device. In other words, having a relatively small p-value allows the user's state to remain largely unchanged under control actions compared to reference actions.Consequently, the sequences of control actions generated by the control algorithm between 0 and p are very close to the actions that would have been taken in real life on this user. h is a second constant corresponding to a number of states, reference actions associated with these states, and reference outputs also associated with said states, large enough that the modifications to insulin delivery during the number p of states, reference actions associated with said states, and reference outputs also associated with said states, have fully impacted the reference output o(i, p+h). Such a configuration allows for the accurate calculation of an evaluation score, because the effect of the control measures taken during the number p of states is fully active at a state in which the time associated with the blood glucose measurement value associated with the state is equal to p+h.

[0157] According to another embodiment, A corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p+h) > predetermined target + high margin;

[0158] B corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p +h) < predetermined objective - low margin;

[0159] C corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p +h) < predetermined objective + high margin;

[0160] D corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p +h) > predetermined target - low margin;

[0161] Such a configuration makes it possible to artificially reduce the evaluation score and therefore to validate only an effective control algorithm.

[0162] According to any one of certain embodiments, the upper margin and the lower margin are variables and vary according to o(i, p+h), the reference output associated with a sequence of states and actions associated i with a state whose associated time of the blood glucose measurement value associated with the state is equal to p+h. Such a configuration allows for more reliable validation of the control algorithm, as it improves process safety by reducing the variance of the evaluation score; it therefore allows for a more precise calculation of an evaluation score and for the validation of only an effective control algorithm.

[0163] According to any one of certain embodiments, p is equal to 0. Such a configuration makes it possible to calculate an evaluation score based solely on the difference in action between a control action and a reference action. Consequently, the evaluation score will accurately determine the algorithm's capability. of control to be more effective than the reference algorithm in various cases, such as meal bolus estimation, for example. According to the present invention, meal bolus estimation means the generation of a control action in response to a carbohydrate intake, such as a meal, for example.

[0164] The plurality of control actions and the plurality of reference actions are associated with the same plurality of states, and at least one glucose measurement value from the same plurality of states is associated with time in at least one continuous time interval. According to the present invention, a continuous time interval is continuous with respect to the time associated with at least one glucose measurement value. In other words, all states comprising at least one glucose measurement value associated with a time in the continuous time interval are included in the same plurality of states. A past glucose measurement value, while having an associated time, is not considered a glucose measurement value in this particular case, since it refers to a glucose measurement performed in the past.

[0165] The plurality of control actions and the plurality of reference actions are associated with the same plurality of states, and at least one glucose measurement value from the same plurality of states is associated with time in a plurality of continuous time intervals linked to at least one particular period of the day. According to the present invention, a particular period of the day could be of any type, such as a postprandial period, a mealtime period, a period of physical activity, or a sleep period, for example. Such a configuration allows the method to validate the control algorithm specifically for a period of the day. Therefore, a user could use several control algorithms, each control algorithm being validated for each particular period of the day. Such a configuration allows for a significant improvement in the regulation of a user's blood glucose.

[0166] According to a certain embodiment, the validation step of the control algorithm 62 consists of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if at least one of a first worst-case score and a second worst-case score satisfies a worst-case score criterion, the first worst-case score being calculated as follows:

[0167] first worst-case score = mean(i such that D(i) > 0 and o(i, p+h) < objective); and the second worst-case score being calculated as follows:

[0168] second worst case score = mean(i such that D(i) < 0 and o(i, p+h) > objective).

[0169] i corresponds to a number of states;

[0170] D(i) corresponds to a sum of the differences in action between the indices 0 and p;

[0171] o(i, p+h) corresponds to a reference output associated with a sequence of states and actions associated with i to a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

[0172] Such a configuration allows the process to validate only a control algorithm that has an acceptable first worst-case score or a second worst-case score and therefore further improves the safety of the process.

[0173] The validation step of control algorithm 62 can also consist of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if the first worst-case score satisfies the worst-case score criterion. Such a configuration allows the process to validate only a control algorithm that has an acceptable first worst-case score corresponding to a glucose value lower than the target, such as hypoglycemia, for example, which further improves process safety.

[0174] The validation step of control algorithm 62 can also consist of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if both the first worst-case score and the second worst-case score satisfy the worst-case scoring criteria. Such a configuration allows the process to validate only a control algorithm whose first and second worst-case scores are acceptable, thus further improving the safety of the process. The first and second worst-case scores can be of any type, such as a predetermined threshold.More specifically, the first worst-case score and the second worst-case score correspond to thresholds ensuring that the control algorithm never generates, during the plurality of states processing step 56, a control parameter worse than the reference parameter, while said reference parameter causes the reference outputs to fall outside a safety range. The safety range can be of any type, such as a range corresponding to normoglycemia, for example.

[0175] According to one embodiment, the validation step of the control algorithm 62 consists of validating the control algorithm if the evaluation score satisfies an evaluation score criterion and if a case score satisfies a case criterion. The case score represents an evaluation of the number of good, bad, and worst control actions, considering states where the control action recommended rescue carbohydrates and the reference action did not. A good case corresponds to a reference output showing hypoglycemia. A bad case corresponds to a reference output showing normoglycemia. The worst case corresponds to a reference output showing hyperglycemia. Such a configuration allows the control algorithm to be evaluated based on specific cases and thus increases the efficiency of the process. The same reasoning can be applied to boluses. rather than emergency carbohydrates. The same reasoning can be applied by taking into account an insulin threshold, such as a basal threshold. The cases considered are those where the reference actions correspond to a lower threshold than the control actions.

[0176] If the control algorithm is validated, the method also includes a step of sending at least one control parameter to a subcutaneous insulin delivery device configured to deliver exogenous insulin into the subcutaneous tissue of a user in response to a control parameter, in particular continuous infusion insulin and / or bolus insulin.

[0177] According to an embodiment in which the control algorithm includes a design parameter influencing the control action generated in the processing step of a plurality of states, the method for validating a control algorithm also includes a step of optimizing the design parameter using a numerical optimization algorithm on the function:

[0178] [Math. 7]

[0179] F(W) = (A + B) / (A + B + C +D)

[0180] W corresponds to the design parameter.

[0181] Such an embodiment makes it possible to optimize and therefore improve the control algorithm.

[0182] The numerical optimization algorithm can be of any type, such as: convex, differentiable, constrained, linear search, gradient descent, Hessian matrix or stochastic optimization, for example.

[0183] The invention also relates to an automated closed-loop blood glucose monitoring system 10 for the controlled administration of insulin to a user comprising: - a continuous glucose monitoring sensor 12 configured to provide a plurality of glucose measurement values ​​representative of a user-measured glucose level at a plurality of associated measurement times; - a subcutaneous insulin delivery device 20 configured to deliver exogenous insulin into a user's subcutaneous tissue in response to a control parameter; - a controller 30 programmed to receive glucose measurement values ​​and provide a control parameter to the subcutaneous insulin delivery device; - the controller being programmed to determine the control parameter using a validated control algorithm in accordance with what has been described above.

[0184] It is evident that the various embodiments and technical effects described above can be applied to the blood glucose control system.

[0185] Although examples of embodiments of the invention have been described, it will be understood by those skilled in the art that various changes, omissions, and / or additions can be made, and that equivalents can be substituted for elements thereof without departing from the spirit of the scope of the invention. Furthermore, numerous modifications can be made to adapt a particular situation or material to the teachings of the invention without departing from its scope. Therefore, it is intended that the invention is not limited to the particular embodiments disclosed for the implementation of this invention, but that the invention includes all embodiments falling within the scope of the supplementary claims. Moreover, unless specifically stated, any use of the terms first, second, etc.does not designate any order or importance, but instead the terms first, second, etc. are used to distinguish one element from another.

Claims

1. Demands A method for validating a control algorithm, the method being implemented by a computer and comprising the steps of: - receive a plurality of states (50), each state comprising at least one glucose measurement value, the glucose measurement value being representative of a user's measured glucose level associated with an associated time; - receive a plurality of reference actions (52) from a reference algorithm, each reference action being associated with a state and corresponding to a control parameter determined by the reference algorithm on the basis of said associated state; - receive several reference outputs (54), each reference output corresponding to at least one glucose measurement value and being associated with at least one state; - to process the plurality of states (56), each state being processed using the control algorithm to generate a control action associated with said processed state and corresponding to a control parameter determined by the control algorithm on the basis of said associated state; - calculate at least one difference of action (58), the calculation of at least one difference of action consisting of calculating at least one difference between at least one control action and at least one reference action; - evaluate at least one difference in action (60), the evaluation of at least one difference in action consisting of calculating at least one evaluation score of at least one difference in action on the basis of at least one reference output and a predetermined target, the predetermined target corresponding to at least one blood glucose value of the user; - validate the control algorithm (62), the control algorithm being validated if at least the evaluation score satisfies one evaluation score criterion, and the step of calculating at least one action difference (58) consists of calculating at least one difference between a sum of a plurality of control actions and a sum of a plurality of reference actions.

2. A method according to claim 1, wherein the evaluation score is calculated using a plurality of action differences.

3. A method according to claim 2, wherein the evaluation score is calculated as follows: evaluation score = (A + B) / (A + B + C + D); wherein: A corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p+h) > predetermined target; B corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p+h) < predetermined target; C corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p+h) < predetermined target; D corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p+h) < predetermined target;, N] such that D(i) < 0 and o(i, p+h) > predetermined target; i corresponds to a sequence of states and associated actions; N corresponds to a sequence of states and associated actions; D(i) corresponds to the sum of the differences in action between the indices 0 and p; p corresponds to a first constant; h corresponds to a second constant; o(i, p+h) corresponds to a reference output associated with a sequence of states and associated actions i to a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

4. Method according to claim 3, wherein: A corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p+h) > predetermined target + high margin; B corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p+h) < predetermined target - low margin; C corresponds to the proportions of i in [0,..., N] such that D(i) > 0 and o(i, p+h) < predetermined target + high margin; D corresponds to the proportions of i in [0,..., N] such that D(i) < 0 and o(i, p+h) > predetermined target - low margin.

5. Method according to claim 4, wherein the upper margin and the lower margin are variables.

6. A method according to any one of claims 3 to 5, wherein p is equal to n

7. V. A method according to any one of claims 1 to 6, wherein the step of calculating at least one action difference (58) consists of calculating an action difference as: D(i) = (Icont " Iref) / Irap D(i) corresponds to an action difference of a state i; Icont corresponds to a sum of a plurality of control actions; Iref corresponds to a sum of a plurality of reference actions; Irap corresponds to a normalization constant.

8. A method according to claim 7, wherein the step of evaluating at least one difference of action (60) consists of calculating an evaluation score of the difference of action based on a difference between a reference output and the predetermined target.

9. A method according to any one of claims 1 to 8, wherein the plurality of control actions and the plurality of reference actions are associated with the same plurality of states and wherein at least one glucose measurement value of the same plurality of states is associated with time in at least one continuous time interval.

10. A method according to claim 9, wherein the plurality of control actions and the plurality of reference actions are associated with the same plurality of states and wherein at least one glucose measurement value of the same plurality of states is associated with time in a plurality of continuous time intervals linked to at least one particular period of a day.

11. A method according to any one of claims 1 to 10, wherein the step of calculating at least one difference of action (58) consists of calculating at least one difference between at least one control action and at least one reference action as follows: D(i) = E(k = j...p)(F(s(i,j)) - a(i,j)) s(i,j) corresponds to the jth state of the ith sequence; a(i,j) corresponds to the jth reference action of the ith sequence; the reference action resulting from the processing of the state s(i,j); and F(s(i,j)) corresponds to the jth control action of the ith sequence; the control action resulting from the processing of the state s(i,j).

12. A method according to any one of claims 1 to 11, wherein the validation step of the control algorithm (62) consists of validating the control algorithm if the rating satisfies a rating criterion and if at least one of a first worst-case score and a second worst-case score satisfies a worst-case score criterion, the first worst-case score being calculated as follows: first worst-case score = mean(i such that D(i) > 0 and o(i, p+h) < target); and the second worst case score is calculated as follows: second worst case score = mean(i such that D(i) < 0 and o(i, p+h) > target). i corresponds to a number of states; D(i) corresponds to a sum of the differences in action between the indices 0 and p; o(i, p+h) corresponds to a reference output associated with a sequence of states and actions associated with i to a state whose associated time of the measurement value of the blood glucose associated with the state is equal to p+h.

13. An automated closed-loop blood glucose monitoring system (10) for controlled insulin delivery to a user, comprising: - a continuous glucose monitoring sensor (12) configured to provide a plurality of glucose measurement values ​​representative of a user-measured glucose level at a plurality of associated measurement times; - a subcutaneous insulin delivery device (20) configured to deliver exogenous insulin into a user's subcutaneous tissue in response to a control parameter; - a controller (30) programmed to receive glucose measurement values ​​and provide a control parameter to the subcutaneous insulin delivery device; in which the controller is programmed to determine the control parameter using a validated control algorithm according to any one of claims 1 to 12.

Images