AUTOMATED SYSTEM FOR CONTROLLING THE BLOOD GLUCOSE LEVEL OF A PATIENT

The present application claims the benefit of French Patent fr16/58881 which will be considered as part of the present disclosure.

The present application relates to the field of automated systems for regulating blood sugar, also called artificial pancreas.

An artificial pancreas is a system for automatically regulating the supply of insulin to a diabetic patient from its history of blood sugar, its history of food intake, and its history insulin injection.

Another example of special regulating systems MPCs (English "model where a microchip for predictive control is"), also called feedforward systems, wherein the regulation of the insulin dose administered accounts for a prediction of the future evolution of the patient's blood glucose level, made from a physiological model describing the uptake of insulin by the patient's body and its impact on the patient's blood glucose level.

It would be desirable to be able to improve the performance of the artificial pancreas to feedforward control, and, more particularly, to the improved quality of the prediction of the future blood glucose levels of the patient, so as to be able to monitor with higher relevancy feeds insulin and the risks of placing the patient in a situation of hyperglycemia or hypoglycemia.

It would be further desirable to be able to limit the risks for the patient related to a possible fault of the physiological model used to predict future blood glucose levels of the patient.

Thus, one embodiment provides an automated system for glycemic control of a patient, comprising:

a blood glucose sensor;

an insulin injection device; and

a processing unit adapted to predict and control the future evolution of the patient's blood glucose level from a physiological model and to control the insulin injection taking account of this prediction,

wherein:

the physiological model includes a system of differential equations describing the evolution of a plurality of state variables as a function of time; and

the processing and control unit is adapted to perform a step of automatic calibration of the physiological model comprising a step of estimating initial values of state variables by minimizing a quantity representing the error, during an observation period passed, between blood glucose estimated from the physiological model and the blood sugar level measured by the sensor.

In one embodiment, the magnitude is representative of the area between a first curve g at representative time evolution of glycemia estimated from the model on the observation period, and a second curve g as representative of the time evolution of the blood sugar level measured by the sensor on the observation period.

In one embodiment, the magnitude is defined as follows:

T-₀ + ATS

^{the m=} ~And Σ 9 - (T-) VBE1²

t=t₀

where T is a variable time discretized, Tg is the start time of the observation phase, and tg + 2\t is the end time of the observing step.

In one embodiment, the calibration method further includes a step of estimating differential equation system by minimizing said magnitude.

In one embodiment, the calibration method comprises a plurality of successive iterations of steps a) and b) the following:

a) estimate the parameters of the system of differential equations by minimizing said quantity by setting the initial values of the state variables; and

b) estimating initial values of state variables by minimizing said magnitude by fixing the parameters of the system of differential equations.

In one embodiment, at the first iteration of step a), the initial values of the state variables are determined analytically assuming that all derived from the differential equation system are zero.

In one embodiment, to simulate the evolution of the patient's blood glucose level from the physiological model, the processing and control unit takes into account the history of insulin injected to the patient by the injection device and history of glucose ingested by the patient.

In one embodiment, the physiological model is the model of ideas about.

Another embodiment provides a method for automated regulation of blood glucose levels in a patient, comprising:

a step of calculating, using a processing and control unit, a prediction of the future evolution of the patient's blood glucose level from a physiological model comprising a system of differential equations describing the evolution of a plurality of state variables as a function of time;

a step of operating a device insulin injection taking account of this prediction; and

a step of automatic calibration of the physiological model comprising a step of estimating initial values of state variables by minimizing a quantity representing the error, during an observation period passed, between blood glucose estimated from the physiological model and the measured blood glucose on the patient by a blood glucose sensor.

In one embodiment, the method further includes a step of estimating differential equation system by minimizing said magnitude.

In one embodiment, the calibration step comprises a plurality of successive iterations of steps a) and b) the following:

a) estimate the parameters of the system of differential equations by minimizing said quantity by setting the initial values of the state variables; and

b) estimating initial values of state variables by minimizing said quantity by setting the parameters of the system of differential equations.

Another embodiment provides an automated system for glycemic control of a patient, comprising:

a blood glucose sensor;

an insulin injection device; and

a processing unit adapted to predict and control the future evolution of the patient's blood glucose level from a physiological model and to control the insulin injection taking account of this prediction,

wherein the processing and control unit is adapted to:

a) a step of performing automatic calibration of the physiological model taking into account a history of blood glucose level measured by the sensor during a period of records;

b) at the end of the calibration step, determining whether the model is satisfactory or not from the at least one numeric indicator representative of the error between the estimated blood glucose from the model and the actual blood glucose level measured by the sensor;

and

c) if the quality of the model is not satisfactory, control the insulin injection regardless of the prediction made from the model.

In one embodiment, the digital indicator comprises an indicator representative of m and the area between a first curve g at representative time evolution of glycemia estimated from the model on the observation period, and a second curve g as representative of the time evolution of the blood sugar level measured by the sensor on the observation period.

In one embodiment, the indicator m is defined as follows:

T-₀ + ATS

^{the m=} ~And Σ 9 - (T-) VBE1²

t=t₀

where T is a variable time discretized, Tg is the start time of the observation phase, and tg + 2\t is the end time of the observing step.

In one embodiment, the digital indicator comprises an indicator IRQs representative of the difference between the estimated blood glucose from the model and the blood glucose levels measured by the sensor at a given time.

In one embodiment, the digital indicator comprises an indicator m2 representative of the difference between the derivative of the blood glucose estimated from the model and the derivative of the blood sugar level measured by the sensor at a given time.

In one embodiment, in step c), the control of the insulin injection is a feedforward control based on a physiological model simplified.

In one embodiment, in step c), the insulin injection is controlled to deliver insulin doses preprogrammed basal rate corresponding to a reference given to the patient.

In one embodiment, the physiological model includes a system of differential equations describing the evolution of a plurality of state variables as a function of time, and step a) automatic calibration of the model comprises a step of estimating differential equation system by minimizing a quantity representing the error, during an observation period passed, between blood glucose estimated from the physiological model and the blood sugar level measured by the sensor.

In one embodiment, step a) automatic calibration of the model further comprises a step of determining initial values of the state variables.

Another embodiment provides a method for automated regulation of blood glucose levels in a patient, comprising:

a step of calculating, using a processing and control unit, a prediction of the future evolution of the patient's blood glucose level from a physiological model; and

a step of operating a device insulin injection taking account of this prediction, the method further comprising:

a) a step of automatic calibration of the physiological model taking into account a history of measured blood glucose by a blood glucose sensor during a period of records;

b) at the end of the calibration step, a step of determining the quality of the physiological model from at least one digital indicator representative of the error between the estimated blood glucose from the model and the actual blood glucose level measured by the sensor; and

c) if the quality of the model is deemed unsatisfactory, a step device control insulin injection regardless of the prediction made from the model.

These features and advantages, as well as other, will be exposed in detail in the following description of particular embodiments made not exclusively in connection with the accompanying drawings of which:

figure 1 schematically represents, in block form, an example of one embodiment of an automated system for glycemic control of a patient;

figure 2 is a simplified representation of a physiological model used in the system of Figure 1 for predicting the future evolution of the patient's blood glucose level;

figure 3 is a diagram showing in more detail an embodiment of a physiological model of Figure 2;

figure 4 is a chart illustrating an example of an automated method for regulating blood glucose implemented by the system of Figure 1;

figure 5 is a diagram illustrating one example of an embodiment of an automated method calibration performed by the system of Figure 1; and

figure 6 is a chart illustrating an example of an embodiment of an automated method for regulating blood glucose implemented by the system of Figure 1.

The same elements have been designated with same references to the different drawings. For clarity, only the elements which are useful to the understanding of embodiments disclosed herein have been shown and are detailed. In particular, the blood sugar measuring device and the device insulin injection control system described have not been detailed, the embodiments described herein is compatible with all or most of the blood sugar measuring devices and known insulin injection. In addition, the hardware implementation of the processing and control unit of the control system described has not been detailed, the implementation of the control and processing unit being within the reach of the skilled person from operative indicia described.

Figure 1 schematically represents, in block form, an example of one embodiment of an automated system for glycemic control of a patient.

The system of Figure 1 includes a sensor 101 (GC) adapted to measure the patient's blood glucose level. In normal operation, the sensor 101 may be positioned permanently on or in the patient's body, for example at the height of his abdomen. The sensor 101 is for example a sensor type CMM (English "continu-glucose networked" - continuous monitoring blood glucose), c'est to say a sensor arranged to measure continuously (for example at least once per minute) the patient's blood glucose level. The sensor 101 is for example a blood glucose sensor in subcutaneously.

The system of Figure 1 further comprises an insulin injection (PMPs) 103, for example an injection device in subcutaneously. The device 103 is for example an automatic injection device type insulin pump, comprising an insulin tank connected to an injection needle implanted under the skin of the patient, the pump electrically controllable to automatically inject insulin doses determined at specific points in time. In normal operation, the injection device 103 may be positioned fixed in or on the patient's body, for example at his abdomen.

The system of Figure 1 further comprises a processing and control unit 105 (CTR) linked to the blood glucose sensor 101, such as by wired connection or by wireless connection (wireless), and secondly to the injection device 103, such as by wired or wireless. In operation, the processing unit 105 and control is adapted to receive the patient's blood glucose level measured by the sensor 101, and to electrically control the device 103 for injecting the patient insulin doses determined at specific points in time. In this example, the processing and control unit 105 is further adapted to receive, via a user interface not detailed, data of CHO (T-) representative of the evolution, as a function of time, of the amount of glucose ingested by the patient.

The processing and control unit 105 is adapted to determine the doses of insulin to be injected to the patient taking into account especially the history of blood glucose level measured by the sensor 101, the history of insulin injected by the device 103, and history of ingestion of glucose by the patient. For this purpose, the processing and control unit 105 includes a digital computer circuit (not detailed), comprising for example a microprocessor. The processing and control unit 105 is for example a mobile device carried by the patient throughout the course of a day and/or night, e.g. a smartphone device configured to implement a control method of the type described below.

In the embodiment of Figure 1, the processing and control unit 105 is adapted to determine the amount of insulin to be administered to the patient with consideration of a prediction of the future change its blood glucose as a function of time. More particularly, the processing and control unit 105 is adapted, from the history of insulin injected and history of glucose ingested, and based on a physiological model describing the uptake of insulin by the patient's body and its impact on glycemia, determining a curve representative of an expected course of the patient's blood glucose level as a function of time, over a future time period, for example a period of 1 to 10 hours. Taking account of this curve, the processing and control unit 105 determines the doses of insulin to be injected to the patient during the upcoming time period, blood glucose for actual (as opposed to blood glucose estimated from the physiological model) of the patient remains within acceptable limits, and in particular to limit the risks of hyperglycemia or hypoglycemia. In this mode of operation, as will be explained in more detail below, the actual blood glucose data measured by the sensor 101 are used mainly for calibration purposes the physiological model.

Figure 2 is a simplified representation of a physiological model MPCs used in the system of Figure 1 for predicting the future evolution of the patient's blood glucose level. In Figure 2, the model is represented in the form of a processing block comprising:

an input EL applied thereto an I signal (a T) representing the development, as a function of time T, the amount of insulin injected to the patient;

an input ℮ 2 on which is applied a signal of CHO (T-) representing the development, as a function of the time T, of the amount of glucose ingested by the patient; and

an output s providing a G signal (a T) representing the development, as a function of the time T, of the patient's blood glucose level.

The physiological model MPC is a compartmentalized model comprising, in addition to the variables input I (T-) and CHO (a T) and the output variable grams (a T), a plurality of state variables corresponding to patient physiological variables, evolving from the time. The time evolution of state variables is governed by a system of differential equations having a plurality of parameters represented in Figure 2 by a vector supplied to an input block of the MPCs in PL. The response of the physiological model is further conditioned by the initial conditions or initial values assigned to the state variables, represented in Figure 2 by a vector supplied to an input block p2 MPCs.

Figure 3 is a diagram showing in more detail an example (non-limiting) of the MPCs physiological model used in the system of Figure 1 for predicting the future evolution of the patient's blood glucose level. This example model, known as model ideas about, is described in more detail in the article entitled "Nonlinear model the feedforward control is OC of glucose concentrations in-subjects Thus spouses of 1 diabetes therapy" of children's ideas about and Al (J. Comp Meas. 2004;25: 905 - 920), and in the article entitled "partitioning on the glucose delivery/transportation, Entente, NDA endogenous muscle during IVGTT", of children's ideas about and Al (Obstet J. Comp EndocrinolMetab 282: Ε 992 - Ε 1007, 2002).

The physiological model of Figure 3 includes a first subsystem model two-compartmental 301 describing the effect of food intake of glucose on the rate of appearance of glucose in the blood plasma. The submodel 301 takes as input the amount of glucose ingested a CHO (T-), e.g. mmole/min., and provided at its output a rate Uq glucose absorption in blood plasma, e.g. mmole/minutes. The submodel 301 includes two state variables'd]_ and d2 respectively corresponding to masses of glucose, for example mmole, in first and second compartments.

The model of Figure 3 further comprises a second model element two-compartmental 303 describing absorption, in blood plasma, of insulin delivered to the patient. The submodel 303 takes as input the amount of insulin the I (T-) injected to the patient, for example in the mu/min., and provided at its output a rate UJ absorption of insulin in the blood plasma, for example in the mu/minutes. The submodel 303 includes two variables of state s] and s2 respectively corresponding to masses of insulin, e.g. mmole, in first and second compartments.

The model of Figure 3 further comprises third submodel 305 describing glucose regulation by the patient's body. The submodel 305 takes input absorption levels Uq glucose and uj insulin, and provided at its output the blood sugar grams (a T), c'est i.e. the concentration of glucose in the blood plasma, e.g. mmole/1. The submodel 305 includes six state variables Q-] _, q2, χ 3, the X] _, the X2< The variables q1 q2 and respectively correspond to masses of glucose, for example mmole, in first and second compartments.

The variables X] _, 2 χ, x3 are variables without unit representing each of the actions of insulin on glucose kinetics. The variable I represents the insulinemia, c'est say the insulin concentration in the blood plasma, for example mu/1.

The model of ideas about is governed by the equation system:

GM (=0

As Qi (0

RV

QD1

TD

QD₂

TD

dx and_T-

TD

dx and₂

TD

dx and₃

TD

the DS₁

TD

the DS₂

TD

LID

TD

HD₁

TD

HD₂

TD

A =

01

+ the X_± (a T)

there grams• grams ω

The L *=(0•<? I (0 - [^ 12 + 2 * (0]*<?₂ (0

= - K._{BL in} - XiCt) + the K_ai• (a T) /

= the K~_b2 2 · ^ (Ο ^ 2 + α - Kbyte

= - K._b3• the X₃ (a T) + K._a3 (a T)/-

(A T)=- T- the Ka-s-T-^)

=K._has• E ^ (T-) - K._has - E₂ (a T)

the K_has - E₂ (a T)

As Qi (0 + the K₁₂Q₂ (a T) F._{the R} PGM +_Q. • [1 the X₃(a T)] + U_GÇt)

THE V,

- K._I (a T)/-

the CHO (T-) -

The di (T-)

₌D₁ (a T) P-₂ (a T)

in Tmax or Tmax

the O₂ (0

,_C.F₀₁ • Gm (T-)

⁰¹ 0.85•(Grams (a T) + 1.0)

With:

F.=[^R (^G GIS> 9

^R 0 * - otherwise

In this system of equations, the variables VQ codevectors, Fqi, K.]_ 2, RF, EGPq, KJoule Q, K._has ] _, KJoule-a, 2, *% 2 '*% 3' *% 3 '*%' *%^I ^ T-max are parameters. VQ codevectors corresponds to the volume distribution of glucose, e.g. quart, Fqi corresponds to a transfer rate of glucose non-insulin-dependent, e.g. mmole/min., the K]_ 2 corresponds to a constant transfer ratio between the two compartments of the under model 305, e.g. in min^- - *, K._has ] _, K._has 2, K._has 3 rate constants correspond to disable insulin, e.g. in min^- - *, RF corresponds to a urinary excretion of glucose, for example mmole/min., EGPq corresponds to an endogenous glucose production, for example in min^- - *, kJoule Q, KJoule Q and K ^ grams correspond to rate constants of activation of insulin, for example in min^- - * -, K._has corresponds to a constant rate of absorption of insulin injected into subcutaneous, e.g. in min^- - * -, Vj corresponds to the volume of dispensing of insulin, e.g. quart, K._I corresponds to a removal rate of plasma insulin, e.g. in min^- ** -, and T_max. corresponds to a time passed up to the absorption peak of glucose ingested by the patient, for example in min. These fifteen parameters correspond to the vector of the representation of Figure 2. The vector turn includes ten values corresponding to the initial values (at a time Tg of start of a simulation of the behavior of the patient from the model) assigned to the ten state variables'd] _, d2, e] _, s2, Q-] _, 0.2 '^{the X} the^{the X} 2'^X 1^and ** of the model.

Among the parameters of the vector, some may be considered to be constant for a given patient. It consists for example of parameters K]_ 2, K._has ] _, K._has 2, K._has gm, K._has , K._I , Vj, VQ codevectors and T_max. . Other parameters, hereafter referred to as time-dependent parameters, are however likely to evolve over time, for example the parameters K ^, the K ^, the K ^ grams, EGPq, Fqi and FR. Because of this variability of certain system parameters, it is in practice necessary to re-calibrate or recalibrate regularly the model in use, e.g. every 1 to 20 min, to ensure that model predictions remain relevant. The update model, also called a customization of the model, it should be carried out automatically by the system of Figure 1, c'est to say without physically measured parameters of time-dependent of the system on the patient and then transmit them to the processing and control unit 105.

Figure 4 is a chart illustrating an example of an automated method for regulating blood glucose implemented by the system of Figure 1.

This method comprises a step of re-calibration or 401 model update, which e.g. can be repeated at regular intervals, e.g. every 1 to 20 min. In this step, the processing and control unit 105 performs a method of re-parameter estimation time-dependent model considering insulin data actually injected by the device 103 and actual blood glucose data measured by the sensor 101 during an observation period passed, for example a period of 1 to 10 hours previous calibration step. More particularly, during the calibration step, the processing and control unit 105 simulates the behavior of the patient over the observation period passed from the physiological model (taking into account possible intakes of glucose and insulin injections during this period), and compares the blood glucose curve model estimated blood glucose curve to the actual measured by the sensor during this same period. The processing and control unit 105 then searches, for time-dependent parameters of the model, a set of values leading to minimize a magnitude representative of the error between the blood glucose curve model estimated curve and the actual blood glucose during the observation period. As an example, the processing and control unit searches for a set of parameters leading to minimize an indicator m is representative of the area between the curve model estimated blood glucose curve and the actual blood glucose during the observation period, e.g. defined as follows:

T-₀ + ATS

^{the m=} ~And Σ 9 - (T-) VBE1²

t=t₀

where T is the time variable discretized, GB corresponds to the start time of the observation phase passed, tan VBE1 + Z-T corresponds to the end time of the observation phase passed (corresponding for example to the start time of the step of calibrating the model), g is the curve temporal variation of the actual blood glucose level measured by the sensor 101 during the period [to verbs, tan + ATs], and g is the blood glucose curve estimated from the model during the period [tan, tan + ATs]. The search algorithm of optimal parameters used in this step is not detailed in the present application, the disclosed embodiments consistent with the usual algorithms used in various fields for optimization problems of parameters by minimizing a cost function.

The method of Figure 4 further comprises, after step 401, a step 403 prediction, by the processing and control unit 105, the temporal course of the patient's blood glucose level over a future time period, from the physiological model updated in step 401 and taking into account the history of insulin injected to the patient and history of glucose ingested by the patient.

The method of Figure 4 further comprises, after step 403, a step 405 for determining, by the processing unit 105 and control, taking into account the predicted future blood glucose curve at step 403, doses of insulin to be injected to the patient during a future time period. Upon completion of this step, the processing and control unit 105 can program the injection device 103 for administering doses determined during the upcoming time period.

The steps 403 blood glucose prediction and 405 and determining future doses of insulin to be administered can for example be reiterated every update of the physiological model (c'est to say after each iteration of step 401), each new ingesting glucose reported by the patient, and/or at each new administration of a dose of insulin from the injection device 103.

One problem in the operation described above is that, when the updating of the physiological model in step 401, the processing and control unit 105 should define a vector initial conditions (conditions Tg) state variables of the model, to be able to simulate the behavior of the patient from the model. These initial conditions are required not only to be able to predict the future movement of the patient's blood glucose level (step 403), but also during the step of updating the model itself (step 401), in order to simulate the evolution of the patient's blood glucose level during the observation period has passed, so as to compare the simulated blood glucose measured blood glucose.

To define the initial status of the state variables of the model, a first possibility is to make the assumption that, in the period preceding the observation period [to verbs, ATs + tan] on which is based the calibration of the model, the patient is in a stationary state, with a flow rate of insulin injected steady, and a food intake of glucose zero. Under this assumption, all derived from the differential equation system can be regarded as being zero at the initial time tan•Tg values of state variables of the system can then be calculated analytically. A disadvantage of this solution is that the output of the model (estimated blood glucose) is not constrained. In particular, the estimated blood glucose at the time tan can be different from the actual blood glucose measured at the instant tan. In this case, the algorithm implemented in step 401 search parameters time-dependent model by minimizing the error between the simulated and the measured blood glucose blood glucose may have hard to converge.

To improve the initialization, a second possibility is to make the same assumptions as above, but by constraining variable Q] _ (tan) so that the blood glucose level estimated at the time of Tg equals the actual blood glucose level measured by the sensor. This improves the relevance of the initialization at the time Tg's. However, at the time the Tg, the derivative of the estimated glucose level and the derivative of the actual blood glucose may diverge. Accordingly, the search algorithm parameters time-dependent system may again have yet hard to converge.

In practice, the two methods described above for determining the initial conditions of the physiological model are often unsatisfactory, which makes it hard to search for a set of values for the parameters relevant time-dependent model. A consequence is that the predictions of the future evolution of the patient's blood glucose level from the model may be erroneous, and lead to poor glycemic control by the system.

To mitigate this problem, in one aspect of an embodiment, there is provided, during the calibration phase or model update (step 401), to consider the initial states model as random variables, and performing, as to estimate parameters of time-dependent model, a search for an optimum set of initial state values by minimizing a quantity representative of the error between the blood glucose curve estimated from the model and the actual blood glucose curve during the observation period on which is based the calibration.

If the accumulated number of time-dependent parameters and variables of state of the physiological model is sufficiently low, the optimum values of the parameters time-dependent and initial states state variables can be determined simultaneously, in a single step of optimizing the model by minimizing the error between the estimated and actual blood glucose blood glucose over the observation period has passed.

In practice, in the model of ideas about, as well as in most physiological models describing the uptake of insulin and glucose by the body and their impact on glycemia, the accumulated number of time-dependent parameters and variables of state is relatively large, which can result in a numerical instability during the search phase optimum values. In other words, certain values may be difficult or impossible to estimate in a single search, the number of unknowns is too large. In this case, the problem can be decomposed into two subproblems, respectively corresponding to the estimation of the time-dependent parameters of the model and estimation of initial states of the model, as will now be described in connection with fig. 5.

Figure 5 is a diagram illustrating one example of an embodiment of an automated method of calibration or upgrade of the system of Figure 1, corresponding to an example of implementation of step 401 of Figure 4.

This method comprises a step 501 in which the parameter vector (reduced here to only those parameters time-dependent model) is initialized to a first set of values pi. The game PI corresponds for example to the values assumed by the parameters before the beginning of the updating phase of the model. Alternatively, the set of values PI is a set of predetermined reference corresponding e.g. to average values assumed by the parameters on a reference period. In step 501, the initial state vector of state variables is further initialized to a first set of value-II. The set of values for example II is determined analytically as described above, assuming a steady-state condition of the patient in the period preceding the calibration phase, and by matching the estimated blood glucose at the time Tg and the measured actual blood glucose at that instant.

In a step 503 after step 501, the processing and control unit 105 search, by fixing the set of initial states to its current state, a set of values of parameters time-dependent model leading to minimize a magnitude representative of the error between the blood glucose curve estimated from the model and the actual blood glucose curve during the observation period, for example the indicator of m defines above. Upon completion of this step, the vector is updated with the new estimated values.

In a step 505 follows step 503, the processing and control unit 105 search, by setting the parameter set to its current state, a set of initial state values of state variables leading to minimize a magnitude representative of the error between the blood glucose curve estimated from the model and the actual blood glucose curve during the observation period, e.g. the indicator m is defined above, or any other indicator representative of the error between the two curves, for example an indicator based upon standard wears. Upon completion of this step, the vector is updated with the new estimated values.

In this example, the steps 503 and 505 are repeated a predetermined number n of times, where n is an integer greater than 1. The values of the parameters time-dependent and initial conditions of the updated model correspond then to the values of the vectors and at the end of the nth iteration of steps 503 and 505. Alternatively the number of iterations of steps 503 and 505 may not be predetermined, and be adjusted taking into account the evolution of the indicator m of error between the estimated blood glucose from the model and the actual blood glucose over the observation period.

Search algorithms optimum values used in the steps of 503 and 505 are not detailed in the present application, the disclosed embodiments consistent with the usual algorithms used in various fields for optimization problems of parameters by minimizing a cost function.

An advantage of the mode of operation described above, wherein the initial values of the state variables of the physiological model are determined by minimizing a quantity representative of the error between the measured blood glucose data and estimated blood glucose during an observation period passed, is that it improves the quality of the prediction of the future blood glucose levels of the patient, and thus monitor with higher relevancy the supply of insulin.

An object of another embodiment is to limit the risks for the patient related to a possible fault of the physiological model used to predict future blood glucose levels of the patient.

For this purpose, according to an aspect of an embodiment, the control and processing device 105 of the control system is adapted, after each update or re-calibration of the physiological model (step 401), to estimate the quality of the physiological model updated by means of one or more numeric indicators of quality, and, if the quality of the model is deemed unsatisfactory, to discontinue using the model to regulate blood glucose of the patient.

Figure 6 is a chart illustrating an example of an embodiment of an automated method for regulating blood glucose implemented by the system of Figure 1.

The method includes the same steps 401, 403 and 405 that in the example in Figure 4. However, the method of Figure 6 further comprises, after each step 401 of updating the physiological model operated by the control system and before carrying out the following steps 403 for predicting future blood glucose levels of the patient from the model and 405 control insulin delivery from the prediction of blood glucose, a step 601 for quality testing of the updated model.

In step 601, the processing and control unit 105 determines one or more numeric indicators of the quality of the model updated in step 401. As an example, the processing and control unit calculates a quality digital indicator representative of the area between the curve blood glucose estimated from the model and the actual blood glucose curve measured by the sensor 101 during an observation period has passed. The indicator corresponds for example to the variable m is defined above.

Instead, or in addition, an indicator representative of the surface between the curves estimated and actual blood glucose blood glucose during an observation period passed, the processing and control unit 105 can calculate the a and/or 1' other quality indicators IRQs m2 and following:

^{the m} the L (T-_NECK curing the)⁼ 9 (T-_COEXTRUDED Y) "§(T-_COEXTRUDED Y)

^{the m} 2 (T-_NECK curing the)⁼ The c (T-_couran -|-) - Gm ' (T-_NECK curing the)'

where T_couran -|- Denotes a present time implementation of step 601 for quality testing of the model, the g corresponds to the function of time evolution of the actual blood glucose level measured by the sensor 101, the g corresponds to the function of time evolution of the simulated blood glucose from the model, gm 'corresponds to the derivative of the function of time evolution of the actual blood glucose, and g' corresponds to the derivative of the function of time evolution of the simulated blood glucose.

By way of example, the quality of the model can be considered satisfactory by the processing and control unit 105 when the m values, and IRQs m2 are less than predetermined threshold values. More generally, any other quality criterion or any other combination of quality criteria may be used in step 601 to determine whether the physiological model to re-calibrated in step 401 can be considered reliable.

If the physiological model is considered to be reliable in step 601 (4 O), 403 and 405 steps may be implemented similarly to what has been previously described, c'est to say that the processing and control unit 105 continues to rely on the predictions made by the physiological model to control administration of insulin to the patient.

If the physiological model is deemed insufficiently reliable in step 601 (d), the processing and control unit 105 ceases to use this model to control administration of insulin to the patient, and implements a method for regulation of substitution in a step 603.

By way of example, in the step 603, the processing and control unit 105 uses a physiological model simplified, e.g. a compartmentalized model having a number of state variables and a number of parameters is reduced relative to the initial model, predicting the progress of the patient's blood glucose level and suitably adjusted insulin injection.

Alternatively, in step 603, the processing and control unit 105 ceases to perform feedforward control, c'est to say it stops using a physiological model to predict future blood glucose levels of the patient and suitably adjusted insulin injection. In this case, the processing and control unit 105 controls for example the insulin injection device 103 for administering doses preprogrammed insulin, corresponding for example to a basal rate reference given to the patient.

Such a method of substitution may for example be used for a predetermined period of time. At the end of this period, the steps 401 for calibrating the model physiological main and 601 for estimating the quality of the physiological model main steps can be repeated, for, if the quality of the physiological model main is deemed satisfactory, reactivate using this model to control administration of insulin to the patient.

It should be noted that the method of Figure 6 is not limited to the embodiment described with reference to Figures 4 and 5, wherein calibrating the physiological model comprises a step of determining initial values of the state variables of the model by minimizing a quantity representative of the error between the measured blood glucose data and estimated blood glucose during an observation period, but can be used regardless of which method is chosen for determining initial values of the state variables of the model.

Particular embodiments have been described. Various variations and modifications will be apparent to the ordinarily skilled artisan. In particular, the described embodiments are not limited to the particular example of physiological model detailed in the present specification, namely the model of ideas about, but are compatible with any physiological model describing the uptake of insulin by a patient's body and its effect on the patient's blood glucose level, for example the model called Cobelli, described in the article entitled "model where an oral glucose-OC-system-absorbing: validating Gold sequences is the standard data file", Chiara Al MAN and attached to children (the IEEE transaction is solar patterns, v. 53, Number 12, for January 2006).