AUTOMATED SYSTEM FOR REGULATING A PATIENT'S BLOOD GLUCOSE LEVEL

Abstract

The invention relates to an automated system for regulating the blood sugar level of a patient, comprising: a blood sugar sensor (101); an insulin injection device (103); and a processing and control unit (105) adapted to predict, from a physiological model, the future evolution of the blood sugar level of the patient over a prediction period, and to control the insulin injection device (103) taking into account this prediction, in which the processing and control unit (105) is adapted to: a) calibrate the physiological model taking into account the blood sugar level measured by the sensor (101) during a past observation period; b) at the end of the calibration, calculate an indicator representative of the error between the blood sugar level estimated from the model and the actual blood sugar level measured by the sensor; and c) adjust the prediction period taking into account the value of the indicator.

Description

AUTOMATED SYSTEM FOR REGULATING A PATIENT'S BLOOD SUGAR Domain This application concerns the field of systems automated blood glucose regulation systems, also called pancreas artificial. Presentation of 1 prior art An artificial pancreas is a system that allows automatically regulate a patient's insulin intake diabetic based on his blood sugar history, his Meal intake history and injection history of insulin. We are particularly interested here in systems of MPC-type regulation (from English "Model-based Predictive") Control") also called predictive control systems, in which the regulation of the administered insulin dose depends on taking into account a prediction of the future evolution of blood glucose levels patient, created from a physiological model describing 1. Assimilation of 1 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 predictive artificial pancreases, and, more specifically, to be able to improve the quality of the prediction of the patient's future blood glucose levels, so as to be able to to control insulin intake with greater relevance and limit the risks of placing the patient in a situation of hyperglycemia or hypoglycemia. Summary Thus, one embodiment provides for a system automated blood glucose regulation system for a patient, comprising a blood glucose sensor; an insulin injection device; and a processing and control unit adapted to predicting, based on a physiological model, a future evolution the patient's blood glucose level over a predicted period, and has order the insulin injection device taking into account of this prediction, in which the processing and control unit is adapted to a) implement an automatic calibration step of the physiological model taking into account the measured blood glucose by the sensor during a past observation period; b) a 1 outcome of 1 calibration step, calculate at minus a numerical indicator representative of 1 error between the estimated blood glucose level from the model and the actual measured blood glucose level by the sensor; and c) adjust the duration of the prediction period taking into account account of the value of 1 numerical indicator. According to one embodiment, 1 numerical indicator includes the root mean square deviation between the estimated blood glucose level based on the model and the actual blood glucose measured by the sensor during the past observation period. According to one embodiment, 1 numerical indicator includes the difference between the actual blood glucose measured by the capture and the blood glucose level estimated by the model at a given moment. According to one embodiment, 1 numerical indicator includes the difference between the derivative of the actual blood glucose measured by the sensor and the derivative of the blood glucose estimated by the model at a given moment. According to one embodiment, the processing unit and The control is configured for, in step 1 c) compare the value from 1 numerical indicator to initial thresholds, and select the duration of the prediction period among a plurality of durations predefined based on the result of the comparison. According to one embodiment, the processing unit and The control method is further adapted, after step 1 b) aTo be determined, based on the value of the numerical indicator, if the model is reliable enough to serve as the basis for controlling the insulin injection device, and, if not, has order the insulin injection device according to a method substitution, without taking into account the prediction made a starting from the model. According to one embodiment, to determine if the the model is sufficiently reliable, the processing unit and control compares the value of 1 digital indicator to a second threshold. According to one embodiment, the processing unit and control is adapted to determine and adjust automatically the second threshold based on past data measured on the patient, so that the injection device is controlled insulin is based on predictions made by the model at least a certain percentage P of the time. According to one embodiment, the substitution method is a predictive control method based on a model Simplified physiology. According to one embodiment, the substitution method consists of ordering the insulin injection device for deliver pre-programmed doses of insulin corresponding to a basal reference flow prescribed to the patient. According to one embodiment, the substitution method consists of ordering the insulin injection device for administer insulin doses determined by the unit of processing and control depending on the current level of Blood glucose measured by the sensor and / or the rate of change of the blood glucose measured by the sensor. Brief description of the drawings These features and advantages, among others, will be described in detail in the following description of modes of specific work carried out, for non-exhaustive purposes, in relation to with the attached figures among which Figure 1 represents schematically, in the form of blocks, an example of a way to implement a system automated blood glucose regulation for a patient; Figure 2 is a simplified representation of a physiological model used in the system of Figure 1 for predict future blood glucose levels of the patient; Figure 3 is a diagram illustrating an example of a automated blood glucose regulation process that can be put into work by the system of figure 1; and Figure 4 is a diagram illustrating in more detail an example of an embodiment of an automated process of blood glucose regulation implemented by the system in Figure 1. Detailed description The same elements have been designated by the same references in the various figures. For the sake of clarity, only the elements that are useful for understanding the modes The implementation details described have been presented and are detailed. In particular, the blood glucose measuring device and the device insulin injections from the described regulatory system did not The methods described were detailed and found to be compatible. with all or most blood glucose monitoring devices and of known insulin injections. Furthermore, the physical implementation of the processing and control unit of the regulatory system The described process has not been detailed; the realization of such a unit of processing and control being within the reach of a person skilled in the art based on the functional indications of the present description. Figure 1 schematically represents, in the form of blocks, an example of a way to implement a system automated blood glucose regulation for a patient. The system in Figure 1 includes a sensor 101 (CG) adapted to measure the patient's blood glucose level. In normal operation, The 101 sensor can be permanently positioned on or in the body of the patient, for example at the level of their abdomen. The 101 sensor is, for example, a CGM type sensor (from the English "Continuous") Glucose Monitoring (continuous blood glucose monitoring) is— a—that is, a sensor adapted to measure continuously (for example, at (at least once every five minutes) the patient's blood glucose. The 101 sensor, for example, is a sub-glucose sensor. cutaneous. The system in Figure 1 further includes a device insulin injection device 103 (PMP) for example subcutaneous injection. Device 103 is, for example, a automatic insulin pump-type injection device, comprising an insulin reservoir connected to a keel injection pump implanted under the patient's skin, the pump being to be electrically controlled to automatically inject insulin doses are determined at specific times. During normal operation, the injection device 103 can be positioned permanently in or on the patient's body, for example at the level of his abdomen. The system in Figure 1 also includes a unit processing and control 105 (CTRL) linked on the one hand to blood glucose sensor 101, for example via wired connection or by radio (wireless) connection and on the other hand to the device injection 103, for example by wired or radio link. In In operation, the processing and control unit 105 is adapted to receive the patient's measured blood glucose data via sensor 101, and to electrically control the device 103 to inject the patient with determined doses of insulin specific moments. In this example, the processing unit and control 105 is further adapted to receive, by 1. Intermediate user interface not detailed, of representative cho(t) data of the evolution, as a function of the time, the amount of glucose ingested by the patient. The user interface can also be designed to allow to capture additional information that may to facilitate blood glucose regulation, for example information relating to the patient's physical activity, or his stress, or any other information related to metabolism of the patient, or the types of food ingested by the patient (bold or not, for example) The 105 processing and control unit is adapted to determine the insulin doses to be injected into the patient taking into account taking into account, in particular, the blood glucose history measured by the sensor 101, of the history of insulin injected by the device 103, and the patient's glucose ingestion history (as well as any additional information) (mentioned above) For this purpose, the processing and control unit 105 includes a digital calculation circuit (not detailed) including, for example, a microprocessor. The processing unit and control 105 is, for example, a mobile device transported by the patient throughout the day and / or the night, for example a smartphone-type device configured for implement a regulation process of the type described below. In the embodiment of Figure 1, the unit of treatment and control 105 is suitable for determining the quantity insulin to be administered to the patient, taking into account a prediction of the future evolution of his blood sugar levels based on the time. More specifically, the processing and control unit 105 is adapted, based on the history of injected insulin and of the history of glucose intake (as well as any possible additional information mentioned above) and based on a physiological model describing the assimilation of insulin by the patient's body and its impact on blood glucose, to be determined a curve representing the expected evolution of blood glucose of the patient depending on time, over a future period called prediction period or prediction horizon, for example a period of 1 to 10 hours. Taking this curve into account, the unit treatment and control 105 determines the insulin doses that should be injected into the patient during the period of prediction to come, so that the actual blood glucose (as opposed to blood glucose estimated from the patient's physiological model) remains within acceptable limits, and in particular to limit the risks of hyperglycemia or hypoglycemia. In this mode of how it works, as will be explained in more detail below, the actual blood glucose data measured by sensor 101 are used primarily for model calibration purposes physiological. Figure 2 is a simplified representation of a MPC physiological model used in the system of Figure 1 to predict the future evolution of the patient's blood glucose levels. On the Figure 2 shows the model represented as a block of treatment involving an input el to which a signal i (t) is applied representative of the evolution, as a function of time t, of the quantity of insulin injected into the patient; an input e2 to which a signal is applied cho(t) represents the evolution, as a function of time t, of the amount of glucose ingested by the patient; and an output providing a representative signal G(t) of the evolution, as a function of time t, of the patient's blood glucose. The MPC physiological model is, for example, a model compartmentalized including, in addition to the input variables i(t) and cho(t) and the output variable G(t) a plurality of variables states corresponding to physiological variables of the patient, evolving over time. The temporal evolution of state variables and the output variable G(t) are governed by a system of differential equations comprising a plurality of parameters represented in figure 2 by a vector [PARAM] applied to a PL input of the MPC block. The model's response physiological is further conditioned by the initial states or initial values ​​assigned to the state variables, represented on Figure 2 by a vector [INIT] applied to an input p2 of the MPC block. As an example, the MPC physiological model used In the system shown in Figure 1, this is the so-called Hovorka model. described in one article entitled "Nonlinear model predictive control of glucose concentration in subjects with type 1 diabetes" Roman Hovorka et al. (Physiol Meas. 2004;25:905–920) and in 1 article entitled "Partitioning glucose distribution / transport, disposal, and endogenous production during IVGTT" by Roman Hovorka et al. (Am J Physiol Endocrinol Metab 282: E992—E1007 2002) More generally, any other physiological model describing the assimilation of insulin by a patient's body and its effect on the patient's blood glucose can be used, by for example the so-called Cobelli model, described in an article entitled "A System Model of Oral Glucose Absorption: Validation on Gold Standard Data" by Chiara Dalla Man et al. (IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 12, DECEMBER 2006) Among the parameters of the vector [PARAM] some can be considered constant for a given patient. Other parameters, referred to hereafter as time parameters— dependents, on the other hand, are likely to change over time. Due to this variability in certain system parameters, In practice, it is necessary to recalibrate. regularly update the model currently in use, for example all every 1 to 20 minutes, for example every 5 minutes, to to ensure that the model's predictions remain relevant. This Model update, also called model customization, must be able to be performed automatically by the system Figure 1, that is to say, without it being necessary to measure physically the time parameters—dependent on the system on the patient and then transmit them to the treatment unit and control 105. Figure 3 is a diagram illustrating an example of a automated blood glucose regulation process that can be put into work by the system of figure 1. This process includes a recalibration step 301 or model update, which can for example be repeated at at regular intervals, for example every 1 to 20 minutes. During From this stage, the processing and control unit 105 puts into covers a process for re-estimating time-dependent parameters of the model taking into account the actual insulin data injected by device 103 and actual blood glucose data measured by sensor 101 during an observation period past duration AT, for example a period of 1 to 10 hours preceding the calibration step. More specifically, during the calibration step, the processing and control unit 105 simulates the patient's behavior over the observation period based on the physiological model (taking into account the possible glucose ingestions and insulin injections during this period) and compares the blood glucose curve estimated by the model a the actual blood glucose curve measured by the sensor during this same period. The processing and control unit 105 search then, for the time parameters—dependent on the model, a set of values ​​leading to the minimization of a quantity representative of the error between the estimated blood glucose curve by the model and the actual blood glucose curve during the period observation. For example, the processing and control searches for a set of parameters leading to minimizing a indicator m representative of 1 area between the blood glucose curve estimated by the model and the actual blood glucose curve during the observation period, also called mean square deviation between estimated blood glucose and actual blood glucose, for example defined as follows m = fittztzmm — go to where t is the discretized time variable, tp—AT corresponds to 1. The observation phase has begun, typ corresponds a 1 instant at the end of the past observation phase (corresponding for example, at the start of the calibration step of the (model) g is the time evolution curve of blood glucose actual measured by sensor 101 during the period tol and f is the blood glucose curve estimated from the model during the period tol As an alternative, for the calculation of 1 mean squared deviation, the variable AT can be replaced by the number of measurements taken during the observation period past. The optimal parameter search algorithm used This step is not detailed in this application. the described embodiments being compatible with the common algorithms used in various fields to solve parameter optimization problems through minimization of a cofut function. Note that during step 301, in addition to the parameters time—dependent on the model, the processing and control unit 105 defines an [INIT] vector of initial states (states a 1 instant tp—AT) of the model's state variables, in order to be able to simulate the Patient behavior based on the model. To define the states initial values ​​of the model's state variables, a first possibility consists of making the assumption that, in the period preceding the observation period [tq—AT, tol on which is based the Model calibration; the patient was in a state stationary, with a constant rate of injected insulin, and a No glucose intake during meals. Under this assumption, all the derivatives of the system of differential equations can be considered as zero at initial time tp—AT The values ​​a 1 instant to—AT system state variables can then be calculated analytically. To improve initialization, a Another possibility is to make the same assumptions as previously, but adding the constraint that blood glucose estimated at 1 typical instant—AT is equal to the actual measured blood glucose by the sensor. To further improve 1 initialization, another One possibility is to consider the initial states of the variables the state of the model as well as random variables, in the same way that the time parameters—dependent on the model. The initial states state variables are then determined in the same way as the time parameters—dependent on the model, that is to say that The processing and control unit 105 is searching for a set of initial state values ​​[INIT] leading to minimizing a representative quantity of 1 error between the blood glucose curve estimated by the model and the actual blood glucose curve during the observation period completed. The process in Figure 3 further includes, after 1 step 301, a prediction step 303, by the unit of processing and control 105, of the temporal evolution of the patient blood glucose over a future prediction period [t0, totTpreql of duration Tpreg, for example between 1 and 10 hours, based on the physiological model updated at step 301 and taking into account the patient's history of insulin injections and the patient's history of glucose intake. The process in Figure 3 further includes, after the step 303, a determination step 305, by the processing unit and control 105, taking into account the future blood glucose curve predicted at step 303, doses of insulin to be injected into the patient during the upcoming prediction period [t0, tqptITpregl A 1 issue From this step, the processing and control unit 105 can program the injection device 103 to administer the doses determined during the prediction period [t0, t0+Tpred] Steps 303 of blood glucose prediction and 305 and determination of future doses of insulin to be administered can for example, to be reiterated with each update of the model physiological (that is, after each iteration of the step 301) each new glucose ingestion signaled by the patient, and / or with each new administration of a dose insulin via injection device 103. In the aforementioned process, the duration Tpreg is 1a prediction period of 1 future evolution of blood glucose patient is an important parameter, conditioning performance of the regulatory system. Given the relatively dynamics slowness of the system that one seeks to regulate, it would be desirable that the Tored prediction period is relatively long, by for example, on the order of 4 hours or more, so as to be able to to anticipate and assess the patient's insulin needs as accurately as possible. However, in practice, the imperfections of the model used constrain the limitation of the prediction horizon considered. According to one aspect of an embodiment, the unit of treatment and control 105 is adapted, after each update day of the physiological model (step 301) to calculate one or several numerical indicators representative of reliability of the updated model, and to adjust the prediction duration Tored based on these indicators. More specifically, if the The updated model is deemed reliable; the prediction timeframe is Tored. will be chosen relatively high, and, if the model is deemed low reliable, the prediction time Tprag will be chosen relatively weak. Compared to a system in which the prediction time Tored is fixed; one advantage of this operating mode is that it allows for improved accuracy in blood glucose prediction future of the patient, and thus to control with greater relevance of insulin intake. Figure 4 is a diagram illustrating in more detail an example of an automated blood glucose regulation process in ostre by the system of figure 1, in which the duration of The Tpreg prediction is adjusted based on an estimate of the reliability of the physiological model. This process includes the same steps 301, 303 and 305 as in the example of Figure 3. However, the process of the figure 4 further includes, after each update step 301 of the physiological model and before the implementation of the steps following 303 predictions of the patient's future blood glucose levels, and 305 of control of insulin delivery from the blood glucose prediction, a step 411 calculation of one or more numerical reliability indicators of the updated model, and of a step 413 of adjusting the prediction duration Tpred &D function of the reliability indicator(s) calculated in step 1 411. During step 411, the processing and control 105 calculates one or more numerical indicators representative of the reliability of the model updated at step 301 As an example, the processing and control unit calculates three digital reliability indicators MM, GD and SD. The MM indicator corresponds to the root mean square deviation between the estimated blood glucose from the updated model and the curve of Actual blood glucose measured by sensor 101 over a period of past observation, for example a period of 1 to 10 hours preceding 1 instant tp, for example the period [tp—AT, tol The GD indicator corresponds to the difference between blood glucose actual measured by sensor 101 and the blood glucose estimated by the model updated at a given moment, for example at time tp 1, and 1 SD indicator corresponds to the difference between the slope or derived from the actual blood glucose measured by sensor 101 and the slope or derivative of the blood glucose estimated by the updated model at a given instant, for example at 1 instant tO. During step 413, the processing and control determines, based on the numerical indicator(s) of reliability calculated at step 411, the prediction time Tpreqg a use for the implementation of step 303. For example, The prediction duration Tpregq is chosen from values predefined D1, In decreasing, with integer greater than or equal to 2, depending on the value of the numerical indicator(s) reliability calculated in step 411. As an example, for each of the m reliability indicators I; calculated at step 411, with j an integer from 1 to a and m an integer greater than or equal to 1, The value of the indicator is compared to a set of thresholds. Predefined SIj1, SI;, of increasing values. The unit of processing and control 105 then searches for the smallest index threshold k such that, for each of the m indicators I; calculated a 1 step 411, the value of indicator Ij is less than the threshold The prediction horizon Tpyag is then chosen to be equal to the Duration Dk. More generally, depending on the objective sought, other functions and / or decision rules enabling the determination of prediction duration Tpeg 4 based on the indicator(s) reliability calculated in 1 step 411 can be implemented. After step 413, steps 303 and 305 can be implementations similar to what has been described previously. It should be noted that in some cases, the reliability of the model physiological update at step 301 may be so low that it it is preferable to stop using the model to regulate the patient's blood glucose level. In the example in Figure 4, the control unit and The 105 treatment of the regulatory system is further adapted, after each update or recalibration of the physiological model (step 301) based on the calculated reliability indicator(s) a 1 step 411, to determine if the updated model is reliable enough to be used to regulate the patient's blood glucose level. More specifically, the process shown in Figure 4 includes, between steps 411 and 413, a step 451 of verification of the reliability of the updated model in step 301. As an example, the reliability of the model can be considered as sufficient by the processing and control unit 105 when the values ​​of the indicators calculated in step 1, 411 are below predefined thresholds, and insufficient in the case On the contrary. For example, using the defined notations As above, the reliability of the model can be considered as sufficient by the processing and control unit 105 when for each of the m reliability indicators Ij calculated at the stage 411, the value of 1 indicator is less than the S$Ijp threshold corresponding, and insufficient when for at least one of the indicators Ij the value of the indicator is greater than the threshold SIjn_ corresponding. More generally, any other criterion of quality or any other combination of quality criteria may to be used at step 451 to determine if the model physiological re-calibrated at step 301 is sufficiently reliable. If the physiological model is considered — as sufficiently reliable at step 451 (0) steps 413, 303 and 305 can be implemented in a similar way to what has been as described previously, that is to say, the processing unit and control 105 continues to be based on the predictions made by the physiological model to regulate 1 administration of 1 insulin to the patient, adjusting the Tored prediction horizon depending on the reliability of the model. If the physiological model is deemed insufficient reliable 1-step 451 (N) — the processing and control unit 105 ceases to use this model to request the administration of 1 insulin to the patient, and implements a regulation method substitution during step 453. For example, during step 453, the unit of treatment and control 105 uses a physiological model simplified, for example a compartmentalized model comprising a number of state variables and a reduced number of parameters by compared to the initial model, to predict 1 evolution of blood glucose of the patient and regulate the insulin injection accordingly. As an alternative, in step 453, the unit of processing and control 105 ceases to implement a predictive ordering, that is to say, it stops using a physiological model to predict the patient's future blood glucose levels and adjust the insulin injection accordingly. In this case, the processing and control unit 105 commands, for example, the insulin injection device 103 for administering doses pre-programmed insulin doses, corresponding for example to a flow rate basal reference dose prescribed to the patient. Alternatively, the unit Processing and control 105 uses a type algorithm decision matrix for determining insulin doses a administer to the patient, depending on various observed parameters such as the current blood glucose level measured by sensor 101, or the rate of change (or slope) of blood glucose on a past period. Such a substitution method could, for example, be used for a predetermined period of time. A 1 issue of During this period, the model calibration steps 301 main physiological, 411 calculation of the indicator(s) of reliability of the main physiological model, and 451 of estimation of the quality of the main physiological model, may be repeated, in order to, if the quality of the main physiological model is deemed sufficient, reactivate the use of the main model to regulate the administration of insulin to the patient. As an example, the thresholds used in step 451 to determine if the main physiological model is sufficiently reliable to be used are chosen from way to maximize the probability that the regulatory system works at least a certain percentage P of the time, for example at least 70% of the time, based on the physiological model main. The thresholds used in steps 451 and 413 are by examples determined based on an analysis of past data measured on a sample of several patients. As For example, we can replay the regulation algorithm for a plurality of patients on a test bench, and, for each patient, with each update of the physiological model, for each of the Possible values ​​D1, In of the prediction duration Tpredr calculate 1 mean squared deviation, over the prediction period Tpreq, between the blood glucose estimated from the updated model and the actual blood glucose curve measured by sensor 101. At each In the model update, we also calculate the m indicators of reliability of the updated model I1, Iy. Thus, with each update With the model updated, we have a set of m corresponding values to the model reliability indicators as defined above, and a set of values ​​corresponding to actual measurements model reliability for prediction durations D1, Not considered. The study of correlations between indicators of model reliability and effective reliability measures allow to determine the thresholds to use in step 413 to choose the duration of the prediction period after each update of the model, and / or step 451 to decide whether or not it is appropriate to switch to an alternative regulatory method. Determining the thresholds based on the aforementioned values reliability indicators and effective reliability measures can be fully or partially automated. As an alternative, the thresholds used in steps 451 and 413 are determined in a similar way to what has just been described, but only on the basis of past measured data on the patient using the system, which allows for customize the operation of the control system. In this In this case, the processing and control unit 105 can be configured to regularly recalculate the thresholds used in step 1, step 413 and / or at step 451, taking into account the new data measured on the patient since the last update of the thresholds.

Claims

1. Automated blood glucose regulation system patient, including a blood glucose sensor (101); an insulin injection device (103); and a processing and control unit (105) adapted to predicting, based on a physiological model, a future evolution the patient's blood glucose level over a predicted period, and has order the insulin injection device (103) holding taking this prediction into account, in which the processing and control unit (105) is suitable for a) implement an automatic calibration step of the physiological model taking into account the measured blood glucose by the sensor (101) during a past observation period; b) a 1 outcome of 1 calibration step, calculate at minus a numerical indicator representative of 1 error between the estimated blood glucose level from the model and the actual measured blood glucose level by the sensor and c) adjust the duration of the prediction period taking into account taking into account the value of said at least one numerical indicator.

2. A system according to claim 1, wherein said at least one numerical indicator includes the root mean square deviation average between the blood glucose estimated from the model and the blood glucose actual measured by the sensor (101) during the period from past observation.

3. System according to claim 1 or 2, wherein said at least one numerical indicator includes the difference between the actual blood glucose measured by the sensor (101) and the blood glucose estimated by the model at a given moment.

4. System according to any one of claims 1 a 3 in which said at least one numeric indicator comprises the difference between the derivative of the actual blood glucose measured by the sensor (101) and the derivative of the blood glucose estimated by the model at a given moment. DEMANDS 5. System according to any one of the claims 1 a 4, in which the processing and control unit (105) is configured to, at step c) compare the value of said at least a numerical indicator has initial thresholds, and selecting the durations of the prediction period among a plurality of durations predefined based on the result of the comparison.

6. System according to any one of claims 1 a 5, in which the processing and control unit (105) is further adapted, after step b) to be determined, from the value of said at least one numerical indicator, if the model is reliable enough to serve as the basis for controlling the insulin injection device, and, if not, has order the insulin injection device (103) according to a substitution method, without taking the prediction into account made from the model. 7 — System according to claim 6, in which The processing and control unit (105) is configured to, To determine if the model is sufficiently reliable, compare the value of said at least one numerical indicator has a second threshold.

8. System according to claim 7 in which the processing and control unit (105) is adapted to automatically determine and adjust the second threshold from of past data measured on the patient, so that the control of the insulin injection device (103) is based on the predictions made by the model at least a certain percentage of time.

9. System according to any one of claims 6 a 8, in which the substitution method is a method of predictive control based on a simplified physiological model.

10. System according to any one of claims 6 a 8, in which the substitution method consists of ordering the insulin injection device (103) for delivering doses pre-programmed insulin corresponding to a basal rate of reference prescribed to the patient. 11 System according to any one of the claims 6 a 8, in which the substitution method consists of ordering the insulin injection device (103) for administering insulin doses determined by the treatment unit and control (105) based on the current measured blood glucose level by the sensor (101) and / or the rate of change of the blood glucose measured by the sensor (101)