CA2884999C - Safeguarding measures for a closed-loop insulin infusion system - Google Patents

Abstract

Processor-implemented methods of controlling an insulin infusion device for a user are provided here. A first method obtains a current insulin on board (IOB) value that estimates active insulin in the user, and compensates a calculated insulin infusion rate in response to the obtained IOB value. A second method supervises the operation of a glucose sensor by obtaining and processing insulin-delivered data and glucose sensor data for the user. An alert is generated if the second method determines that a current glucose sensor value has deviated from a predicted sensor glucose value by at least a threshold amount.

Description

SAFEGUARDING MEASURES FOR A CLOSED-LOOP INSULIN INFUSION SYSTEM
[0001]
TECHNICAL FIELD
100021 Embodiments of the subject matter described herein relate generally to drug delivery systems and more specifically to systems for controlling the infusion rate of insulin based on state variable feedback.
BACKGROUND
[0003] The pancreas of a normal healthy person produces and releases insulin into the blood stream in response to elevated blood plasma glucose levels. Beta cells (I3-cells), which reside in the pancreas, produce and secrete the insulin into the blood stream, as it is needed. If I3-cells become incapacitated or die, a condition known as Type I
diabetes mellitus (or in some cases if I3-cells produce insufficient quantities of insulin, Type II
diabetes), then insulin must be provided to the body from another source.
[0004] Traditionally, since insulin cannot be taken orally, insulin has been injected with a syringe. More recently, use of infusion pump therapy has been increasing, especially for delivering insulin for diabetics. For example, external infusion pumps are worn on a belt, in a pocket, or the like, and deliver insulin into the body via an infusion tube with a percutaneous needle or a cannula placed in the subcutaneous tissue. As of 1995, less than 5% of Type I diabetics in the United States were using infusion pump therapy. Presently, over 7% of the more than 900,000 Type I diabetics in the United States are using infusion pump therapy, and the percentage of Type I diabetics that use an infusion pump is growing at an absolute rate of over 2% each year. Moreover, the number of Type I diabetics is growing at 3% or more per year. In addition, growing numbers of insulin-using Type II diabetics are also using infusion pumps. Physicians have recognized that continuous infusion provides greater control of a diabetic's condition, and are also increasingly prescribing it for patients. Although offering control, pump therapy can suffer from several complications that make use of traditional external infusion pumps less desirable for the user.
[0005] In insulin pumps, it is common to use fast acting insulin as opposed to the slower acting insulin that is used for injections, because pumps allow changing of insulin profiles. As insulin companies develop faster acting insulin, the faster acting insulin is often adopted quickly. However, current pumps are still limited by the speed of the insulin they are using.
BRIEF SUMMARY
[0006] A processor-implemented method is presented here. The method can be used to control an insulin infusion device for a user. Certain embodiments of the method involve the operation of a processor architecture having at least one processor device to obtain a current insulin on board (JOB) value that represents an estimate of active insulin in the body of the user. The method continues by calculating, by the processor architecture, an 10B rate based at least in part on the obtained current 10B
value. The method continues by determining, by the processor architecture, an adjusted insulin infusion rate based at least in part on the calculated JOB rate and an uncompensated insulin infusion rate. The processor architecture selects a final insulin infusion rate for the insulin infusion device, wherein either the determined adjusted insulin infusion rate, the uncompensated insulin infusion rate, or a current basal rate is selected as the final insulin infusion rate.
[0007] Also presented here is a processor-implemented method of controlling an insulin infusion device for a user. Certain embodiments of the method begin by generating a current IOB value that represents an estimate of active insulin in the body of the user. The method continues by calculating an JOB rate based at least in part on the generated current IOB value, obtaining an uncompensated insulin infusion rate, and determining an adjusted insulin infusion rate in accordance with the expression AdjustedRate(n)= max(0; PIDRate(n)¨ 10BRate(n)). The method continues by selecting a final insulin infusion rate in accordance with the expression

2 max(Basal; AdjustedRate(n)), P1DRate > Basal. In this FinalRate(n) =
PID Rate(n), PIDRate < Basal expression: AdjustedRate(n) is the determined adjusted insulin infusion rate;
PIDRate(n) is the obtained uncompensated insulin infusion rate; IOBRate(n) is the calculated LOB
rate; FinalRate(n) is the selected final insulin infusion rate; and Basal is a current basal rate maintained by the insulin infusion device for the user.
[0008] Also presented here is a tangible and non-transitory electronic storage medium having processor-executable instructions that, when executed by a processor architecture comprising at least one processor device, perform a method of controlling an insulin infusion device for a user. In certain embodiments, the method begins by estimating a current JOB value that indicates an amount of active insulin in the body of the user. The method continues by calculating an JOB rate based at least in part on the estimated current JOB value, determining an adjusted insulin infusion rate based at least in part on the calculated IOB rate and an uncompensated insulin infusion rate, and selecting a final insulin infusion rate for the insulin infusion device, wherein either the determined adjusted insulin infusion rate, the uncompensated insulin infusion rate, or a current basal rate is selected as the final insulin infusion rate. The method then provides the selected final insulin infusion rate to regulate delivery of insulin by the insulin infusion device.
[0009] An electronic device is also presented here. Certain embodiments of the electronic device include a processor architecture and at least one memory element associated with the processor architecture. The at least one memory element stores processor-executable instructions that, when executed by the processor architecture, perform a method of controlling an insulin infusion device for a user. The method involves: computing a current IOB value that indicates an amount of active insulin in the body of the user; calculating an JOB rate based at least in part on the computed JOB
value; determining an adjusted insulin infusion rate based at least in part on the calculated IOB rate and an uncompensated insulin infusion rate; and selecting a final insulin infusion rate for the insulin infusion device. The selecting step selects either the determined adjusted insulin infusion rate, the uncompensated insulin infusion rate, or a current basal rate as the final insulin infusion rate.
[OM An electronic controller for an insulin infusion device is also presented here.
The electronic controller includes a processor architecture comprising at least one processor device, and at least one memory clement associated with the processor architecture. The at least one memory element stores processor-executable instructions

3 that, when executed by the processor architecture, provide an JOB compensation module that estimates a current JOB value that indicates an amount of active insulin in the body of the user, calculates an JOB rate based at least in part on the estimated current JOB
value, and determines an adjusted insulin infusion rate based at least in part on the calculated JOB rate and an uncompensated insulin infusion rate. The JOB
compensation module selects a final insulin infusion rate for the insulin infusion device, wherein the final insulin infusion rate is selected as either the determined adjusted insulin infusion rate, the uncompensated insulin infusion rate, or a current basal rate. The JOB
compensation module then provides the selected final insulin infusion rate to regulate delivery of insulin by the insulin infusion device.
[0011] An exemplary embodiment of an electronic device is also provided here. The electronic device includes a processor architecture having at least one processor device, and at least one memory element associated with the processor architecture.
The at least one memory element stores processor-executable instructions that, when executed by the processor architecture, perform a method of controlling an insulin infusion device for a user. The method operates the insulin infusion device in a closed-loop mode to deliver insulin to the body of the user, obtains current insulin-delivered data that indicates an amount of insulin delivered by the insulin infusion device during a most recent sampling period, obtains current sensor data that indicates a current sensor glucose value for the user corresponding to the most recent sampling period, and processes historical insulin-delivered data and historical sensor data, for a plurality of historical sampling periods prior to the most recent sampling period, to obtain predicted sensor glucose values for a historical time period. The method continues by calculating a difference between the current sensor glucose value and a predicted current sensor glucose value for the most recent sampling period, wherein the predicted sensor glucose values for the historical time period include the predicted current sensor glucose value. The method continues by generating an alert when the difference exceeds a threshold error amount.
[0012] The following detailed description also relates to a tangible and non-transitory electronic storage medium having processor executable instructions that, when executed by a processor architecture comprising at least one processor device, perform a method of controlling an insulin infusion device for a user. The method involves operation of the insulin infusion device in a closed-loop mode to deliver insulin to the body of the user.
The method continues by identifying, from historical sensor glucose values for the user, a baseline historical sensor glucose value obtained during a begin-training sampling period.

4 The method calculates a plurality of candidate solutions to a sensor glucose prediction model, wherein each of the plurality of candidate solutions is calculated as a function of a bounded initial condition and historical insulin delivered data for the user, and wherein the bounded initial condition is influenced by the baseline sensor glucose value. The method continues by selecting a best-matched solution from the calculated plurality of candidate solutions, based on a comparison of predicted sensor glucose values from the calculated plurality of candidate solutions to a first portion of the historical sensor glucose values. The predicted sensor glucose values from the best-matched solution are compared to a second portion of the historical sensor glucose values, wherein the first portion of the historical sensor glucose values corresponds to a distant history period, the second portion of the historical sensor glucose values corresponds to a recent history period, and the distant history period occurred before the recent history period that data samples. The method continues by generating an alert, in response to the comparing, when the second portion of the historical sensor glucose values deviates from the best-matched solution by at least a threshold error amount.
100131 Also presented here is an embodiment of an electronic controller for an insulin infusion device. The electronic controller includes a processor architecture comprising at least one processor device, and at least one memory element associated with the processor architecture. The at least one memory element stores processor-executable instructions that, when executed by the processor architecture, provide a model supervisor module to obtain, during closed-loop operation of the insulin infusion device, insulin-delivered data that indicates an amount of insulin delivered by the insulin infusion device during a most recent sampling period, and current sensor data that indicates a current sensor glucose value for the user corresponding to the most recent sampling period. The model supervisor module defines a model training period and a model prediction period for a historical period of time, and finds a best-matched solution to a sensor glucose prediction model, relative to historical sensor glucose values obtained during the model training period, wherein the best-matched solution is a function of a baseline sensor glucose value obtained during the model training period, and is a function of historical insulin-delivered data for the user obtained during the historical period of time. The model supervisor module compares at least one predicted sensor glucose value from the best-matched solution to at least one historical sensor glucose value corresponding only to the model prediction period, and generates an alert, in response to the comparing, when the at least one historical sensor glucose value deviates from the at least one predicted sensor glucose value by at least a threshold error amount.
10014] Also included below is a detailed description of a processor-implemented method of controlling an insulin infusion device for a user. The method may begin by operating the insulin infusion device in a closed-loop mode to deliver insulin to the body of the user. The method continues by obtaining current insulin-delivered data that indicates an amount of insulin delivered by the insulin infusion device during a most recent sampling period, obtaining current sensor data that indicates a current sensor glucose value for the user corresponding to the most recent sampling period, and processing historical insulin-delivered data and historical sensor data, for a plurality of historical sampling periods prior to the most recent sampling period, to obtain predicted sensor glucose values for a historical time period. The method then calculates a difference between the current sensor glucose value and a predicted current sensor glucose value for the most recent sampling period, wherein the predicted sensor glucose values for the historical time period include the predicted current sensor glucose value. An alert is generated when the difference exceeds a threshold error amount.
100151 Also included below is a detailed description of a processor-implemented method of controlling an insulin infusion device for a user. The method may begin by operating the insulin infusion device in a closed-loop mode to deliver insulin to the body of the user. The method continues by identifying, from historical sensor glucose values for the user, a baseline historical sensor glucose value obtained during a begin-training sampling period. Next, a plurality of candidate solutions to a sensor glucose prediction model is calculated, wherein each of the plurality of candidate solutions is calculated as a function of a bounded initial condition and historical insulin delivered data for the user, and wherein the bounded initial condition is influenced by the baseline sensor glucose value. The method continues by selecting a best-matched solution from the calculated plurality of candidate solutions, based on a comparison of predicted sensor glucose values from the calculated plurality of candidate solutions to a first portion of the historical sensor glucose values. At least one predicted sensor glucose value from the best-matched solution is compared to a second portion of the historical sensor glucose values, wherein the first portion of the historical sensor glucose values corresponds to a distant history period, the second portion of the historical sensor glucose values corresponds to a recent history period, and the distant history period occurred before the recent history period that data samples. An alert is generated, in response to the comparing, when the second portion of the historical sensor glucose values deviates from the best-matched solution by at least a threshold error amount.
[0016] Another embodiment of a processor-implemented method of controlling an insulin infusion device for a user is also presented below. The method involves operating the insulin infusion device in a closed-loop mode to deliver insulin to the body of the user, defining a model training period and a model prediction period for a historical period of time, and finding a best-matched solution to a sensor glucose prediction model, relative to historical sensor glucose values obtained during the model training period, wherein the best-matched solution is a function of a baseline sensor glucose value obtained during the model training period, and is a function of historical insulin-delivered data for the user obtained during the historical period of time. The method continues by comparing at least one predicted sensor glucose value from the best-matched solution to at least one historical sensor glucose value corresponding only to the model prediction period. An alert is generated, in response to the comparing, when the at least one historical sensor glucose value deviates from the at least one predicted sensor glucose value by at least a threshold error amount [0017] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.
BRIEF DESCRIPTION OF THE DRAWINGS
[0018] A more complete understanding of the subject matter may be derived by referring to the detailed description and claims when considered in conjunction with the following figures, wherein like reference numbers refer to similar elements throughout the figures.
[0019] FIG. 1 is a block diagram of a closed loop glucose control system in accordance with an embodiment of the present invention.
[0020] FIG. 2 is a front view of closed loop hardware located on a body in accordance with an embodiment of the present invention.
[0021] FIG. 3A is a perspective view of a glucose sensor system for use in an embodiment of the present invention.
[0022] FIG. 3B is a side cross-sectional view of the glucose sensor system of FIG.
3A.

100231 FIG. 3C is a perspective view of a sensor set of the glucose sensor system of FIG. 3A for use in an embodiment of the present invention.
[0024] FIG. 3D is a side cross-sectional view of the sensor set of FIG. 3C.
[0025] FIG. 4 is a cross sectional view of a sensing end of the sensor of FIG. 3D.
[0026] FIG. 5 is a top view of an infusion device with a reservoir door in the open position, for use in an embodiment of the present invention.
[0027] FIG. 6 is a side view of an infusion set with the insertion needle pulled out, for use in an embodiment of the present invention.
[0028] FIG. 7 is a circuit diagram of a sensor and its power supply in accordance with an embodiment of the present invention.
[0029] FIG. 8A is a diagram of a single device and its components in accordance with an embodiment of the present invention.
[0030] FIG. 8B is a diagram of two devices and their components in accordance with an embodiment of the present invention.
[0031] FIG. 8C is another diagram of two devices and their components in accordance with an embodiment of the present invention.
[0032] FIG. 8D is a diagram of three devices and their components in accordance with an embodiment of the present invention.
[0033] FIG. 9 is a table listing the devices of FIGS. 8A-D and their components.
[0034] FIG. 10 is a block diagram of the glucose sensor system of FIG. 3A.
[0035] FIG. 11A is a detailed block diagram of an A/D converter for the glucose sensor system of FIG. 10 in accordance with an embodiment of the present invention.
100361 FIG. 11B is a detailed block diagram of the A/D converter for the glucose sensor system of FIG. 10 with a pulse duration output selection option in accordance with an embodiment of the present invention.
[0037] FIG. 12 is a circuit diagram of an I-F A/D converter of FIG. 10 accompanied by charts of node signals in accordance with an embodiment of the present invention.
[0038] FIG. 13 is another circuit diagram of an I-F A/D converter of FIG.

accompanied by charts of node signals in accordance with an embodiment of the present invention.
[0039] FIG. 14 is still another circuit diagram of an I-F A/D converter of FIG. 10 accompanied by charts of node signals in accordance with an embodiment of the present invention.

[0040] FIG. 15 is a circuit diagram of an I-V A/D converter of FIG. 10 in accordance with an embodiment of the present invention.
[0041] FIG. 16 is a block diagram of the glucose sensor system of FIG. 10 with a pre-filter and a filter in accordance with an embodiment of the present invention.
[0042] FIG. 17 is a chart of an example of a pre-filter of FIG. 16 and its effects on digital sensor values Dsig in accordance with an embodiment of the present invention.
[0043] FIG. 18 is frequency response chart for a filter of FIG. 16 in accordance with an embodiment of the present invention.
[0044] FIG. 19A is a plot of a filtered and an unfiltered sensor signal over time in accordance with an embodiment of the present invention.
[0045] FIG. 19B is close up of a section of the plot of FIG. 19A in accordance with an embodiment of the present invention.
[0046] FIG. 20 is a cross-sectional view of a sensor set and an infusion set attached to the body in accordance with an embodiment of the present invention.
[0047] FIG. 21 is a frequency response chart of a time delay correcting Weiner filter in accordance with an embodiment of the present invention.
[0048] FIG. 22 is a plot of a digital sensor values Dsig before and after time delay correction compared to actual glucose measurements over time in accordance with an embodiment of the present invention.
[0049] FIG. 23A is a diagram of a glucose clamp (glucose level with respect to time).
[0050] FIG. 23B is a plot of insulin concentration in a normal glucose tolerant (NGT) individual in response to various magnitudes of glucose clamps of FIG. 23A.
100511 FIG. 24A is a diagram of a glucose clamp.
[0052] FIG. 24B is a diagram of a proportional insulin response to the glucose clamp of FIG. 24A in accordance with an embodiment of the present invention.
[0053] FIG. 24C is a diagram of an integral insulin response to the glucose clamp of FIG. 24A in accordance with an embodiment of the present invention.
[0054] FIG. 24D is a diagram of a derivative insulin response to the glucose clamp of FIG. 24A in accordance with an embodiment of the present invention.
[0055] FIG. 24E is a diagram of a combined proportional, integral, and derivative insulin response to the glucose clamp of FIG. 24A in accordance with an embodiment of the present invention.
[0056] FIG. 25A is a plot of insulin responses to a glucose clamp for exercise trained and normal individuals.

100571 FIG. 25B is a bar chart of glucose uptake rates for exercise trained and normal individuals.
[0058] FIG. 26 is a block diagram of a closed loop system to control blood glucose levels through insulin infusion based on glucose level feedback in accordance with an embodiment of the present invention.
[0059] FIG. 27 is a detailed block diagram of the portion of the control loop of FIG.
26 that is in the body in accordance with an embodiment of the present invention.
[0060] FIGS. 28A and 28B are plots of measured insulin responses of two different normal glucose tolerant (NGT) individuals to a glucose clamp for use with an embodiment of the present invention.
[0061] FIG. 29A is a plot of two different glucose sensor outputs compared to glucose meter readings during a glucose clamp in accordance with an embodiment of the present invention.
10062] FIG. 29B is a plot of actual insulin concentration in blood compared to a controller commanded insulin concentration in response to the glucose clamp of FIG.
29A in accordance with an embodiment of the present invention.
[0063] FIG. 30 is a top view of an end of a multi-sensor for measuring both glucose concentration and pH in accordance with an embodiment of the present invention.
[0064] FIG. 31A is a representative drawing of blood glucose compared to sensor measured blood glucose over time in accordance with an embodiment of the present invention.
[00651 FIG. 31B is a representative drawing of sensor sensitivity over the same period of time as FIG. 31A in accordance with an embodiment of the present invention.
[0066] FIG. 31C is a representative drawing of sensor resistance over the same period of time as FIG. 31A in accordance with an embodiment of the present invention.
[0067] FIG. 32 is a block diagram using the derivative of sensor resistance to determine when to recalibrate or replace the sensor in accordance with an embodiment of the present invention.
[0068] FIG. 33A is a plot of an analog sensor signal Isig over time in accordance with an embodiment of the present invention.
10069] FIG. 33B is a plot of sensor resistance over the same period of time as FIG.
32A in accordance with an embodiment of the present invention.
[0070] FIG. 33C is a plot of the derivative of the sensor resistance of FIG. 32B in accordance with an embodiment of the present invention.

[0071] FIG. 34A is a bottom view of a telemetered characteristic monitor in accordance with an embodiment of the present invention.
[0072] FIG. 34B is a bottom view of a different telemetered characteristic monitor in accordance with an embodiment of the present invention.
[0073] FIG. 35A is a drawing of a blood plasma insulin response to a glucose clamp in a normal glucose tolerant (NGT) individual in accordance with an embodiment of the present invention.
[0074] FIG. 35B is a drawing of the blood plasma insulin response of FIG.
35A when delayed due to insulin being delivered to the subcutaneous tissue instead of directly into the blood stream in accordance with an embodiment of the present invention.
[0075] FIG. 36A is a drawing of blood plasma insulin concentration over time after an insulin bolus is delivered directly into the blood stream in accordance with an embodiment of the present invention.
[0076] FIG. 36B is a drawing of a blood plasma insulin concentration over time after an insulin bolus is delivered into the subcutaneous tissue in accordance with an embodiment of the present invention.
[0077] FIG. 37 is a block diagram of the closed loop system of FIG. 26 with the addition of a post-controller compensator and a derivative filter in accordance with an embodiment of the present invention.
[0078] FIG. 38A is a plot of sensor signal measurements and Via measurements with respect to time in accordance with an embodiment of the present invention.
[0079] FIG. 38B is a plot of a measured counter electrode voltage Vent with respect to time in accordance with an embodiment of the present invention.
[0080] FIG. 38C is a plot of calculated sensor sensitivity with respect to time in accordance with an embodiment of the present invention.
100811 FIG. 38D is a plot of a calculation of sensor resistance Rsi with respect to time in accordance with an embodiment of the present invention.
100821 FIG. 38E is a plot of another calculation of sensor resistance Rs2 with respect to time in accordance with an embodiment of the present invention.
100831 FIG. 38F is a plot of the derivative of sensor resistance Rsi of FIG. 38D with respect to time in accordance with an embodiment of the present invention.
[0084] FIG. 38G is a plot of the derivative of the sensor resistance Rs2 of FIG. 38E
with respect to time in accordance with an embodiment of the present invention.

[0085] FIG. 38H is a plot of when sensors were replaced with respect to time in accordance with an embodiment of the present invention.
[0086] FIGS. 39A and 39B are a block diagrams of a closed loop glucose control system in accordance with embodiments of the present invention.
[0087] FIG. 40 is a block diagram of auto blood withdrawal and return in accordance with an embodiment of the present invention.
100881 FIG. 41A is a plot actual blood glucose concentration in accordance with an embodiment of the present invention.
[0089] FIG. 41B is a plot of actual insulin concentration in blood compared to a controller commanded insulin concentration in response to the blood glucose in FIG. 41A
in accordance with an embodiment of the present invention.
100901 FIG. 42 illustrates a control feedback block diagram of state variable feedback and in accordance with an embodiment of the present invention.
[0091] FIG. 43 is a plot of basal insulin delivery rate over time using different control gains in accordance with embodiments of the present invention.
[0092] FIG. 44 is a plot of subcutaneous insulin over time using different control gains in accordance with embodiments of the present invention.
[0093] FIG. 45 is a plot of plasma insulin over time using different control gains in accordance with embodiments of the present invention.
[0094] FIG. 46 is a plot of insulin effect over time using different control gains in accordance with embodiments of the present invention.
[0095] FIG. 47 is a plot of simulated glucose concentration over time using a PID
controller with state variable feedback and a PID controller without state variable feedback in accordance with embodiments of the present invention.
[0096] FTG. 48 is a plot of simulated insulin delivery over time using a PID controller with state variable feedback and a PTD controller without state variable feedback in accordance with embodiments of the present invention.
[0097] FIG. 49 is a block diagram that illustrates processing modules and algorithms of an exemplary embodiment of a closed-loop system controller.
100981 FIG. 50 is a flow chart that illustrates an exemplary embodiment of a control process for an insulin infusion device.
[0099] FIG. 51 is a graph of integral clip value versus sensor glucose levels.
[00100] FIG. 52 is a block diagram that schematically illustrates an exemplary embodiment of an insulin on board (I0B) compensation module.

100101] FIG. 53 is a flow chart that illustrates an exemplary embodiment of an compensation process.
[00102] FIG. 54 is a diagram that depicts certain time events associated with the operation of a model supervisor module.
[00103] FIG. 55 is a flow chart that illustrates an exemplary embodiment of a sensor model supervision process.
[00104] FIG. 56 is a flow chart that illustrates an exemplary embodiment of a sensor model training process, which may be performed in conjunction with the sensor model supervision process depicted in FIG. 55.
[00105] FIG. 57 is a diagram that illustrates two exemplary fault conditions that can be detected by the model supervisor module.
DETAILED DESCRIPTION
[00106] The following detailed description is merely illustrative in nature and is not intended to limit the embodiments of the subject matter or the application and uses of such embodiments. As used herein, the word "exemplary" means "serving as an example, instance, or illustration." Any implementation described herein as exemplary is not necessarily to be construed as preferred or advantageous over other implementations.
Furthermore, there is no intention to be bound by any expressed or implied theory presented in the preceding technical field, background, brief summary or the following detailed description.
[00107] Techniques and technologies may be described herein in terms of functional and/or logical block components, and with reference to symbolic representations of operations, processing tasks, and functions that may be performed by various computing components or devices. Such operations, tasks, and functions are sometimes referred to as being computer-executed, computerized, software-implemented, or computer-implemented. It should be appreciated that the various block components shown in the figures may be realized by any number of hardware, software, and/or firmware components configured to perform the specified functions. For example, an embodiment of a system or a component may employ various integrated circuit components, e.g., memory elements, digital signal processing elements, logic elements, look-up tables, or the like, which may carry out a variety of functions under the control of one or more microprocessors or other control devices.

[00108] When implemented in software or firmware, various elements of the systems described herein are essentially the code segments or instructions that perform the various tasks. The program or code segments can be stored in any tangible and non-transitory processor-readable medium. The "processor-readable medium" or "machine-readable medium" may include any medium that can store or transfer information.
Examples of the processor-readable medium include an electronic circuit, a semiconductor memory device, a ROM, a flash memory, an erasable ROM (EROM), a floppy diskette, a CD-ROM, an optical disk, a hard disk, or the like.
1001091 The various tasks performed in connection with a process described herein may be performed by software, hardware, firmware, or any combination thereof.
It should be appreciated that a described process may include any number of additional or alternative tasks, the tasks shown in a particular figure need not be performed in the illustrated order, and a described process may be incorporated into a more comprehensive procedure or process having additional functionality not described in detail herein.
Moreover, one or more of the tasks shown in the figures could be omitted from an embodiment of a described process as long as the intended overall functionality remains intact.
1001101 As shown in the drawings for purposes of illustration, the invention is embodied in a closed loop infusion system for regulating the rate of fluid infusion into a body of a user based on feedback from an analyte concentration measurement taken from the body. In particular embodiments, the invention is embodied in a control system for regulating the rate of insulin infusion into the body of a user based on a glucose concentration measurement taken from the body. In preferred embodiments, the system is designed to model a pancreatic beta cell (0-cell). In other words, the system controls an infusion device to release insulin into a body of a user in a similar concentration profile as would be created by fully functioning human 0-cells when responding to changes in blood glucose concentrations in the body.
100111] Thus, the system simulates the body's natural insulin response to blood glucose levels and not only makes efficient use of insulin, but also accounts for other bodily functions as well since insulin has both metabolic and mitogcnic effects. However, the algorithms must model the 13-cells closely, since algorithms that are designed to minimize glucose excursions in the body, without regard for how much insulin is delivered, may cause excessive weight gain, hypertension, and atherosclerosis.
In preferred embodiments of the present invention, the system is intended to emulate the in vivo insulin secretion pattern and to adjust this pattern consistent with the in vivo 13-cell adaptation experienced by normal healthy individuals. The in vivo 3-cell response in subjects with normal glucose tolerance (NGT), with widely varying insulin sensitivity (SI), is the optimal insulin response for the maintenance of glucose homeostasis.
1001121 Preferred embodiments include a glucose sensor system 10, a controller and an insulin delivery system 14, as shown in FIG. 1. The glucose sensor system 10 generates a sensor signal 16 representative of blood glucose levels 18 in the body 20, and provides the sensor signal 16 to the controller 12. The controller 12 receives the sensor signal 16 and generates commands 22 that are communicated to the insulin delivery system 14. The insulin delivery system 14 receives the commands 22 and infuses insulin 24 into the body 20 in response to the commands 22.
100113] Generally, the glucose sensor system 10 includes a glucose sensor, sensor electrical components to provide power to the sensor and generate the sensor signal 16, a sensor communication system to carry the sensor signal 16 to the controller 12, and a sensor system housing for the electrical components and the sensor communication system.
[00114] Typically, the controller 12 includes controller electrical components and software to generate commands for the insulin delivery system 14 based on the sensor signal 16, and a controller communication system to receive the sensor signal 16 and carry commands to the insulin delivery system 14.
[00115] Generally, the insulin delivery system 14 includes an infusion device and an infusion tube to infuse insulin 24 into the body 20. In particular embodiments, the infusion device includes infusion electrical components to activate an infusion motor according to the commands 22, an infusion communication system to receive the commands 22 from the controller 12, and an infusion device housing to hold the infusion device.
[00116] In preferred embodiments, the controller 12 is housed in the infusion device housing and the infusion communication system is an electrical trace or a wire that carries the commands 22 from the controller 12 to the infusion device. In alternative embodiments, the controller 12 is housed in the sensor system housing and the sensor communication system is an electrical trace or a wire that carries the sensor signal 16 from the sensor electrical components to the controller electrical components.
In other alternative embodiments, the controller 12 has its own housing or is included in a supplemental device. In another alternative embodiment, the controller is located with the infusion device and the sensor system all within one housing. In further alternative embodiments, the sensor, controller, and/or infusion communication systems may utilize a cable, a wire, fiber optic lines, RF, IR, or ultrasonic transmitters and receivers, or the like instead of the electrical traces.
[00117] System Overview [00118] Preferred embodiments of the invention include a sensor 26, a sensor set 28, a telemetered characteristic monitor 30, a sensor cable 32, an infusion device 34, an infusion tube 36, and an infusion set 38, all worn on the body 20 of a user, as shown in FIG. 2. The telemetered characteristic monitor 30 includes a monitor housing 31 that supports a printed circuit board 33, batteries 35, antenna (not shown), and a sensor cable connector (not shown), as seen in FIGS. 3A and 3B. A sensing end 40 of the sensor 26 has exposed electrodes 42 and is inserted through skin 46 into a subcutaneous tissue 44 of a user's body 20, as shown in FIGS. 3D and 4. The electrodes 42 are in contact with interstitial fluid (ISF) that is present throughout the subcutaneous tissue 44. The sensor 26 is held in place by the sensor set 28, which is adhesively secured to the user's skin 46, as shown in FIGS. 3C and 3D. The sensor set 28 provides for a connector end 27 of the sensor 26 to connect to a first end 29 of the sensor cable 32. A second end 37 of the sensor cable 32 connects to the monitor housing 31. The batteries 35 included in the monitor housing 31 provide power for the sensor 26 and electrical components 39 on the printed circuit board 33. The electrical components 39 sample the sensor signal 16 and store digital sensor values (Dsig) in a memory and then periodically transmit the digital sensor values Dsig from the memory to the controller 12, which is included in the infusion device.
[00119] The controller 12 processes the digital sensor values Dsig and generates commands 22 for the infusion device 34. Preferably, the infusion device 34 responds to the commands 22 and actuates a plunger 48 that forces insulin 24 out of a reservoir 50 located inside the infusion device 34, as shown in FIG. 5. In particular embodiments, a connector tip 54 of the reservoir 50 extends through the infusion device housing 52 and a first end 51 of the infusion tube 36 is attached to the connector tip 54. A
second end 53 of the infusion tube 36 connects to the infusion set 38. Insulin 24 is forced through the infusion tube 36 into the infusion set 38 and into the body 20. The infusion set 38 is adhesively attached to the user's skin 46, as shown in FIG. 6. As part of the infusion set 38, a cannula 56 extends through the skin 46 and terminates in the subcutaneous tissue 44 completing fluid communication between the reservoir 50 and the subcutaneous tissue 44 of the user's body 20.
[00120] In alternative embodiments, the closed-loop system can be a part of a hospital-based glucose management system. Given that insulin therapy during intensive care has been shown to dramatically improve wound healing, reduce blood stream infections, renal failure, and polyneuropathy mortality, irrespective of whether subjects previously had diabetes (See Van den Berghe G. et al., NEJM 345: 1359-67, 2001), the present invention can be used in this hospital setting to control the blood glucose level of a patient in intensive care. In these alternative embodiments, since an intravenous (IV) hookup is typically implanted into a patient's arm while the patient is in an intensive care setting (e.g., ICU), a closed loop glucose control can be established which piggy-backs off the existing IV connection. Thus, in a hospital based system, IV catheters which are directly connected to a patient vascular system for purposes of quickly delivering IV fluids, can also be used to facilitate blood sampling and direct infusion of substances (e.g., insulin, anticoagulants) into the intra-vascular space.
Moreover, glucose sensors may be inserted through the IV line to give real-time glucose levels from the blood stream. Therefore, depending on the type of hospital based system, the alternative embodiments would not necessarily need the described system components such as the sensor 26, the sensor set 28, the telemetered characteristic monitor 30, the sensor cable 32, the infusion tube 36, and the infusion set 38 as described in the preferred embodiments. Instead, standard blood glucose meters or vascular glucose sensors as described in provisional application entitled "Multi-lumen Catheter," filed Sep. 27, 2002, Ser. No. 60/414,248, can be used to provide the blood glucose values to the infusion pump control and the existing IV
connection can be used to administer the insulin to the patient.
[00121] It is important to appreciate that numerous combinations of devices in the hospital-based system can be used with the closed loop controller of the present invention. For example, as described in FIG. 39B compared to the preferred system in FIG. 39A, an auto blood glucose/intravenous insulin infusion system can automatically withdraw and analyze blood for glucose concentration at fixed intervals (preferably 5-20 minutes), extrapolate the blood glucose values at a more frequent interval (preferably 1 minute), and use the extrapolated signal for calculating an IV-insulin infusion according to the controller described below. The modified auto blood glucose/intravenous insulin infusion system would eliminate the need for subcutaneous sensor compensation and subcutaneous insulin compensation (as described with regards to the lead-lag compensator below). The automatic withdrawal of blood, and subsequent glucose determination can be accomplished with existing technology (e.g. VIA or Biostator like blood glucose analyzer) or by the system described in FIG. 40. The system in FIG. 40 uses a peristaltic pump 420 to withdraw blood across an amperometric sensor 410 (the same technology as used in sensor 26) and then return the blood with added flush (0.5 to 1.0 ml) from the reservoir 400. The flush can consist of any makeup of saline, heparin, glucose solution and/or the like. If the blood samples are obtained at intervals longer than 1 minute but less than 20 minutes, the blood glucose determinations can be extrapolated on a minute-to-minute basis with extrapolation based on the present (n) and previous values (n-1) to work with the logic of the controller as described in detail below. For blood samples obtained at intervals greater than 20 minutes, a zero-order-hold would be used for the extrapolation. Based on these blood glucose values, the infusion device can administer insulin based on the closed loop controller described in greater detail below.
[00122] In other modifications to the system, a manual blood glucose/intravenous insulin infusion system can be used where frequent manual entry of blood glucose values from a standard blood glucose meter (e.g., YSI, Beckman, etc.) and extrapolate the values at more frequent intervals (preferably 1 min) to create a surrogate signal for calculating IV-insulin infusion. Alternatively, a sensor blood glucose/intravenous insulin infusion system can use a continuous glucose sensor (e.g., vascular, subcutaneous, etc.) for frequent blood glucose determination. Moreover, the insulin infusion can be administered subcutaneously rather than intravenously in any one of the previous examples according to the controller described below.
[00123] In still further alternative embodiments, the system components may be combined in a smaller or greater number of devices and/or the functions of each device may be allocated differently to suit the needs of the user.
[00124] Controller [00125] Once the hardware for a closed loop system is configured, such as in the preferred embodiments described above, the effects of the hardware on a human body are determined by the controller. In preferred embodiments, the controller 12 is designed to model a pancreatic beta cell (P-cell). In other words, the controller 12 commands the infusion device 34 to release insulin 24 into the body 20 at a rate that causes the insulin concentration in the blood to follow a similar concentration profile as would be caused by fully functioning human 13-cells responding to blood glucose concentrations in the body 20. In further embodiments, a "semi-closed-loop" system may be used, in which the user is prompted to confirm insulin delivery before any insulin is actually delivered.
[00126] A controller that simulates the body's natural insulin response to blood glucose levels not only makes efficient use of insulin but also accounts for other bodily functions as well since insulin has both metabolic and mitogenic effects.
Controller algorithms that are designed to minimize glucose excursions in the body without regard for how much insulin is delivered may cause excessive weight gain, hypertension, and atherosclerosis. In preferred embodiments, of the present invention, the controller 12 is intended to emulate the in vivo insulin secretion pattern and to adjust this pattern to be consistent with in vivo 13-cell adaptation. The in vivo 13-cell response in subjects with normal glucose tolerance (NGT), with widely varying insulin sensitivity (Si), is the optimal insulin response for the maintenance of glucose homeostasis.
[00127] The 13-Cell and PID Control [00128] Generally, the in vivo 13-cell response to changes in glucose is characterized by "first" and "second" phase insulin responses. This biphasic insulin response is clearly seen during hyperglycemic clamps applied to NGT subjects, as shown in FIG.
23B.
During a hyperglycemic clamp the glucose level is rapidly increased from a basal level GB to a new higher level Gc and then held constant at the higher-level Gc as shown in FIG. 23A. The magnitude of the increase in glucose (AG) affects the insulin response.
Four insulin response curves are shown for four different glucose clamp levels in FIG.
23B.
[00129] The biphasic insulin response of a 13-cell can be modeled using components of a proportional, plus integral, plus derivative (PID) controller. A PID
controller is selected since PID algorithms are stable for a wide variety of non-medical dynamic systems, and PID algorithms have been found to be stable over widely varying disturbances and changes in system dynamics.
[00130] The insulin response of13-cells during a hyperglycemic clamp is diagrammed in FIGS. 24A-E using the components of a PID controller to model the 13-cell.
A
proportional component Up and a derivative component Up of the ND controller may be combined to represent a first phase insulin response 440, which lasts several minutes. An integral component U1 of the PID controller represents a second phase insulin response 442, which is a steady increase in insulin release under hyperglycemic clamp conditions.
The magnitude of each component's contribution to the insulin response is described by the following equations:

[00131] Proportional Component Response: Up = Kp(G ¨ GB) [00132] Integral Component Response: U, itt (G ¨ GB) dt + 'B , and [00133] Derivative Component Response: UD =

K D daGt [00134] Where 1001351 Up is the proportional component of the command sent to the insulin delivery system, [00136] Uf is the integral component of the command sent to the insulin delivery system, 1001371 Up is the derivative component of the command sent to the insulin delivery system, [00138] Kp is a proportional gain coefficient, [00139] K, is an integral gain coefficient, 1001401 Kp is a derivative gain coefficient, [00141] G is a present blood glucose level, [00142] GB is a desired basal glucose level, [00143[ t is the time that has passed since the last sensor calibration, [00144] to is the time of the last sensor calibration, and [00145] IB is a basal insulin concentration at to, or can also be described as U4I0).
[00146] The combination of the PID components that model the two phases of insulin response by a n-cell is shown in FIG. 24E as it responds to the hyperglycemic clamp of FIG. 24A. FIG. 24E shows that the magnitude of the first phase response 440 is driven by the derivative and proportional gains, Kp and Kp. And the magnitude of the second phase response 442 is driven by the integral gain 1(1.
[001471 The components of the PID controller can also be expressed in its discrete form:
[00148] Proportional Component Response: P
- conn = K p(SG7 ¨ G5p) [00149] Integral Component Response: icõõn = icon (sGf n _ Lisp), i , con =
[00150] Derivative Component Response: Dconn = KDdGdefl [00151] Where Kp, K1, and Kp are the proportional, integral, and derivative gain coefficients, SGf and dGdtf are the filtered sensor glucose and derivative respectively, and the superscript n refers to discrete time.

[00152] An acute insulin response is essential for preventing wide postprandial glycemic excursions. Generally, an early insulin response to a sudden increase in glucose level results in less total insulin being needed to bring the glucose level back to a desired basal glucose level. This is because the infusion of insulin increases the percentage of glucose that is taken up by the body. Infusing a large amount of insulin to increase the percentage of glucose uptake while the glucose concentration is high results in an efficient use of insulin. Conversely, infusing a large amount of insulin while the glucose concentration is low results in using a large amount of insulin to remove a relatively small amount of glucose. In other words, a larger percentage of a big number is more than a larger percentage of a small number. The infusion of less total insulin helps to avoid development of insulin resistance in the user. As well, first-phase insulin is thought to result in an early suppression of hepatic glucose output.
1001531 Insulin sensitivity is not fixed and can change dramatically in a body depending on the amount of exercise by the body. In one study, for example, insulin responses in highly exercise-trained individuals (individuals who trained more than 5 days a week) were compared to the insulin responses in subjects with normal glucose tolerance (NGT) during a hyperglycemic clamp. The insulin response in exercise-trained individuals 444 was about one-half of the insulin response of the NGT subjects 446, as shown in FIG. 25A. But the glucose uptake rate for each of the individuals (exercise-trained 448 or normal 450) was virtually identical, as shown in FIG. 25B.
Thus, it can be speculated that the exercise-trained individuals have twice the insulin sensitivity and half of the insulin response leading to the same glucose uptake as the NGT
individuals. Not only is the first phase insulin response 440 reduced due to the effects of exercise, but the second phase insulin response 442 has also been shown to adjust to insulin sensitivity, as can be seen in FIG. 25A.
1001541 In preferred embodiments, a closed loop control system may be used for delivering insulin to a body to compensate for 0-cells that perform inadequately. There is a desired basal blood glucose level GB for each body. The difference between the desired basal blood glucose level GB and an estimate of the present blood glucose level G is the glucose level error GE that must be corrected. The glucose level error GE is provided as an input to the controller 12, as shown in FIG. 26.
[00155] If the glucose level error GE is positive (meaning that the present estimate of the blood glucose level G is higher than the desired basal blood glucose level GB) then the controller 12 generates an insulin delivery command 22 to drive the infusion device 34 to provide insulin 24 to the body 20. In terms of the control loop, glucose is considered to be positive, and therefore insulin is negative. The sensor 26 senses the 1SF
glucose level and generates a sensor signal 16. The sensor signal 16 is filtered and calibrated to create an estimate of the present blood glucose level 452. In particular embodiments, the estimate of the present blood glucose level G is adjusted with correction algorithms 454 before it is compared to the desired basal blood glucose level GB to calculate a new glucose level error GE to start the loop again.
[00156] If the glucose level error GE is negative (meaning that the present estimate of the blood glucose level is lower than the desired basal blood glucose level GB) then the controller 12 reduces or stops the insulin delivery depending on whether the integral component response of the glucose error GE is still positive.
1001571 If the glucose level error GE is zero, (meaning that the present estimate of the blood glucose level is equal to the desired basal blood glucose level GB) then the controller 12 may or may not issue commands to infuse insulin depending on the derivative component (whether the glucose level is raising or falling) and the integral component (how long and by how much glucose level has been above or below the basal blood glucose level GB). In "semi-closed loop" embodiments, the user is prompted before the controller 12 issues the commands to infuse insulin. The prompts may be displayed to the user on a display, sounded to the user, or otherwise provide an indication to the user that the system is ready to deliver insulin, for example a vibration or other tactile indication. In addition, the amount of insulin to be delivered may be displayed, with or without other information, such as the total amount infused for the day or the potential effect on the user's blood glucose level by the insulin delivery. In response, the user may indicate that the insulin should or should not be delivered, for example by selecting a button, key, or other input. In further embodiments, there must be at least two keystrokes so that insulin is not delivered by accident.
[00158] To more clearly understand the effects that the body has on the control loop, a more detailed description of the physiological effects that insulin has on the glucose concentration in the interstitial fluid (ISF) is needed. In preferred embodiments, the infusion device 34 delivers insulin through the cannula 56 of the infusion set 38 into the ISF of the subcutaneous tissue 44 of the body 20. And the insulin 24 diffuses from the local ISF surrounding the cannula into the blood plasma and then spreads throughout the body 20 in the main circulatory system, as described in the block diagram of FIG. 27. The insulin then diffuses from the blood plasma into the interstitial fluid ISF
substantially throughout the entire body. The insulin 24 binds with and activates membrane receptor proteins on cells of body tissues. This facilitates glucose permeation into the activated cells. In this way, the tissues of the body 20 take up the glucose from the ISF. As the ISF
glucose level decreases, glucose diffuses from the blood plasma into the ISF
to maintain glucose concentration equilibrium. Finally, the glucose in the ISF permeates the sensor membrane and affects the sensor signal 16.
[00159] In addition, insulin has direct and indirect effects on liver glucose production.
Increased insulin concentration decreases liver glucose production. Therefore, acute and immediate insulin response not only helps the body to efficiently take up glucose but also substantially stops the liver from adding to the glucose in the blood stream.
In alternative embodiments, insulin is delivered more directly into the blood stream instead of into the interstitial fluid, such as delivery into veins, arteries, the peritoneal cavity, or the like.
And therefore, any time delay associated with moving the insulin from the interstitial fluid into the blood plasma is diminished. In other alternative embodiments, the glucose sensor is in contact with blood or body fluids other than interstitial fluid, or the glucose sensor is outside of the body and measures glucose through a non-invasive means. The embodiments that use alternative glucose sensors may have shorter or longer delays between the blood glucose level and the measured blood glucose level.
[00160] Selecting Controller Gains [00161] In preferred embodiments, the controller gains Kp, lc, and KD, are selected so that the commands from the controller 12 cause the infusion device 34 to release insulin 24 into the body 20 at a rate, that causes the insulin concentration in the blood to follow a similar concentration profile, as would be caused by fully functioning human 13-cells responding to blood glucose concentrations in the body. In preferred embodiments, the gains may be selected by observing the insulin response of several normal glucose tolerant (NGT) individuals, with healthy normally functioning 13-cells. The first step in determining a set of controller gains is to take periodic measurements of blood glucose and blood insulin concentrations from the group of NGT individuals. Second, each individual in the group is subjected to a hyperglycemic clamp, while continuing to periodically measure and record the blood glucose and blood insulin concentrations.
Third, a least squares curve fit is applied to the recorded blood insulin concentrations measured over time for each individual. The result is a set of curves representing the insulin responses to the hyperglycemic clamp for each individual of the group.
Fourth, the curves are used to calculate the controller gains Kp, KJ, and KD, for each individual. And finally, the proportional gains from each of the individuals are averaged together to obtain an average proportional gain, Kp, to be used in a controller 12. Similarly, the integral gains, K.1, and the derivative gains, KD, are averaged to obtain an average integral gain, Ict, and an average derivative gain, KD, for the controller 12. Alternatively, other statistical values may be used instead of averages such as, maximums, minimums, the high or low one, two or three sigma standard deviation values, or the like.
The gains calculated for various individuals in a group may be filtered to remove anomalous data points before statistically calculating the gains to be used in a controller.
[00162] In an example, a least squares curve-fitting method is used to generate representative insulin response curves from two fasted individuals in a group, as shown in FIGS. 28A and B. Then the controller gains were calculated from the insulin response curves of the two representative individuals and are shown in Table 1. When calculating the controller gains, the insulin clearance rate (k), was assumed to be 10 (ml of insulin)/min/(kg. of body weight). The insulin clearance rate k is the rate that insulin is taken out of the blood stream in a body. Finally, the average value for each type of gain is calculated using the measurements from the group, as shown in Table 1.
Individuals Proportional Gain, Kp Integral Gain, K1 Derivative Gain, KD
a 0.000406 0.005650 0.052672 0.000723 0.003397 0.040403 Average 0.000564 0.004523 0.046537 Table 1. MD Controller Gains Calculated From The Insulin Response Curves Of Two NGT Individuals [00163] The controller gains may be expressed in various units and/or may be modified by conversion factors depending on preferences for British or S. I.
Units, floating-point or integer software implementation, the software memory available, or the like. The set of units for the controller gains in Table I is:
[00164] Kp: (mU of insulin)/min/(Kg of body weight) per (mg of glucose)/(d1 of plasma);
[00165] Kr: (mU of insulin)/minl(Kg of body weight) per (mg of glucose)i(d1 of plasma) min.; and 100166] KD: (mU of insulin)/min/(Kg of body weight) per (mg of glucose)/(d1 of plasma)/min.

1001671 In alternative embodiments, other curve fitting methods are used to generate the insulin response curves from the measurements of blood insulin concentrations.
[00168] An estimate of an insulin clearance rate (k), the individual's body weight (W), and the insulin sensitivity SI are needed to calculate the controller gains from the insulin response curves for each NGT individual. The insulin clearance rate (k) is generally proportional to body weight and is well documented in literature. The individual's insulin sensitivity S1 may be measured using an intravenous glucose tolerance test, a hyperinsulinemic clamp, or in the case of a diabetic, comparing the individual's daily insulin requirement to their daily carbohydrate intake.
[00169] In particular embodiments, two parameters, the insulin sensitivity Si and the insulin clearance rate k, arc measured for each individual. In other embodiments, the insulin clearance rate k is estimated from literature given the individual's body weight. In other particular embodiments, longer or shorter insulin clearance times are used. In still other embodiments, all of the parameters are estimated. In additional embodiments, one or more parameters are measured, while at least one parameter is estimated from literature.
[00170] In other alternative embodiments, the controller gains are calculated using a group of individuals with similar body types. For example, the insulin response to a hyperglycemic clamp may be measured for several tall, thin, NGT, males in order to calculate the controller insulin response gains for each individual in the group. Then the gains are statistically combined to generate a set of representative controller gains for tall, thin, NGT, males. The same could be done for other groups such as, but not limited to, short, heavy, NGT, females; medium height, medium weight, highly exercised trained, females; average height and weight 10 year olds; or the like. Then the controller gains are selected for each individual user based on the group that best represents them. In further alternative embodiments, controller gains are uniquely selected for each individual user.
In particular embodiments, the controller gains for a user are selected based on measurements of insulin sensitivity, insulin clearing time, insulin appearance time, insulin concentration, body weight, body fat percentage, body metabolism, or other body characteristics such as pregnancy, age, heart conditions, or the like.
[00171] In other alternative embodiments, the controller gains are estimated as a function of a user's body weight W and insulin sensitivity S1. A series of observations are used to justify this method. The first observation is that the controller gains are proportional to each other. In other words, small changes in glucose concentration cause a small derivative response UD, a small proportional response Up and a small integral response Ur. And larger changes in glucose concentration cause a proportionally larger derivative response UD, a proportionally larger proportional Up response and a proportionally larger integral response Ur, as shown in FIG. 23B. Changes in the glucose concentration proportionally affect all three components of the controller response UpLo.
The second observation is that the first phase insulin response (0) is proportional to the derivative gain KD. And the third observation is that two constants may be readily obtained from information in published literature or may be measured from a cross-section of the general population. The two constants are the insulin clearance rate (k) for a human given a body weight and the disposition index (DI) for a human given a change in glucose concentration.
[00172] While there are multiple sources for the information needed to calculate the insulin clearance rate k, one source is the article "Insulin clearance during hypoglycemia in patients with insulin-dependent diabetes mellitus", written by Kollind M et al., published in Horm Metab Res, 1991 July; 23(7):333-5. The insulin clearance rate k is obtained from the insulin infused divided by the steady state plasma insulin concentration. An insulin clearance constant Ak, which is independent of an individual's body weight, may be obtained by dividing the insulin clearance rate k (measured from a particular individual) by the individual's body weight. The insulin clearance constant Ak is generally the same for all humans, except under extenuating circumstances such as after an individual has contracted HIV, other metabolic affecting diseases, or the like.
[00173] The disposition index (DI) for a human given a change in glucose concentration is available from information presented in the article "Quantification of the relationship between insulin sensitivity and beta-cell function in human subjects.
Evidence for a hyperbolic function", written by Khan S E et al., published in Diabetes, 1993 November; 42(11):1663-72.
[00174] Both the disposition index DI and the insulin clearance rate k may be measured directly from tests. The disposition index DI may be calculated given the first phase insulin response measured form a glucose clamp test and the individual's insulin sensitivity measured from an insulin sensitivity test. The insulin clearance rate k may be measured from an insulin clearance test. The glucose clamp test and the insulin clearance test are described in the above-mentioned articles and are well known in the art. The insulin sensitivity S1 may be measured using an intravenous glucose tolerance test or a hyperinsulinemic clamp test.
[00175] Given these observations, then the following parameters may be measured from an NGT individual's insulin response to a glucose clamp: a desired first phase insulin response 4)1, the ratio of KD to Kp, and the ratio of KD to lc. Then the derivative gain KD may be calculated from the first phase insulin response 4)1 using the constants k and DI. And finally Kp and KI may be calculated using the ratios of KD to Kp and KD to K1.
[00176] The first phase insulin response 4)1 may be observed in a NGT
individual as the area under the insulin response curve during approximately the first 10 minutes of a glucose clamp. The increase in the glucose concentration during the glucose clamp is AG=(G¨GB), where G is equal to Gc, the glucose concentration during the clamp, and GB
is the basal glucose concentration before the clamp.
[00177] The importance of the first phase insulin response 4)1 has been emphasized by studies indicating that, in subjects with normal glucose tolerance (NGT), the product of first phase insulin response 411 and insulin sensitivity (Si) is a constant known as the disposition index, DI=4)1S1. Therefore, 01 =
[00178] For a different AG there is a different (1)1 and therefore a different DI. But, the ratio DIIAG is substantially constant even for different individuals with different insulin sensitivities.
[00179] The insulin sensitivity SI is defined as the percentage of the glucose concentration that the body tissues will take up for a given amount of insulin. The 3-cell naturally adapts to changes in insulin sensitivity by adjusting the amount of insulin it secretes during the first phase insulin response 4)1. This suggests that the body naturally seeks an optimal level of glucose tolerance. A controller that mimics this characteristic of the p-cell more accurately simulates the body's natural insulin response.
100180] The instantaneous insulin response (RI) may be calculated given the insulin clearance rate (k) and the first phase insulin response 4)1, RI=141.
[00181] The insulin clearance rate k is proportional to body weight (W), therefore substituting a proportional constant Ak and the user's body weight W fork and replacing DI
4)1 with the ratio of DI over S1 yields the following equation: R, =
St [00182] The instantaneous insulin response R1 may also be expressed as the product of the derivative gain KD and the change in glucose concentration AG, RI=KDAG.
[00183] Setting the two equations for R1 equal to each other and solving for KD yields, = w AkD1 KD
Si AG
[00184] As mentioned above, DI/AG and Ak are constants available or calculated from AkDi data in published literature. Combining the constants into a single constant, Q, Q
AG ' yields an equation for the derivative gain KD that is a function of the user's body weight W and the user's insulin sensitivity Si, KD = Q
[00185] Once the derivative gain KD is calculated, the proportional and integral gains arc calculated using ratios. The ratio of KD/Kp can be set to the dominant time constant for insulin action, ranging from 10-60 minutes, but more typically 20-40 minutes and preferably 30 minutes. For example, calculating Kp given KD using a time constant of 30 KD KD
minutes, yields the following relationship: ¨ = 30 Kp = ¨ . In a similar fashion, Kp 30 the ratio of K0/K.1 can be set to the average ratio measured from a population of NGT
individuals. And K1can be calculated from KD.
[00186] In particular embodiments, the user enters their body weight W and insulin sensitivity Si into the device that contains the controller. Then the controller gains are automatically calculated and used by the controller. In alternative embodiments, an individual enters the user's body weight W and insulin sensitivity S1 into a device and the device provides the information to the controller to calculate the gains.
[00187] A study was conducted to confirm that the insulin response for an individual could be reproduced using the glucose sensor as an input. In the study, glucose and insulin measurements were taken while a hyperglycemic clamp was applied to a NGT
individual. The glucose level measurements, shown in FIG. 29A, were used as the inputs to a mathematical model created to simulate a PID insulin response controller.
The insulin dosing commanded by the controller in response to the glucose clamp very closely approximates the actual insulin appearance in the NGT individual, as shown in FIG. 29B.
The insulin concentration measured from periodic blood samples 456 taken from the individual during the test are represented by dots in FIG. 29B. The output from the mathematical model simulating the insulin response commanded by the controller is shown as a solid line 458 in FIG. 29B.

[00188] Three different devices were used to measure the individual's blood glucose during the study. Blood glucose meter readings 460 from periodic blood samples taken from the individual are represented by the dots in FIG, 29A. Two MiniMecl sensors (such as those described in the section entitled "sensor", below) were placed in the individual's subcutaneous tissue, and the sensor readings 462, 464 are shown as lines in FIG. 29A.
The sensor readings 462, 464 are slightly delayed compared to the meter readings 460.
The delay is most likely due to the delay between blood glucose and interstitial fluid (ISF) glucose and can be substantially corrected through the use of a filter if needed. In this study, the delay was not corrected by a filter and did not significantly affect the controller's ability to command an insulin response that matches the natural response of the NGT individual. This study indicates that the PID insulin response controller model is a good minimal model of insulin secretion that captures the biphasic response of healthy I3-cells. Correction of the delay is only expected to increase the accuracy of the model.
[00189] Fuzzy Logic to Select Between Multiple Sets of Controller Gains [00190] In preferred embodiments, one set of controller gains is used for a particular individual. In alternative embodiments, more than one set of controller gains is used, and fuzzy logic is used to select between sets of controller gains and to determine when to change from one set of controller gains to another. In particular alternative embodiments, the controller gains are different if the glucose level is above or below the desired glucose basal level. In other alternative embodiments, the controller gains are different if the glucose level is increasing or decreasing. A justification for different sets of gains comes from physiological studies that indicate that ft-cells turn off faster than they turn on. In still other alternative embodiments, the controller gains are different depending on whether the glucose level is above or below the desired glucose basal level and whether the glucose level is increasing or decreasing, which results in four sets of controller gains.
In additional alternative embodiments, the controller gains change depending on the magnitude of the hypoglycemic excursion. In other words, the controller gains for small changes in glucose are different than those for large changes in glucose.
[00191] Self-Tuning Controller Gains [00192] Further embodiments may include a controller that self-tunes one or more the gains, Kp, K1, KD to accommodate changes in insulin sensitivity. In particular embodiments, previous measurements of glucose levels are compared to the desired basal glucose level GB. For example, the desired basal glucose level GB is subtracted from the previous glucose level measurements. Then any negative values, within a predefined time window, are summed (in essence integrating the glucose level measurements that were below the basal glucose level GB). If the resulting sum is greater than a pre-selected hypoglycemic integral threshold, then the controller gains are increased by a factor (1+a).
Conversely, if the integral of the glucose level measurements that were measured above the basal glucose level GB within the predefined time window is greater than a pre-selected hyperglycemic integral threshold, then the controller gains are decreased by a factor (1¨a).
[00193] In particular embodiments, the predefined time window over which the glucose concentration integrals are evaluated is generally 24 hours, and the controller gains are adjusted if needed at the end of each predefined time window. In alternative embodiments, the integrals of the glucose level measurements are continuously calculated over a moving window of time, and if either integral exceeds a threshold, the gains are immediately adjusted. In particular embodiments, the moving time window is one hour, and the time window may be restarted whenever the gains are adjusted. In other alternative embodiments, the time window is longer or shorter depending on the sensor accuracy, the rate at which an individual's insulin sensitivity changes, the computational capabilities of the hardware, or the like.
[00194] In particular embodiments, the adjustment amount (a) is 0.01. In alternative embodiments, the adjustment amount a is greater or smaller depending on the sensor accuracy, the rate at which an individual's insulin sensitivity changes, the rate at which the sensor sensitivity Si changes, or the like. In still other alternative embodiments, the adjustment amount a is made larger or smaller depending on the amount that the integral of the measured glucose levels exceeds a threshold. In this way, the gains are adjusted by greater amounts if the measured glucose level G is significantly deviating from the desired blood glucose level GB and less if the measured glucose level G is closer to the desired blood glucose level G. In additional alternative embodiments, the controller employs a Kalman filter.
[00195] State Variable Feedback [00196] While the primary signal determining the 3-cell's insulin response is glucose, there also exists a putative effect of insulin per se to inhibit insulin secretion. This effect may be directly related to the concentration of insulin in plasma (IP(t)), or mediated through some signal proportional to insulin effect (IEFF(t)). The 13-cell can likely directly sense these signals (i.e., directly sense insulin concentration and secondary signals proportional to insulin effect such as free fatty acid). Feedback from these intermediary signals is analogous to what is known as state variable feedback; that is feedback, whereby the variable being controlled (glucose in this case) is used together with feedback of each intermediary signal that affects the variable (insulin concentration in plasma and interstitial fluid). With this type of feedback, undesirable slow kinetic process can be made to appear much faster than they are. For example, if 13-cell insulin secretion were inhibited by a signal proportional to insulin concentration in the interstitial fluid where it acts, the delay between plasma and interstitial insulin could be made to appear to be shorter. For the artificial closed-loop algorithm, or for "semi-closed-loop" algorithms, this beneficial effect can be achieved by using "state observers"
(mathematical equations that predict the insulin concentration in various parts of the body knowing the history of past insulin delivery). In "semi-closed loop" algorithms, the algorithms arc the same as for closed loop algorithms but there is a user confirmation step before any insulin is actually administered. By using state variable feedback, it is possible to make the insulin in an insulin pump act faster than the insulin actually is.
[00197] To estimate subcutaneous insulin concentration, plasma insulin concentration, and insulin effect, the following equations may be used:
1001981 ¨disc = ¨ isc) dt dlp 1001991 ¨ = a2(Jsc ¨ p) at dl 1002001 ¨ = a( 1p IEF) dt 1002011 Wherein 'Sc is the estimate of normalized insulin concentration in the subcutaneous space, Ip is the estimate of normalized insulin concentration in the plasma, IEF is the estimate of insulin effect on glucose, al is the rate constant between insulin delivery and the subcutaneous insulin compartment, a2 is the rate constant between subcutaneous insulin and plasma compartments, a3 is the rate constant between the plasma compartment and the insulin effect. ID is the delivered insulin, which can be a function of the three state variables OK, 1p, and [00202] In particular embodiments, an open loop fixed base rate plus user requested bolus would result in the bolus being increased a certain amount and the basal rate subsequently decreased the same amount in accordance to the following formula:
[00203] /D' = (1 + +Y2 + y3)/D ¨
-sc Ygp Y3IEF
[00204] Wherein ID is the user requested basal (Ur%) plus bolus (U) profile and ID' is the state feedback adjusted profiles. Note that for a given kinetic excursion the total amount of insulin requested (area under curve of ID) and delivered (area under curve of ID') is identical. Here, 71, 72, and y3 are state-feedback gains (scalars).
Careful choice of these gains the pump to correct its delivery rate to compensate for delays associated with the dispersion of insulin from the bolus injection into the subcutaneous layer of the patient, to the plasma, and to its actual insulin effect/action on the body.
Thus, by estimating how much insulin from a bolus is in the subcutaneous layer, the plasma, or is actually acting on the patient's glucose level (state variables Isc, Ip and 'SF), it is possible to optimize delivery of insulin over time to the patient. Using state feedback the bolus is increased by an amount (1+y1+72+73) that is gradually taken away from future insulin delivery (-Ow - 72Ip - TtIH.). As a result, the apparent insulin pharmolcinetic curve appears faster. This is akin to developing a faster acting insulin, but it is achieved algorithmically by rearranging the distribution of the insulin delivery per unit bolus by delivering more upfront and removing the extra amount at a later time. The three gains can be chosen to move the time delays (Put, 1/02, and 1/03) to any arbitrary locations. In control theory, this is known as pole placement.
1002051 State feedback can be used in open loop and closed loop insulin delivery algorithms and with "semi-closed-loop" delivery algorithms. State feedback can be used in conjunction with a Proportional-Integral-Derivative (Pa) or any other type of closed loop controller. yl is the feedback gain multiplied to ISO, 72 is the feedback gain multiplied to Ip, and 73 is the feedback gain multiplied to IEF-[00206] The physical state space form directly taken from the equations above is:
{[iiscl = raa, ¨0a2 0 H,r1+,71 3] . ID
r 0 a3 -a3 i [/EF] [ 0 -I or (i. = Ax + Bu i EF
'Sc [1 [y = Cx + du ID = [0 0 01 = [ Ip + 1:11. ID

[00207] The finite difference form is calculated as follows (wherein cx indicates an exponential function):
Define: k1 = e'ir , k2 =. e-a2r, k3 = e-air = (1 - ki)(4)(i - 1)) + k1/5(i - 1) (eq lb) 113(0 = (1¨ k 2)(1.5 c (0) + k 21 p(i ¨ 1) (eq 2b) 47(0 = (1 ¨ k3)(4,(0) k3/EF(i ¨1) (eq 3b) 1002081 The Laplace Form is as follows, wherein s represents the Stackel determinant used in Laplace equations:
'Sc - ¨ (eq 1c) /D s+al az - = (eq 2c) isc s-Faz IEFF a3 (eq 3c) 1p s+a3 1p aia =2 (eq 4) ID (s+ai)(s+crz) IEFF a1a2a3 (eq 5) (s+ai)(s+ ce2)(s+ a3) [00209] To obtain the transfer function of insulin delivery with state feedback, the control equation is as follows, wherein E represents the error between the actual glucose concentration and the desired glucose concentration (G¨G):
/D = PID=E¨Y
,i-sc Y2lp Y31EFF (eq 6) Substituting equations (eq lc), (eq 4) and (eq 5) into (eq 6) and rearranging, the following transfer functions are obtained, wherein GM is a gain multiplier:
= (GM)(PID)(s+ai)(s+a2)(s+a3) E (s+a1)(s+az)(s+a3)+a1ri(s+a2)(s+a3)+a1a2y2(s+a3)+a1a2a3y3 (eq 7) 'Sc (GM)(PID)a1(s+a2)(s+a3) ¨

E = (s+a1)(s+a2)(s+a3)+a1lt1 (s+a2)(s+a3)+aiazY2 (s+a3)+a1a2a3Y3 (eq 8) p = (G WW1 D)ala2(s+a3) E (s+a1)(s+a2)(s+a3)+a1n(s+a2)(s+a3)+a1a272 (s+a3)+a1a2 az Y3 (eq 9) IEFF (GM)(PID)a1a2a3 ¨ = (eq 10) (s+a1)(s+a2)(s+a3)+a1r1(s+a2)(s+a3)+cr1a2Y2(s+a3)+a1a2(r3Y3 [00210] The computation of the gain multiplier is also obtained in the state variable feedback method. When state variable feedback is used, the gain multiplier (GM) is a scalar that forces a step response to reach the same steady value whether state feedback is used or not. In other words, GM ensures that the total amount given per unit of bolus will be the same in both cases. In the case of state feedback, more insulin is given up front, but this extra insulin is taken away later. To calculate GM in particular embodiments, the "final value theorem" from control systems is used. The final value theorem states that to evaluate the steady state of any transfer function T(s) given any input X(s), the steady state output response to the input is given by:
Yss(t 00) = lims,o(sT(s)X(s)) 1002111 The Laplace form of a step input is given by X(s) = is and the steady state solution of the final value theorem simplifies to:
Yss(t ¨" ) = lims-,o(T(s)).
In the case when there is no state feedback, (R, 72 and y3 = 0), the steady state solution may be obtained from equation (eq 7) to be as follows:
(t ¨> 00) = 1 (eq 11) With state feedback without the gain correction factor, the steady state solution is:
(eq 12) 1.-4-y1+),2+Y3 GM is then evaluated as the ratio of equation (eq 12) to equation (eq 11) to obtain the following: GM=1-1-71+72+y3.
[00212] Using state variable feedback, a closed loop control equation and state feedback gain are determined for pole placement. Specifically, the gains are calculated for the insulin delivery equation shown above. In particular embodiments, they are determined as follows: First, with state feedback, the denominator of equations (eq 7), (eq 8), (eq 9), and (eq 10) is:
D = s3 + (al + a2 + a3 + yial)s2 +
(a1a2 + (a1 + a2)a3 + y2a1a2 + (a2 + a3)y1 ai)s +
(a1a2a3 + y3ai a2 a3 + y2ai a2a3 + ai a2a3) (eq 14) 100213] To get the poles of the system in the equations (eq 7), (eq 8), (eq 9), or (eq 10), D may be set equal to zero yield the characteristic equation:
53 + (a1 + a2 + a3 + y1a1)s2 +
(a1a2 + (al + a2)a3 + y2a1a2 + (a2 + a3)y1a1)s +
(a1a2a3 + y3a1a2a3 + y2aia2a3 + ylaia2a3) = 0 (eq 16) If the desired system poles or roots of (eq 16) are defined by eigenvalues ki, kz and X3, then the characteristic equation can be written as:
(s ¨ A1)(s ¨ A2)(s ¨ 23) = 0 Expanding and collecting like powers of s, (eq 16) can be written as:
¨(At + A2 +113)s2 +(111A2 +111113 + A2A3).3 A1/12A3 = 0 (eq 17) [00214] Setting the coefficients of like powers of s equal to each other we have the system of equations:
al + a2 + a3 + yiai = ¨(A1 + 112+2.3) (eq 18) a2 + a1a3 + a2a3 + na1a2 + al (a2 + a3) =
A.1.12 + A1.1.3 + A2A3 (eq 19) a1a2a3 + y3a1a2a3 + y2a1a2a3 + y1a1a2a3 ¨ 111112113 (eq 20) [00215] This results in three equations and three unknowns, 71, 72 and 73.
Therefore, the unknown gains can be solved for in terms of the desired poles 2i,A.2, X3, and system time constants, al, az and al. These formulas enable us to control the desired pharmacokinetics of insulin as it appears in the different compartments:
= ¨(Ai+A2+A,i-crii-a2-Fa3) Yi al Ava2+Ai,13 +A22.3 -a1a2 -al a3-az a3 +(Ai +.12 +.13 +al +a2+er3)(a2+a3) Y2 =
cri az -111213 AiA2+AiA3+,12A3-aia2-aia3-a2a3+Gti +12 -F13+ai +a2 +a3)(a2 +a3) y3¨ a1a2a3 a1a2 (11+12+13+a1+a2+(x3) al [00216] Thus, through the above calculations, the gains can be calculated and used in the control equation for insulin delivery of:
/D = PID=E¨v i-I
sc Y2IP Y3IEF
[00217] PID is the output of a PID controller of any other closed loop (or "semi-closed-loop") controller. The gains are generally calculated just once, but could be calculated more often if desired. The control equation may be calculated on a repeated basis, after predetermined periods of time or continuously. For example, and without limitation, it may be calculated every five, ten, thirty or sixty minutes.
Just the state variable portion (v t-/
sc Ygp Y3-1EF) may be updated or the entire equation may be updated. By updating the control equation, it is possible to continually improve the delivery of insulin to the patient.
[00218] A control feedback block diagram of an embodiment of a pump using state variable feedback is shown in FIG. 42. As shown, the desired glucose Go 600 of the patient is entered into the PID Controller 610. The output of the PID
controller is an insulin delivery value ID 601. The block then calculates how much insulin should actually be delivered to the patient as a bolus in addition to the insulin delivery value and how much should be taken away from the basal rate, as discussed above. At each discrete time interval, Ti, (Ti 620, T2 630, and T3 640), the amount of insulin that has entered into the subcutaneous layer from the pump is calculated to provide Isc 602. That value will be multiplied (or otherwise factored by) 71 605 and subtracted from the output of the PID
controller to provide an improved desired insulin value based on the subcutaneous insulin concentration (with the other calculations following). At each discrete time interval Ti, the amount of insulin that has entered into the plasma from the subcutaneous compartment is calculated to provide Ip 603. That value will be multiplied (or otherwise factored by) 72 606 and subtracted from the output of the P1D controller to determine an improved desired insulin value based on the plasma insulin concentration. At each discrete time interval Ti, the amount of insulin actually going into action or the effective insulin compartment from the insulin in the plasma is calculated to provide IEF 604. That value will be multiplied (or otherwise factored by) 73 607 and subtracted from the output of the PID controller to determine an improved desired insulin value based on the effective insulin. The insulin actually delivered to the subject 650 will then change the blood glucose G of the user 608, which will then be measured by the sensor 660 and compared to the desired glucose 600.
1002191 FIGS. 43-46 show graphs with the effect of state feedback. FIG. 43 shows the effect on the basal insulin delivery rate achieved using the algorithm described above. A
bolus is given at time zero. Line 700 shows the insulin delivery when no state feedback is used. This line would be the same as a regular delivery of an insulin bolus and is indicated as 0.0000, because it does not change the amount of basal rate being delivered.
The other three lines illustrate the change in insulin delivery rate over time when all of the state feedback is placed in one of the gains 71, 72, or 73. As can be seen, if all the state feedback is placed in the gain 71 (for the subcutaneous layer), the basal insulin delivery rate 701 (in relation to the standard basal rate) starts out low and gradually moves to a limit of zero, or the rate without state feedback, as steady state is reached.
If all of the state feedback is placed in the gain 72 (for the plasma layer), the basal insulin delivery rate 702 starts at zero, dips lower, and then gradually returns up to a limit of zero as steady state is reached. If all of the state feedback is placed in the gain 73 (for the insulin action/effect), the basal insulin delivery rate 703 starts at zero, dips lower, but more slowly than for the all 72 delivery rate, and then gradually returns up to a limit of zero as steady state is reached. In all cases, the total delivery of insulin is the same.
1002201 FIG. 44 shows the effect of state feedback per unit bolus on the subcutaneous insulin. In other words, a bolus of insulin is given to a patient at time zero and the figure shows the rate in which the amount of insulin in the subcutaneous layer, from that bolus, decreases to zero. Line 705 shows the amount of insulin in the subcutaneous layer over time with no state feedback. Line 706 shows the amount of insulin in the subcutaneous layer over time when all of the state feedback is placed in gain 71. Line 707 shows the amount of insulin in the subcutaneous layer over time when all of the state feedback is placed in gain 72. Line 708 shows the amount of insulin in the subcutaneous layer over time when all of the state feedback is placed in gain 73.

[00221] FIG. 45 shows the effect of state feedback per unit bolus on the plasma insulin. In other words, a bolus of insulin is given to a patient at time zero and the figure shows the rate in which the amount of insulin in the plasma layer, from that bolus, increases from zero (there is a slight delay from injecting insulin to when the insulin moves into the plasma from the subcutaneous layer), reaches its peak and then returns to zero. Line 710 shows the amount of insulin in the plasma over time with no state feedback. Line 711 shows the amount of insulin in the plasma over time when all of the state feedback is placed in gain YI. Line 712 shows the amount of insulin in the plasma over time when all of the state feedback is placed in gain 72. Line 713 shows the amount of insulin in the plasma over time when all of the state feedback is placed in gain 73.
[00222] FIG. 46 shows the effect of state feedback per unit bolus on the insulin effect.
In other words, a bolus of insulin is given to a patient at time zero and the figure shows the rate in which the amount of insulin from that bolus creates the insulin effect on the body, starting at zero (there is a delay from the injection of insulin into the subcutaneous layer and through the plasma to the insulin effect), rising to its maximum point, and decreasing to zero. Line 715 shows the insulin effect over time with no state feedback.
Line 716 shows the insulin effect over time when all of the state feedback is placed in gain 'yr. Line 717 shows the insulin effect over time when all of the state feedback is placed in gain 72. Line 718 shows the insulin effect over time when all of the state feedback is placed in gain 73.
[00223] FIGS. 47 and 48 compare insulin state variable feedback used in conjunction with a PID closed loop controller as opposed to use of a PID closed loop controller alone (with no insulin state variable feedback). FIG. 47 shows the simulated glucose concentration of a patient over time. Meals are given at 8, 13, 18, 22, and 32 hours. The glucose concentration using the PID with insulin state feedback is shown as line 800. The glucose concentration using the PID without insulin state feedback is shown as line 801.
With glucose concentrations, it is always preferable to keep a patient's concentrations from being too high or too low, so the more that the closed loop program can avoid high and low values, the better. As can be seen in FIG. 47, as time goes on, the glucose concentration using the PID with insulin state feedback improves over time (versus the glucose concentration using the PID without insulin state feedback) in that it varies less as time goes on, keeping the patient with a more steady glucose level that will greatly reduce hyper- and hypoglycemic events. FIG. 48 shows average simulated insulin delivery profiles from the same system as FIG. 47. Line 810 represents the insulin delivery using the PID with insulin state feedback. Line 811 represents the insulin delivery using the PID without insulin state feedback. As can be seen, the insulin delivery using the PID
with insulin state feedback contains more spikes and dips, resulting from the state feedback.
[00224] Modifying the PID Controller to Incorporate an Integrator Leak In preferred embodiments, the PID control response was described with constant gain components, Kp, K1, KD. Although the preferred control response guarantees zero steady-state error (i.e., steady state glucose minus a desired basal glucose (GB)41), inherently, the integral component Ul = K 5(G ¨ GB)dt + U1(t0) destabilizes feedback control because there is no temporal wind down of the insulin response while the integral component models the increase in the insulin response. Without any correction, the integral component has a tendency to over-estimate the increase in the insulin response.
Since a small difference between steady-state glucose and GB is typically acceptable in insulin response control, an alternative modeling of the integral component can incorporate an integrator leak to reduce the magnitude of the destabilizing effect.
Specifically, changes in WO can be described by a term proportional to the error in glucose and a term that leaks in proportion to the magnitude of U1. This can be expressed in the formula:
di Ki(G ¨GB) ¨ KLEAK ;
dt with initial condition Ui(to).
[00225] The parameter KLEAK is the reciprocal time constant of the rate of leaking (TT FAK in min=1/KLEAK), where TT PAK is a tuning parameter that can be set based on empirical data, and be tied with the other gain components Kp, K1, K0.
However, the current realization of the artificial 13-cell has TLIAK as a user input. U1 can also be expressed in discrete form by standard methods.
[00226] Post-Controller (Lead/Lag) Compensator 1002271 In preferred embodiments, commands are issued from the controller without regard to where in the body the insulin delivery system will infuse the insulin. In essence, the assumption is that the insulin is either delivered directly into the blood stream for immediate use by the body, or that any time delays caused by delivering the insulin somewhere in the body other than the blood stream can be compensated for by adjusting Kp, K1, and KD. In this case, the commands generally model a13-cell insulin secretion profile, an example of which is shown in FIG. 35A. And since the 3-cells secrete insulin directly into the blood stream, thef3-cell insulin secretion profile is the intended blood plasma insulin concentration profile. However, an insulin delivery delay may distort the intended blood plasma insulin concentration profile, as shown in FIG. 35B. The insulin delivery delay is the amount of time between the instant that the command is given to the insulin delivery system to infuse insulin and the time that insulin reaches the blood plasma. An insulin delivery delay may be caused by a diffusion delay, represented by a circle with an arrow 528 in FIG. 20, which is the time required for insulin that has been infused into a tissue to diffuse into the blood stream. Other contributors to insulin delivery delay may include, time for the delivery system to deliver the insulin to the body after receiving a command to infuse insulin, time for the insulin to spread throughout the circulatory system once it has entered the blood stream, and/or by other mechanical or physiological causes. In addition, the body clears insulin even while an insulin dose is being delivered from the insulin delivery system into the body. Since insulin is continuously cleared from the blood plasma by the body, an insulin dose that is delivered to the blood plasma too slowly or is delayed is at least partially, if not significantly, cleared before the entire insulin dose fully reaches the blood plasma. And therefore, the insulin concentration profile in the blood plasma never achieves the same peak (nor follows the same profile) it would have achieved if there were no delay. Given an insulin dose delivered all at once into the blood plasma at time zero, the insulin concentration in the blood plasma is raised virtually instantaneously (not shown) and then would decrease exponentially over time as the body clears (uses or filters out) the insulin, as shown in FIG. 36A per equation: Cp = ,where:
Cp is is the concentration of insulin in the blood plasma, To is a mass of the insulin dose delivered directly to the blood plasma at time zero, VP is a volume of the blood plasma in the body, P1 is a reciprocal time constant for insulin clearance, and t is the time that has passed since the delivery of the insulin dose directly into the blood plasma.
1002281 The time constant for insulin clearance Pi may be calculated using the following equation: P1 = ¨, where:

k is the volume insulin clearance rate, and V, is a volume of the blood plasma in the body.
[00229] Or the time constant for insulin clearance Pi may be obtained by providing insulin to an individual that does not generate his own insulin, and then periodically testing blood samples from the individual for insulin concentration. Then, using an exponential curve fitting routine, generate a mathematical expression for a best-fit curve for the insulin concentration measurements, and observe the time constant in the mathematical expression.
[00230] Given the same insulin dose (delivered at time zero all at once) into the subcutaneous tissue, instead of directly into the blood plasma, the concentration of insulin in the blood plasma would begin to rise slowly as insulin diffuses from the interstitial fluid ISF into the blood plasma, as shown in FIG. 36B. At the same time that insulin is entering the blood plasma, the body is clearing insulin from the blood. While the rate at which insulin is entering the blood plasma exceeds the insulin clearance rate, the insulin concentration in the blood plasma continues to increase. When the insulin clearance rate exceeds the rate at which insulin is entering the blood plasma from the interstitial fluid ISF, the insulin concentration in the blood plasma begins to decrease. So, the result of delivering insulin into the interstitial fluid ISF instead of directly into the blood stream is that the insulin concentration in the blood plasma is spread over time rather than increased virtually instantaneously to a peak followed by a decay.
[00231] A bi-exponential equation may be used to model the insulin concentration in blood plasma given an insulin dose delivered to the subcutaneous tissue:
Cp = ______ 10D (e-'2t e-)3t) , where:
VpVISF(P3-P2) Cp is the concentration of insulin in the blood plasma, is the mass of the insulin dose delivered to the subcutaneous tissue at time zero, D is a diffusion coefficient (the rate at which insulin diffuses from the interstitial fluid ISF into the blood glucose) V, is a volume of the blood plasma in the body, VisF is a volume of interstitial fluid ISF that the insulin is delivered to, P2 is a time constant, P3 is a time constant greater than or equal to P2, and t is time since the delivery of the insulin dose into the interstitial fluid ISF.
[00232] The time constants may be calculated using the quadratic formula:
al+ jai-4a0 P2/ P3 = __ 2 ,where:
D+K
al = ¨ ¨ , and VP VISF
ao (D+x) D D2 VP !SF) V ISFVP
1002331 In alternative embodiments, a post-controller lead-lag compensator 522 is used to modify the commands (Um)) to compensate for the insulin delivery delay and/or the insulin clearance rate k, as shown in FIG. 37. The post-controller lead-lag compensator 522 is of the form Ucomp = ¨s+a where 1/a and 1/)' are the lead and lag constants UPID S-Fy respectively, s is the Laplace variable, and UCOMP is the compensated commands calculated by the lead-lag compensator 522.
[00234] The PID controller generates commands (Upm) for a desired insulin delivery rate into the blood plasma. The commands Um are calculated and issued periodically depending on the update rate for the control loop, which is selected based on a maximum anticipated rate of change of the blood glucose level, an insulin delivery system minimum insulin dosage, insulin sensitivity, a maximum and a minimum acceptable glucose concentration, or the like. The commands Um) are used as inputs to the post-controller lead-lag compensator 522.
[00235] In particular embodiments, the compensated commands (Ucomp) issued from the post-controller lead-lag compensator 522 uses more than one value from the controller. In particular embodiments, post-controller lead-lag compensator 522 uses the present command (UpiDn) and the previous command (Uppn-1) to calculate a compensated command 1.1comP per a compensation equation:

UCOMPn = (1 Y)IICOMPn I UPIDn + (1 a)UPIDn-1 , where:
UnDn is the present command, UpID"1 is the previous command, Ucompn I is the previous compensated control output, a is the reciprocal lead time constant in min I, and is the reciprocal lag time constant in min-1.
1002361 This is a first forward difference equation. However, other forms can be used alternatively (e.g., first backward or bilinear), but all result in a compensated control output (Ucomp) that is comprised of a weighted history of both past MD outputs (Um), and past compensated outputs (UcomP).
1002371 An alternative method of modifying the commands (Upm) to compensate for the insulin delivery delay and/or the insulin clearance can be performed based on a weighted history of past insulin delivery. By giving the most recent delivery history more weight, the weighted history of the previous insulin delivered can then be subtracted from the present PID control output to yield a compensated control output.
Expressed in Laplace domain this results in: Ucomp = PID E ¨aticoMP , where E is the Laplace transformed error signal (G¨GB), k determines how much the PID output is reduce in proportion to the weighted history of past control outputs, and a is the reciprocal time constant determining how long a history is weighted (the preferred value of a would be equal to the reciprocal dominant time constant or subcutaneous insulin appearance, P2).
Solving the compensated signals as a function of the error results in:
u(s) s+avv s+aw ¨ = PID - = PID
E(s) s+(a+A) s+y which is identical to the previously described lead-lag compensation.
1002381 In other alternative embodiments, additional previous command values may be used. In still other alternative embodiments, the compensation equation compensates for both time constants P2 and P3.
[00239] In still more alternative embodiments, the controller gains are modified to include the effects of the post-controller lead/lag compensator so that the post-controller leadilag compensator is not needed to modify the commands to account for the insulin delivery delay.
[00240] In particular embodiments, the insulin delivery system provides finite insulin doses into the body in response to commands from the controller. The smallest amount of insulin that the insulin delivery system can deliver is the minimum finite insulin dose.
The controller may generate commands for a dose of insulin to be delivered that is not a whole number multiple of the minimum finite insulin dose. Therefore, either too much or too little insulin is delivered by the insulin delivery system in response to the commands.
In particular alternative embodiments, the post-controller lead-lag compensator truncates the command to the nearest whole number multiple of the minimum finite insulin dose and adds the remaining commanded volume of insulin to the next command. In other alternative embodiments, a compensator rounds the command to the nearest whole number multiple of the minimum finite insulin dose. In still other alternative embodiments, other methods are used to compensate for the difference between the commands and the nearest whole number multiple of the minimum finite insulin dose. In other embodiments, no compensation is needed.
[00241] Eliminating the Lead-Lag Compensator with Feedback of Predicted Plasma Insulin 1002421 Yet in another alternative embodiment, the Pl D control commands may be modified to emulate the effect of plasma insulin on a 13-cell to determine optimal insulin administration by feeding back a predicted plasma insulin based on the subcutaneous insulin infusion. The net effect of such feedback is to replace an undesired dynamic with a more desirable one and achieve a plasma insulin profile that a I3-cell would achieve.
This can be seen as follows (using Laplace transformed variables). Assume the relation between glucose above basal (G¨GB) and insulin delivery (ID) is described by a linear transfer function (s) = C(s)(G(s) ¨ GB) , where C(s) may be, but is not necessarily, described by the PID controller transfer function. If the 13-cell is using peripheral insulin (In(s)) levels to suppress insulin secretion the predicted rate of insulin delivery would be modified as: (s) = C(s)(G(s) ¨ GB) ¨ klp(s) .
[00243] For portal insulin delivery the relation between ID(s) and plasma insulin Ip(s) is known to be approximated by a single time delay:
I (s) = ¨I D (s) .
s+ct Substituting In(s) value into the previous formula and making k large results in:
õ kk, I D (s) ¨ C(s)(G(s)-GB) kk, C ¨ (tr( ) ¨ GB); 1 CC
1+¨ kki s+a s+a which would completely cancel the undesirable time constant I /a. In practice a lower value of k would be used resulting in:
I D (s) = C (s)(G (s) GB) _kki m(s) = G(s)_s+a (G(s) GB) , s+a s+y where 7=a+kk1 (i.e., something greater than a). Thus, the effect for the 13-cell, of adding a plasma insulin feedback is to replace the portal insulin delivery time constant (a) with a faster time constant (7=a+kki; 7>a). In block diagram form:
s+a ID Ip G ¨ GB ¨> C(S)'¨ ¨> ¨>
s+y s+a which is equivalent to:

G ¨ GB ¨> C(S) ¨ ¨>
S-Fy [00244] To apply this mechanism to subcutaneous insulin delivery all that is needed is the transfer function between sc insulin delivery and plasma insulin. This transfer function is well approximated by a hi-exponential time course (bolus response) or:
(s) k2 ID(s) (s+a1)(s+a2) thus, kk, I D (s) = C (s)(G (s) ¨ GB) I D (s) = C(S) kk2 (G (S) ¨ GB) l+(s+a)(s+a,) in the limiting case as kk2/(s+al)(s+a2)>>1 this is approximately equal to I D (S) = (s)(s+ai)(s+a2) (G (S) ¨ B) k k2 where again, the undesirable time constants associated with subcutaneous insulin delivery have been eliminated. In practice they would just be replaced with more desirable rate constants (i.e., faster time constants).
1002451 Correction of Hypoglycemic Excursion Around -200 Minutes (Wind-Down) 1002461 Previous modeling of 13-cells using a PID controller gave excellent predictability of the "first" and "second" phase insulin responses during prolonged periods of increased glucose appearance. However, if the periods of increased glucose appearance is followed by a rapid decrease in glucose appearance, the PID
controller would not be able to correctly predict the wind down of the insulin response to lower glucose levels. FIG. 41B illustrates the insulin response to the blood glucose level of FIG.
4IA based on the clinical data (shown as data points), the PID modeling (shown as a solid line), and correction of the PID for the hypoglycemic excursion (shown as a dashed line).
1002471 In preferred embodiments, the hypoglycemic excursion is corrected by modifying the PID controller to a PD control with Adaptive Proportional Gain (or Bilinear P1D controller), which is modified form of the original PID
equations. As described previously, the discrete PID algorithm is as follows:
Proportional Component Response:
Pn ¨ G
con = Kp m)) , Integral Component Response:
icon= 1conn-1 KASGT ¨ G5p); I
-con = , and Derivative Component Response:
Dconn = KDdGdel where Kp, K1, and Ko are the proportional, integral, and derivative gain coefficients, SCif and dGdtf are the filtered sensor glucose and derivative respectively, and the superscript n refers to discrete time.
[00248] In the Bilinear PID controller, the proportional gain Kp is based on the integrated error term. The magnitude of each component's contribution to the insulin response is described by the following equations:
= Kp n (SGF ¨ INT) Pconn = KDdGdtn Dconn Kpn = Kpn-1 Ki(sGp ¨ Gsp) , where Kp = Kpo Where the proportional gain now integrates at rate K1(initial value Kpo) and the proportional component is related to an intercept value (INT) where (INT<Gsp).
The modified formulation can be seen to fit the hypoglycemic glucose excursion without systematic error as the adaptive PD line shown as a dashed line in FIG. 39.
[00249] In additional embodiments, the Bilinear PID controller can also incorporate an integrator leak by modifying the formula to multiply the previous Kp with a value such as as follows:
Kpn = aKpn-1 KI(SGT ¨ G.,p) , where a 0.99 [00250] An alternative method of correcting the hypoglycemic glucose excursion can be performed by integrator clip into the PID control. PID controllers generally have integrator-reset rules that prevent excessive "winding" and such a rule can be used to correct the hypoglycemic glucose excursion. For example, the integrator can be clipped as follows:
If (SG < 60 mg/d1 AND /n-1 > Kp(SP-60)) then /n-1 = Kp(SP ¨ 60) This equation resets the integrator such that if the sensor glucose falls below 60 mg/di the insulin delivery is zero for all stable or falling sensor glucose signals. The clipping limit represents an absolute threshold, similar to the human counter regulatory response.
[00251] However, other approaches that may emulate the 13-cell more accurately include the use of piecewise continuous functions. For example, the following function allows for progressive clipping to be tuned:
y(SG) = yo + (1 ¨
u' r,-6o if (SG T1 mgicil AND 4%1- > yKp (SP ¨ 60)) then I gin" = K p (SP ¨ 60) This equation introduces two additional tuning parameters (70 and Ti) and starts to check the integrator output at a higher threshold. For example, if 70=5 and T1=100 mg/di, the integrator output would be clipped to 4 Kp60 if glucose fell to 90 mg/di, 3 Kp60 if glucose fell to 80 mg/d1 and so forth until glucose reached 60 where it would be clipped at Kp60.
Other functions than that proposed in the above equation (e.g. functions based on the rate of fall of glucose, or percent decrease in Ic.) may alternatively be used.
[00252] System Configurations [00253] The following sections provide exemplary, but not limiting, illustrations of components that can be utilized with the controller described above. Various changes in components, layout of various components, combinations of elements, or the like may be made without departing from the scope of the embodiments of the invention.
[00254] Before it is provided as an input to the controller 12, the sensor signal 16 is generally subjected to signal conditioning such as pre-filtering, filtering, calibrating, or the like. Components such as a pre-filter, one or more filters, a calibrator and the controller 12 may be split up or physically located together, and may be included with a telemetered characteristic monitor transmitter 30, the infusion device 34, or a supplemental device. In preferred embodiments, the pre-filter, filters and the calibrator are included as part of the telemetered characteristic monitor transmitter 30, and the controller 12 is included with the infusion device 34, as shown in FIG. 8B. In alternative embodiments, the pre-filter is included with the telemetered characteristic monitor transmitter 30 and the filter and calibrator are included with the controller 12 in the infusion device, as shown in FIG. 8C. In other alternative embodiments, the pre-filter may be included with the telemetered characteristic monitor transmitter 30, while the filter and calibrator are included in the supplemental device 41, and the controller is included in the infusion device, as shown in FIG. 8D. To illustrate the various embodiments in another way, FIG. 9 shows a table of the groupings of components (pre-filter, filters, calibrator, and controller) in various devices (telemetered characteristic monitor transmitter, supplemental device, and infusion device) from FIGS. 8A-D. In other alternative embodiments, a supplemental device contains some of (or all of) the components.
[00255] In preferred embodiments, the sensor system generates a message that includes information based on the sensor signal such as digital sensor values, pre-filtered digital sensor values, filtered digital sensor values, calibrated digital sensor values, commands, or the like. The message may include other types of information as well such as a serial number, an ID code, a check value, values for other sensed parameters, diagnostic signals, other signals, or the like. In particular embodiments, the digital sensor values Dsig may be filtered in the telemetered characteristic monitor transmitter 30, and then the filtered digital sensor values may be included in the message sent to the infusion device 34 where the filtered digital sensor values are calibrated and used in the controller.
In other embodiments, the digital sensor values Dsig may be filtered and calibrated before being sent to the controller 12 in the infusion device 34. Alternatively, the digital sensor values Dsig may be filtered, and calibrated and used in the controller to generate commands 22 that are then sent from the telemetered characteristic monitor transmitter 30 to the infusion device 34.
1002561 In further embodiments, additional optional components, such as a post-calibration filter, a display, a recorder, and a blood glucose meter may be included in the devices with any of the other components or they may stand-alone. Generally, if a blood glucose meter is built into one of the devices, it will be co-located in the device that contains the calibrator. In alternative embodiments, one or more of the components are not used.
[00257] In preferred embodiments, RE telemetry is used to communicate between devices, such as the telemetered characteristic monitor transmitter 30 and the infusion device 34, which contain groups of components. In alternative embodiments, other communication mediums may be employed between devices such as wires, cables, IR
signals, laser signals, fiber optics, ultrasonic signals, or the like.
1002581 Filtering 1002591 In preferred embodiments, the digital sensor values Dsig and/or the derivative of the digital sensor values are processed, filtered, modified, analyzed, smoothed, combined, averaged, clipped, scaled, calibrated, or the like, to minimize the effects of anomalous data points before they are provided as an input to the controller.
In particular embodiments, the digital sensor values Dsig are passed through a pre-filter 400 and then a filter 402 before they are passed to the transmitter 70, as shown in FIG. 16.
The filters are used to detect and minimize the effects of anomalous digital sensor values Dsig. Some causes of anomalous digital sensor values Dsig may include temporary signal transients caused by sensor separation from the subcutaneous tissue, sensor noise, power supply noise, temporary disconnects or shorts, and the like. In particular embodiments, each individual digital sensor value Dsig is compared to maximum and minimum value-thresholds. In other particular embodiments, the differences between consecutive pairs of digital sensor values Dsig are compared with rate-of-change-thresholds for increasing or decreasing values.
1002601 Pre-Filter 1002611 In particular embodiments, the pre-filter 400 uses fuzzy logic to determine if individual digital sensor values Dsig need to be adjusted. The pre-filter 400 uses a subset of a group of digital sensor values Dsig to calculate a parameter and then uses the parameter to determine if individual digital sensor values Dsig need to be adjusted in comparison to the group as a whole. For example, the average of a subset of a group of digital sensor values Dsig may be calculated, and then noise thresholds may be placed above and below the average. Then individual digital sensor values Dsig within the group are compared to noise thresholds and eliminated or modified if they are outside of the noise thresholds.
1002621 A more detailed example is provided below to more clearly illustrate, but not limit, an embodiment of a pre-filter. A group of eight digital sensor values Dsig are shown in FIG. 17 including a most recently sampled value, labeled L, sampled from the analog sensor signal Tsig at time i, and the seven previous values K, H, G, F, E, D, and C
sampled at times (i-1) through (i-7). An average value is calculated using the four temporally middle values in the group, H, G, F, and E sampled at times (i-2) through (i-5). The calculated average value is represented as a dashed/dotted average line 404. A

high noise threshold 406 is established at 100% above the average line 404. In other words, the magnitude of the high noise threshold 406 is two times the magnitude of the =
average line 404. A negative noise threshold 408 is established at 50% below the average line 404. In other words, the magnitude of the negative noise threshold 408 is one half of the magnitude of the average line 404. The individual magnitudes of each of the eight values, L, K, H, G, F, E, D, and C are compared to the high and negative noise thresholds 406 and 408. If a value is above the high noise threshold 406 or below the negative noise threshold 408 then the value is considered anomalous and the anomalous value is replaced with the magnitude of the average line 404. In the example shown in FIG. 17, the value K is above the high noise threshold 406 so it is replaced with the average value M. Also, the value D is below the negative noise threshold 408 so it is replaced with the average value N. In this way noisy signal spikes are reduced. Therefore, in the example, values L, K, H, G, F, E, D, and C are inputs to the pre-filter 400 and values L, M, H, G, F, E, N, and C are outputs from the pre-filter 400. In alternative embodiments, other noise threshold levels (or percentages) may be used. In other alternative embodiments, values outside of the thresholds may be replaced with values other than the average value, such as the previous value, the value of the closest threshold, a value calculated by extrapolating a trend line through previous data, a value that is calculated by interpolation between other values that are inside the thresholds, or the like.
[00263] In preferred embodiments, when any of a group's values are outside of the noise thresholds 406 or 408 then a warning flag is set. If one to three values are outside of the noise thresholds 406 or 408, a "noise" flag is set. If more than three values are outside of the noise thresholds 406 or 408, a "discard" flag is set which indicates that the whole group of values should be ignored and not used. In alternative embodiments, more or less values need be outside of the thresholds 406 or 408 to trigger the "noise"
flag or the "discard" flag.
[00264] In preferred embodiments, each digital sensor value Dsig is checked for saturation and disconnection. To continue with the example of FIG. 17, each individual value is compared to a saturation threshold 410. If a value is equal to or above the saturation threshold 410 then a "saturation" flag is set. In particular embodiments, when the "saturation" flag is set, a warning is provided to the user that the sensor 26 may need calibration or replacement. In further particular embodiments, if an individual digital sensor value Dsig is at or above the saturation threshold 410, the individual digital sensor value Dsig may be ignored, changed to a value equal to the average line 404, or the entire group of values associated with the individual digital sensor value Dsig may be ignored.
In preferred embodiments, the saturation threshold 410 is set at about 16%
below the maximum value of the range of digital sensor values that may be generated. In preferred embodiments, the maximum digital sensor value represents a glucose concentration greater than 150 mg/d1. In alternative embodiments, the maximum digital sensor value may represent larger or smaller a glucose concentrations depending on the range of expected glucose concentrations to be measured, the sensor accuracy, the sensor system resolution needed for closed loop control, or the like. The full range of values is the difference between the maximum and the minimum digital sensor value that may be generated. Higher or lower saturation threshold levels may be used depending on an expected signal range of the sensor, sensor noise, sensor gains, or the like.
[00265] Similarly, in preferred embodiments, if a digital signal value Dsig is below a disconnect threshold 412, then a "disconnect" flag is set indicating to a user that the sensor is not properly connected to the power supply and that the power supply or sensor may need replacement or recalibration. In further particular embodiments, if a digital sensor value Dsig is below the disconnect threshold 412, the individual value may be ignored, changed to a value equal to the average line 404, or the entire group of values associated with the individual digital sensor value Dsig may be ignored. In preferred embodiments, the disconnect threshold 410 is set at about 20% of the full range of values.
Higher or lower disconnect threshold levels may be used depending on an expected signal range of the sensor, sensor system noise, sensor gains, or the like.
[00266] In alternative embodiments, other methods are used to pre-filter the digital sensor values Dsig such as rate-of-change thresholds, rate-of-change squared thresholds, noise thresholds about a least squares fit line rather than about the average of a subset of a group's values, higher or lower noise threshold lines, or the like.
[00267] Noise Filter [00268] After the digital sensor values Dsig are evaluated, and if necessary, modified by the pre-filter 400, the digital sensor values Dsig are passed to the filter 402. The filter 402 may be used to reduce noise in particular frequency bands. Generally the body's blood glucose level 18 changes relatively slowly compared to a rate at which digital sensor values Dsig are collected. Therefore, high frequency signal components are typically noise, and a low pass filter may be used to improve the signal to noise ratio.
[00269] In preferred embodiments, the filter 402 is a finite impulse response (FIR) filter used to reduce noise. In particular embodiments, the FIR filter is a 7th order filter tuned with a pass band for frequencies from zero to 3 cycles per hour (c/hr) and a stop band for frequencies greater than about 6 c/hr, as shown in an example frequency response curve 414 in FIG. 18. However, typically FIR filters tuned with a pass band for frequencies from zero up to between about 2 c/hr and 5 c/hr and a stop band beginning at 1.2 to three times the selected pass band frequency will sufficiently reduce noise while passing the sensor signal. In particular embodiments, FIR filters tuned with a pass band for frequencies from zero up to between about 2 c/hr and 10 c/hr and a stop band beginning at 1.2 to three times the selected pass band frequency will sufficiently reduce noise. In the 7th order filter, unique weighting factors are applied to each of eight digital sensor values Dsig. The digital sensor values Dsig include the most recently sampled value and the seven previous values. The effects of a low pass filter on a digital sensor values collected at one minute intervals is shown in FIGS. 19A and B. An unfiltered sensor signal curve 416 of digital sensor values is contrasted with a curve of the same signal after the effects of a 7th order FIR filter 418. The filtered signal curve 418 is delayed and the peaks are smoother compared to the unfiltered sensor signal curve 416. In other particular embodiments, higher or lower order filters may be used. In still other particular embodiments, filter weighting coefficients may be applied to digital sensor values Dsig collected at time intervals shorter or longer than one minute depending on the desired sensor sample rate based on the body's physiology, the computational capabilities of the telemetered characteristic monitor transmitter 30, the sensor's response time, or the like. In alternative embodiments, filters with other frequency responses may be used to eliminate other noise frequencies depending on the type of sensor, noise from the power supply or other electronics, the sensor's interaction with the body, the effects of body motion on the sensor signal, or the like. In still other alternative embodiments, the filter is an infinite impulse response (IIR) filter.
1002701 In alternative embodiments, other methods are used to pre-filter the digital sensor values Dsig such as rate-of-change thresholds, rate-of-change squared thresholds, noise thresholds about a least squares fit line rather than about the average of a subset of a group's values, higher or lower noise threshold lines, or the like.
1002711 Delay Compensation Filter [00272] Aside from noise reduction, a filter may be used to compensate for time delays. Ideally, a sensor would provide a real time, noise-free measurement of a parameter that a control system is intended to control, such as a blood glucose measurement. However, realistically there are physiological, chemical, electrical, and algorithmic sources of time delays that cause the sensor measurement to lag behind the present value of blood glucose.
1002731 A physiological delay 422 is due to the time required for glucose to move between blood plasma 420 and interstitial fluid (ISF). The delay is represented by the circled double headed arrow 422 in FIG. 20. Generally, as discussed above, the sensor 26 is inserted into the subcutaneous tissue 44 of the body 20 and the electrodes 42 near the tip of the sensor 40 are in contact with interstitial fluid (ISF). But the desired parameter to bc measured is the concentration of blood glucose. Glucose is carried throughout the body in blood plasma 420. Through the process of diffusion, glucose moves from the blood plasma 420 into the ISF of the subcutaneous tissue 44 and vice versa. As the blood glucose level 18 changes so does the glucose level in the ISF. But the glucose level in the ISF lags behind the blood glucose level 18 due to the time required for the body to achieve glucose concentration equilibrium between the blood plasma 420 and the ISF.
Studies show the glucose lag times between blood plasma 420 and ISF vary between 0 to 30 minutes. Some parameters that may affect the glucose lag time between blood plasma 420 and 1SF are the individual's metabolism, the current blood glucose level, whether the glucose level is rising, or falling, or the like.
[00274] A chemical reaction delay 424 is introduced by the sensor response time, represented by the circle 424 surrounding the tip of the sensor 26 in FIG. 20.
The sensor electrodes 42 are coated with protective membranes that keep the electrodes 42 wetted with ISF, attenuate the glucose concentration, and reduce glucose concentration fluctuations on the electrode surface. As glucose levels change, the protective membranes slow the rate of glucose exchange between the ISF and the electrode surface.
In addition, there is a chemical reaction delay simply due to the reaction time for glucose to react with glucose oxidase GOX to generate hydrogen peroxide, and the reaction time for a secondary reaction, the reduction of hydrogen peroxide to water, oxygen and free electrons.
[00275] There is also a processing delay as the analog sensor signal Isig is converted to digital sensor values Dsig. In preferred embodiments, the analog sensor signal Isig is integrated over one-minute intervals and then converted to a number of counts.
In essence an A/D conversion time results in an average delay of 30 seconds. In particular embodiments, the one-minute values are averaged into 5-minute values before they are sent to the controller. The resulting average delay is two and one half minutes. In alternative embodiments, longer or shorter integration times are used resulting in longer or shorter delay times. In other embodiments the analog sensor signal current Isig is continuously converted to an analog voltage Vsig and a AID converter samples the voltage Vsig every 10 seconds. Then six 10-second values are pre-filtered and averaged to create a one-minute value. Finally, five 1-minute values are filtered and then averaged creating a five-minute value resulting in an average delay of two and one half minutes.
Other embodiments use other electrical components or other sampling rates and result in other delay periods.
1002761 Filters also introduce a delay due to the time required to acquire a sufficient number of digital sensor values Dsig to operate the filter. Higher order filters, by definition, require more digital sensor values Dsig. Aside from the most recent digital sensor value Dsig, FIR filters use a number of previous values equal to the order of the filter. For example, a 7th order filter uses 8 digital sensor values Dsig.
There is a time interval between each digital sensor value Dsig. To continue with the example, if the time interval between digital sensor values Dsig is one minute, then the oldest digital sensor value Dsig used in a 7th order FIR filter would be seven minutes old.
Therefore, the average time delay for all of the values used in the filter is three and a half minutes.
However, if the weighting factors associated with each of the values are not equal then the time delay may be longer or shorter than three and one half minutes depending on the effects of the coefficients.
[00277] Preferred embodiments of the invention include a FIR filter that compensates for both the various time delays, of up to about 30 minutes as discussed above, and high frequency noise, greater than about 10 c/hr also discussed above. Particular embodiments employ a 7th order Weiner type FIR filter. The coefficients for the filter are selected to correct for time lags while simultaneously reducing high frequency noise. An example of a frequency response curve 426 is shown in FIG. 21. The example frequency response curve 416 is generated for a Weiner filter with a pass band for frequencies from zero up to about 8 c/hr and a stop band for frequencies greater than about 15 c/hr for a sensor with a sensitivity of about 20 pA/100 mg/d1. A study conducted with sensors in dogs demonstrates that a FIR filter may be used to compensate for time delays.
During the study a filter was used to compensate for a time delay of about 12 minutes.
The results, presented in FIG. 22, show dots 428 representing actual blood plasma glucose levels measured with a blood glucose meter, a broken line 430 representing sensor measurements without delay compensation, and a solid line 432 representing sensor measurements with delay compensation. The sensor in the test was abnormally low in sensitivity. Studies with average sensitivity sensors in humans are indicating a time delay of about 3 to 10 minutes is more normal. Other filter coefficients and other orders of filters may be used to compensate for the time delay and/or noise.
[00278] In alternative embodiments, other types of filters may be used as long as they remove a sufficient portion of the noise from the sensor signal. In other alternative embodiments, no time compensation is used if the rate of change in the blood glucose level is slow compared to the time delay. For example, a five-minute delay between blood plasma glucose and a sensor measurement does not have to be corrected for a closed loop glucose control system to function.
[002791 Derivative Filter [00280] Further embodiments may include a filter to remove noise from the derivative of the sensor signal before the controller uses it. A derivative is taken from the digital sensor values Dsig, which results in digital derivative sensor values (dDsig/dt). The digital derivative sensor values dDsig/dt are passed through a FIR filter. In particular embodiments, the derivative filter is at least a 7th order FIR filter tuned to remove high frequency noise. In alternative embodiments, higher or lower order filters may be used and the filters may be tuned to remove various frequencies of noise. In other alternative embodiments, a derivative is taken from the glucose level error GE values and then passed through a derivative filter 526, as shown in FIG. 37. In further alternative embodiments, a derivative is taken of an analog sensor signal Isig and a hardware filter is used to remove noise.
100281] Calibration [00282] In preferred embodiments, after filtering, the digital sensor values Dsig are calibrated with respect to one or more glucose reference values. The glucose reference values are entered into the calibrator and compared to the digital sensor values Dsig. The calibrator applies a calibration algorithm to convert the digital sensor values Dsig, which are typically in counts into blood glucose values. In particular embodiments, the calibration method is of the type described in U.S. patent application Ser.
No. 09/511,580, filed on Feb. 23, 2000, entitled "GLUCOSE MONITOR CALIBRATION METHODS".
In particular embodiments, the calibrator is included as part of the infusion device 34 and the glucose reference values are entered by the user into the infusion device 34. In other embodiments, the glucose reference values are entered into the telemetered characteristic monitor transmitter 30 and the calibrator calibrates the digital sensor values Dsig and transmits calibrated digital sensor values to the infusion device 34. In further embodiments, the glucose reference values are entered into a supplemental device where the calibration is executed. In alternative embodiments, a blood glucose meter is in communication with the infusion device 34, telemetered characteristic monitor transmittcr 30 or supplemental device so that glucose reference values may be transmitted directly into the device that the blood glucose meter is in communication with. In additional alternative embodiments, the blood glucose meter is part of the infusion device 34, telemetered characteristic monitor transmitter 30 or supplemental device such as that shown in U.S. patent application Ser. No.
09/334,996, filed on Jun. 17, 1999, entitled "CHARACTERISTIC MONITOR WITH A
CHARACTERISTIC METER AND METHOD OF USING THE SAME".
[00283] In preferred embodiments, to obtain blood glucose reference values, one or more blood samples are extracted from the body 20, and a common, over-the-counter, blood glucose meter is used to measure the blood plasma glucose concentration of the samples. Then a digital sensor value Dsig is compared to the blood glucose measurement from the meter and a mathematical correction is applied to convert the digital sensor values Dsig to blood glucose values. In alternative embodiments, a solution of a known glucose concentration is introduced into the subcutaneous tissue surrounding the sensor 26 by using methods and apparatus such as described in U.S. patent application Ser. No.
09/395,530, filed on Sep. 14, 1999, entitled "METHOD AND KIT FOR SUPPLYING A
FLUID TO A SUBCUTANEOUS PLACEMENT SITE", or by using injection, infusion, jet pressure, introduction through a lumen, or the like. A digital sensor value Dsig is collected while the sensor 26 is bathed in the solution of known glucose concentration. A mathematical formula such as a factor, an offset, an equation, or the like, is derived to convert the digital sensor value Dsig to the known glucose concentration. The mathematical formula is then applied to subsequent digital sensors values Dsig to obtain blood glucose values. In alternative embodiments, the digital sensor values Dsig are calibrated before filtering. In additional alternative embodiments, the digital sensor values Dsig are calibrated after pre-filtering and before filtering. In other alternative embodiments, the sensors are calibrated before they are used in the body or do not require calibration at all.
[00284] Sensor Signal Processing Systems [00285] Before filtering and calibrating, generally the sensor signal is processed to convert the sensor signal from a raw form into a form acceptable for use in the filters and/or calibrator. In preferred embodiments, as shown in FIG. 10, an analog sensor signal Isig is digitally quantified through an A/D converter 68 resulting in digital sensor values Dsig that are transmitted by a transmitter 70 from the telemetered characteristic monitor transmitter 30 to another device. In particular embodiments, the analog sensor signal Isig is an analog current value that is converted to a digital sensor value Dsig in the form of a digital frequency measurement, as shown in FIG. 11(a). The general circuit includes an integrator 72, a comparator 74, a counter 76, a buffer 78, a clock 80 and the transmitter 70. The integrator 72 generates a substantially ramped voltage signal (A), and the instantaneous slope of the ramped voltage signal is proportional to the magnitude of the instantaneous analog sensor signal Isig. The comparator 74 converts the ramped voltage signal (A) from the integrator 72 into square wave pulses (B). Each pulse from the comparator 74 increments the counter 76 and also resets the integrator 72. The clock 80 periodically triggers the buffer 78 to store the present value from the counter 76 and then resets the counter 76. The values stored in the buffer 78 are the digital sensor values Dsig.
The clock 80 may also periodically signal the transmitter 70 to send a value from the buffer 78. In preferred embodiments, the clock period is one minute. However, in alternative embodiments, the clock period may be adjusted based on how often measurements are needed, sensor signal noise, sensor sensitivity, required measurement resolution, the type of signal to be transmitted, or the like. In alternative embodiments, a buffer is not used.
[00286] A/D Converters [00287] Various A/D converter designs may be used in embodiments of the present invention. The following examples are illustrative, and not limiting, since other AID
converters may be used.
[00288] Ito F (Current to Frequency (Counts)), Single Capacitor, Quick Discharge 1002891 In preferred embodiments, the integrator 72 consists of a first Op-Amp 92 and a capacitor 82, shown in FIG. 12. The integrator 72 sums the analog sensor signal Isig current by charging the capacitor 82 until the capacitor voltage (A') achieves a high reference voltage (Vreft-I). The capacitor voltage (A') is measured at the output of the first Op-Amp 92. A second Op-Amp 94 is used as a comparator. When the capacitor voltage (A') reaches VrefH, the comparator output (B') changes from low to high. The high comparator output (B') closes a reset switch 84 that discharges the capacitor 82 through a voltage source (V+). The high comparator output (B') also triggers a reference voltage switch 88 to close, while substantially simultaneously an inverter 86 inverts the comparator output (B'). And the inverter output (C') triggers a reference voltage switch 90 to open. The result is that the reference voltage of the comparator is changed from VrefH
to the low reference voltage (VrefL).
[00290] When the capacitor voltage (A') is discharged to VrefL, the comparator output (B') returns to low, thus forming a pulse. The low comparator output (B') opens the reset switch 84 allowing the capacitor 82 to begin charging again.
[00291] Virtually simultaneously, the low comparator output (B') also triggers the reference voltage switch 88 to open and the inverter output (C') triggers reference voltage switch 90 to close resulting in changing the comparator reference voltage from VrefL
back to VrefH.
1002921 Ito F, Single Reversible Capacitor [00293] In alternative embodiments, two or more integrator switches are used to control the polarity of one or more capacitors. A particular embodiment is shown in FIG.
13. Generally, only one of the two integrator-switches 110 and 112 is closed and the other integrator switch is open. When the first integrator switch 110 is closed, the second integrator switch 112 is open and an integrator Op-Amp 114 sums the analog sensor signal Isig current by charging a capacitor 116 until the capacitor voltage (A") achieves a high reference voltage (VrefH). The comparator 120 compares the integrator output (A") to the reference voltage VrefH. And when the capacitor voltage (A") reaches VrefH, the comparator output (B") shifts from low to high, initiating a pulse.
[00294] The high comparator output (B") pulse causes the capacitor polarity to reverse using the following method. The high comparator output (B") triggers the second integrator switch 112 to close while virtually simultaneously the inverter 118 inverts the comparator output (B"). And the low inverter output (C") pulse triggers the first integrator switch 110 to open. Once the capacitor's polarity is reversed, the capacitor 116 discharges at a rate proportional to the analog sensor signal Isig. The high comparator output (B") pulse also triggers the reference voltage of the comparator to change form VrefH the low reference voltage (VrefL). When the capacitor voltage (A") is discharged to VrefL, the comparator output (B") returns to low. The low comparator output (B") opens the second integrator switch 112 and virtually simultaneously the high inverter output (C") closes the first integrator switch 110 allowing the capacitor 116 to begin charging again. The low comparator output (B") also triggers the comparator reference voltage to change from VrefL back to Vreft-1.

[00295] An advantage of this embodiment is that sensor signal errors, which may be created due to capacitor discharge time, are reduced since the magnitude of the analog sensor signal Isig drives both the charging and the discharging rates of the capacitor 116.
[00296] Ito F, Dual Capacitor [00297] In further alternative embodiments, more than one capacitor is used such that as one capacitor is charging, at a rate proportional to the magnitude of the analog sensor signal Isig, another capacitor is discharging. An example of this embodiment is shown in FIG. 14. A series of three switches are used for each capacitor. A first group of switches 210 is controlled by a latch voltage C", and a second group of switches 212 are controlled by voltage D", which is the inverse of C". Substantially, only one group of switches is closed at a time. When the first group of switches 210 is closed, the voltage across a first capacitor 216 increases at a rate proportional to the analog sensor signal Isig until the integrator voltage (A") at the output of Op-Amp 214 achieves a reference voltage (Vref). At the same time one of the switches shorts the circuit across a second capacitor 222 causing it to discharge. A comparator 220 compares the integrator output (A") to the reference voltage Vref. And when the integrator output (A") reaches Vref, the comparator output (Fr) generates a pulse. The comparator output pulse increments a counter 76, and triggers the latch output voltage C" from a latch 221 to toggle from a low voltage to a high voltage. The change in the latch voltage C' causes the second group of switches 212 to close and the first group of switches 210 to open. One of the switches from the second group of switches 212 shorts the circuit across the first capacitor 216 causing it to discharge. At the same time the voltage across the second capacitor 222 increases at a rate proportional to the analog sensor signal Isig until the integrator voltage (A") at the output of Op-Amp 214 achieves a reference voltage (Vref). Again, the comparator 220 compares the integrator output (A") to the reference voltage Vref. And when the integrator output (A') reaches Vref, the comparator output (B'") generates a pulse. The comparator output pulse increments the counter 76, and triggers the latch output voltage C'" to toggle from a high voltage to a low voltage, which causes the switches to return to their initial position with the first group of switches 210 closed and the second group of switches 212 to open.
[00298] In summary, as the blood glucose level 18 increases, the analog sensor signal Isig increases, which causes the voltage coming out of the integrator 72 to ramp up faster to the high reference voltage VrefH, which causes the comparator 74 to generate pulses more often, which adds counts to the counter 76 faster. Therefore, higher blood glucose levels generate more counts per minute.
[00299] The charge storage capacity for the capacitors used in the integrator 72, and the reference voltages VrefH, and VrefL are selected such that the count resolution for counts collected in a one-minute period at a glucose level of 200 mg/d1 represents a blood glucose measurement error of less than 1 mg/d1. In particular embodiments, VrefH is 1.1 volts and VrefL is 0.1 volts. Higher or lower reference voltages may be selected based on the magnitude of the analog sensor signal Isig, the capacity of the capacitors, and the desired measurement resolution. The source voltage V+ is set to a voltage sufficiently high to discharge one or more capacitors quickly enough that the discharge times do not significantly reduce the number of counts per minute at a blood glucose level of 200 [00300] Pulse Duration Output Feature [00301] In preferred embodiments, the transmitter 70 transmits the digital sensor values Dsig from the buffer 78 whenever triggered by the clock 80. However, in particular embodiments, the user or another individual may use a selector 96 to choose other outputs to be transmitted from the transmitter 70, as shown in FIG. 11B.
In preferred embodiments, the selector 96 is in the form of a menu displayed on a screen that is accessed by the user or another individual by using buttons on the surface of the telemetered characteristic monitor transmitter 30. In other embodiments, a dial selector, dedicated buttons, a touch screen, a signal transmitted to the telemetered characteristic monitor transmitter 30, or the like, may be used. Signals that may be selected to be transmitted, other than the digital sensor values Dsig, include, but are not limited to, a single pulse duration, digital sensor values before pre-filtering, digital sensor values after pre-filtering but before filtering, digital sensor values after filtering, or the like.
[00302] In particular embodiments, a pulse duration counter 98 counts clock pulses from a pulse duration clock 100 until the pulse duration counter 98 is reset by a rising or falling edge of a pulse from the comparator 74, as shown in FIG. 11B. The accumulated count at the time that the pulse duration counter 98 is reset represents the pulse duration for a portion of a single pulse from the comparator 74. The accumulated count from the pulse duration counter 98 is stored in the single pulse buffer 102 when triggered by the reset signal. When an individual selects the single pulse output, the transmitter 70 transmits the values from the single pulse buffer 102. The pulse duration clock 100 period must be sufficiently shorter than the period between individual pulse edges from the comparator 74 given a high analog sensor signal Isig to have sufficient resolution to quantify different pulse durations from the comparator 74.
[00303] Ito V (Current to Voltage), Voltage AID
[00304] Alternative methods may be used to convert the analog sensor signal Isig from an analog current signal to a digital voltage signal. The analog sensor signal Isig is converted to an analog voltage Vsig using an Op Amp 302 and a resistor 304, as shown in FIG. 15. And then periodically a clock 308 triggers an AID converter 306 to take a sample value from the analog voltage Vsig and convert it to a digital signal representing the magnitude of the voltage. The output values of the AID converter 306 are digital sensor values Dsig. The digital sensor values Dsig are sent to a buffer 310 and then to the transmitter 70. In particular embodiments, the resistor 304 may be adjusted to scale the Vsig to use a significant portion of the range of the voltage AID converter 306 depending on the sensor sensitivity, the maximum glucose concentration to be measured, the desired resolution from the voltage AID converter 306, or the like.
[00305] In alternative embodiments, a buffer 310 is not needed and the digital sensor values Dsig are sent from the AiD converter directly to the transmitter 70. In other alternative embodiments, the digital sensor values Dsig are processed, filtered, modified, analyzed, smoothed, combined, averaged, clipped, scaled, calibrated, or the like, before being sent to the transmitter 70. In preferred embodiments, the clock 308 triggers a measurement every 10 seconds. Ti alternative embodiments, the clock 308 runs faster or slower triggering measurements more or less frequently depending on how quickly the blood glucose level can change, the sensor sensitivity, how often new measurements are needed to control the delivery system 14, or the like.
[00306] Finally, in other alternative embodiments, other sensor signals from other types of sensors, as discussed in the section "Sensor and Sensor Set" below, are converted to digital sensor values Dsig if necessary before transmitting the digital sensor values Dsig to another device.
[00307] Additional Controller Inputs [00308] Generally, the proportional plus, integral plus, derivative (PID) insulin response controller uses only glucose (digital sensor values Dsig) as an input. Conversely, in a normally glucose tolerant human body, healthy 13-cells benefit from additional inputs such as neural stimulation, gut hormone stimulation, changes in free fatty acid (FFA) and protein stimulation etc. Thus in other alternative embodiments, the PID
controller, as discussed above, can be augmented with one or more additional inputs. In particular alternative embodiments, the user may manually input supplemental information such as a start of a meal, an anticipated carbohydrate content of the meal, a start of a sleep cycle, an anticipated sleep duration, a start of an exercise period, an anticipated exercise duration, an exercise intensity estimation, or the like. Then, a model predictive control feature assists the controller to use the supplemental information to anticipate changes in glucose concentration and modify the output commands accordingly. For example, in a NOT individual, neural stimulation triggers the I3-cells to begin to secrete insulin into the blood stream before a meal begins, which is well before the blood glucose concentration begins to rise. So, in alternative embodiments, the user can tell the controller that a meal is beginning and the controller will begin to secrete insulin in anticipation of the meal.
[00309] In other alternative embodiments, the user or another individual may manually override the control system or select a different controller algorithm. For instance, in particular alternative embodiments, an individual may select to normalize to a basal glucose level immediately, and instead of using the 13-cell emulating PID
controller another controller would take over such as a PID controller with different gains, a PD
controller for rapid glucose adjustment, or the like. Additional alternative embodiments allow an individual to turn off the integral component of the PID controller once the glucose level is normalized and no meals are anticipated. In other particular alternative embodiments, the user may select to turn off the controller entirely, therefore disengaging the closed loop system. Once the closed loop system is not controlling insulin dosing, the user may program the infusion device with a basal rate, variable basal rates, boluses, or the like, or the user may manually enter each individual dosage when it is needed.
1003101 In still other alternative embodiments, more than one body characteristic is measured, and the measurements are provided as inputs to a controller.
Measured body characteristics that may be used by the controller include, but are not limited to, the blood glucose level, blood and/or ISF pH, body temperature, the concentration of amino acids in blood (including arginine and/or lysine, and the like), the concentration of gastrointestinal hormones in blood or ISF (including gastrin, secretin, cholecystokinin, and/or gastro inhibitory peptide, and the like), the concentration of other hormones in blood or ISF
(including glucagons, growth hormone, cortisol, progesterone and/or estrogen, and the like), blood pressure, body motion, respiratory rate, heart rate, and other parameters.
[00311] In NOT individuals, the glucose-induced secretion of insulin by healthy P-eens may be as much as doubled in the presence of excess amino acids. Yet, the presence of excess amino acids alone, without elevated blood glucose, only mildly increases insulin secretions according to the Textbook of Medical Physiology, Eighth Edition, written by Arthur C. Guyton, published by W. B. Saunders Company, 1991, Ch.
78, pg.
861, section "Other Factors That Stimulate Insulin Secretion". In particular alternative embodiments, amino acid concentrations are estimated or measured, and the controller's insulin response increases when amino acid concentrations are sufficiently high.
[00312] In NGT individuals, the presence of sufficient quantities of gastrointestinal hormones in the blood causes an anticipatory increase in blood insulin, which suggests that 13-cells release insulin before increases in blood glucose due to an individual's anticipation of a meal. In particular alternative embodiments, the concentration of gastrointestinal hormones is measured or estimated, and when concentrations are high enough to indicate that a meal is anticipated, the controller commands are adjusted to cause insulin introduction into the body even before the blood glucose level changes. In other alternative embodiments, the controller uses measurements or estimates of other hormones to modify the rate of insulin secretion.
[00313] In NGT individuals, the body's cells take up glucose during periods of heavy exercise with significantly lower levels of insulin. In alternative embodiments, physiologic parameters such as body motion, blood pressure, pulse rate, respiratory rate, or the like, are used to detect periods of heavy exercise by the body and therefore provide inputs to the controller that decreases (or eliminates) the amount of insulin infused into the body to compensate for glucose concentrations.
[00314] Sensor Compensation and End-of-Life Detection [00315] In particular embodiments, the sensor sensitivity 510 may degrade over time, as shown in FIG. 31B. As the sensor sensitivity 510 changes the sensor signal accuracy degrades. If the sensor sensitivity 510 changes significantly then the sensor must be recalibrated or replaced. A diagnostic signal may be used to evaluate whether sensor signal accuracy has changed and/or may be used to adjust the signal or to indicate when to recalibrate or replace the sensor. As the sensor sensitivity 510 decreases, the measured glucose level 512 using the sensor signal underestimates the actual blood glucose level 514, and the measurement error 516 between the measured glucose level 512 and the actual blood glucose level 514 becomes greater over time, as shown in FIG.
31A. The sensor sensitivity 510 decreases due to increases in sensor resistance Rs, as shown in FIG.
3 IC. The sensor resistance Rs is the resistance provided by the body between the working electrode WRK and the counter electrode CNT, shown as the sum or RI and R2 in the circuit diagram of FIG. 7. The sensor resistance Rs can be obtained indirectly by measuring the analog sensor signal Isig and the counter electrode voltage Vent and then calculating the resistance, Rs=Vent/Isig.
[00316] As the sensor resistance Rs increases, the analog sensor signal Isig response to a given glucose concentration decreases. In preferred embodiments, the decrease in the analog sensor signal Isig may be compensated for by identifying the amount that the sensor resistance Rs has changed since the last calibration and then using the change in resistance in a correction algorithm 454 to adjust the analog sensor signal value. A
compensation value calculated by the correction algorithm 454 is used to increase the sensor analog signal value. The compensation value increases over time as the sensor resistance Rs increases. The correction algorithm 454 includes at least one value that varies with changes in sensor resistance Rs. In particular embodiments, a low pass filter is applied to the sensor resistance Rs measurement to decrease high frequency noise before evaluating how much the sensor resistance Rs has changed since the last calibration.
[00317] In alternative embodiments, the sensor resistance Rs may be calculated using different equations. For instance, a sensor resistance Rs2 may be calculated as:
Rs2 = (V0 ¨ Vcntlisig) [00318] In particular embodiments, Vo is the same voltage as Vset. An advantage of this approach is that it accounts for the voltage level Vset, which can vary from sensor to sensor and/or monitor to monitor, and/or as the analog sensor signal changes.
This removes the noise and/or offset associated with variations in Vset, and can provide a more accurate indication of sensor resistance. In other particular embodiments, Vo is set at ¨0.535 volts, which is a commonly used voltage for Vset. In further embodiments, Vo is calculated from paired measurements of Vent and Isig. Using least squares or another curve fitting method, a mathematical equation representing the curve (typically a straight line equation) is derived from the relationship between Vent and Isig. Then, Vo is obtained by extrapolating the curve to find the value for Vent when Isig is zero.
[00319] FIGS. 38A-H show a comparison between calculating the sensor resistance with Vo and without Vo. The plot of the derivative of Rs2 shown in FIG. 38G is cleaner and indicates the sensor failure more clearly than the plot of the derivative of Rs shown in FIG. 38F. Hence sensor resistance Rs2 may be used instead of, or in conjunction with, sensor resistance Rs described above.

[00320] In preferred embodiments, the sensor is recalibrated or replaced when the change in the sensor resistance Rs since the last calibration exceeds a threshold, or the rate of change of the sensor resistance dRs/dt exceeds another threshold. In particular embodiments, the rate of change of the sensor resistance dRs/dt may be compared to two thresholds as shown in FIG. 32. If dRs/dt exceeds a "replacement" threshold then a warning is provided to the user to replace the sensor. If dRs/dt exceeds a "recalibrate"
threshold then a warning is provided to the user to recalibrate the sensor.
[00321] In an example shown in FIGS. 33A-C, the analog sensor signal Isig decreases dramatically at approximately 0.3 days, as seen in FIG. 33A. Given only the analog sensor signal Isig, the user would believe that the decrease in the analog sensor signal Isig is due to a decrease in blood glucose. But in reality the drop in the analog sensor signal Isig is due to a sudden change in sensor sensitivity. The sensor resistance Rs, shown in FIG. 33A increases as the analog sensor signal Isig drops at about 0.3 days.
The derivative of the sensor resistance dRs/dt, shown in FIG. 33C, clearly shows a spike 522 at about 0.3 days when the analog sensor signal Isig dropped. The spike 522 in the change in sensor resistance dRs/dt indicates a sensor anomaly rather than a realistic drop in blood glucose. If a threshold were placed at +/-4 on the dRs/dt, the user would have received a warning to replace the sensor at about 0.3 days. As seen in FIG. 33A, the sensor was not replaced until about 1.4 days. The analog sensor signal lsig was under estimating the true glucose level from about 0.3 days until the sensor was replaced at about 1.4 days.
[00322] In particular embodiments, the amount of time dt over which the derivative of the sensor resistance Rs is taken is the entire time since the last calibration. In other embodiments, the amount of time dt over which the derivative is taken is fixed, for example over the last hour, 90 minutes, 2 hours, or the like.
1003231 In alternative embodiments, the sensor is recalibrated or replaced when the integral of the sensor resistance Rs over a predetermined time window (5 Rs d/dt) exceeds a predetermined resistance integral threshold. An advantage to this approach is that it tends to filter out potential noise that could be encountered from a signal that includes occasional spikes, sudden variations in voltage levels, or the like.
Preferably, the integral of the sensor resistance Rs is calculated over a time window (such as 15 minutes, or the like) based on Rs measurements obtained at set rates (such as 1 minute,

5 minutes, or the like) during the time window. In alternative embodiments, the time windows may be longer or shorter and different sampling rates may be used, with the selection being dependent on noise, response of the system, sampling rate used in the controller, or the like. In further embodiments, the time windows and sampling rates may change over time, such as when approaching the end of the expected sensor life, or as the equations indicate that the sensor is degrading, or the like.
[00324] Like above, multiple thresholds may be used. For instance, if IRs d/dt exceeds a "replacement" threshold then a warning is provided to the user to replace the sensor.
And Wilts didt exceeds a "recalibrate" threshold then a warning is provided to the user to recalibrate the sensor. In further alternative embodiments, the counter electrode voltage Vcnt is used to evaluate other characteristics such as, sensor accuracy, sensor bio-fouling, sensor function, sensor voltage operating range, sensor attachment, or the like.
[00325] pH Controller Input [00326] In alternative embodiments, the controller uses measurements of both the interstitial fluid (ISF) glucose level and a local pH in the ISF surrounding the sensor to generate commands for the infusion device. In particular alternative embodiments, a single multi-sensor 508 located in the subcutaneous tissue is used to measure both the glucose level and the pH. The tip of the multi-sensor 508 that is placed into the subcutaneous tissue with three electrodes is shown in FIG. 30. The working electrode 502 is plated with platinum black and coated with glucose oxidase (GOX). The reference electrode 506 is coated with silver-silver chloride. And the counter electrode 504 is coated with iridium oxide (Jr Ox). The analog sensor signal Isig is generated at the working electrode 502 due to the reaction between glucose oxidase GOX and the ISF
glucose as described with the preferred sensor embodiment. In this alternative embodiment however, as glucose in the ISF reacts with the glucose oxidase GOX
on the working electrode and gluconic acid is generated, the local pH in the ISF
surrounding the sensor decreases, which changes the potential of the iridium oxide on the counter electrode 504, with respect to the reference electrode REF. So, as the pH
decreases, the voltage at the counter electrode 504 increases. Therefore, as the glucose concentration increases, the local pH decreases, which causes the counter electrode voltage to increase.
So, the glucose concentration may be estimated based on the counter electrode voltage.
The counter electrode voltage estimate of glucose concentration can be compared to the estimate of glucose level from the analog sensor signal Isig. The two estimates of the glucose level may be combined by a weighted average or one estimate may simply be used as a check to verify that the other sensing method is functioning properly. For example, if the difference between the two estimates is 10% for a period of time and then suddenly the difference increased to 50%, a warning would be issued indicating to the user that the sensor may need to be replaced or recalibrated.
1003271 In additional alternative embodiments, the pH level near the sensor may be used to detect infection. By tracking trends in the pH over time, a dramatic change in pH
may be used to identify that an infection has developed in proximity to the sensor. A
warning is used to notify the user to replace the sensor.
[00328] The pH sensor may be used in other embodiments. When insulin is not available to assist the body to use glucose, the body shifts to consuming fat for energy. As the body shifts from using glucose to using almost exclusively fat for energy, concentrations of keto acids (acetoacetic acid and fl-hydroxybutyric acid) increase from about 1 mEq/liter to as high as 10 mEceliter. In particular alternative embodiments, the pH level is measured to detect increases in keto acids in the body. In embodiments of the present invention, a warning is provided to the user when the ISF pH level is too low.
[00329] A side effect of the increased of keto acid concentrations is that sodium is drawn from the body's extra cellular fluid to combine with the acids so that the body can excrete the acids. This leads to increased quantities of hydrogen ions, which greatly increases the acidosis. Severe cases lead to rapid deep breathing, acidotic coma and even death. In other alternative embodiments, an ion-selective electrode (ISE) is used to detect changes in sodium concentration. A special membrane is used to coat the ISE so that it only senses changes in sodium concentration. In particular alternative embodiments, the ISE is a fourth electrode added to the glucose sensor. hi another alternative embodiment, a three-electrode system is used with a silver-silver chloride reference electrode REF, an Jr Ox counter electrode CNT, and a sodium ion-selective (Na ISE) working electrode WRK.
[00330] While pH measurements, end-of-life measurements, hormone measurements, or the like, add inputs to the controller that can significantly affect the accuracy of insulin delivery, the basic input to the controller is generally a glucose measurement. The glucose measurement is provided by the sensor system. And once the controller uses the glucose measurement to generate commands, the delivery system executes the commands.
The following is a detailed description of several apparatus embodiments for the sensor system and the delivery system.
[00331] Sensor System [00332] The sensor system provides the glucose measurements used by the controller.
The sensor system includes a sensor, a sensor set to hold the sensor if needed, a telemetered characteristic monitor transmitter, and a cable if needed to carry power and/or the sensor signal between the sensor and the telemetered characteristic monitor transmitter.
[00333] Sensor and Sensor Set [00334] In preferred embodiments, the glucose sensor system 10 includes a thin film electrochemical sensor such as the type disclosed in U.S. Pat. No. 5,391,250, entitled "METHOD OF FABRICATING THIN FILM SENSORS"; U.S. patent application Ser.
No. 09/502,204, filed on Feb. 10, 2000, entitled "IMPROVED ANALYTE SENSOR
AND METHOD OF MAKING THE SAME"; or other typical thin film sensors such as described in commonly assigned U.S. Pat. Nos. 5,390,671; 5,482,473; and 5,586,553.
See also U.S. Pat. No. 5,299,571.
[00335] The glucose sensor system 10 also includes a sensor set 28 to support the sensor 26 such as described in U.S. Pat. No. 5,586,553, entitled "TRANSCUTANEOUS
SENSOR INSERTION SET" (published as PCT Application WO 96/25088); and U.S.
Pat. No. 5,954,643, entitled "INSERTION SET FOR A TRANSCUTANEOUS
SENSOR" (published as PCT Application WO 98/56293); and U.S. Pat. No.
5,951,521, entitled "A SUBCUTANEOUS IMPLANTABLE SENSOR SET HAVING. THE
CAPABILITY TO REMOVE OR DELIVER FLUIDS TO AN INSERTION SITE".
[00336] In preferred embodiments, the sensor 26 is inserted through the user's skin 46 using an insertion needle 58, which is removed and disposed of once the sensor is positioned in the subcutaneous tissue 44. The insertion needle 58 has a sharpened tip 59 and an open slot 60 to hold the sensor during insertion into the skin 46, as shown in FIGS.
3C and D and FIG. 4. Further description of the needle 58 and the sensor set 28 are found in U.S. Pat. No. 5,586,553, entitled "TRANSCUTANEOUS SENSOR INSERTION
SET" (published as PCT Application WO 96/25088); and U.S. Pat. No. 5,954,643, entitled "INSERTION SET FOR A TRANSCUTANEOUS SENSOR" (published as PCT
Application WO 98/5629).
[00337] In preferred embodiments, the sensor 26 has three electrodes 42 that are exposed to the interstitial fluid (ISF) in the subcutaneous tissue 44 as shown in FIGS. 3D
and 4. A working electrode WRK, a reference electrode REF and a counter electrode CNT are used to form a circuit, as shown in FIG. 7. When an appropriate voltage is supplied across the working electrode WRK and the reference electrode REF, the ISF
provides impedance (R1 and R2) between the electrodes 42. And an analog current signal Isig flows from the working electrode WRK through the body (RI and R2, which sum to Rs) and to the counter electrode CNT. Preferably, the working electrode WRK is plated with platinum black and coated with glucose oxidase (GOX), the reference electrode REF
is coated with silver-silver chloride, and the counter electrode is plated with platinum black. The voltage at the working electrode WRK is generally held to ground, and the voltage at the reference electrode REF is substantially held at a set voltage Vset. Vset is between 300 and 700 mV, and preferably to about 535 mV.
[00338] The most prominent reaction stimulated by the voltage difference between the electrodes is the reduction of glucose as it first reacts with GOX to generate gluconic acid and hydrogen peroxide (H202). Then the H202 is reduced to water (H20) and (01 at the surface of the working electrode WRK. The 0- draws a positive charge from the sensor electrical components, thus repelling an electron and causing an electrical current flow.
This results in the analog current signal Isig being proportional to the concentration of glucose in the ISF that is in contact with the sensor electrodes 42. The analog current signal Isig flows from the working electrode WRK, to the counter electrode CNT, typically through a filter and back to the low rail of an op-amp 66. An input to the op-amp 66 is the set voltage Vset. The output of the op-amp 66 adjusts the counter voltage Vent at the counter electrode CNT as Isig changes with glucose concentration. The voltage at the working electrode WRK is generally held to ground, the voltage at the reference electrode REF is generally equal to Vset, and the voltage Vent at the counter electrode CNT varies as needed.
[00339] In alternative embodiments, more than one sensor is used to measure blood glucose. In particular embodiments, redundant sensors are used. The user is notified when a sensor fails by the telemetered characteristic monitor transmitter electronics. An indicator may also inform the user of which sensors are still functioning and/or the number of sensors still functioning. In other particular embodiments, sensor signals are combined through averaging or other means. If the difference between the sensor signals exceeds a threshold then the user is warned to recalibrate or replace at least one sensor. In other alternative embodiments, more than one glucose sensor is used, and the glucose sensors are not of the same design. For example, an internal glucose sensor and an external glucose sensor may be used to measure blood glucose at the same time.

1003401 In alternative embodiments, other continuous blood glucose sensors and sensor sets may be used. In particular alternative embodiments, the sensor system is a micro needle analyte sampling device such as described in U.S. patent application Ser.

No. 09/460,121, filed on Dec. 13, 1999, entitled "INSERTION SET WITH
MTCROPTERCING MEMBERS AND METHODS OF USING THE SAME", or an internal glucose sensor as described in U.S. Pat.
Nos. 5,497,772; 5,660,163; 5,791,344; and 5,569,186, and/or a glucose sensor that uses florescence such as described in U.S. Pat. No. 6,011,984.
In other alternative embodiments, the sensor system uses other sensing technologies such as described in Patent Cooperation Treaty publication No. WO

99/29230, light beams, conductivity, jet sampling, micro dialysis, micro-poration, ultra sonic sampling, reverse iontophoresis, or the like. In still other alternative embodiments, only the working electrode WRK is located in the subcutaneous tissue and in contact with the ISF, and the counter CNT and reference REF electrodes are located external to the body and in contact with the skin. In particular embodiments, the counter electrode CNT
and the reference electrode REF are located on the surface of a monitor housing 518 and are held to the skin as part of the telemetered characteristic monitor, as shown in FIG.
34A. In other particular embodiments, the counter electrode CNT and the reference electrode REF are held to the skin using other devices such as running a wire to the electrodes and taping the electrodes to the skin, incorporating the electrodes on the underside of a watch touching the skin, or the like. In more alternative embodiments, more than one working electrode WRK is placed into the subcutaneous tissue for redundancy. In additional alternative embodiments, a counter electrode is not used, a reference electrode REF is located outside of the body in contact with the skin, and one or more working electrodes WRK are located in the ISF. An example of this embodiment implemented by locating the reference electrode REF on a monitor housing 520 is shown in FIG. 34B. In other embodiments, ISF is harvested from the body of an individual and flowed over an external sensor that is not implanted in the body.
[00341] Sensor Cable [00342] In preferred embodiments, the sensor cable 32 is of the type described in U.S.
patent application Ser. No. 60/121,656, filed on Feb. 25, 1999, entitled "TEST
PLUG
AND CABLE FOR A GLUCOSE MONITOR".
In other embodiments, other cables may be used such as shielded, low noise cables for carrying nA currents, fiber optic cables, or the like. In alternative embodiments, a short cable may be used or the sensor may be directly connected to a device without the need of a cable.

[00343] Telemetered Characteristic Monitor Transmitter [00344] In preferred embodiments, the telemetered characteristic monitor transmitter 30 is of the type described in U.S. patent application Ser. No. 09/465,715, filed on Dec.
17, 1999, entitled "TELEMETERED CHARACTERISTIC MONITOR SYSTEM AND
METHOD OF USING THE SAME" (published as PCT Application WO 00/19887 and entitled, "TELEMETERED CHARACTERISTIC MONITOR SYSTEM"), and is connected to the sensor set 28 as shown in FIGS.
3A and B.
[003451 In alternative embodiments, the sensor cable 32 is connected directly to the infusion device housing, as shown in FIG. 8A, which eliminates the need for a telemetered characteristic monitor transmitter 30. The infusion device contains a power supply and electrical components to operate the sensor 26 and store sensor signal values.
[00346] In other alternative embodiments, the telemetered characteristic monitor transmitter includes a receiver to receive updates or requests for additional sensor data or to receive a confirmation (a hand-shake signal) indicating that information has been received correctly. Specifically, if the telemetered characteristic monitor transmitter does not receive a confirmation signal from the infusion device, then it re-sends the information. In particular alternative embodiments, the infusion device anticipates receiving blood glucose values or other information on a periodic basis. If the expected information is not supplied when required, the infusion device sends a "wake-up" signal to the telemetered characteristic monitor transmitter to cause it to re-send the information.
[00347] Insulin Delivery System [00348] Infusion device [00349] Once a sensor signal 16 is received and processed through the controller 12, commands 22 are generated to operate the infusion device 34. In preferred embodiments, semi-automated medication infusion devices of the external type are used, as generally described in U.S. Pat. Nos. 4,562,751; 4,678,408; 4,685,903; and U.S. patent application Ser. No. 09/334,858, filed on Jun. 17, 1999, entitled "EXTERNAL INFUSION
DEVICE
WITH REMOTE PROGRAMMING, BOLUS ESTIMATOR AND/OR VIBRATION
CAPABILITIES" (published as PCT application WO 00/10628).
In alternative embodiments, automated implantable medication infusion devices, as generally described in U.S. Pat. Nos. 4,373,527 and 4,573,994, are used.

[00350] Insulin [00351] In preferred embodiments, the infusion device reservoir 50 contains HUMALOG lispro insulin to be infused into the body 20. Alternatively, other forms of insulin may be used such as HUMALINg, human insulin, bovine insulin, porcine insulin, analogs, or other insulins such as insulin types described in U.S. Pat. No.
5,807,315, entitled "METHOD AND COMPOSITIONS FOR THE DELIVERY OF MONOMERIC
PROTEINS", and U.S. Patent Application Ser. No. 60/177,897, filed on Jan. 24, 2000, entitled "MIXED BUFFER SYSTEM FOR STABILIZING POLYPEPTIDE
FORMULATIONS", or the like. In further alternative embodiments, other components are added to the insulin such as polypeptides described in U.S. patent application Ser. No. 09/334,676, filed on Jun. 25, 1999, entitled "MULTIPLE AGENT DIABETES THERAPY", small molecule insulin mimetic materials such as described in U.S. patent application Ser. No. 09/566,877, filed on May 8,2000, entitled "DEVICE AND METHOD FOR INFUSION OF SMALL MOLECULE
INSULIN MIMETIC MATERIALS", or the like.
[00352] Infusion Tube [00353] In preferred embodiments, an infusion tube 36 is used to carry the insulin 24 from the infusion device 34 to the infusion set 38. In alternative embodiments, the infusion tube carries the insulin 24 from infusion device 34 directly into the body 20. in further alternative embodiments, no infusion tube is needed, for example if the infusion device is attached directly to the skin and the insulin 24 flows from the infusion device, through a cannula or needle directly into the body. In other alternative embodiments, the infusion device is internal to the body and an infusion tube may or may not be used to carry insulin away from the infusion device location.
[00354] Infusion Set [00355] In preferred embodiments, the infusion set 38 is of the type described in U.S.
Pat. No. 4,755,173, entitled "SOFT CANNULA SUBCUTANEOUS INJECTION SET".
In alternative embodiments, other infusion sets, such described in U.S. Pat. Nos. 4,373,527 and 4,573,994, are used.
In alternative embodiments, other infusion sets, such as the Rapidlet from Disetronic, the Silhouedetrom MiniMed, or the like, may be used. In further alternative embodiments, no infusion set is required, for example if the infusion device is an internal infusion device or if the infusion device is attached directly to the skin.
1003561 Configurations With Supplemental Devices 1003571 In further alternative embodiments, the pre-filter, filters, calibrator and/or = controller 12 are located in a supplemental device that is in communication with both the telemetered characteristic monitor transmitter 30 and the infusion device 34.
Examples of supplemental devices include, a hand held personal digital assistant such as described in U.S. patent application Ser. No. 09/487,423, filed on Jan. 20, 2000, entitled "HANDHELD PERSONAL DATA ASSISTANT (PDA) WITH A MEDICAL DEVICE
AND METHOD OF USING THE SAME", a computer, a module that may be attached to the telemetered characteristic monitor transmitter 30, a module that may be attached to the infusion device 34, a RF
programmer such as described in U.S. patent application Ser. No. 09/334,858, filed on Jun. 17, 1999, entitled EXTERNAL INFUSION DEVICE WITH REMOTE PROGRAMMING, BOLUS ESTIMATOR AND/OR VIBRATION CAPABILITIES (published as PCT
application WO 00/10628), or the like. In particular embodiments, the supplemental device includes a post-calibration filter, a display, a recorder, and/or a blood glucose meter. In further alternative embodiments, the supplemental device includes a method for a user to add or modify information to be communicated to the infusion device 34 and/or the telemetered characteristic monitor transmitter 30 such as buttons, a keyboard, a touch screen, and the like.
[00358] In particular alternative embodiments, the supplemental device is a computer in combination with an analyte monitor and a RF programmer. The analyte monitor receives RF signals from the telemetered characteristic monitor transmitter 30, stores the signals and down loads them to a computer when needed. The RF programmer sends control signals to the infusion device 34 to reprogram the rate of insulin infusion. Both the analyte monitor and the RF programmer are placed into separate communication stations. The communication stations include IR transmitters and IR receivers to communicate with the analyte monitor and the RF programmer. The sensor signal values are transmitted via the telemetered characteristic monitor transmitter 30 to the analyte monitor located in one of the communication stations. Then the sensor signal values are communicated through the IR receiver in a first communication station and to the computer. The computer processes the sensor signal values through one or more filters, calibrators, and controllers to generate commands 22. The commands arc sent to a second communication station and sent to an RF programmer by the IR transmitter in the communication station. Finally the RF programmer transmits the commands 22 to the infusion device 34. The communication station, analyte monitor and infusion device 34 may be of the type described in U.S. patent application Ser. No. 09/409,014, filed on Sep.
29, 1999 entitled COMMUNICATION STATION FOR INTERFACING WITH AN
INFUSION PUMP, ANALYTE MONITOR, ANALYTE METER OR THE LIKE
(published as a PCT application WO 00/18449).
Alternatively, the RF programmer may be omitted and the infusion device may be placed in a communication station, or the infusion device may receive the commands without the use of an RF programmer and/or a communication station.
[00359] Overnight Closed-Loop System [00360] A closed-loop insulin delivery system of the type described herein may utilize a variety of control algorithms to regulate the delivery of insulin to the body of the patient in a safe and predictable manner. Overnight operation of a closed-loop insulin infusion system should be carefully controlled in an automated way that need not depend on patient, user, or caregiver interaction. In this regard, a number of safeguards can be implemented with the system. These safeguards are intended to provide actionable sensor glucose readings, assess the accuracy of sensor readings, and constrain insulin delivery based upon possible sensor over-read conditions. These safeguards will alert the user and allow the patient to take appropriate actions. Therefore, these safeguards will mitigate the potential risks of overnight closed-loop control.
[00361] The control algorithm utilized by the system may be considered to be one type of safeguard in that it emulates the effect of insulin inhibiting insulin secretion. The system may also implement sensor performance safeguards. For example, a closed-loop initiation algorithm determines if the system can enter the closed-loop mode by calculating a recent calibration factor. The initiation algorithm checks the time between recent and prior calibration factors and determines the relative sensor error between the readings. As another example of a sensor safeguard, the system may employ a model supervisor during the closed-loop mode. The model supervisor checks that the sensor glucose readings are adequate for use during overnight closed-loop mode by comparing model-predicted sensor glucose values in real-time against actual sensor glucose values.
If the model-predicted glucose values and the actual values differ significantly, the system triggers a fail-safe alert indicating a faulty sensor. This fail-safe alert can be generated in response to a number of sensor issues, such as sensor drift, sensor dislodgement, sensor compression artifact, etc.
1003621 The system may also implement a target glucose level safeguard. In this regard, a start-up algorithm can be deployed to provide a smooth transition between the open-loop mode and the closed-loop mode by gradually adjusting the target glucose level while in the closed-loop mode. The adjusted target glucose is used by the closed-loop control algorithm until the adjusted target glucose converges to a particular setpoint. At that time, the setpoint can be used for future dosing calculations during the closed-loop mode.
1003631 The system may also utilize at least one insulin limit as an insulin delivery and sensor performance safeguard. In this context, the insulin limit constrains the maximum amount of insulin delivered to the patient at any time in order to avoid over-delivery of insulin by the closed-loop control system due to potential sensor faults. In practice, the insulin limit is a value that is specific to each patient and is calculated based on the patient's basal rate, fasting blood glucose, and insulin sensitivity.
1003641 The system may also employ one or more insulin delivery safeguards.
For example, an insulin delivery timeout continuously monitors (during closed-loop operation) if the patient is receiving insulin at the insulin limit for a prolonged period of time and, if so, triggers a fail-safe alert. This safeguard also monitors if the system is not delivering insulin for a prolonged period of time and, if so, triggers a fail-safe alert. A
correction bolus is another insulin delivery safeguard. The system calculates an insulin bolus dosage for mitigating hyperglycemia at the commencement of closed-loop mode if the patient is above a designated blood glucose threshold. The determination can be achieved by acquiring a blood glucose meter reading at the initiation of closed-loop mode. The correction bolus is calculated based on the patient's insulin sensitivity, the amount of insulin on board, and a glucose target. Insulin on board (I013) compensation is yet another insulin delivery safeguard. IOB compensation estimates the amount of insulin on board based on manual boluses administered, such that the system can effectively account for the JOB. In this regard, the manual boluses may be subtracted from the insulin dose that is calculated by the PID-IFB control algorithm.
1003651 The system may also implement one or more communication safeguards.
For example, a "missed sensor transmission" feature continuously monitors data being received by the controller. For missed data packets totaling less than 15 minutes of operating time, the system remains in closed-loop mode. During this time, however, the system continues to calculate the insulin dose using the closed loop control algorithm based on the last valid sensor glucose value. For missed data packets totaling minutes, the safeguard will switch to a pre-programmed safe basal rate, defined as half the patient's night time basal rate. If the controller starts receiving data packets during the safe basal rate timeframe, the system will again switch to the closed-loop mode. For missed data packets totaling more than 60 minutes, the system will switch to the open-loop mode where it will deliver a pre-programmed basal rate (which may be set by a caregiver).
1003661 The exemplary closed-loop control algorithms, methodologies, and techniques described in more detail below may be based around a PID control algorithm of the type presented in the preceding sections of this disclosure. In certain embodiments, the closed-loop control algorithms utilize a PID insulin feedback (PID-IFB) control algorithm. More specifically, the PID-IFB control algorithm cooperates with other algorithms, processes, and controls that represent additional safeguards that may apply during overnight use (and/or during other periods of use). These additional safeguards may include, without limitation: the use of an "Insulin Limits" parameter; a closed-loop initiation circuit that is based on glucose sensor calibration; an insulin on board (JOB) compensation algorithm;
monitoring missed transmissions; and monitoring sensor glucose against predicted sensor glucose.
1003671 In practice, optimal or desired settings for the Insulin Limits parameter should be determined. In this regard, the Insulin Limits parameter serves as an input to the controller logic for each patient, and it imposes an upper limit to the insulin delivery rate as an additional safety feature to avoid over-delivery of insulin by the controller due to potential sensor error. In certain embodiments, the Insulin Limits parameter is calculated from the patient's basal rate, fasting blood glucose, and insulin sensitivity.
[00368] Referring again to FIG. 1, a closed-loop system generally includes a glucose sensor system 10, a controller 12, and an insulin delivery system 14. Although FIG. 1 depicts these primary elements as separate blocks, embodiments of the system may combine two or more of the illustrated blocks into a single physical component. For example, an investigational test configuration of the closed-loop system may include a traditional patient-worn infusion pump (corresponding to the insulin delivery system 14), a conventional continuous glucose sensor/transmitter assembly (corresponding to the glucose sensor system 10), and a mobile computing device with a suitably written software application installed thereon (corresponding to the controller 12).
The mobile computing device may be, for example: a smartphone; a tablet computer; a netbook computer; a digital media player; a handheld video game device; or the like.
It should be appreciated that the desired closed-loop control functionality can be carried out by way of one or more computer-executable programs or applications designed to run on the mobile computing device. An investigational test configuration may also include a translator device that serves as a data communication interface between the mobile computing device (which may utilize standard wireless data communication technologies such as the Wi-Fi or BLUETOOTH data communication protocol) and the glucose sensor system (which may use a proprietary data communication protocol that is usually incompatible with the mobile computing device).
100369] In other embodiments, the functionality of the glucose sensor system 10 could be integrated into the insulin delivery system 14, perhaps as an interchangeable disposable module that attaches to the housing of the insulin delivery system 14. In yet other embodiments, the functionality of the controller 12 could be incorporated into the insulin delivery system 14 such that a separate and distinct controller device need not be carried by the patient. Indeed, the control software utilized by the controller 12 can be ported for installation in an insulin infusion pump, a pump monitor device, or the like to implement the functionality of the controller 12 in those devices if so desired. In further embodiments, a single hardware device platform could be suitably designed to accommodate the functionality of the insulin delivery system 14, the glucose sensor system 10, and the controller 12. These and other possible implementations are contemplated by this disclosure, and the particular manner in which the closed-loop system is configured and deployed is not intended to limit or otherwise restrict the scope or application of the closed-loop control techniques described herein.
[00370] Although not shown in FIG. 1, the closed-loop system may include or cooperate with a conventional blood glucose meter (e.g., a finger stick device) that provides measured BG values to the controller 12 and/or to the insulin delivery system 14, such that the glucose sensor system 10 can be calibrated. In certain embodiments, the measured BG values are sent to the insulin delivery system 14, which in turn sends the BG value, sensor calibration factor, and calibration time to the controller 12. The controller 12 can process and analyze the received information to determine whether or not the system can enter the closed-loop operating mode. In this regard, the controller 12 may check to ensure that the calibration of the glucose sensor system 10 is within an acceptable range before allowing the system to enter the closed-loop mode.

1003711 After entering the closed-loop mode, the insulin delivery system 14 sends sensor glucose (SG) values, sensor Isig values, calibration factors, "insulin delivered"
values, and other data as needed to the controller 12 in accordance with a predetermined schedule, e.g., at five minute intervals. The controller 12 determines the desired insulin dose based on the closed-loop algorithm to maintain the patient at a target glucose setpoint, and communicates suitable control data and instructions to the insulin delivery system 14. The insulin delivery system 14 responds to deliver the insulin dose specified by the controller 12 to the user.
1003721 FIG. 49 is a block diagram that illustrates processing modules and algorithms of an exemplary embodiment of a closed-loop system controller 900, and FIG. 50 is a flow chart that illustrates an exemplary embodiment of a control process 1000 that may be performed at least in part by the controller 900 to control the insulin delivery system 14. The controller 12 shown in FIG. 1 may be configured in accordance with that shown in FIG. 49. FIG. 49 schematically depicts certain inputs and outputs of the controller 900, where the parallelograms represent the inputs, the ovals represent the outputs, and the rectangles represent the various functional modules of the controller 900. In the context of this description, a "functional module" may be any process, technique, method, algorithm, computer-executable program logic, or the like. In this regard, the controller 900 could be realized as any electronic device having a processor architecture with at least one processor device, and at least one memory element that is cooperatively associated with the processor architecture. The processor architecture is suitably configured to execute processor-executable instructions stored in the at least one memory element such that the controller 900 can perform the various control operations and methods described in detail herein.
[00373] The host electronic device that implements the controller 900 may be realized as a monitor device for an insulin infusion device, where the monitor device and the insulin infusion device are two physically distinct hardware devices. In another embodiment of the system, the host electronic device that implements the controller 900 may be realized as a portable wireless device, where the portable wireless device and the insulin infusion device are two physically distinct hardware devices. The portable wireless device in this context may be, without limitation: a mobile telephone device; a tablet computer device; a laptop computer device; a portable video game device; a digital media player device; a portable medical device; or the like. In yet other system embodiments, the host electronic device and the insulin infusion device are physically and functionally integrated into a single hardware device. In such embodiments, the insulin infusion device will include the functionality of the controller 900 as presented here.
[00374] Certain embodiments of the controller 900 include a plurality of cooperating functional modules that are designed and configured to determine the insulin dose to be delivered to keep the patient at the target glucose setpoint during an overnight closed-loop operating mode. In this regard, the illustrated embodiment of the controller 900 may include the following functional modules, without limitation: a closed-loop initiation module 902; a start-up module 904; a proportional integral derivative insulin feedback (PID-IFB) control module 906; an insulin limit module 908; an insulin on board (I0B) compensation module 910; an insulin delivery timeout module 912; a model supervisor module 914; and a missed transmission module 916.
[00375] Referring to FIG. 50, the control process 1000 may begin at any time when it is desired to enter the closed-loop operating mode. Accordingly, the control process 1000 may begin in response to a user-initiated command, automatically in response to the detection of operating conditions that are usually indicative of closed-loop operation (e.g., sleeping), or the like. Certain embodiments of the control process 1000 may begin with one or more system checks (task 1002) to confirm whether or not the system is allowed to enter the closed-loop operating mode. This particular example employs a sensor calibration check before allowing the system to proceed to the closed-loop mode.
Referring to FIG. 49, the closed-loop initiation module 902 may be involved during task 1002.
[00376] In some embodiments, the closed-loop initiation module 902 may consider certain sensor performance criteria that prevents closed-loop initiation. Such criteria may include, without limitation: (1) during start-up when the calibration is not stable; (2) when the sensor sensitivity changes significantly; (3) when sensors may be calibrated with a potentially invalid meter reading thereby changing the sensor sensitivity significantly; (4) any other situation that could cause a mismatch between the sensor and meter for a number of most recent calibrations spaced over a designated period of time (e.g., the two most recent calibrations).
[00377] The illustrated embodiment of the closed-loop initiation module 902 receives at least the following items as inputs: a meter (measured) BG value 920; at least one sensor calibration factor 922 (i.e., calibration measurements, calibration data, etc.); the current sensor lsig value 924; and timestamp data 926 that indicates the calibration time associated with the BG value 920 and the sensor calibration factor 922. Some or all of this input data may be provided directly or indirectly by the insulin delivery system 14 (see FIG. 1), a translator device, a monitor device, or any device in the closed-loop system. This description assumes that a new sensor calibration factor 922 and new timestamp data 926 is generated for each measured BG value 920, wherein the sensor calibration factor 922 is associated with the calibration of the glucose sensor system 10 (see FIG. 1) that is being used to monitor the patient. In particular, the sensor calibration factor may be based on the meter BG value 920 and the corresponding sensor Isig value 924.
[00378] The closed-loop initiation module 902 analyzes the input data (both current values and historical values) to determine whether or not the system is allowed to enter into the closed-loop mode. For example, the closed-loop initiation module 902 may:
check the period between two consecutive calibration timestamp values; compare recent and prior calibration factor values; and the like. The "outputs" of the closed-loop initiation module 902 correspond to two operating modes of the system. More specifically, the closed-loop initiation module 902 controls whether the system remains operating in the open-loop mode 928 or whether the system starts the closed-loop mode 930.
1003791 Referring to FIG. 50, if the closed-loop mode is not permitted (the "No"
branch of query task 1004), then the control process 1000 operates the system such that it remains in the open-loop mode (task 1006). On the other hand, if the closed-loop mode is permitted (the "Yes" branch of query task 1004), then the control process 1000 can initiate and start the closed-loop mode in an appropriate manner (task 1008).
Referring again to FIG. 49, a correction bolus 932 can be calculated and delivered (if needed) to mitigate hyperglycemia at the commencement of the closed-loop mode. This correction bolus 932 serves as an additional safeguard to achieve a target blood glucose level if a measured meter reading is greater than a threshold value. If the control process 1000 determines that a correction bolus is required, then an appropriate insulin dose instruction is generated for execution by the insulin delivery system at the outset of the closed-loop mode.
[00380] Referring to FIG. 49, the start-up module 904 may be called in response to a determination that the system can proceed to the closed-loop operating mode.
Once the system is in the closed-loop mode, the controller retrieves historical data that can be processed and used as described in more detail below. In certain embodiments, for example, the controller obtains data for the last 24 hours (from the insulin delivery system, from a monitor, or the like). Thereafter, the controller retrieves data packets once every sampling period to obtain, without limitation: sensor glucose (SG) values; sensor Isig values; sensor calibration factors; information related to the amount of insulin delivered; information related to manual boluses delivered; and sensor calibration factors.
As explained in more detail below, the received information can be used in the various safeguards, and to determine the final insulin dose.
1003811 The start-up module 904 receives sensor glucose (SG) values 940 as an input, and the functionality of the start-up module 904 may be initiated in response to the start of the closed-loop mode 930 (this trigger mechanism is represented by the dashed arrow 942 in FIG. 49). The SG values 940 may be provided directly by the glucose sensor system 10 or indirectly via the insulin delivery system 14, a translator device, or any device in the closed-loop system (see FIG. 1). This description assumes that SG values 940 are received by the start-up module 904 in an ongoing manner as they become available. The start-up module 904 may also utilize a target glucose setpoint value 944, which may be internally maintained, generated, andior provided by the controller 900.
For the implementation presented here, the target glucose setpoint value 944 represents a fixed (constant) value that the user can specify (FIG. 49 depicts the target glucose setpoint value 944 in dashed lines to indicate that the value is a user-specified parameter rather than a functional module or data received by the system).
1003821 In certain embodiments, the start-up module 904 calculates a final target glucose value 946, which serves as an input to the PID-IFB control module 906.
The final target glucose value 946 enables the system to make a smoother transition between open-loop and closed-loop modes (by gradually adjusting the final target glucose value 946).
The start-up module 904 may utilize the target glucose setpoint value 944 to calculate the final target glucose value 946. In this regard, the start-up module 904 elevates the final target glucose value 946 to the same level as the sensor glucose value at the start of the closed-loop mode, provided the sensor glucose is above a certain threshold. As time progresses, the final target glucose value 946 gradually decreases back to the target glucose setpoint value 944 (usually in approximately two hours). Referring to FIG. 50, the control process 1000 calculates the final target glucose value (task 1010) and continues by calculating an uncompensated insulin infusion rate, PIDRate(n), based at least in part on the final target glucose value (task 1012). For this example, the start-up module 904 may be involved during task 1010, and the PID-IFB control module 906 may be involved during task 1012.
[00383] As an additional safeguard, the insulin limit module 908 cooperates with the PID-IFB control module 906 to provide an upper insulin limit that is calculated based on the patient's basal rate, fasting blood glucose, and insulin sensitivity. This upper insulin limit imposes an upper limit to the insulin delivery rate to avoid over-delivery of insulin by the system due to potential sensor error.
[00384] The PID-IFB control module 906 may be configured to carry out the control processes described in more detail above with reference to FIGS. 1-48. In some embodiments, the PID-IFB control module 906 receives at least the following items as inputs: the SG value 940 (which may be used to calculate a rate of change value that indicates the rate of change of the SG value); the current sensor Isig value 950; the current sensor calibration factor 952; and an amount of insulin delivered 954.
The inputs to the PID-IFB control module 906 may be provided directly or indirectly by the insulin delivery system 14, the glucose sensor system 10, a translator device, a monitor device, and/or any device in the closed-loop system (see FIG. 1). The PID-IFB control module 906 is suitably configured to calculate the insulin infusion rate based on the current and past SG values 940, the SG rate of change, the sensor Isig value 950, the sensor calibration factor 952, the final target glucose value 946, and the insulin delivered 954 in order to achieve euglycemia. These (and possibly other) values may be received by the P1D-IFB control module 906 in an ongoing manner as they become available, e.g., in five minute intervals or in accordance with any desired schedule.
[00385] The insulin delivered 954 is a parameter or value that indicates the amount of insulin that has been delivered to the patient by the insulin delivery system.
Thus, the insulin delivered 954 may indicate recent boluses (typically by Units) delivered over a period of time. In certain implementations, the insulin delivered 954 corresponds to the amount of insulin delivered in the last sampling time, which may be, without limitation:
one minute; five minutes; thirty seconds; or any designated sampling time. The insulin delivered 954 may also indicate the amount of insulin delivered by the delivery system as basal or boluses in any defined period of time in the past (e.g., the last N
hours) or the amount of insulin delivered by the system in the last sampling cycle. In practice, the PID-IFB control module 906 (and the JOB compensation module 910) may be "initialized" to collect and save historical values for the insulin delivered 954 as needed.
Thereafter, the insulin delivered 954 can simply indicate an amount of insulin administered by the system during the last sampling time period if by a bolus or basal channels.
[003861 As mentioned above, the PID-IFB control module 906 may utilize an upper insulin limit, which is a patient-specific parameter. In certain embodiments, the upper insulin limit may be entered by the user, a caregiver, or the like.
Alternatively, the insulin limit module 908 may be responsible for calculating or otherwise managing the upper insulin limit if so desired. The upper insulin limit imposes an upper limit to the insulin delivery rate as an additional safety feature to avoid over-delivery of insulin by the controller 900 due to potential sensor error. Thus, if the PID-IFB control module 906 recommends a dose higher than the insulin limit, the insulin limit will constrain the insulin delivered to the insulin limit value. In addition, the insulin limit will "freeze" the integral component of the PID to its previous value to prevent integral windup, which can cause continuous integrating of the glucose error until it reaches maximum values. In certain embodiments, the upper insulin limit has a default value set at five times the patient's basal rate. Hence, if the maximum value is reached, the PID-IFB
control algorithm will be fairly aggressive in calculating an insulin dose.
Accordingly, to minimize integral windup, the insulin limit is fed back to the PID-IFB control module 906 (as depicted in FIG. 49) for use in the next insulin dose calculation.
1003871 The PID-IFB control module 906 operates as described previously to calculate a current insulin dose 958 as an output value (the current insulin dose 958 is also referred to herein as the uncompensated insulin infusion rate, PIDRate(n)). Tn practice, the current insulin dose 958 is typically expressed as an infusion rate (Units/Hour). In the context of this description, the current insulin dose 958 represents a baseline closed-loop infusion rate, which may be subjected to further adjustment or compensation by the JOB
compensation module 910. Referring again to FIG. 50, the control process 1000 may compensate for the insulin "on board" the patient by calculating an adjusted insulin infusion rate, AdjustedRate(n), based at least in part on the uncompensated insulin infusion rate (task 1014). For this example, the JOB compensation module 910 may be involved during task 1014.
1003881 The 10B compensation module 910 receives at least the following items as inputs: the current insulin dose 958; and information regarding manual boluses delivered 960. The manual boluses delivered 960 may be provided directly or indirectly by the insulin delivery system 14, a translator device, a monitor device, and/or any device in the closed-loop system (see FIG. 1). This description assumes that the manual boluses delivered 960 is received by the 10B compensation module 910 in an ongoing manner as it becomes available, e.g., in five minute intervals or in accordance with any desired schedule. The JOB compensation module 910 is suitably configured to estimate insulin on board based on manual boluses delivered, before or during closed-loop operation, in order to compensate the final infusion rate to help avoid over-delivery of insulin by the controller 900. Accordingly, the output of the 10B compensation module 910 may be a final insulin dose 962 expressed as a final infusion rate (Units/Hour). The final insulin dose 962 is also referred to herein as the adjusted insulin infusion rate, AdjustedRate(n).
[00389] Referring to FIG. 50, the control process 1000 uses the adjusted insulin infusion rate, AdjustedRate(n), to control the insulin infusion device, which in turn regulates the delivery of insulin to the body of the user (task 1016). In certain embodiments, the adjusted insulin infusion rate is communicated to the insulin infusion device in an appropriate manner (such as wireless data communication). The control process 1000 may continue as described above in an iterative and ongoing manner to monitor the condition of the user and deliver insulin as needed without user involvement.
That said, if the control process 1000 determines that the closed-loop operating mode should be terminated (the "Yes" branch of query task 1018), then the control process 1000 causes the system to switch back to the open-loop mode (task 1020). The closed-loop mode may be ended in response to a user-initiated command, automatically in response to the detection of operating conditions that are usually indicative of open-loop operation, or the like.
[00390] If query task 1018 determines that the closed-loop mode should continue (the "No" branch of query task 1018), then the control process 1000 may check whether it is time to perform another iteration of the control routine. In other words, the control process 1000 may check for the next sampling time (query task 1022). If it is time for the next iteration, then the control process 1000 may return to task 1010 and repeat the computations with the next set of data values. For example, the next iteration of the control routine may obtain and process the current values of some or all of the following parameters, without limitation: the SG value 940; the SG rate of change; the sensor Isig value 924; the amount of insulin delivered 954; and the manual boluses delivered 960.
This allows the control process 1000 to adjust the final insulin infusion rate in an ongoing manner in accordance with a predetermined schedule, a designated sampling rate, or the like.

[00391] The insulin delivery timeout module 912 monitors if the patient is receiving continuous delivery of insulin at the maximum insulin limit or the minimum allowable infusion of zero Units/Hour for a time specified by the controller.
Accordingly, the insulin delivery timeout module 912 may receive the insulin delivered 954 as an input. If the specified time is exceeded, the system will trigger a fail-safe alert 966.
Otherwise, the system remains in the closed-loop operating mode 968.
[00392] Referring back to FIG. 49, the model supervisor module 914 receives at least the following as inputs: the insulin delivered 954; sensor Isig values 950;
and one or more sensor calibration factors 952. The inputs to the model supervisor module 914 may be provided directly or indirectly by the insulin delivery system 14, the glucose sensor system 10, a translator device, a monitor device, and/or any device in the closed-loop system (sec FIG. 1). The model supervisor module 914 is suitably designed and configured to estimate the user's glucose concentration in real time (or substantially real time) based on the insulin delivered 954, the sensor Isig values 950, and the sensor calibration factors 952. The sensor calibration factors 952 used by the model supervisor module 914 are equal to the sensor calibration factors 922 used by the closed-loop initiation module 902. That said, the closed-loop initiation module 902 utilizes the sensor calibration factors 922 at one particular time, whereas the model supervisor module 914 considers the sensor calibration factors 952 in an ongoing and continuous manner during operation in the closed-loop mode. Should the model-predicted glucose and the sensor glucose values differ significantly, the system will exit closed loop mode.
Accordingly, the model supervisor module 914 regulates whether the system remains in the closed-loop mode 974 or switches to the open-loop mode 976.
[00393] The missed transmission module 916 is suitably configured to monitor the following, without limitation: the sensor Isig values 950; the SG values 940;
and the sensor calibration factors 952. More particularly, the missed transmission module 916 continuously monitors to check whether the system is receiving data packets that convey the necessary information and input values. For missed data packets totaling less than a lower threshold of time (e.g., 15 minutes), the system remains in the closed-loop mode, as indicated by block 980 in FIG. 49. During this time, the system will continue to calculate the insulin dose using the closed-loop control methodology based on the last valid sensor glucose value. For missed data packets totaling a time longer than the lower threshold and shorter than an upper threshold of time (e.g., 60 minutes), the missed transmission module 916 will switch the system to a pre-programmed safe basal rate, as indicated by block 982 in FIG. 49. In certain embodiments, this safe basal rate is defined as half the patient's overnight basal rate, and this parameter may be programmed by a caregiver or physician. If the missed transmission module 916 starts receiving data packets while the safe basal rate is being administered, the system will switch back to the closed-loop mode. For missed data packets totaling more than the upper threshold of time, the system will switch to the open-loop mode, as indicated by block 984 in FIG. 49. At this point, the system will be controlled to deliver a pre-programmed open-loop overnight basal rate.
[00394] To summarize, the controller 900 determines whether to enter into the closed-loop mode in response to at least the recent meter BG values 920, the sensor calibration factors 922, and the calibration timestamp data 926. The controller 900 utilizes the closed-loop initiation module 902 to check if the sensor calibration time between the last two calibration values is within an acceptable range, and whether any change between the two calibration values (recent and prior value) is acceptable. If so, the controller 900 will switch the system into the closed-loop mode. Once the system is in the closed-loop mode, the controller 900 will periodically receive data packets (e.g., every five minutes) that include the current SG value 940, the current sensor lsig values 950, the insulin delivered 954, the sensor calibration factors 952, and manual boluses delivered 960. In certain embodiments, each of the data packets received by the controller 900 includes data collected during the previous 24-hour period.
[00395] The start-up module 904 utilizes the SG values 940 and the target glucose setpoint value 944 to calculate the final target glucose value 946. In some embodiments, the target glucose setpoint value 944 is set to 120 mg/dL, although other settings could be used if so desired (a typical range of settings may be, for example 70-300 mg/dL). This results in a smoother transition between open-loop and closed-loop modes by gradually adjusting the final target glucose value 946. The final target glucose value 946 is sent to the PID-IFB control module 906 for use as one input to calculate the final insulin dose 962.
[00396] The PID-IFB control module 906 utilizes the final target glucose value 946, the current and past SG values 940, the SG rate of change values, and the insulin delivered 954 to determine the insulin infusion rate (final insulin dose 962) in order to achieve euglycemia. As an additional safeguard, the upper insulin limit (calculated based on the patient's basal rate, fasting blood glucose, and insulin sensitivity) from the insulin limit module 908 is input into the controller 900 for each patient to impose an upper limit to the insulin delivery rate to avoid over-delivery of insulin by the controller 900. The PID-IFB control module 906 considers the upper insulin limit before sending the final insulin dose 962 to the 10B compensation module 910, which estimates insulin on board from manual boluses, before or during closed-loop operation, in order to calculate the final insulin dose 962. The final insulin dose 962 may be communicated from the controller 900 directly or indirectly to the insulin delivery system 14 such that the final insulin dose 962 can be delivered to the patient during closed-loop operation.

[00397] Additional safeguards could be implemented to monitor the system during closed-loop operation, such that the system exits the closed-loop mode when certain criteria are not met. For example, the controller 900 may cause the system to exit the closed-loop mode if more than a designated number of consecutive data packets are missed. This assumes that the controller 900 usually receives data packets (from the insulin delivery system 14, from a monitor, from a translation device, or the like) in a continuous manner during closed-loop operation. Thus, if the controller 900 detects that more than a threshold number of consecutive data packets are not received as expected, the system will be commanded to exit the closed-loop mode. This functionality is associated with the missed transmission module 916, as described previously.
[00398] Moreover, the model supervisor module 914 estimates the user's glucose concentration in an ongoing manner, based on the insulin delivered 954, the sensor Isig values 950, and the sensor calibration factors 952. If the difference between the model-predicted glucose and the sensor glucose value is greater than a stated threshold, the controller 900 may cause the system to exit the closed-loop mode.
1003991 As summarized above, the controller 900 employs a number of modules or functions that cooperate to regulate the delivery of insulin during closed-loop operation:
the closed-loop initiation module 902; the start-up module 904; the PID-IFB
control module 906; the insulin limit module 908; and the 10B compensation module 910.

Moreover, the controller 900 may employ a number of modules that perform various safeguarding functions during closed-loop operation. These safeguarding modules may include: the insulin delivery timeout module 912; the model supervisor module 914; and the missed transmission module 916.
[00400] Closed-Loop Initiation Module: First Representation [00401] The closed-loop initiation module 902 checks for changes in sensor sensitivity and determines whether or not the system is allowed to enter the closed-loop mode.
Referring again to FIG. 49, the inputs to the closed-loop initiation module 902 include the meter BG value 920, the sensor calibration factor 922, and the calibration timestamp data 926. The closed-loop initiation module 902 checks for a series of conditions pertaining to the sensor calibration factor values 922 and the times when the sensor calibration factor values 922 were obtained. If all the conditions are met, the controller 900 initiates the closed-loop operating mode. If this criteria is not met, the system remains in the open-loop operating mode and the controller 900 requests a new sensor calibration.
[00402] Certain embodiments of the closed-loop initiation module 902 execute one or more functions, algorithms, or methods to determine whether or not the system can proceed to the closed-loop mode. The following are the parameters and variables used by an exemplary embodiment of the closed-loop initiation module 902:
t = time when attempting to enter the closed-loop mode;
Recent Calibration Factor (CFR) = the most recent sensor calibration factor (CF) value;
tR = the time when the CFR was obtained;
Prior Calibration Factor (CFP) = the last CF value before the CFR;
tP = the time when the CFP was obtained;
CFchange = percentage change in CF from a previous CF to a current CF, for any pair of CFs. The CFchange can be calculated according to the following equation:
CFchange = (abs(CFcurrent ¨ CFprevious)/CFprevious)* 100 (eq 50) tRecent = time window for the most recent calibration before attempting to start closed-loop mode (minutes) tDiffmin = minimum time difference between the recent calibration and the calibration prior to the recent calibration (minutes) tDiffmax = maximum time difference between the recent calibration and the prior calibration (minutes) CFmin = minimum acceptable CF (mg/dL per nA) CFmax = maximum acceptable CF (mg/dL per nA) CFprevious = the CF value before CFcurrent in the pair of CF values CFchangeTh = threshold for acceptable CFchange in % (mg/dL per nA) [00403] In some embodiments, the closed-loop initiation module 902 is implemented in the form of a series of processing steps. Using the logic described below, the closed-loop initiation module 902 decides whether to let the system enter the closed-loop mode.

100404.1 Case A
If (tP is not in the time window (tR-tDiffmin : tR)), the following logic is checked.
If (CFmin < CFR < CFmax) If (t-tRecent < tR < t) If (tR-tDiffmax < tP < tR-tDiffmin) If (CFmax < CFP < CFmax) Calculate CFchange with CFR as the CFcurrent and with CFP as the CFprevious in Equation 50 stated above If (CFchange < CFchangeTh) Enter Closed Loop Else Cannot enter closed loop at that time Else Cannot enter closed loop at that time Else Cannot enter closed loop at that time Else Cannot enter closed loop at that time Else Cannot Enter Closed Loop 1004051 If any of the above conditions is not met, the system remains in the open-loop mode. Thus, in order to enter the closed-loop mode, new calibration(s) that satisfy all of the conditions in Case A (or Case B as described below) will be required.
1004061 Case B
If (tP is in the time window (tR-tDiffmin : tR)), CFV = most recent CF value in the time window of tR-tDiffmax : tR-tDiffmin tV = time when CFV was obtained If (CFmin < CFR < CFmax) if there is a CFV available Calculate CFchange with CFR as the CFcurrent and with CFV
as the CFprevious in Equation 50 stated above If (CFV, CFR and all CF values between times tV and tR lie in the range of (CFmin:CFmax) AND CFchange between CFV
and CFR is < CFchangeTh) Enter closed loop Else Cannot enter closed loop at that time Else Cannot enter closed loop at that time Else Cannot enter closed loop at that time Else Cannot Enter Closed Loop 1004071 If any of the above conditions is not met, the system remains in the open-loop mode. Thus, in order to enter the closed-loop mode, new calibration(s) that satisfy all of the conditions in Case A or Case B will be required.
100408] In accordance with certain variations of the closed-loop initiation module 902, the system requests a meter BG and related calibration when entering the closed-loop mode. In such alternative embodiments, therefore, the closed-loop initiation module 902 uses the meter BG and Isig to calculate CFR. Thus, in such an implementation, the sensor current would also be an input to the closed-loop initiation module 902.
Accordingly, because CFR is now being calculated by the closed-loop initiation module 902 itself, the conditions in Case A and Case B (i.e., checking whether t-tRecent < tR < t) will always be met.
[00409] In particular implementations, some of the parameters mentioned above can be fixed. In this regard, the following values may be utilized in an exemplary embodiment. It should be appreciated that these values are provided here for illustrative purposes only, and that an implementation of the closed-loop initiation module 902 may utilize different values if so desired.
tRecent =120 minutes tDiffmin = 120 minutes tDiffmax = 480 minutes CFmin = 2.5 mg/dL per nA
CFmax =6 mg/dL per nA
1004101 Closed-Loop initiation Module: Second Representation [00411] In accordance with some embodiments, the functionality of the closed-loop initiation module 902 can be represented as follows. The closed-loop initiation module 902 may be implemented in the form of a series of case steps. In this regard, the closed-loop initiation module 902 first calculates the recent calibration factor value (CFR) using the most recent meter BG and Isig values as shown in the following Equation Al:
CFR = meterBGI(Isig ¨2) (eq Al) Here, CFR is the recent calibration factor value, meterBG is the meter BG
value, and Isig is the sensor Isig value. The "-2" in Equation Al represents a constant offset that is used by the calibration algorithm when calculating calibration factors and sensor glucose.
[00412] Using the logic described below for Case C or Case D, the closed-loop initiation module 902 decides whether to let the system enter the closed-loop mode. Each case condition is dependent upon the time at which the most recent prior calibration factor (CFP) was obtained.
[00413] Case C
[00414] Case C corresponds to a scenario where the time of prior calibration is greater than 120 minutes before the most recent calibration. In addition, the recent calibration factor (CFR) and the prior calibration factor (CFP) are within limits as shown in the following logical expressions.
CFmin < CFR < CFmax (eq A2) CFmin < CFP < CFmax (eq A3) Here, CFR is the recent calibration factor value, CFP is the prior calibration value, CFmin is the minimum value for the calibration factor that is set as 2.5 mg/dL
per nA, and CFmax is the maximum value for the calibration factor that is set as 6 mg/dL per nA.
[00415] For Case C, the time of recent calibration (tR) is within two hours of the start of closed-loop initiation as shown in the following logical expression.
t ¨ tRecent < tR < t (eq A4) Here, tR is the time of recent calibration, t is the time when attempting to enter the closed-loop mode, and tRecent is the time window for the most recent calibration before attempting to start the closed-loop mode (which is set to 120 minutes).
[00416] For Case C, the time of prior calibration (tP) occurs between two and eight hours before time of recent calibration factor as shown in the following logical expression.
tR ¨ tDif fmax < tP tR ¨ tDif fmin (eq A5) Here, tP is the time of prior calibration, tR is the time when CFR was obtained, tDiffmax is the maximum time difference between the recent calibration and the prior calibration (which is set as 480 minutes (8 hours)), and tDiffmin is the minimum time difference between the recent calibration and the calibration prior to the recent calibration (which is set as 120 minutes (2 hours)).
1004171 For Case C, the calibration change (CFchange) is less than 35% as shown in the following logical expression, where CFchange is calculated according to Equation A6.
CFchange = (abs(CFR ¨ CFP)ICFP) x 100 (eq A6) CFchange CFchangeTh (eq A7) Here, CFchange is the percentage change in calibration factor from the previous calibration factor to a current calibration factor for any pair of calibration factors, CFchangeTh is the threshold for acceptable CFchange, (which is set to 35% for this example), CFR is the most recent calibration factor value, and CFP is the last calibration factor value before CFR.
[00418] If all of the foregoing conditions are met for Case C (Equations A2-A7), the closed-loop initiation module 902 can initiate the methodology for calculating the correction bolus (if needed). If, however, any condition is not met, the controller 900 remains in the open-loop mode. Thus, in order to enter the closed-loop mode, new calibration(s) that satisfy all of the conditions in Case C or Case D will be required.
[00419] Case D
[004201 Case D corresponds to a scenario where the time of prior calibration is less than 120 minutes before the most recent calibration. If the prior calibration is less than two hours before the recent calibration, an additional prior calibration factor (CFP2) is included in the analysis. This allows the closed-loop initiation module 902 to evaluate sensor sensitivity that has at least a two hour time span.
[00421] For Case D, the closed-loop initiation module 902 finds an earlier second prior calibration factor (CFP2) that occurs within two to eight hours before the time of the recent calibration factor (CFR) as shown in the following logical expression.
tR ¨ tDif fmax < tP2 tR ¨ tDif fmin (eq A8) Here, tP2 is the time when the second prior calibration factor (CFP2) is obtained, tR is the time when CFR was obtained, tDiffmax is the maximum time difference between tP2 and tR (which is set as 480 minutes (8 hours) for this example), and tDiffinin is the minimum time difference between tP2 and tR (which is set as 120 minutes (2 hours) for this example).
[00422] For Case D, the closed-loop initiation module 902 also determines if more than one calibration factor (CF1...CFn) is available between the time of the second prior calibration factor (CFP2) and the time of the recent calibration factor (CFR), as shown in the following logical expression.
tP2 < t1 ...tn < tR (eq A9) Here, ii... In is the time when more calibration factors (CF1...CFn) are observed, tR is the time when CFR was obtained, and tP2 is the time when CFP2 is obtained.
1004231 For Case D, the time of recent calibration (tR) is within two hours of the start of closed-loop initiation, as shown in the following logical expression.
t ¨ tRecent < tR < t (eq A10) Here, tR is the time of recent calibration, t is time when attempting to enter the closed-loop mode, and tRecent is the time window for the most recent calibration before attempting to start the closed-loop mode (which is set to 120 minutes for this example).
[00424] For Case D, all calibration factors including recent calibration factor (CFR), prior calibration factor (CFP), second prior calibration factor (CFP2) and CFI
...CFn are within limits as shown in the following logical expressions.
CFmin < CFR < CFmax (eq A11) CFmin < CFP < CFmax (eq Al2) CFmin < CFP2 < CFmax (eq A13) CFmin < CF1 ...CFn < CFmax (eq A14) Here, CFR is the recent calibration factor, CEP is the prior calibration factor, CFP2 is the second prior calibration factor, CFI CFn are calibration factors obtained between tP2 and tR, CFmin is the minimum value for the calibration factor (which is set as 2.5 mg/dL
per nA for this example), and CFmax is the maximum value for the calibration factor (which is set as 6 mg/dL per nA for this example).
[00425] For Case D, the calibration change (CFchange) between CFR and CFP2 is less than 35%, as shown in the following logical expression where CFchange is calculated according to Equation A15.
CFchange = (abs(CFR ¨ CFP2)/CFP2) X 100 (eq A15) CFchange CFchangeTh (eq A16) Here, CFchange is the percentage change in calibration factor from the previous calibration factor to a current calibration factor for any pair of calibration factors, CFchangeTh is the threshold for acceptable CFchange (which is set to 35% for this example), CFR is the most recent calibration factor value, and CFP2 is the most recent calibration factor value in the time range described in Equation A8.
[00426] If all of the foregoing conditions are met for Case D (Equations A8-A16), the closed-loop initiation module 902 can initiate the methodology for calculating the correction bolus (if needed). If, however, any condition is not met, the controller 900 remains in the open-loop mode. Thus, in order to enter the closed-loop mode, new calibration(s) that satisfy all of the conditions in Case C or Case D will be required.
[00427] Correction Bolus With JOB Compensation [00428] As described above, a correction bolus 932 may be commanded at the beginning of the closed-loop mode. The purpose of the correction bolus is to provide an insulin dose for mitigating hyperglycemia at the initiation of the closed-loop mode. This can be achieved by first acquiring a blood glucose meter reading value immediately prior to closed-loop initiation. If that BG meter reading value is above a certain correction threshold (CTH, which is 180 mg/dL for this example), the controller 900 will deliver an insulin dose based on the patient's insulin sensitivity (ISF, mg/dL/Units), insulin on board, and the desired target glucose level (TG, mg/dL) in order to bring the subject's glucose level to the target glucose level.

1004291 In accordance with certain implementations, the correction bolus (CB) is delivered based on the meter BG value (in mg/dL) acquired at the start of the closed-loop mode, as shown below:
BG-TG
I0B(n), if (BG > CTH) CB ={ IsF (eq A17) 0, if (BG 5_ CTH) Here, CB is the correction bolus, BG is the blood glucose meter value (mg/dL), TG is the target glucose (mg/dL), ISF (see Equation A18) is the patient's adjusted insulin sensitivity factor (mg/dLUnits), CTI-I is the correction threshold for blood glucose beyond which a correction bolus will be delivered (mg/dL), and I0B(n) is the active insulin on board from manual boluses (Units) where n is the current sampling point as described previously.
ISF = !SF factor x ISF0 (eq A18) Here, ISF is the patient's adjusted insulin sensitivity factor (mg/dL/Units), /SF0 is the patient's established insulin sensitivity factor (mg/dLIUnits) and ISFfactor is an ISF
adjusting factor (unit less). The default value for ISFfactor is set to one, which makes ISF
= ISF0. However, for this particular example, ISFfactor has been assigned as an adjustable parameter with a range of 0.5 to 2 in order to provide greater flexibility for optimizing the patient's insulin sensitivity factor.
CB_ 1CB, if (CB 0) A19 0, if (CB <0) (eq ) Here, CB is the correction bolus expressed in Units. It should be appreciated that Equation A19 is utilized because the controller 900 can only deliver positive boluses.
1004301 Start-Up Module [00431] The start-up module 904 processes sensor glucose (SG) values 940 and the target glucose setpoint value 944 (which is set to 120 mg/dL in certain embodiments) to calculate the final target glucose value 946, which in turn serves as an input to the PID-IFB control module 906. Accordingly, the final target glucose value 946 is sent to the PID-IFB control module 90610 calculate the final insulin dose 962. Referring again to FIG. 49, the start-up module 904 is "activated" or initiated in response to the starting of the closed-loop mode.
1[004321 At the beginning of the closed-loop operating mode, the start-up module 904 calculates the difference between the current SG value 940 and the target glucose setpoint value 944, as indicated by Equation 51 below.
= 1SG (n) ¨ Setpoint, m = 1 DeltaGlu(n) (eq 51) (0, m > 1 In Equation 51, SG is the sensor glucose value, n is the current sampling point, Setpoint is the target glucose setpoint value defined by user, and m is the sampling time during closed-loop operation (m=1 indicates the start of the closed-loop mode, and m increases with each sample received during the closed-loop mode). DeltaGlu(n) is forced to zero when m> 1 as well as in the following situation described in Equation 52.
[00433] Force DeltaGlu to zero if it is less than a certain threshold set in the controller 900 (referred to as MinDeltaGlu). Otherwise, it remains as DeltaGlu(n) if it is more than the threshold set in the controller 900, as described in Equation 52:
DeltaGlu < MinDeltaGlu DeltaGlu(n) = (eq 52) DeltaGlu(n), DeltaGlu > MinDeltaGlu Here DeltaGlu is the difference between the current SG value 940 and the defined target glucose setpoint value 944, calculated from Equation 51 above, and MinDeltaGlu is the minimum difference (set in the controller 900) allowable between the current SG value 940 and the target glucose setpoint value 944.
[00434] The dynamic setpoint (DynSP) is calculated based on a discretized second order transfer function model as indicated in Equation 53:
DynSP(n) = cdi = DynSP(n ¨ 1) + c d2 = DynSP(n ¨ 2) + en = DeltaGlu(n) +
cni = DeltaGlu(n ¨ 1) (eq 53) Here, DynSP is the dynamic setpoint value, n is the current sampling point, n-1 is the last sampling point, and n-2 is the second to last sampling point. The parameters ccli, cd2, cno, and cni are the coefficients of the setpoint model. These parameters are calculated based on the two time constants (1õ1 and isp2) of the setpoint model, as indicated below:
cdi = eaxxl + eaxx2 cd2 = ¨eaxxl x eaxx2 cno = 1 (axxixeaxxl-axx2xeaxx2) cni =
daxx2i Where:
axxl =
axx2 = 1 It sp2 eaxxl = e-axxl=Ts eaxx2 = e-axx2.7'5 daxx21 = axx2 ¨ axxl In the above equations 7's indicates the sampling interval in minutes, and -Espi and Tsp2 are the time constants of the setpoint model. Moreover, axxl is the reciprocal of the time constant rspi, axx2 is the reciprocal of the time constant zsp2, eaxxl is the exponential decay factor for rspi, eaxx2 is the exponential decay factor for 1,2, and da;cx21 is the difference between the reciprocal of tspi and isp2.
100435] The final target glucose value 946 is obtained by adding the dynamic setpoint value (calculated in Equation 53) with the target glucose setpoint value 944 as shown in Equation 54:
Final Target = Setpoint + DynSP(n) (eq 54) 100436] In particular implementations, some of the parameters mentioned above for the start-up module 904 can be fixed. In this regard, the following values may be utilized in an exemplary embodiment. It should be appreciated that these values are provided here for illustrative purposes only, and that an implementation of the start-up module 904 may utilize different values if so desired.
Setpoint = 120 mg/dL

MinDeltaGlu = 30 mg/dL (nominal); 0 mg/dL (lower bound); 600 mg/dL (upper bound) Tv' =25 minutes (nominal); 0.1 minute (lower bound); 250 minutes (upper bound) Tsp2 =30 minutes (nominal); 0.1 minute (lower bound); 250 minutes (upper bound) 1004371 PID-IFB Control Module 1004381 The PID-IFB control module 906 calculates the insulin infusion rate (final insulin dose 962) based on the current and past sensor glucose values 940, the sensor Isig values 950, the rate of change of the sensor glucose values, the sensor calibration factor 952, the final target glucose value 946, the target glucose setpoint value 944, insulin limits such as the upper insulin limit, and the insulin delivered 954 in order to achieve euglycemia. In certain embodiments, the PID-IFB control module 906 receives its inputs every five minutes, therefore the calculation of the insulin feedback components considers how often the input data is being received by the controller 900.
[00439] The PID-IFB control module 906 calculates the final insulin dose 962 in accordance with Equation 55:
u(n) = P (n) + 1(n) + D (n) ¨ Y ilsc ¨Y2lp Y3IEFF (eq 55) Note that P I D Rate (n) E u(n). In Equation 55, the insulin infusion rate u(n) represents the final insulin dose 962 shown in FIG. 49. In Equation 55, P (n) , I (n) , and D (n) are, respectively, the proportional, integral, and derivative components of the PID
controller.
The insulin feedback components correspond to the remaining terms. The variables 71, 72, and 73 represent tuning coefficients. In accordance with some embodiments, yoD.64935, 72=0.34128, and =y3=0.0093667, although other values may be used. The parameters /õ(n), i,,(n), and IEFF(n) correspond to the states of the insulin pharmacokinetic model corresponding to, respectively, the subcutaneous, plasma, and effective site compartments. Therefore, the amount of insulin delivered is reduced in proportion to the predicted insulin concentration in the different compartments.
[00440] The exemplary implementation of the control algorithm employed by the PID-IFB control module 906 is implemented in discrete (sampled) time. In the notation used, n is the current time step, where = nT, , and t is the continuous time (in minutes) when using a sampling period of Tõ.

[00441] The proportional component P(n) is calculated as follows:
P(n) Kp[SG(n) ¨ Final Target(n)] (eq 56) Here, Kp is the overall controller gain (expressed in Units/hour per mg/dL), SG(n) is the current sensor glucose, n indicates the current sampling point, and Final Target(n) is the calculated final target glucose setpoint from Equation 54. It should be appreciated that Kp is a patient-specific value and, therefore, the actual value of Kp will vary from one person to another. Although the range can vary depending upon the patient, in most typical scenarios, the value of Kp may be within the range of 0.008 to 0.200.
[00442] The integral component 1(n) can be calculated as follows:
1(n) = 1(n ¨ 1) + [SG(n) ¨ Final Target(n)] (eq 57) Here, 1(n-1) is the integral component from the previous sampling point, n indicates the current sampling point, n-1 indicates the previous sampling point, Kp is the overall controller gain, Ts is the sampling period, T1 is the integral time constant, SG(n) is the current sensor glucose, and Final Target(n) is the calculated final target glucose setpoint from Equation 54.
[00443] The derivative component D (n) can be calculated as follows:
D(n) = Kp X TD X dSGdt(n) (eq 58) Here, Kp is the overall controller gain, TD is the derivative time constant, dSGdt(n) is derivative of the sensor glucose value (pre-filtered to remove noise), and n indicates the current sampling point.
[00444] The parameters that need to be set (tuned) for the controller 900 are:
Kp, 17, and Tr) (see Equations 56, 57, and 58). For the first three PID parameters, they will be calculated in accordance with the previous studies mentioned below (these previous approaches have resulted in good controller performance). The nominal (i.e., for the case with no insulin feedback) controller gain Kpg is calculated based on the subject's insulin total daily dose IIDD, (in Units/day), as indicated in Equation 59:

Kpo =

(90)(1500) ITDD (eq 59) Here, Kpo is the default controller gain, and bap is the subject's insulin total daily dose in Units/Day.
[00445] The purpose of insulin feedback is to allow the controller 900 to deliver more insulin up-front (e.g., at the onset of a meal disturbance), but to prevent over-infusion of insulin, in a similar fashion to the bolus estimator calculations used in existing insulin pumps to prevent the stacking of boluses. Therefore, when insulin feedback is used, the controller gain Kp can be adjusted so that at steady state (i.e., basal delivery conditions) the insulin delivery system will deliver the same amount of insulin as in the nominal case.
This is accomplished by multiplying the nominal controller gain KpO
(calculated for no insulin feedback) by (1 + 71 + 72 + 73) as indicated below:
Kp = Kp factor = KR) = (1 + yi + y2 + y3) (eq 60) Here, Kp is the overall controller gain, KpfaCtor is the gain factor for Kp, Kpo is the default controller gain, 71 (0.64935) is a tuning coefficient for the subcutaneous insulin concentration, 72 (0.34128) is a tuning coefficient for plasma insulin concentration, and 73 (0.0093667) is a tuning coefficient for effective insulin concentration.
[00446] The integral component 1(n) is also equipped with anti-windup and saturation constraints to address integral windup issues. This is achieved by calculating an integral clip value (upper limit for the integral component, 'Clip) as indicated by the following equations:
Imax(n), (SG(n) > UnwindHigh) !Clip(n) = Iramp(n), (SG(n) > UnwindLow) and (SG(n) < UnwindH Oh) I low(n), (SG(n) < UnwindLow) (eq 61) Imax(n) = Imax f actor X Basal X (1 + yi + Y2 + Y3) (eq 61a) Here, Imaxfactor is the gain factor for Imax, and Basal is the subject's night time basal rate. In accordance with these expressions, /Clip becomes equal to the lmax (which is a constant value) when the sensor glucose value is greater than an upper threshold (UnwindHigh). In certain embodiments, the value of (expressed in Units/hour) may reach about 15. In a typical case, Imax may have a default value of 5Ø When the sensor glucose value is between the upper and lower threshold (UnwindLow), then IClip becomes equal to Iramp(n), which is calculated as indicated in Equation 62.
Iramp(n) = Kp = [SG(n) ¨ UnwindLow] ( SG(n)-UnwindLow ("max Kp Unw indH igh-UnwindLow 1 retpoint(n) ¨ UnwindLow]) (eq 62) Here, Kp is is the overall controller gain, SG(n) is the current sensor glucose, UnwindLow is the sensor glucose lower threshold, UnwindHigh is the sensor glucose upper threshold, Imax is a constant value, and Setpoint(n) is a user-defined target glucose setpoint.
[00447] Finally, if sensor glucose is below the UnwindLow threshold, IClip assumes the value of Ilow(n), which can be calculated by Equation 61:
flow(n) = Kp[Setpoint(n) ¨ Unwindlow] (eq 63) Here, Kp is the overall controller gain, Setpoint is the target glucose defined by the user, and UnwindLow is the sensor glucose lower threshold.
[00448] FIG. 51 is a graph of 'cup (in Units/Hour) versus sensor glucose level (in mg/dL) in accordance with one example. FIG. 51 depicts the relationships between Imax, UnwindLow, and UnvvindHigh for this particular example.
[00449] The integral component /(n) as calculated in Equation 57 must be less than or equal to the IClip value as shown in Equation 64:
1( ) EIClip(n), 1(n) > IClip(n) n = (eq 64) / (n), I(n) 1Clip(n) [00450] The insulin feedback components correspond to the remaining terms. As mentioned above, for this particular example yM).64935, y2=0.34128, and 73=0.0093667 (the tuning coefficients), while the parameters 'Sc (n), Ip(n), and IEFF(71) correspond to the states of the insulin pharmacokinetic model corresponding to, respectively, the subcutaneous, plasma, and effective site compartments. Therefore, the amount of insulin delivered is reduced in proportion to the predicted insulin concentration in the different compartments.

1004511 The model describing the insulin pharmacokinetics (insulin PK) is given by the following expressions:
isc(n) = an X isc(n ¨1) + A x (n) (eq 65) where:
= exp (¨ ¨Ts) (eq 65a) (eq 65b) ks Here, Ts is the sampling time (which is five minutes for this example), and zs. is the time constant for estimated subcutaneous insulin level in minutes (which is set to 50 minutes for this example).
f(n) = a21 x /sc(n ¨1) + a22 x /p(rt ¨1) + fl2 x ID (n) (eq 65c) where:
s Tp) an = Ts = [exp(-7-1) ¨ exp (_ L)1 (Ts (eq 65d) Ts Tp a22 = exp --Ts (eq 65e) /60 \ TS
P2 = ) = (ts = 11 exp VI ¨ Tp = {1 exp(--Z)1)1(7-5. ¨ rp) (eq 65f) Here, Ts is the sampling time, which is five minutes for this example, ts is the time constant for estimated subcutaneous insulin level in minutes, which is set to 50 for this example, and ri, is the time constant for estimated plasma insulin level in minutes, which is set to 70 for this example.
/EFF(n) = a31 x isc(n ¨ 1) + a32 x ip(n ¨ 1) + a33 X IEFF(n ¨1) x 1 D (71) (eq 66) For Equation 66:
Ls Ts a31 = Ts = [Ts = (Tp ¨ re) = exp(_) Tr (Tp ¨ T e) = exp(--)+ Te = (Ts ¨ Tp) =
\ rs / Tp exp (--Ts)1/[(Ts Tp). (Ts Te) (Tv Te)] (eq 66a) re Ts s a32 = rp = [exp exp (_!)]/(T T e) (eq 66b) Ty TS) a33 = exp /--) (eq 66c) Te P3 = (-61)' (F4 ' (Tp ¨ re ) = (1 exp (¨ ¨Z))1¨ Tf, = (Ts ¨ re) = [1 ¨
exp (_)1+

4 .(r-rp) [1 exp (¨ ¨
Ts)i) Ars _ Tp) (Ts _ re) (Tp Te) rp Te)/
(eq 66d) Here, Ts is the sampling time, which is five minutes for this example, rs is the time constant for estimated subcutaneous insulin level in minutes, which is set to 50 for this example, Tr, is the time constant for estimated plasma insulin level in minutes, which is set to 70 for this example, and Te is the time constant for estimated interstitial insulin level in minutes, which is set to 55 for this example.
1004521 ID(n) is the calculated and administered insulin infusion.
1004531 The notation (n-1) indicates values at the previous time step.
[00454] /sc is the subcutaneous insulin model estimate/prediction.
[00455] Ip is the plasma insulin model estimate/prediction.
1004561 JEFF is the effective site insulin model estimate/prediction.
[00457] For this particular example, the insulin PK model parameters all, azt, ot,z, cc31, 11-32, 033, pi, 132, and 133 are set to 0.9802, 0.014043, 0.98582, 0.000127, 0.017889, 0.98198, 1.1881, 0.0084741, and 0.00005, respectively. These values were calculated from PK-PD
data for insulin as part of an empirical study and investigation. It should be appreciated that the specific values presented here merely reflect one possible set of suitable values, and that any one or more of these values may be adjusted as appropriate for the particular embodiment.
[00458] Insulin Limit [00459) The final infusion rate as calculated by Equation 55 is limited such that is does not exceed the maximum upper insulin limit (Umax) as indicated below in Equation 67:
u(n) = I/ max , u(n) > Umax u(n) Umax (eq 67) Umax is calculated as indicated by Equation 68:
BGLBL¨FBG0 Umax = I +
basa1,0 K1 (eq 68) Here, Umax is the upper insulin limit bounded at BGLBL (see Equation 68a below), and ibasai3O is the basal rate defined for the user to bring the patient to a Fasting Blood Glucose (FBG) of the value FBG0 mg/dL. BGLBL (mg/dL) is the lower buffer limit BG when reaching Umax, FBG0 is the estimated blood glucose using meter blood glucose readings at the end of the night period, and IC/ is insulin gain as calculated by Equation 69 below.
BGLBL = Setpoint ¨ ILL? (eq 68a) Here, BGLBL (mg/dL) is the lower buffer limit BG when reaching Umax, Setpoint is the target glucose setpoint value 944 (FIG. 49) that is defined by the user, and ILB is the insulin limit buffer, which is an amount (in mg/dL) that the system allows as an additional buffer to handle higher insulin needs. For example, an ILB of 50 allows the system to deliver additional insulin to lower by 50 mg/dL from the Setpoint.
KI = ¨IS * 3 (n21 per ¨u) (eq 69) dL
Here, K/ is insulin gain, and IS is the insulin sensitivity estimated by the 1800 rule as IS = 1800/TD/ (Total Daily Insulin). In certain embodiments, the value of Umax may reach 150 Units/hour. In a typical implementation, the value of Umax is calculated on a per-patient basis, and the typical range for Umax is between 0.5 to 3.0 Units/hour.

1004601 Moreover, if the insulin limit is active, then the integral component of the PID
algorithm is frozen to its previous value. This feature can be employed to assist in the prevention of integral wind up. This information is passed back to the PID-IFB
control module 906 for use during the next PID calculation.
1 1(n ¨ 1), u(n) = Umax 1(n) =
/(n), u(n) < Umax (cq 70) [00461] As mentioned above, previous studies may be leveraged for purposes of configuring and/or operating the PID-IFB control module 906. A first study (Study 1;
Panteleon et al., 2006) looked at the effect of changing the controller gain in a study on eight diabetic dogs. The nominal gain was calculated based on the total daily dose of insulin for one experiment; in duplicate experiments this nominal gain was increased and decreased by fifty percent. Insulin delivery in the six-hour postprandial period tended to increase with increasing gain, but this did not achieve statistical significance. This was due to feedback, with lower insulin delivered as glucose levels fell. In essence, the higher gains tended to give more insulin earlier (achieving much better glucose regulation), and reducing the infusion later, whereas the lower gains tended to keep insulin infusion above basal for a longer period of time due to the glucose levels remaining significantly above target. Note that the controller gain affects all of the components of the P1D
algorithm, including the integral and derivative terms.
1004621 A second study (Study 2; Steil et al., 2006) applied a PID controller to ten human subjects. In this study the nominal controller gain was about 42 percent higher than that proposed here, with the integral time constants the same (therefore also a higher integral response), but slightly lower derivative time constants (defined in terms of rising or falling blood glucose, instead of day or night response). In the proposed embodiment presented here, the night-time derivative time constant is just slightly lower than that used in Study 2.
[00463] A third study (Study 3; Weinzimer et al., 2008) applied a PID
controller to seventeen human subjects. For a subset of eight subjects, no pre-meal bolus was given, while in the remaining nine subjects about 50 percent of the standard meal bolus was given about fifteen minutes before the meal. The controller tuning was the same as that proposed herein. In both cases, performance was acceptable, with the pre-meal bolus helping to lower the postprandial peak glucose excursion. When compared with home treatment using standard pump therapy, both closed-loop algorithms were superior, reducing both glucose excursions above 180 mg/di and below 70 mg/d1.
100464] One of the observations from Study 3 was that the meal-related insulin infusion persisted above pre-meal levels for at least four hours after the meal. This result led to the introduction of insulin feedback to the algorithm, which serves to compensate for insulin already delivered, while at the same time allowing for more aggressive action at the start of meals, where it is needed.
1004651 In particular implementations, some of the parameters mentioned above for the PID-IFB control module 906 can be fixed. In this regard, the following values may be utilized in an exemplary embodiment. It should be appreciated that these values are provided here for illustrative purposes only, and that an implementation of the PID-IFB
control module 906 may utilize different values if so desired.
i= 0.64935 72= 0.34128 -y3= 0.0093667 = 0.90483741803596 0,21 = 0.065563404170158 a22 = 0.931062779704023 a31 = 0.00297495042963571 0.32 = 0.083822962634882 o.33 = 0.083822962634882 = 5.70975491784243 132 = 0.202428967549153 133 = 0.202428967549153 [00466] The values of the PID parameters mentioned above (Kp, rj, and ID) are not expected to change, but it may be desirable to adjust them if doing so would improve the response of the glucose excursion. The nominal values for these P1D
parameters, together with exemplary allowable ranges, are shown in Table 2.
Parameter Nominal Value Lower Bound Upper Bound Controller gain: x x (1 + + Y2 0.05 x X (1+ y, + yi 20 x Kõ X (1+ + Y2 Kp +3) +3) +3) Integral time 250 min 10 min 1500 min constant: r1 Derivative time 75 min 10 min 500 min constant: VD
UnwindLow 80 0 200 (mg/dL) Unwindifigh 100 0 200 (mg/dL) Imax 2.5 x (1 + y, + y, + y3) 0 10 x (1 + y, +
y2 + y3) x Basal Rate x Basal Rate Table 2: Adjustable Parameters For the PID-IFB Control Module 1004671 The lower bound for the controller gain Kpo would lower it by 95 percent from the nominal, resulting in an overall less aggressive response of the controller 900, and resulting in less insulin delivered for the same glucose level and trend. The upper bound allows this gain to be increased up to 20 times from the nominal. Although this would result in more insulin being given, this is always with respect to the actual glucose level and its derivative, rapidly falling blood glucose, even with an elevated blood glucose level, results in a decreased delivery of insulin due to the derivative component.
[00468] The integral time constant ri determines how fast a deviation in the blood glucose from the desired glucose target accumulates. A higher value results in a slower response to persistent deviations in glucose from target. Lastly, the derivative time constant JD can safely be decreased all the way to zero, as the lower it is the less insulin will be delivered. Anything much higher than the upper bound of 500 minutes would make the controller too sensitive to small changes in the rate of change of the sensor glucose, which even though it is not necessarily dangerous, is still undesirable in terms of overall performance.
[00469] IOB Compensation Module [00470] FIG. 52 is a block diagram that schematically depicts one suitable embodiment of the JOB compensation module 910 in additional detail. As mentioned briefly above, the JOB compensation module 910 adjusts the current insulin dose 958 (i.e., the insulin infusion rate as computed by the PID-IFB control module 906) as needed to generate the final insulin dose 962 (i.e., the final adjusted infusion rate used by the insulin infusion device). The 10B compensation module 910 may also receive basal rate data 990 and the information related to the manual boluses delivered 960 as input for purposes of calculating the current value of the final insulin dose 962 and/or for purposes of calculating future values of the final insulin dose 962. The basal rate data 990 may indicate the current basal rate of insulin being delivered to the user, and the manual boluses delivered 960 may indicate the amount for each bolus administered to the user, along with timestamp data corresponding to the delivery date/time of each bolus. In this regard, the manual boluses delivered 960 may include information for any number of boluses delivered in the past, and the manual boluses delivered 960 can be updated as needed in response to each bolus that is delivered going forward. Moreover, the basal rate could be dynamically adjusted if needed or desired (automatically by the system, by the user, by a caregiver, etc.).
[00471] The manual boluses delivered 960 may be collected and saved in association with the JOB compensation module 910 for use as bolus history 992. In this regard, the bolus history 992 may include any number of past bolus amounts administered during a given period of time. The JOB compensation module 910 may also use a number of constants, parameters, coefficients, configuration settings, gain values, or the like (for simplicity, the constants 994 shown in FIG. 52 are intended to encompass these and any other quantities that might be used by the IOB compensation module 910 to calculate the final insulin dose 962. FIG. 52 also depicts JOB history 996, which represents previously calculated 10B values (i.e., historical 10B values calculated for past sampling times). As explained in more detail below, the JOB history 996 and the bolus history 992 may influence the determination of the final insulin dose 962. It should be appreciated that the bolus history 992, the constants 994, and the IOB history 996 can be stored and maintained in one or more memory storage elements of the host system. FIG. 52 shows these data items "inside" the JOB compensation module 910 for simplicity and ease of description.
[00472] The 10B compensation module 910 provides an additional safeguard that estimates the insulin on board the body of the patient from manual boluses that were delivered prior to activation of the closed-loop mode, to compensate the final insulin dose 962 and help avoid over-delivery of insulin. When the system initially enters the closed-loop mode, the IOB compensation module 910 considers the manual boluses delivered 960 that have been administered over a defined period of time (e.g., the last eight hours), and subtracts the manual boluses. Thereafter, during the closed-loop mode, the JOB
compensation module adjusts for manual boluses that were delivered during each sampling period (e.g., every five minutes).

1004731 FIG. 53 is a flow chart that illustrates an exemplary embodiment of an compensation process 1100, which may be performed by the JOB compensation module 910. The process 1100 represents one iteration that is performed for a current sampling point or time, n. Accordingly, the process 1100 receives, obtains, or accesses a variety of inputs that may have an influence on the output of the JOB compensation module (task 1102). For example, the process 1100 may utilize the current value of the uncompensated insulin infusion rate, PIDRate(n), as generated by the PID-IFB control module 906. The process 1100 may also use the current basal rate (provided with the basal rate data 990), some of the bolus history 992, and/or some of the 10B history 996 as needed.
[00474] In the context of the process 1100, the JOB compensation module 910 calculates insulin on board from manual boluses at each cycle and compensates the controller output rate (insulin infusion rate) if active JOB is greater than a certain threshold. Accordingly, the process 1100 calculates, generates, or otherwise obtains a current JOB value (task 1104) that represents an estimate of active insulin in the body of the user. In certain embodiments, the active IOB is estimated in accordance with a discretized three-compartment insulin pharrnacolcinetic (PK) model, as indicated below:
108(n) = ci1= 10B(n¨ 1) + c12 = 10B(n ¨ 2) + c13= 10B(n¨ 3) + cbo =
Ubolus(n) + cb, = Ubolus(n ¨ 1) + cb2 = Ubolus(n ¨2) (eq 71) [00475] Here, JOB is the active insulin on board, Ubolus is the amount of manual bolus delivered in units per sample, n is the current sampling point, n-1 is the last sampling point, n-2 is the second to last sampling point, and n-3 is the third to last sampling point. Accordingly, the process 1100 obtains the current JOB value, 10B(n), based at least in part on historical bolus delivery data for the user (see the manual boluses delivered 960 and the bolus history 992 in FIG. 52). The parameters cii, ci2, ci3, cbo, and cb2 are the coefficients of the insulin absorption model. These parameters are calculated based on the three time constants (r(., rp, and 7-cif) of the insulin pharmacokinetic model, as indicated below:
= eaxx3 + eaxx4 + eaxx5 (eq 71a) ci2 =--(eaxx3 X eaxx4 + (eaxx3 + eaxx4)x eaxx5) (eq 71b) ci3 = eaxx3 X eaxx4 X eaxx5 (eq 71c) cbo = 1 cb, = dprod x (¨(daxx22 X eaxx3 + daxx22 x eaxx4) x axx3 x axx4 +
(daxx31 X eaxx3 + daxx31 X eaxx5) X axx3 x axx5 ¨ (daxx32 x eaxx4 +
daxx32 x eaxx5) X axx4 X axx5) (eq 71d) cb2 = dprod x (daxx22 X eaxx3 x eaxx4 x axx3 x axx4 + daxx32 X eaxx4 x eaxx5 x axx4 x axx5 ¨ daxx31 X eaxx3 x eaxx5 X axx3 X axx5) (eq 71e) Where:
axx3 = 1/z5c (cq 71f) axx4 = 1/zp (eq 71g) axx5 = ltreff (eq 71h) eaxx3 = e'xx3=TsC (eq 71i) eaxx4 = e-axx+TsC (eq 71j) eaxx5 = e-axx5=T5C (eq 71k) daxx22 = axx4 ¨ axx3 (eq 711) daxx31 = axx5 ¨ axx3 (eq 71m) daxx32 = axx5 ¨axx4 (eq 71n) dprod = ¨1/(daxx22 x daxx31 x daxx32) (eq 710) In the above equations TsC indicates a modified sampling interval in minutes, which can be calculated as TsC = Ts * 6/CurveSpeed , where Ts is the sampling interval and CurveSpeed is the insulin on board decay speed rate in hours. zs( , zp, and Tell are the respective time constants of the subcutaneous, plasma, and effective compartments of the insulin PK model.
[00476] JOB (in Units) as calculated by Equation 71 represents the residual active insulin in the body from manual boluses (that may have been administered before the start of the closed-loop mode or during closed-loop operation) which must be taken into account for calculation of the final insulin delivery rate. This is achieved by first calculating the JOB rate (task 1106) and then deducting the 1013 rate from the PID-IFB
calculated infusion rate, as indicated below, Accordingly, in certain situations the process 1100 determines an adjusted insulin infusion rate (task 1108) based at least in part on the calculated 10B rate and the uncompensated insulin infusion rate, PIDRate(n):

r0ainlOB X 10B(n), 10B(n)> MinlOB
10BRate(n)= (eq 72) , 10B(n)< MinlOB
AdjustedRate(n)=max(0, PIDRate(n)-10BRate(n)) (eq 73) Note that the process 1100 calculates the JOB rate, IOBRate(n), based at least in part on the current JOB value, I0B(n). The JOB rate, which is expressed in Units per hour (IA), represents the amount of active insulin accumulated from manual boluses in the body per unit of time. Hence, this additional insulin which is already present in the body is deducted from the controller delivery rate (PIDRate). This accounts for all the manual boluses that have been administered by the user, and minimizes the possibility of over-delivery by the controller. Here, GainIOB is the IOB decay rate in h-1, and MinlOB
is the minimum JOB required to compensate the P1DRate (where MinlOB is expressed in Units). Thus, the JOB rate is calculated to be equal to the current 10B value multiplied by an IOB decay rate when the current JOB value is greater than a minimum IOB
value, and is calculated to be equal to zero when the current JOB value is less than or equal to the minimum JOB value. In this regard, MinlOB is a minimum threshold for JOB, below which the effect of JOB on glucose is considered to be negligible; hence not needed to be compensated.
1004771 As reflected in Equation 73, the process 1100 selects the adjusted insulin infusion rate to be the maximum of zero or the difference between the uncompensated insulin infusion rate and the calculated JOB rate (from task 1106). Note that the difference between PIDRate and 10BRate could be negative, as these values are calculated from different sources. PIDRate is the controller calculated infusion rate and IOBRate is the accumulated active insulin in the body obtained from manual boluses.
Accordingly, Equation 73 ensures that the AdjustedRate does not fall below zero.
1004781 Next, the process 1100 may calculate, select, or otherwise determine a final insulin infusion rate (task 1110). In certain embodiments, the final insulin infusion rate (the final insulin dose 962 in FIG. 49) is calculated as indicated below:
FinalRate(n)=_ max(Basal,AdjustedRate(n)), PIDRate > Basal (eq 74) PIDRate(n), PIDRate <Basal As indicated by this expression, the process 1100 selects either the adjusted insulin infusion rate (AdjustedRate(n)), the uncompensated insulin infusion rate (PIDRate(n)), or the current basal rate (Basal) to serve as the final insulin infusion rate (FinalRate(n)) for the insulin infusion device. Here, PIDRate is the insulin infusion rate as calculated by the PID-IFB control module 906 and Basal is the current pre-programmed pump basal rate.
Accordingly, task 1110 selects the final insulin infusion rate to be equal to the uncompensated insulin infusion rate when the current basal rate is greater than or equal to the uncompensated insulin infusion rate. In contrast, when the current basal rate is less than the uncompensated insulin infusion rate, task 1110 selects the final insulin infusion rate to be either the current basal rate or the adjusted insulin infusion rate, whichever is higher.
1004791 In the context of task 1110, the PIDRate is used as the FinalRate (when PIDRate is less than or equal to Basal) to allow the controller to "apply brakes" (in other words, suppress insulin delivery rate) in order to prevent any potential hypoglycemia. On the other hand, when PIDRate is greater than Basal, the FinalRate will be the maximum of Basal or AdjustedRate, which ensures that the insulin adjustment only accounts for the insulin coming from boluses, and not the basal. A lower bound (i.e., the value of Basal) is applied to the FinalRate when PIDRate is greater than Basal; this lower bound is utilized to prevent over-compensation of insulin on board under these circumstances.
1004801 The process 1100 may continue by communicating or otherwise providing the final insulin infusion rate, FinalRate(n), to the insulin infusion device (task 1112). For embodiments where the process 1100 is executed natively by the insulin infusion device itself, then the process 1100 may simply provide the final insulin infusion rate to the processing logic or fluid delivery control module of the infusion device. In turn, the insulin infusion device responds by regulating delivery of insulin in accordance with the final insulin infusion rate.
1004811 This description assumes that the process 1100 is repeated for each sampling time. For the next sampling time, therefore, the value of 17 may be incremented by one (or by any desired amount) to establish an index for the next iteration of the process 1100 (task 1114). Thereafter, the process 1100 may return to task 1102 to obtain the newest input data values and to repeat the various tasks described above.
Accordingly, the process 1100 facilitates regulation of insulin delivery to the body of the user in a controlled and ongoing manner by continuously adjusting the final insulin infusion rate while the system is operating in the closed-loop mode.

[00482] In certain embodiments, it may be desirable to adjust some of the parameters utilized by the JOB compensation module 910 if doing so would improve performance.
The nominal values for these parameters, together with exemplary allowable ranges, are shown in Table 3.
Parameter Nominal Value Lower Bound Upper Bound CurveSpeed 6 1 8 GainlOB 1.25 0 5 MinIOB 1 0 500 Table 3: Adjustable Parameters For the IOB Compensation Module [00483] Insulin Delivery Timeout Module [00484] The insulin delivery timeout module 912 is suitably designed and configured to continuously monitor (during the closed-loop mode) if the patient is receiving insulin at the insulin limit (Umax) or no insulin (Urn/n, which may be defined as insulin delivery at little to no Units/Hour) for a prolonged period of time. If one of these insulin delivery conditions is detected, the system will issue a warning and continue operating under the closed-loop mode. As mentioned previously, the insulin delivery timeout module may process the insulin delivered 960 as an input.
[00485] Accordingly, the insulin delivery timeout module 912 introduces an additional safeguard that checks for a series of steps as described below for the delivery of insulin at the insulin limit (Um ax Timeout) or no insulin (Umin Timeout) for a prolonged period of time. This is achieved by calculating the total amount of insulin delivered by the system during the closed-loop mode in a pre-specified moving window that is identified as the insulin time window.
[00486] Regarding the Umin Timeout condition, once the insulin time window for Umin (which may be defined as delivering insulin at zero Units/Hour) has been reached from the start of the closed-loop mode, the system will monitor the amount of insulin being delivered at the user specified insulin time window and compare it with the amount that could have been delivered if operating at the patient's basal rate for the same time span, as shown in the following logical expression.

Pump Delivery Rate =
FinalRate, if nmin > (MinDeliveryTo11100) x Urals%lin f 14 Alert, if tq,iinmin < (MinDeliveryTo1/100)x Urasinar (eq 75) Here, Pump Delivery Rate (Units/Hour) is the infusion rate, which is either equal to the FinalRate from Equation 74 (i.e., the infusion rate calculated by the controller during the closed-loop mode), or a pre-programmed overnight basal rate that is used during open-loop operation. The quantity UrZmin is the total amount of insulin delivered (in Units) by the closed-loop control algorithm in a user-specified insulin time window for Umin, and the quantity Csnaftlin is the total amount of insulin that could be delivered if operating at a pre-programmed overnight basal rate in the same insulin time window for Urn/n. The parameter MinDeliveryTol is a user-specified tolerance, in percentage of U 1Bgaisnal i n , that the system has to deliver in order to stay in the closed-loop mode.
[00487] In accordance with this particular example, closed-loop control continues as long as the total amount of insulin delivered during the insulin time window (which is set to 120 minutes for this example) by the system is greater than the total amount of insulin that might have been delivered if operating at five percent (which is the default minimum tolerance for this example) of basal. Moreover, a fail-safe alert is triggered once the total amount of insulin delivered during the insulin time window (120 minutes) by the system is less than five percent of basal.
[00488] Regarding the Umax Timeout condition, once the insulin time window for Umax has been reached from the start of the closed-loop mode, the system will monitor the amount of insulin being delivered at the user specified insulin time window and compare it with the amount that might be delivered if operating at the Unica rate for the same time span, as shown in the following logical expression.
Pump Delivery Rate =
{ .- FinalRate, if U71/Yoirkfax -(MaxDeliveryTo11100) x 14,tr"
(MaxDeliveryTo1/100) x UAlax (eq 76) Alert, if uroitnmax >
Here, Pump Delivery Rate is the infusion rate, which is equal to either the FinalRate or a pre-programmed overnight basal rate that is used during operation in the open-loop mode.
i The quantity UrWnMaxot is the total amount of insulin delivered (in Units) by the closed-loop control algorithm in a user-specified insulin time window for Umax, and the quantity UwinaXmax is the total amount of insulin that could have been delivered in the M
user-specified moving window for Umax if operating at a calculated Umax rate.
The parameter MaxDelivegTol is a user-specified tolerance, in percentage of Umwainxmax, that the system must adhere to in order to stay in the closed-loop mode.
[00489] In accordance with this particular example, closed-loop control continues as long as the total amount of insulin delivered during the insulin time window (which is set to 600 minutes for this example) by the system is less than the total amount of insulin that might have been delivered if operating at 95% (which is the default maximum tolerance for this example) of Umax. Moreover, a fail-safe alert is triggered once the total amount of insulin delivered during the insulin time window (600 minutes) by the system is greater than the total amount that might have been delivered if operating at 95% of Umax.
[00490] In certain embodiments, it may be desirable to adjust some of the parameters utilized by the insulin delivery timeout module 912 if doing so would improve performance. The nominal values for these parameters, together with exemplary allowable ranges, are shown in Table 4.
Parameter Nominal Value Lower Bound Upper Bound Insulin Time Window for Untin 120 minutes 30 minutes 600 minutes Insulin Time Window for Umax 600 minutes 30 minutes 600 minutes MinDeliveryTol 5% 0% 100%
MaxDeliveryTol 95% 0% 100%
Table 4: Adjustable Parameters For the Insulin Delivery Timeout Module [00491] Model Supervisor Module 1004921 The model supervisor module 914 is similar to the insulin delivery timeout module 912 in that it monitors and polices the system during closed-loop operation. In practice, closed-loop systems are only aware of signals (inputs) that are being provided by the measurement device(s). If measurements deviate from true values, the control system may react to the deviation. When using continuous glucose sensors for diabetes, the sensors provide the measurements to the closed-loop control system and, based on these measurements, insulin is delivered to the subject. Accordingly, sensor performance and integrity should be closely monitored. Fortunately, there is a relationship between the insulin and meal intake to the glycemic response. This relationship can be translated into a mathematical model that is able to predict the sensor glucose response based upon the insulin delivered. The sensitivity of sensor glucose to insulin delivered is patient-specific (the sensitivity can usually be learned over a period of three to six days for each patient).
[00493] The model supervisor module 914 uses a mathematical model that enables individualization of a patient's unique blood glucose time-dependent response.
The model describes the sensor glucose time-dependent response as a function of insulin and meal intake. The exemplary mathematical model described herein has a number of benefits and advantages: it is linear, physiologically-based, and includes only parameters that have direct connection to measurable data (sensor glucose and insulin delivered).
These features are important because a linear model is easy to analyze and predict.
Moreover, a physiological-based model facilitates an understanding of the origin of the predictions (e.g., insulin sensitivity, meal intake, etc.), and use of measureable data reduces the need to estimate unobserved variables (e.g., metabolism, cell activity, etc.).
[00494] FIG. 54 is a diagram that defines certain time events for the model supervision. The label "present" indicates the most recent sampling time or sampling period 1120, and k is equal to the present sampling time minus a period corresponding to a length of prediction horizon (LPH) in sampling times. FIG. 54 also indicates a period corresponding to a length of training horizon (LTH) in sampling times, which refers to a model training period. Insulin history is defined as the length of data needed to estimate the plasma insulin. In order for the model supervisor module 914 to be able to estimate a fault, it considers the records of the insulin delivered over the past insulin history plus the LTH and the LPH sampling periods, and at least 80 percent of the Isig (electrical signal) measurements from k-LTH and k.
[00495] As described in more detail below, the model supervisor module 914 considers a "moving window" that includes a historical time period 1122 that is defined from the present (most recent sampling period 1120) back to the beginning of the LTH.
The moving window considered by the model supervisor module 914 may also include insulin history that precedes the LTH, as depicted in FIG. 54. Data obtained during each window of time is processed and analyzed at or near the present time and preferably before the next sampling period has ended. Thus, at the end of each new sampling period, the "moving window" shifts by one sampling period such that the model supervisor module 914 can consider the most recently obtained data for the current sampling period while disregarding the data that no longer appears within the updated window of time (i.e., the oldest data is no longer considered). The historical time period 1122 may be defined by the LTH and the LPH, which for this example immediately follows the LTH (as shown in FIG. 54). The LPH may also be referred to herein as the "recent history period" or the "model prediction period" for reasons that will become apparent from the following description. The LTH may also be referred to herein as the "distant history period" or the "model training period" for reasons that will become apparent from the following description. In this regard, the LTH (distant history period) corresponds to a period of time from a begin-training sampling period 1124 to an end-training sampling period 1126, inclusive, while the LPH (recent history period) corresponds to a period of time from a begin-prediction sampling period 1128 to the most recent sampling period 1120, inclusive. Accordingly, by definition the current sampling period (i.e., the most recent sampling period 1120) resides within the LPH. For this particular example, the begin-prediction sampling period 1128 corresponds to the end-training sampling period 1126.
Alternatively, the begin-prediction sampling period 1128 may immediately follow the end-training sampling period 1126.
[00496] FIG. 55 is a flow chart that illustrates an exemplary embodiment of a sensor model supervision process 1150, which may be performed by the model supervisor module 914. The process 1150 is shown and described in a simplified manner that focuses on functionality for ease of understanding. Certain aspects of the process 1150 arc addressed in more detail below with reference to particular representations of the model supervisor module 914.
[00497] The process 1150 represents one iteration that is performed for a current sampling point or time, which corresponds to the most recent sampling period.
This example assumes that the insulin infusion device is already operating in the closed-loop mode (task 1152) to deliver insulin to the body of the user, and that the process 1150 receives relevant data in accordance with a predetermined schedule (e.g., a sampling period of five minutes). Accordingly, the process 1150 receives, obtains, or accesses a variety of inputs that may have an influence on the operation of the model supervisor module 914 (task 1154). For example, the process 1150 may receive at least the following data for the current sampling period: current insulin-delivered data that indicates an amount of insulin delivered by the insulin infusion device during the most recent sampling period; current sensor data that indicates a current sensor glucose value for the user, which corresponds to the most recent sampling period; and a current sensor calibration factor, which may be needed to compensate for recent meter-based calibrations. Any amount of historical data could also be received during task 1154 if so desired. Thus, some amount of redundancy may be built into the system (which may be desirable to account for missed transmissions, lost packets, or the like). The sensor data may be received and processed in any suitable form. For example, a continuous glucose sensor may generate Isig (electrical current) values that can be mapped to sensor glucose values. The model supervisor module 914 may be suitably configured to process Isig values directly, or it could translate or map the raw big values into any desired representation.
1004981 The process 1150 may also access or retrieve historical data that was received for past sampling periods (task 1156). Task 1156 may represent an initialization routine that populates a grid, matrix, or other type of database structure as needed to prepare the model supervisor module 914 for the various calculations, analyses, and functions described in more detail below. It should be appreciated that subsequent iterations of the process 1150 (which are performed in an ongoing manner during the closed-loop mode) need not repeat the initialization of the historical data. Rather, task 1156 may simply adjust the data history to reflect newly received data. For the embodiments described here, the following historical data may be processed by the model supervisor module 914, without limitation: historical insulin-delivered data for the user; and historical sensor glucose values for the user. The historical insulin-delivered data may correspond to amounts of insulin delivered by the insulin infusion device during each historical sampling period of interest, and the historical sensor glucose values may correspond to respective sensor glucose measurements obtained during each historical sampling period of interest. In certain implementations, each historical sensor glucose value may be associated with or derived from a historical Isig value and a sensor calibration factor.
[00499] The process 1150 is iterative in nature, and each iteration considers data associated with the defined historical period of time (see FIG. 54).
Accordingly, the process 1150 may define the model training period and the model prediction period for the historical period of time (task 1158). In this regard, task 1158 may identify or designate which data samples fall within the model training period (the LTH in FIG. 54) and/or which data samples fall within the model prediction period (the LPH in FIG. 54).
Task 1158 may also serve to identify or designate "stale" data samples that need not be considered going forward. In practice, if data for the oldest sampling period is missing for some reason, then the process 1150 can make appropriate adjustments (e.g., search for the closest available data sample, wait for the next sampling period, or the like).

1005001 Next, the process 1150 processes at least some of the historical data to determine a best-matched solution to a sensor glucose prediction model (task 1160). Task 1160 may be considered to be a training procedure that finds the best-fitting sensor glucose prediction function, which in turn can be used to check (predict) the integrity and quality of the glucose sensor. In certain embodiments, the sensor glucose prediction model is expressed as a fourth order ordinary differential equation that, when solved given the initial conditions, provides model-predicted sensor glucose values.
Notably, task 1160 uses the actual sensor glucose values obtained during the model training period (and does not use any of the actual sensor glucose values obtained during the model prediction period) to determine which candidate solution will be selected as the best-matched solution. Conceptually, task 1160 generates a plurality of curves (or discrete values that may be used to visualize curves for purposes of this explanation) and compares the portion of the curves within the model training period to the actual sensor glucose values obtained during the model training period. In an ideal scenario with a perfect match, one of the generated curves will precisely track the actual sensor glucose values within the model training period. In practice, however, the generated curves will deviate from the actual sensor glucose values. Accordingly, task 1160 identifies the calculated curve that best matches the actual sensor values. It should be appreciated that this best-matching curve also includes model-predicted sensor glucose values that extend beyond the model training period and into the model prediction period.
[00501] The process 1150 may continue by comparing at least one historical sensor glucose value obtained during the model prediction period to at least one corresponding predicted sensor glucose value of the best-matched solution (task 1162). In certain embodiments, task 1162 checks only one actual sensor glucose value: the current sensor glucose value obtained for the most recent sampling period. In other embodiments, any or all of the sensor glucose values obtained during the model prediction period could be analyzed during task 1162. The straightforward example described here only considers the current sensor glucose value such that the comparison in task 1162 is simple and straightforward. In this regard, task 1162 may calculate a difference between the current sensor glucose value (i.e., the most recent historical value) and the predicted current glucose value for the most recent sampling period (the difference may be expressed as an absolute value), and the process 1150 may continue by comparing the calculated difference to a threshold error amount (query task 1164). In other embodiments, the comparison performed during task 1162 may involve a more advanced methodology, e.g., curve-fitting that considers more than one sampling point in the model prediction period, statistical analyses, or the like. For example, rather than calculating error on a point-by-point basis, the process 1150 could utilize any appropriate methodology to determine whether or not the historical sensor glucose values in the model prediction period deviate (by at least a threshold amount) from the corresponding model-predicted values in the best-matched solution.
[00502] If the calculated error between the model-predicted glucose value(s) and the corresponding actual historical sensor glucose value(s) is less than or equal to the error threshold, or otherwise satisfies the predetermined criteria monitored by the model supervisor module 914, then the "No" branch of query task 1164 is followed and the process 1150 continues to the next sampling period (task 1166). At this point, the process 1150 returns to task 1152 such that the core of the process 1150 can be repeated to consider the data received for the next sampling period. Thus, the oldest data considered by the previous iteration of the process 1150 is disregarded, the newly received data is designated as the "most recent" data, and the historical time period or "analysis window"
for the current iteration of the process 1150 shifts by one sampling period (see FIG. 54).
[00503] If the calculated error exceeds the threshold error amount (the "Yes"
branch of query task 1164), then the process 1150 may generate an alert, an alarm, and/or a message (task 1168). In practice, an alert, alarm, or message can be initiated by the model supervisor module 914 for rendering, annunciation, delivery, playback, etc.
For example, an alert could be presented at the insulin infusion device, at a remote monitoring station, at a handheld controller device, or the like. In certain embodiments, the process 1150 switches from the closed-loop mode to the open-loop mode (or to some type of safe operating mode with reduced insulin delivery) when the threshold error amount is exceeded (task 1170).
[00504] One important aspect of the process 1150 relates to the manner in which the best-matched sensor glucose prediction model is chosen (see task 1160). In this regard, FIG. 56 is a flow chart that illustrates an exemplary embodiment of a sensor model training process 1180, which may be performed in conjunction with the sensor model supervision process 1150 depicted in FIG. 55. The process 1180 is shown and described in a simplified manner for ease of understanding. Certain aspects of the process 1180 are addressed in more detail below with reference to particular implementations of the model supervisor module 914.

[00505] As mentioned previously, the exemplary sensor glucose prediction model utilized here is expressed as a fourth order ordinary differential equation.
In accordance with conventional mathematics, the model-predicted sensor glucose values (G) in time are calculated as a function of the two model prediction initial conditions Go and dGo. Here, Go is the estimated sensor glucose value for the begin-training sampling period 1124 (the start of LTH in FIG. 54), and dG0 is the derivative of Go. Therefore, different initial condition values result in different solutions to the sensor glucose prediction model; each distinct set of initial conditions corresponds to a different prediction model. For the sake of processing efficiency, the model supervisor module 914 imposes limits and boundaries on the initial condition values to calculate and analyze a manageable number of candidate solutions. In this regard, the sensor model training process 1180 may begin by calculating a range or boundary for each bounded initial condition (task 1182).
[00506] For the exemplary embodiment presented here, the initial condition dG0 is bounded in a simple manner that is based on a predetermined parameter (which may be adjustable): dG0 = +grad_bound. In contrast, the boundary for the initial condition Go is based on (or is otherwise influenced by) a baseline historical sensor glucose value obtained during the model training period, such as the sensor glucose value that was obtained during the begin-training sampling period 1124. Thus, the process 1180 may identify, from the historical sensor glucose values, the baseline sensor glucose value to be used for purposes of calculating the boundary for the initial condition Go: Go =
SG k-LTH + 0.14 = SG k_LTH, where SG k-LTH is the baseline sensor glucose value obtained for the earliest sampling period in the historical time period under analysis (see FIG. 54).
Notably, the boundary for Go is a function of the baseline sensor glucose value, which may vary in an ongoing manner during operation of the system, and which may vary from one iteration of the process 1180 to another. In practice, if the sensor glucose data is missing for the begin-training sampling period 1124, then the process 1180 can take appropriate measures, e.g., search for the nearest available sensor glucose data point, wait for the next sampling period, etc.
[00507] The process 1180 may then continue by determining, calculating, or otherwise obtaining the next set of initial conditions (task 1184). The manner in which the process 1180 selects and steps through the different initial conditions is unimportant in this context. The current set of initial conditions is used to calculate a candidate solution to the sensor glucose prediction model (task 1186). As mentioned above, each candidate solution is calculated as a function of the two bounded initial conditions.
Moreover, each candidate solution is calculated as a function of estimated plasma insulin for the user, which in turn is calculated as a function of an amount of insulin delivered to the user.
Accordingly, task 1186 may estimate plasma insulin for the user based on: the current insulin-delivered data (obtained for the most recent sampling period); the historical insulin-delivered data; and an insulin basal rate for the user. In practice, task 1186 considers the total insulin (basal, bolus, and any other insulin delivered) for all sampling periods. This allows the process 1180 to obtain the candidate solution to the sensor glucose prediction model based at least in part on the estimated plasma insulin and based at least in part on the baseline sensor glucose value obtained at the earliest sampling period under analysis.
1005081 The process 1180 may continue by generating a training error value, quantity, or function for the candidate solution (task 1188). The training error may be calculated by comparing predicted sensor glucose values from the candidate solution to the corresponding historical sensor glucose values, to obtain a metric that indicates how closely the predicted values match the actual values. In certain embodiments, the training error is based only on predicted values and actual values for the model training period (LTH in FIG. 54), and, therefore, task 1188 does not consider any predicted or actual values for the model prediction period (LPH in FIG. 54).
[00509] If the process 1180 has considered all of the initial condition combinations (the "Yes" branch of query task 1190), then the process 1180 may proceed to a task 1192.
If more sets of initial conditions remain (the "No" branch of query task 1190), then the process 1180 may return to task 1184, retrieve the next set of initial conditions, and continue as described above. Task 1192 is performed after a plurality of different candidate solutions have been calculated, using the different sets of initial conditions.
Task 1192 may be performed to select the best-matching candidate solution from the plurality of calculated solutions. For this particular embodiment, the selection is based on the training errors generated during task 1188. For example, the candidate solution having the lowest training error may be selected as the best-matching solution.
[00510] It should be appreciated that the process 1180 need not be performed in the illustrated sequence, and that some tasks could be performed in parallel. For example, the calculation of the training errors (task 1188) may instead be performed after all of the candidate solutions have been obtained and saved. Moreover, the process 1180 could be designed to immediately eliminate candidate solutions (following completion of task 1188) that have a training error that exceeds a predetermined error threshold.
As another option, the process 1180 could be designed to immediately designate a candidate solution as the best-matched solution if the associated training error satisfies certain criteria.
[00511] The foregoing concepts and methodologies may be implemented in a practical embodiment of the model supervisor module 914. The following description relates to two possible embodiments that implement the general concepts presented above.
It should be appreciated that the particular embodiments described below are not exhaustive, and that the description of the embodiments is not intended to limit or restrict the scope or application of the subject matter presented here.
[00512] Model Supervisor Module: First Representation [00513] The model supervisor module 914 is suitably designed and configured to detect potentially faulty sensor measurements. The model supervisor module 914 may utilize a mathematical model that is trained offline. For example, parameters that could be estimated offline estimated include, without limitation: K1 (insulin gain in mg/dL per U/h;
ri (first insulin time constant, in minutes); r2 (second insulin time constant, in minutes);
Masai (basal insulin, in U/h); and SGbase (blood glucose (BG) at fasting, in mg/dL, when Masai insulin is delivered.
[00514] The model supervisor module 914 trains the model prediction initial conditions, Go and dGo every sampling time. Go and dGo represent the BG
(mg/dL) and BG derivative (mg/dUmin) estimate values at k-LTH (see FIG. 54), where LTH is the length of the training data (sampling times) and k is equal to the present sampling time minus LPH. In this context, LPH is the length of the prediction horizon in sampling times.
Go and dGo estimations are bounded as formulated by the expressions collectively identified below as Equation 77. Note that these initial conditions and their boundaries were also described above with reference to task 1182 of the sensor model training process 1180.
Go = CG k _LTH 0.14 = CG Mk_LTH
(eq 77) dG0 = + grad_bound For Equation 77, CGAlk-mi is the CGM measurement at sampling time k-LTH, and grad bound is a predefined absolute maximum BG derivative in time (mg/dL/min).

1005151 The model supervisor module 914 estimates plasma insulin, 4õ at k-LTH
using the insulin history records from k-LTH-insulin history and k-LTH (see FIG. 54) in accordance with Equation 81. Having the estimated /p, Go, and dGo, a model prediction is generated from present-LTH-LPH, until present (as described above for task 1186 of the sensor model training process 1180). The model prediction enables the calculation of two values: Terr and Perr. Terr is defined as the mean sum square of errors between the model prediction and the CGM records from k-LTH and k (Equation 78). Perr is defined as the absolute mean error between the model prediction and the CGM records from k and ' present (Equation 79). Note that Terr is one type of training error, which was described above for task 1188 of the process 1180, and Perr is one type of prediction error, as described above for task 1162 and query task 1164 of the sensor model supervision process 1150. A fault is defined when Perr < err] and Terr > err2 (Equation 80).
Elic=k_LTH(Modeli¨CGMi) 2 Terr = _________________________________________ (eq 78) LTH
EI-FLPH Modeli¨CGMi) Perr = abs( E¨k LPH (eq 79) 1, if Terr < err1 and Perr > err2 Fault = 0, else if Terr < err2 or Perr > err1 ¨1, if not enough data records available (eq 80) In Equation 80, Fault 1 indicates a faulty sensor, Fault 0 indicates a non-faulty sensor, and Fault -1 indicates that there is not enough information to decide.
Referring again to FIG. 55, Fault 1 corresponds to the "Yes" branch of query task 1164.
1005161 In certain embodiments, some of the parameters used by the model supervisor module 914 may be adjustable. Table 5 identifies the adjustable parameters, along with some exemplary values for the parameters.
Parameter Symbol Typical Starting Minimum Value Maximum Value Value Insulin gain K1 -100 -500 -1 Insulin time ri 30 0.1 150 constant Insulin time r2 150 0.1 150 constant Fasting BG SGbase 120 50 300 Basal insulin at lbasal 1 0.1 3 fasting BG
Training data err! 5 1 100 error Prediction data err2 20 1 100 error Length of LTH 8 5 30 training data Length of LP!! 3 1 20 prediction data Length of insulin Insulin History 48 30 60 data BG maximum grad bound 5 1 8 absolute derivative Table 5: Adjustable Parameters For the Model Supervisor Module [00517] The following equations describe the mathematical model equations in Laplace transform form:
In(s) ¨ (50s+1)(7os-1-1) ipos a a + dlpofl) (eq 81) In this expression, cc=3500, 3-120, and fp is the /p in deviation form. ipo and dip are the ip and derivative initial conditions, respectively.
[00518] All the insulin states are formulated in deviation form from the given insulin value lbasal as it is expressed by the following Equation 82:
ix = ¨ Ibasal (eq 82) In Equation 82, x represents D, in, or P.
1005191 The following Equation 83 expresses BG in deviation form from SGbase.
Note that Equation 83 represents one suitable expression for the sensor glucose prediction model, which is a fourth order ordinary differential equation.

6(S) ¨ __________ (Ki = ip ____ (60 = a + (6o = s + A) = fl +
(ris+1)2s+1) (50s+1)(70s+1) -(k) = S2 CI60 = S) = X + (60 = s3 + d60 = s2) = 8)) (eq 83) In Equation 83, the following relationships hold:
a = 120 + r1 + T2 = 3500+ 120Ti +120r2 + TiT2 x = 3500Ti + 3500T2 + 120T1T2 6 = 3500T1T2 Moreover, in Equation 78, d, Kb do, dGo, rj, and 1-2 are the BG in deviation form from SGbase, the insulin gain, the BG initial conditions in deviation form, the BG
derivative initial conditions, and two time constants, respectively.
[00520] Model Supervisor Module: Second Representation [00521] In accordance with some embodiments, the functionality of the model supervisor module 914 can be represented as follows. As mentioned above, the model supervisor module 914 estimates the user's glucose concentration in real-time based on the insulin delivered, the sensor Isig values, and sensor calibration factors.
If the model-predicted sensor glucose value (SG) and the actual SG value differ significantly, the system will trigger a fail-safe alert that indicates that the collected data contains unexplained behavior, which may in turn be associated with a faulty sensor and/or insulin delivery, or an unannounced meal intake.
[005221 The time frames and reference time periods for the model supervisor module 914 are defined as shown in FIG. 54. The methodology performed by the model supervisor module 914 uses data packets received for past time frames to estimate plasma insulin and model-predicted glucose in order to estimate the fault conditions.
The sampling time is the time interval between two consecutive data packets, which for this particular example is five minutes. The insulin history in FIG. 54 corresponds to a defined past time frame that is needed to estimate plasma insulin (for this example, the insulin history corresponds to four hours or 48 sampling periods). The length training horizon (LTII) for this example includes 24 data packets, which corresponds to a past time frame of 120 minutes. The length of predicted horizon (LH?) for this example includes 24 data packets, which corresponds to a past time frame of 120 minutes. In FIG. 54, k is equal to the present number of data packets minus LPH, and "present" indicates the most recent sampling time.
[00523] The following equations describe the mathematical model in Laplace transform form. Equation 84 provides an estimate of the plasma insulin, and Equation 85 provides the model-predicted SG values. Accordingly, the model supervisor module 914 in accordance with this particular embodiment estimates the plasma insulin as follows:
In(s) ¨ (sasi i)(7os+1) (ID + ipoS E dipoy) (eq 84) For this example, c = 3500, y = 120, ip is the estimated plasma insulin in deviation form, (s) refers to Laplace transform form, and ID is the insulin delivered from the system in deviation form. Moreover, ipo is the estimated plasma insulin in deviation form for the sampling time identified as k-LTH (see FIG. 54), dip is the derivative of the estimated plasma insulin, and a and 13 are constants.
[00524] The insulin states described above are formulated in deviation form from a given insulin value Ibasal as it is expressed by the following equation:
ix =I - Ibasal (eq 85) In Equation 85, x represents D or P (where D refers to insulin delivered and P
refers to plasma insulin), and hasai,o is the estimated basal rate defined for each user to bring the patient to a fasting blood glucose (FBG) of the value FBG0 (in mg/dL).
[00525] For this second embodiment, the model-predicted sensor glucose value, d, in time is calculated in accordance with Equation 83 and the associated relationships, as described for the first embodiment of the model supervisor module 914. In this regard, d is the model-predicted SG value in deviation form from FBG0 (estimated blood glucose using meter blood glucose readings at the end of the night period), (s) refers to Laplace transform form, il, and I are the two insulin time constants identified for each patient, which are related to how fast a patient reacts to insulin, k is the insulin gain, and ip is the estimated plasma insulin in deviation form. Moreover, a is the estimated SG
value (in mg/dL) in deviation form for the sampling time of k-LTH (see FIG. 54), as calculated in accordance with Equation 86 below, and dGo (calculated by Equation 87 below) is the derivative of the estimated SG value (in mg/dL/min) for the sampling time of k-LTH. The constants a, 13, x, and 8 are calculated as set forth above in the context of Equation 83.
[00526] The estimated blood glucose values are calculated as a function of the model prediction initial conditions, Go and dGo. For this particular embodiment, the estimations for Go and dGo are bounded as formulated by the following equations. Note that these initial conditions and their boundaries were also described above with reference to task 1182 of the sensor model training process 1180.
Go = SGk¨LTH 0.14 = SGk_LTH (eq 86) dGo = + grad_bound (eq 87) Here, Go is the estimated SG value (in mg/dL) value for the sampling time of k-LTH, SGk_LTH is the SG measurement for the sampling time of k-LTH, dG 0 is the derivative of the estimated SG value (in mg/dL/min) for the sampling time of k-LTH, and grad_bound is a predefined absolute maximum SG derivative in time (mg/dL/min). For certain embodiments, grad_bound is a fixed parameter. For the example presented here, grad_bound has a value of 5 mg/dL/min.
[00527] The model prediction facilitates the calculation of two values: Terr and Perr.
Terr is defined as the mean absolute error between the model predicted SG
values and the actual SG records for the sampling time identified as k-LTH and k as calculated by Equation 88 below. Perr is defined as the mean absolute error between the model predicted SG values and the actual SG records for sampling times identified as k to present (see FIG. 54) as calculated by Equation 89 below.
abs(Modeli-SGi) t=k-LTH
Terr (eq 88) LTH
Here, Terr is defined as the mean absolute error between the model predicted SG values (Model) and the SG records (SG) for the sampling time identified as k-LTH and k.
abs(Modelpresent¨SGpresent) Perr = = 100% (eq 89) SGpresent Here, Perr is defined as the percentage of error between the model prediction and the SG
measurement at the present (most recent) sampling time.

[00528] In accordance with this particular implementation, the model supervisor module 914 estimates the fault scenario based on Equation 90, where Fault 1 signifies a faulty sensor, Fault 0 indicates a non-faulty sensor, Fault 3 indicates a training error, and Fault -1 indicates that there is not enough data available to make a determination.
1, if Terr < err1 and Perr > err2 0, if Perr err2 or Terr err1 Fault = (eq 90) 3, if Terr > err3 k. ¨1, if not enough data available In Equation 90, err] is the upper threshold for the mean absolute error (see Equation 88).
Thus, if the training error is above this threshold, a fault cannot be triggered because the credibility of the training is suspect. The err2 is the lower threshold for Equation 89. If the prediction value of the model and the sensor measurement at present are above this threshold and the training error is less than err], then a fault will be triggered. The err3 defines a lower threshold for the training period. If Equation 88 indicates a value that is above this threshold, then a warning associated with poor training can be triggered.
[00529] FIG. 57 is a diagram that illustrates exemplary sensor conditions corresponding to a non-faulty sensor (Fault 0) and a faulty sensor (Fault 1).
The common horizontal axis indicates the present sampling time at the far right, along with the time periods identified by LPH and LTH. The sampling time 1202 corresponds to the oldest data considered by the model supervisor module 914 at the present time.
Accordingly, historical data 1204 for sampling times that occurred before the sampling time 1202 is disregarded.
[00530] The top plot 1206 in FIG. 57 is indicative of a non-faulty sensor (Fault 0), the middle plot 1208 is indicative of a faulty sensor (Fault 1), and the bottom plot 1210 depicts the insulin administered, which is needed in order to estimate the plasma insulin and to generate the model predicted SG values. In the plots 1206, 1208, the solid line 1212 represents the model-predicted SG values, and the dots represent the actual SG
measurements. The dashed vertical line 1214 represents the demarcation between the LTH time frame and the LPH time frame. The lines between the solid line 1212 and the dots represent the difference (error) between the model-predicted SG values and the actual SG measurements. Dashed lines are utilized in the LPH time frame, which corresponds to fifteen minutes or three sampling periods for this example.

[00531] Referring to the top plot 1206, there is good agreement between the model predicted SG values (represented by the solid line 1212) and the actual SG
measurements (represented by the dots). In other words, the actual measurements do not deviate significantly from the predicted values. In certain embodiments, the model supervisor module 914 only compares actual measurement values that are within the LPH
time frame. In accordance with one exemplary embodiment, the model supervisor module 914 determines the fault status based solely on the most recently obtained data, i.e., the information received for the last sampling time. For this example depicted in FIG. 55, Perr is less than or equal to err2. Thus, in accordance with Equation 90, the model supervisor module 914 returns Fault 1 and the system is commanded to remain in the closed-loop mode.
[00532] Referring to the middle plot 1208, of FIG. 57, there is good agreement between the model predicted SG values and the actual SG measurements in the LTH time period (for this period, Terr is less than err] in Equation 90). Note, however, that there is a significant difference observed between the model predicted SG last value and the actual last SG measurement 1218 (Perr is greater than err2 in Equation 90). In this scenario, therefore, the model supervisor module 914 will issue a fail-safe alert and/or take other appropriate measures.
1005331 In certain embodiments, some of the parameters used by the model supervisor module 914 may be adjustable. Table 6 identifies some adjustable parameters for this embodiment, along with some exemplary values for the parameters.
Parameter Default Value Lower Bound Upper Bound (mgidL/U/H) -100 -360 -49 FBGo (mat) 120 50 300 lbasal (U/H) 1 0.1 3 errl (mg/dL) 5 1 30 err2 (%) 50 20 100 err3 (mg/dL) 10 1 30 LTH (sampling times) 24 4 48 LPH (sampling times) 24 1 48 Table 6: Adjustable Parameters For the Model Supervisor Module [00534] Missed Transmission Module [00535] The missed transmission module 916 continuously checks whether the controller receiving data packets (including SG values) for processing. The missed transmission module 916 keeps the system operating in the closed-loop mode for when less than a stated number of data packets are missed (e.g., less than four data packets in a row, a total number of data packets that represent a timespan of less than 15 minutes, or the like). During this time, the system will continue to calculate the insulin dose using the closed-loop control algorithm based on the last valid sensor glucose value or sensor big value. For missed data packets that represent a time longer than a lower time threshold and longer than an upper time threshold (e.g., between 15 and 60 minutes), the missed transmission module 916 switches the system to a pre-programmed safe basal rate, which may be defined as half the patient's nighttime basal rate. If the controller starts receiving data packets during the safe basal rate timeframe, the system will switch back to the closed-loop mode. For missed data packets that represent a time longer than the upper time threshold, however, the missed transmission module 916 switches the system to the open-loop mode to deliver a pre-programmed nighttime basal rate, which may be set by a healthcare provider or a caregiver.
[00536] The missed transmission module 916 cheeks for different scenarios pertaining to when and what kind of packet is lost during transmission. Different steps are executed depending on the type of lost transmission. The details of four different scenarios are described below.
[00537] Case 1 [00538] If the sensor Isig value and the SG value are both received by the controller, then:
(a) the sensor Isig is saved by the controller;
(b) the SG value is saved by the controller;
(c) a Zero Order Hold (ZOH) count is set to zero; and (d) the system remains in the closed-loop mode as described previously.
[00539] Case 2 1005401 If the sensor Isig value is not received, but the SG value is received by the controller, then:
(a) the ZOH count is set to zero;

(b) Isig is calculated by Equation 91 (see below) using the SG value and the sensor calibration factor; and (c) the system remains in the closed-loop mode.
isigccac = (SG/CF') + 2 (eq 91) [00541] Case 3 [00542] If the sensor Isig value is received, but the SG value is not received by the controller, then:
(a) the ZOH count is set to zero;
(b) SG is calculated by Equation 92 (see below) using the Isig value and the sensor calibration factor; and (c) the system remains in the closed-loop mode.
SGcaic = (Isig ¨ 2) x CF' (eq 92) [00543] Case 4a 1005441 If neither the sensor Isig value nor the SG value are received by the controller (i.e., both values are not received), and if:
ZOH Count < ZOH Count Max then:
(a) the ZOH count for the sensor Isig and SG is calculated based on previous values;
(b) ZOH Count = ZOH Count +1;
(c) TimeoutCount = 0; and (d) the system remains in the closed-loop mode.
[00545] Case 4b 1005461 If neither the sensor lsig value nor the SG value are received by the controller (i.e., both values are not received), and if:
ZOH Count > ZOH Count Max then:
(a) an "invalid" place holder for the sensor lsig and SG value is saved;
(b) the system remains in the closed-loop mode, but switches to a temporary safe basal rate, which is half the patient's night time basal rate when in the open-loop mode;
(c) if a packet is received by the system while it is delivering the safe basal rate, the system will transition back to the closed-loop mode;
(d) for every minute that the system is delivering the safe basal rate, a TimeoutCount is incremented: TimeoutCount = TimeoutCount + /;
(e) if TimeoutCount > Timeout Count Max, then the system switches to the open-loop mode.
1005471 In accordance with certain embodiments, ZOH Count Max has a fixed value of two, and Timeout Count Max has a fixed value of 45, although different values may be used as appropriate to the particular implementation. Moreover, the safe basal rate used by the missed transmission module 916 may be adjustable. In this regard, the safe basal rate may be adjustable within a range of about zero to five Units/Hour.
1005481 While at least one exemplary embodiment has been presented in the foregoing detailed description, it should be appreciated that a vast number of variations exist. It should also be appreciated that the exemplary embodiment or embodiments described herein are not intended to limit the scope, applicability, or configuration of the claimed subject matter in any way. Rather, the foregoing detailed description will provide those skilled in the art with a convenient road map for implementing the described embodiment or embodiments. It should be understood that various changes can be made in the function and arrangement of elements without departing from the scope defined by the claims, which includes known equivalents and foreseeable equivalents at the time of filing this patent application.

Claims (15)

Claims:

1. A system for controlling an insulin infusion device for infusing insulin into a user including:
a processor architecture having at least one processor; and at least one memory element associated with the processor architecture, the memory element storing processor-executable instructions which when executed by the processor architecture, cause:
a) operating the insulin infusion device in a closed-loop mode to deliver insulin to the body of the user;
b) obtaining current insulin-delivered data that indicates an amount of insulin delivered by the insulin infusion device during a most recent sampling period;
c) obtaining current sensor data that indicates a current sensor glucose value for the user corresponding to the most recent sampling period;
d1) defining a model training period and a model prediction period for a historical period of time wherein the model training period precedes the model prediction period;
d2) obtaining historical insulin delivered data and historical glucose sensor data from said model training period;
d3) obtaining historical insulin delivered data and historical glucose sensor data from said model prediction period;
d4) finding a model to predict said historical glucose sensor data during the model training period from a bounded initial condition comprising a baseline glucose value measured at the first sampling period of the model training period, and the insulin delivered data during the model training period, wherein finding said model comprises calculating a plurality of candidate solutions to a sensor glucose prediction model and selecting a solution from the calculated plurality of candidate solutions that best matches the historical glucose sensor data from the model training period;

d5) using said model for a plurality of sampling periods within the model prediction period, to obtain predicted sensor glucose values, including a predicted current sensor glucose value;
e) calculating a difference between the current sensor glucose value and the predicted current sensor glucose value for the most recent sampling period;
and f) generating an alert when the difference exceeds a threshold error amount.

2. The system of claim 1, wherein at e) respective differences between a plurality of model-predicted sensor glucose values within the model prediction period and corresponding historical measured sensor glucose values are calculated, and the alert is generated when the model-predicted sensor glucose values deviate by more than threshold amounts from the corresponding measured values.

3. The system of claim 1 or 2, wherein a) to f) repeat for successive new current insulin-delivered and sensor data while said difference does not exceed the threshold error amount.

4. The system of claim 1, 2, or 3, wherein the memory element further includes instructions which when executed cause the infusion device to switch from the closed-loop mode to an open-loop mode in response to the alert.

5. The system of claim 1, wherein the first sampling period in the model prediction period immediately follows the last sampling period in the model training period.

6. The system of any one of claims 1 to 5, wherein the model operates by:
estimating plasma insulin for the user based on the current insulin-delivered data, the historical insulin-delivered data, and an insulin basal rate for the user; and obtaining the model-predicted sensor glucose values based on the estimated plasma insulin.

7. The system of any one of claims 1 to 6, wherein each of the plurality of candidate solutions is calculated as a function of the bounded initial condition.

8. The system of any one of claims 1 to 7, wherein the number of historical sampling periods prior to the most recent sampling period remains fixed and the instructions repeatedly cause the processor architecture to obtain new current insulin-delivered values and new current sensor values, thereby fonning a moving window of historical data for processing, and disregarding earlier values than said fixed number of samples.

9. A method of controlling an insulin infusion device, the method being perfomied by a controller comprising a processor architecture comprising at least one processor device and at least one memory element associated with the processor architecture, the memory element storing processor-executable instructions which, when executed by the processor architecture, cause:
a) operating the insulin infusion device in a closed-loop mode;
b) obtaining current insulin-delivered data that indicates an amount of insulin delivered by the insulin infusion device during a most recent sampling period;
c) obtaining current sensor data that indicates a current sensor glucose value corresponding to the most recent sampling period;
d1) defining a model training period and a model prediction period for a historical period of time wherein the model training period precedes the model prediction period;
d2) obtaining historical insulin delivered data and historical glucose sensor data from said model training period;
d3) obtaining historical insulin delivered data and historical glucose sensor data from said model prediction period;
d4) finding a model to predict said historical glucose sensor data during the model training period from a bounded initial condition comprising a baseline glucose value measured at the first sampling period of the model training period, and the insulin delivered data during the model training period, wherein finding said model comprises calculating a plurality of candidate solutions to a sensor glucose prediction model and selecting a solution from the calculated plurality of candidate solutions that best matches the historical glucose sensor data from the model training period;
d5) using said model for a plurality of sampling periods within the model prediction period, to obtain predicted sensor glucose values, including a predicted current sensor glucose value;
e) calculating a difference between the current sensor glucose value and the predicted current sensor glucose value for the most recent sampling period;
and f) generating an alert when the difference exceeds a threshold error amount.

10. The method of claim 9, further comprising:
switching from the closed-loop mode to an open-loop mode in response to generating the alert.

11. The method of claim 9 or 10, wherein at e) respective differences between a plurality of model-predicted sensor glucose values within the model prediction period and corresponding historical measured sensor glucose values are calculated, and the alert is generated when the model-predicted sensor glucose values deviate by more than threshold amounts from the corresponding measured values.

12. The method of claim 9, 10, or 11, wherein the method repeats for successive new current insulin-delivered and new current sensor data, the previous current insulin-delivered and previous current sensor data being included in the historical data, whereby historical data belonging to the earliest sampling period is disregarded, thereby forming a moving window of historical data.

13. The method of any one of claims 9 to 12, wherein the first sampling period in the model prediction period immediately follows the last sampling period in the model training period.

14. The method of any one of claims 9 to 13, wherein the model operates by:
estimating plasma insulin based on the current insulin-delivered data, the historical insulin-delivered data, and an insulin basal rate; and obtaining the model-predicted sensor glucose values based on the estimated plasma insulin.

15. The method of any one of claims 9 to 13, wherein each of the plurality of candidate solutions is calculated as a function of the bounded initial condition.