Method of overnight closed-loop insulin delivery with model predictive control and glucose measurement error model

• US 8,585,637 B2
• Filed: 09/22/2011
• Issued: 11/19/2013
• Est. Priority Date: 03/31/2009
• Status: Active Grant

• Current Assignee: Abbott Diabetes Care Incorporated (Abbott Laboratories Incorporated), Cambridge Enterprises Limited
• Original Assignee: Abbott Diabetes Care Incorporated (Abbott Laboratories Incorporated), Cambridge Enterprises Limited
• Inventors: Wilinska, Malgorzata E., Budiman, Erwin S., Hayter, Gary A., Taub, Marc B., Hovorka, Roman
• Primary Examiner(s): VU, QUYNH-NHU HOANG
• Application Number:
  US13/240,855
• Publication Number:
  US 20120010600A1
• Time in Patent Office: 789 Days
• Field of Search: 604 65- 67, 604/500, 600/300, 600/316, 600/319, 600/347, 600/365
• US Class Current: 604/65
• CPC Class Codes: A61B 5/14532 for measuring glucose, e.g. by tissue impedance measurement
• CPC Class Codes: A61B 5/1495 Calibrating or testing of in-vivo probes
• CPC Class Codes: A61M 2005/14208 with a programmable infusion control system, characterised by the infusion program
• CPC Class Codes: A61M 2005/14292 Computer-based infusion planning or simulation of spatio-temporal infusate distribution
• CPC Class Codes: A61M 2005/14296 Pharmacokinetic models
• CPC Class Codes: A61M 2205/50 with microprocessors or computers
• CPC Class Codes: A61M 2205/52 with memories providing a history of measured variating parameters of apparatus or patient
• CPC Class Codes: A61M 2205/70 with testing or calibration facilities
• CPC Class Codes: A61M 2230/201 Glucose concentration
• CPC Class Codes: A61M 5/142 Pressure infusion, e.g. using pumps
• CPC Class Codes: A61M 5/1723 using feedback of body parameters, e.g. blood-sugar, pressure measurement of body parameters A61B5/00
• CPC Class Codes: G16H 20/17 delivered via infusion or injection

Abstract

A closed-loop method for insulin infusion overnight uses a model predictive control algorithm (“MPC”). Used with the MPC is a glucose measurement error model which was derived from actual glucose sensor error data. That sensor error data included both a sensor artifacts component, including dropouts, and a persistent error component, including calibration error, all of which was obtained experimentally from living subjects. The MPC algorithm advised on insulin infusion every fifteen minutes. Sensor glucose input to the MPC was obtained by combining model-calculated, noise-free interstitial glucose with experimentally-derived transient and persistent sensor artifacts associated with the FreeStyle Navigator® Continuous Glucose Monitor System (“FSN”). The incidence of severe and significant hypoglycemia reduced 2300- and 200-fold, respectively, during simulated overnight closed-loop control with the MPC algorithm using the glucose measurement error model suggesting that the continuous glucose monitoring technologies facilitate safe closed-loop insulin delivery.

Claims

1. A method for delivering insulin to a patient, the method comprising:

sensing a glucose level and providing a glucose measurement signal representative of the sensed glucose;

providing a control signal as a function of the glucose measurement signal in accordance with a control model and a glucose measurement error model, wherein the glucose measurement error model is derived from actual glucose sensor measurement data; and

delivering insulin in response to the control signal.

2. The method for delivering insulin of claim 1, wherein the glucose measurement error model is derived solely from actual glucose sensor measurement data.

3. The method for delivering insulin of claim 1, wherein the glucose measurement error model is derived solely from actual glucose sensor error data, excluding sensor noise data.

4. The method for delivering insulin of claim 1, wherein the glucose measurement error model is derived solely from actual glucose sensor measurement data to the exclusion of randomly-generated variable data.

5. The method for delivering insulin of claim 1, wherein the glucose measurement error model is derived solely from a fixed time history of error data from actual use of a glucose sensor of the same type as the sensor of the system.

6. The method for delivering insulin of claim 1, wherein the glucose measurement error model is derived from actual glucose sensor measurement data from a glucose sensor of the same type as the sensor of the system.

7. The method for delivering insulin of claim 1, wherein the glucose measurement error model is derived solely from a fixed time history of error data from actual use of a glucose sensor of the same type as the sensor of the system, to the exclusion of randomly-generated variable data and to the exclusion of sensor noise data.

8. The method for delivering insulin of claim 1, wherein providing the control signal further comprises producing the control signal in accordance with a model predictive control.

9. The method for delivering insulin of claim 1, further comprising determining a calibration error of a glucose sensor from actual sensor data and deriving the glucose measurement error model therefrom.

10. The method for delivering insulin of claim 9, wherein determining a calibration error comprises determining calibration error based on the difference between a plasma glucose level and the glucose level signal.

11. The method for delivering insulin of claim 1, further comprising determining a glucose sensor dropout reading from actual sensor data and deriving the glucose measurement error model therefrom.

12. The method for delivering insulin of claim 1, wherein providing the control signal is also a function of weight of a patient, a total daily insulin dose, and a basal insulin profile, the method further comprising:

determining, based on the control model, at least one accepted value;

calculating from the glucose level signal at least one inferred value;

adjusting the control model in accordance with the accepted value and inferred value; and

forecasting a future plasma glucose level excursion based on the control model.

13. The method for delivering insulin of claim 12, wherein determining the accepted value comprises basing the determination on an insulin sensitivity of the patient, a glucose distribution volume, and an insulin distribution volume.

14. The method for delivering insulin of claim 12, wherein calculating the inferred value comprises calculating the inferred value also from glucose flux and a carbohydrate bioavailability.

15. The method for delivering insulin of claim 1, further comprising adjusting a value of the control signal in accordance with a safety check.

16. The method for delivering insulin of claim 15 wherein adjusting a value of the control signal comprises at least one of:

imposing a maximum infusion rate related to a basal rate depending on a current sensor glucose level, time since a previous meal, and carbohydrate content of a meal;

shutting off insulin delivery at a predetermined low sensor glucose value;

reducing insulin delivery when sensor glucose is decreasing rapidly; and

capping the insulin infusion to a pre-programmed basal rate if an insulin delivery pump occlusion is inferred.

17. The method for delivering insulin of claim 1, wherein sensing, providing a control signal, and delivering insulin are performed virtually, each occurring for in silico testing of a method for delivery of insulin to a virtual patient.