Patients were recruited from the Oregon Health and Science University (OHSU) outpatient clinics in Portland, Oregon. Patients who were pregnant or had cardiovascular, cerebrovascular, kidney, or liver disease or any other uncontrolled chronic medical conditions were excluded. Other exclusion criteria included oral or parenteral corticosteroid use, immunosuppressant use, visual or physical impairments that impede the use of a continuous glucose-monitoring device, insulin or glucagon allergy, hypoglycemia unawareness or hospitalization within the past 2 years for severe hypoglycemia, serum insulin antibody titer >100 μU/ml, or requirement of >200 units insulin/day. The research protocol was approved by the OHSU Institutional Review Board, and all subjects provided written informed consent. Permission to carry out these studies was granted by the U.S. Food and Drug Administration (FDA) (investigational device exemption no. G080130).
A total of 22 closed-loop studies in 14 subjects were performed. Age was 36.7 ± 3.7 years, with a duration of diabetes of 14.1 ± 3.1 years. A1C was 7.6 ± 0.3% and BMI 27.8 ± 1.5 kg/m2. The study for one patient was stopped early because of repeated intravenous catheter failures. The data from this study were excluded from the analysis, leaving 21 datasets from 13 subjects.
As requested by the FDA, five subjects participated in single 9-h studies with both insulin and glucagon to assess the safety and effectiveness of the study protocol. Eight subjects underwent one study with insulin and placebo and one with insulin and glucagon (see ). Of the 13 studies during which glucagon was given, it was delivered using high-gain parameters in seven studies and using low-gain parameters in six. Low- versus high-gain glucagon is discussed in detail below. The treatment order of each paired study was determined by a randomization scheme. In paired studies, subjects were blinded as to whether they received glucagon or placebo.
Study diagram depicting the number of subjects studied under each condition and the study lengths.
Subjects wore two subcutaneous glucose sensors, either DexCom Seven Plus or Medtronic Guardian Real-Time glucose sensors. Sensors were placed 8–24 h prior to beginning the study. For subjects taking long-acting insulin at night, the dose was reduced by 50% the night prior to the study. The following morning, subjects were admitted to the Oregon Clinical and Translational Research Institute at OHSU. An intravenous catheter was placed in a forearm vein. The forearm was warmed with a heating pad to arterialize the venous blood. Venous glucose was measured every 10 min in duplicate using a HemoCue Glucose 201 Analyzer. Glucose sensor readings were recorded from the receivers every 5 min. For the first 2 h, the insulin and glucagon delivery rates were determined by venous glucose levels. After the first 2 h, the sensed glucose values from the sensor with better accuracy were input into the algorithm every 5 min to determine the hormone delivery rates. If the sensor accuracy became suboptimal, defined as a median absolute relative difference (MARD) exceeding 20% or median absolute difference (MAD) exceeding 20 mg/dl, control was switched to the other sensor. If the accuracy of both sensors was poor, control was switched to venous glucose and the sensors were recalibrated. Sensors were calibrated at a minimum of every 12 h.
The Fading Memory Proportional Derivative (FMPD) algorithm (9
) was used to determine the insulin and subcutaneous glucagon (or placebo) delivery rates. Aspart insulin (Novo Nordisk) was delivered subcutaneously via an Animas IR 1000 insulin pump. Glucagon or saline placebo was given through a subcutaneous catheter via a Medfusion 2001 syringe pump. One milligram of glucagon (Novo Nordisk) was mixed with 3 ml of sterile water. The glucagon preparation was freshly reconstituted every 8 h. A study physician was onsite at all times and had the ability to override the hormone infusion rates called for by the FMPD algorithm, which occurred only 1.7% of the time. Either a registered nurse or physician was responsible for adjusting the insulin delivery rate and glucagon delivery rate every 5 min, based on the controller output.
The FMPD algorithm determined the hormone delivery rates based on proportional error, defined as the difference between the current glucose level and the target level, and the derivative error, defined as the rate of change of the glucose. The “fading memory” designation refers to weighting recent errors more heavily than remote errors. This weighting provides an adaptive component to the algorithm, as described previously (9
). In simple terms, the insulin rate was increased for high or rising glucose levels and glucagon was given for low or falling glucose levels. The basal insulin infusion rate (in units per hour) was given at a rate of 35% of the patient's typical total daily insulin dose, divided by 24.
Determination of insulin delivery
In the FMPD algorithm, the gain factors determined the degree to which proportional or derivative errors led to changes in hormone delivery rates. There were separate gain factors for insulin and glucagon. Positive proportional errors (glucose level above target) and positive derivative errors (rising glucose level) called for an increase in the insulin delivery rate. The overall insulin delivery rate was determined by adding the rates called for by the proportional error (IIRpe), the derivative error (IIRde), and the basal insulin rate.
The proportional error gain factor was 1.2 × 10−3 ± 0.078 × 10−3 units/kg per mg/dl/h for glucagon studies and 1.3 × 10−3 units/kg per mg/dl/h for placebo studies. The derivative error gain factor was 2.0 × 10−3 ± 0.096 × 10−3 units/kg per mg/dl for glucagon studies and was 2.0 × 10−3 units/kg per mg/dl for placebo studies. The mean blood glucose target was 110 ± 1 mg/dl for glucagon studies and 110 mg/dl for placebo studies. There were no significant differences between any of these parameters between the groups. For subjects who underwent two closed-loop studies, the algorithm parameters were identical for both.
Insulin on board, the amount of insulin that had been delivered and was assumed to be active, was continually estimated using a model that we derived from data published by Holmes et al. (12
). To minimize hypoglycemia, the insulin infusion was discontinued if the estimated insulin on board reached 15% of the subject's estimated total daily insulin requirement.
Determination of glucagon delivery
The proportional and derivative error gain factors for glucagon were negative, such that negative proportional and derivative errors called for an increase in the glucagon rate. For glucagon, the average weighted proportional error was calculated over a 15 min interval and the average weighted derivative error was calculated over a 10 min interval. There was no basal glucagon infusion rate.
In this project, we tested two closely related algorithms for administering glucagon. Four subjects completed 9-h studies and two subjects completed 28-h studies with low-gain factor settings. In these low-gain glucagon studies, the mean proportional error gain factor was −0.23 ± 0.04 ml/kg per mg/dl/h, the mean derivative error gain factor was −0.06 ± 0.009 ml/kg per mg/dl, and target glucose for glucagon infusion was 108 ± 3 mg/dl. Two subjects completed 9-h studies and five subjects completed 28-h studies with high-gain factor settings. For all of these high-gain glucagon studies, the proportional error gain factor was −2.70 ml/kg per mg/dl/hour, the derivative gain factor was −0.60 ml/kg per mg/dl, and the target glucose for glucagon infusion was 97 ± 1 mg/dl. To avoid overdelivery of glucagon, when total glucagon delivery over the prior 50 min reached a ceiling of 1.0 μg/kg, the algorithm initiated a refractory period for the subsequent 50 min, during which glucagon could not be delivered. Thus, short pulses of glucagon delivery over 5–10 min were followed by the absence of glucagon delivery for 50 min. The insulin rate was reduced by 75% for 40 min after each maximal glucagon pulse.
Patients were given two meals during each 9-h study and four meals during each 28-h study. Each meal was announced to the controller and an open loop premeal bolus was given. Aspart insulin was given 0–10 min before meals, depending on the subject's premeal glucose level. For low-gain glucagon studies, 53.3 ± 7.0% of usual premeal insulin dose was given. The amount of premeal insulin was increased after the first four studies because of a pattern of postprandial hyperglycemia in those studies. For all placebo and high-gain glucagon studies, 75% of the usual premeal insulin dose was given.
Subjects were treated for hypoglycemia if the venous glucose value fell below 70 mg/dl. For glucose levels 60–69 mg/dl, subjects were given 15 g oral carbohydrate, and the treatment repeated as needed every 15 min. For a glucose value <60 mg/dl, 10 g dextrose was given intravenously.
Arterialized venous glucose values, not sensed glucose values, were used to compare hypoglycemia and glucose control between groups. Glucose area under the curve (AUC) was calculated as published elsewhere (13
). Minutes in the hypoglycemic range, defined as glucose <70 mg/dl, hypoglycemic events, treatments for hypoglycemia, units of insulin delivered, and micrograms of glucagon delivered were normalized to 24 h for data from both 9- and 28-h studies. Data are expressed as means ± SE. Sensor accuracy was calculated by comparing sensor glucose to reference glucose values (14
). Comparisons were made using paired or unpaired t
tests, as appropriate. Calculations were performed using Microsoft Excel 2007 (version 12).