OBJECTIVE Nocturnal hypoglycemia can cause seizures and is a major impediment to tight glycemic control, especially in young children with type 1 diabetes. We conducted an in-home randomized trial to assess the efficacy and safety of a continuous glucose monitor–based overnight predictive low-glucose suspend (PLGS) system. RESEARCH DESIGN AND METHODS In two age-groups of children with type 1 diabetes (11–14 and 4–10 years of age), a 42-night trial for each child was conducted wherein each night was assigned randomly to either having the PLGS system active (intervention night) or inactive (control night). The primary outcome was percent time <70 mg/dL overnight. RESULTS Median time at <70 mg/dL was reduced by 54% from 10.1% on control nights to 4.6% on intervention nights (P < 0.001) in 11–14-year-olds (n = 45) and by 50% from 6.2% to 3.1% (P < 0.001) in 4–10-year-olds (n = 36). Mean overnight glucose was lower on control versus intervention nights in both age-groups (144 ± 18 vs. 152 ± 19 mg/dL [P < 0.001] and 153 ± 14 vs. 160 ± 16 mg/dL [P = 0.004], respectively). Mean morning blood glucose was 159 ± 29 vs. 176 ± 28 mg/dL (P < 0.001) in the 11–14-year-olds and 154 ± 25 vs. 158 ± 22 mg/dL (P = 0.11) in the 4–10-year-olds, respectively. No differences were found between intervention and control in either age-group in morning blood ketosis. CONCLUSIONS In 4–14-year-olds, use of a nocturnal PLGS system can substantially reduce overnight hypoglycemia without an increase in morning ketosis, although overnight mean glucose is slightly higher.

In the print version of the article cited above, the clinical trial registry number should have been reported as NCT01823341, ClinicalTrials.gov, rather than NCT01591681. The online version reflects this change. Bruce A. Buckingham, Dan Raghinaru, Fraser Cameron, B. Wayne Bequette, H. Peter Chase, David M. Maahs, Robert Slover, R. Paul Wadwa, Darrell M. Wilson, Trang Ly, Tandy Aye, Irene Hramiak, Cheril Clarson, Robert Stein, Patricia H. Gallego, John Lum, Judy Sibayan, Craig Kollman, and Roy W. Beck, for the In Home Closed Loop Study Group Diabetes Care Volume 38, September 2015 1813

In this paper, we briefly examine the recent developments in artificial pancreas controllers, that automate the delivery of insulin to patients with type-1 diabetes. We argue the need for offline and online runtime verification for these devices, and discuss challenges that make verification hard. Next, we examine a promising simulation-based falsification approach based on robustness semantics of temporal logics. These ideas are implemented in the tool S-Taliro that automatically searches for violations of metric temporal logic (MTL) requirements for Simulink(tm)/Stateflow(tm) models. We illustrate the use of S-Taliro for finding interesting property violations in a PID-based hybrid closed loop control system.

Individuals with type 1 Diabetes Mellitus (T1DM) must inject insulin to regulate blood glucose concentrations. The artificial pancreas project seeks to automate the delivery of insulin in response to continuous glucose monitor (sensor) signals. Most medical devices must go through extensive animal studies before human studies can be conducted, but regulatory authorities (FDA, in the United States) have allowed investigators to skip animal trials for the artificial pancreas project by conducting exhaustive simulation-based (in silico) clinical trials. Still, current simulators only provide a rough evaluation of prospective algorithms because they cannot accurately model all physiologic processes. In this paper we propose an alternative simulation approach that works directly from clinical data, reducing the number of required assumptions. This approach calculates changes to real data based on changes in the inputs rather than calculating the effect of the entire input. Here, we choose a simple, linear, insulin-glucose model based on published insulin-glucose test data. The simulator with the linear insulin glucose model is validated against an FDA-accepted simulator for various magnitudes of modified insulin dosing. The scenarios include a 0-200% step change to the patients' background insulin delivery rates (basal rates), and a 50-150% scaling of their meal insulin doses (boluses). These studies show that the differential simulator induces less error than patient variability when using other simulators. We also show two illustrative test cases for testing revision to artificial pancreas controllers.

Background: Aerobic exercise can lower blood glucose levels and alter insulin sensitivity both during and several hours after exercise, creating challenges for a closed-loop artificial pancreas. Predictive low glucose suspend (PLGS) algorithms are a first step toward an artificial pancreas, but few of these have been successfully applied to exercise. This study incorporates physical activity measurements from a combined accelerometer/heart rate monitor (HRM) to improve the performance of an existing PLGS algorithm at mitigating exercise-associated hypoglycemia in participants with type 1 diabetes. Methods: In all, 22 subjects with type 1 diabetes on insulin pump therapy were provided a combined accelerometer/HRM and (if not already using one) a continuous glucose monitor (CGM), then instructed to go about their everyday lives while wearing the devices. After the monitoring period, each subject’s insulin pump, CGM, and accelerometer/HRM were downloaded and the data were used to augment an existing PLGS algorithm to incorporate activity. Using a computer simulator, the accelerometer-augmented algorithm was compared to the HRM-augmented algorithm to determine which was most effective at mitigating hypoglycemia. Results: Mean length of monitoring was 4.9 days. Across all subjects, 11 061 CGM readings were recorded during the monitoring period. In the simulator analysis, the PLGS algorithm reduced hypoglycemia by 62%, compared to 71% and 74% reductions for the HRM-augmented and accelerometer-augmented algorithms, respectively; combined accelerometer and HRM augmentation provided a 76% reduction. Conclusions: In a simulated setting, the accelerometer-augmented pump suspension algorithm decreases the incidence of exercise-related hypoglycemia by a meaningful amount compared to the PLGS algorithm alone. Results also failed to justify the additional user burden of a HRM.

BACKGROUND:Closed-loop control of blood glucose levels in people with type 1 diabetes offers the potential to reduce the incidence of diabetes complications and reduce the patients' burden, particularly if meals do not need to be announced. We therefore tested a closed-loop algorithm that does not require meal announcement.MATERIALS AND METHODS:A multiple model probabilistic predictive controller (MMPPC) was assessed on four patients, revised to improve performance, and then assessed on six additional patients. Each inpatient admission lasted for 32 h with five unannounced meals containing approximately 1 g/kg of carbohydrate per admission. The system used an Abbott Diabetes Care (Alameda, CA) Navigator(®) continuous glucose monitor (CGM) and Insulet (Bedford, MA) Omnipod(®) insulin pump, with the MMPPC implemented through the artificial pancreas system platform. The controller was initialized only with the patient's total daily dose and daily basal pattern.RESULTS:On a 24-h basis, the first cohort had mean reference and CGM readings of 179 and 167 mg/dL, respectively, with 53% and 62%, respectively, of readings between 70 and 180 mg/dL and four treatments for glucose values <70 mg/dL. The second cohort had mean reference and CGM readings of 161 and 142 mg/dL, respectively, with 63% and 78%, respectively, of the time spent euglycemic. There was one controller-induced hypoglycemic episode. For the 30 unannounced meals in the second cohort, the mean reference and CGM premeal, postmeal maximum, and 3-h postmeal values were 139 and 132, 223 and 208, and 168 and 156 mg/dL, respectively.CONCLUSIONS:The MMPPC, tested in-clinic against repeated, large, unannounced meals, maintained reasonable glycemic control with a mean blood glucose level that would equate to a mean glycated hemoglobin value of 7.2%, with only one controller-induced hypoglycemic event occurring in the second cohort.

The primary goal of a low glucose suspend system is to reduce the risk of overnight hypoglycemia (low blood glucose) in individuals with type 1 diabetes by reducing/suspending insulin infusion. We have developed a Kalman filter-based algorithm, combined with a number of safety rules, to implement a predictive low glucose suspend system that shuts off an insulin pump based on a prediction of hypoglycemia 30-70 minutes in the future. This system has been studied in over 2,000 nights in an outpatient-home environment. In this paper, based on an analysis of this data, we isolate the effects of the individual rules in part by simulating their removal from the existing data. Specifically, we decompose the basal insulin into small boluses and, using a model of insulin pharmacodynamic action (the time effect of insulin on blood glucose), alter the real data corresponding to the addition or removal of basal insulin via simulation. Our results show that limiting the total suspension to 180 minutes per night prevents excessive suspension in cases where the average calibration is an excessive 58 mg/dl, above the mean of 18 mg/dl. Further, we also show that a simple threshold algorithm that suspends below 100 mg/dl if the glucose level is flat or falling, is comparable in performance. Lastly, we show that the Kalman filter at the heart of this algorithm reduces the time spent below 70 mg/dl by 50% at the expense of a mean rise of 12 mg/dl in morning glucose levels.

Background: Exercise-associated hypoglycemia is a common adverse event in people with type 1 diabetes. Previous in silico testing by our group demonstrated superior exercise-associated hypoglycemia mitigation when a predictive low glucose suspend (PLGS) algorithm was augmented to incorporate activity data. The current study investigates the effectiveness of an accelerometer-augmented PLGS algorithm in an outpatient exercise protocol. Methods: Subjects with type 1 diabetes on insulin pump therapy participated in two structured soccer sessions, one utilizing the algorithm and the other using the subject’s regular basal insulin rate. Each subject wore their own insulin pump and a Dexcom G4™ Platinum continuous glucose monitor (CGM); subjects on-algorithm also wore a Zephyr BioHarness™ 3 accelerometer. The algorithm utilized a Kalman filter with a 30-minute prediction horizon. Activity and CGM readings were manually entered into a spreadsheet and at five-minute intervals, the algorithm indicated whether the basal insulin infusion should be on or suspended; any changes were then implemented by study staff. The rate of hypoglycemia during and after exercise (until the following morning) was compared between groups. Results: Eighteen subjects (mean age 13.4 ± 3.7 years) participated in two separate sessions 7-22 days apart. The difference in meter blood glucose levels between groups at each rest period did not achieve statistical significance at any time point. Hypoglycemia during the session was recorded in three on-algorithm subjects, compared to six off-algorithm subjects. In the postexercise monitoring period, hypoglycemia occurred in two subjects who were on-algorithm during the session and four subjects who were off-algorithm. Conclusions: The accelerometer-augmented algorithm failed to prevent exercise-associated hypoglycemia compared to subjects on their usual basal rates. A larger sample size may have achieved statistical significance. Further research involving an automated system, a larger sample size, and an algorithm design that favors longer periods of pump suspension is necessary.

Background: Continuous glucose monitors (CGMs) provide real-time interstitial glucose concentrations that are essential for automated treatment of individuals with type 1 diabetes. Miscalibration, noise spikes, dropouts, or pressure applied to the site (e.g., lying on the site while sleeping) can cause inaccurate glucose signals, which could lead to inappropriate insulin dosing decisions. These studies focus on the problem of pressure-induced sensor attenuations (PISAs) that occur overnight and can cause undesirable pump shut-offs in a predictive low glucose suspend system. Methods: The algorithm presented here uses real-time CGM readings without knowledge of meals, insulin doses, activity, sensor recalibrations, or fingerstick measurements. The real-time PISA detection technique was tested on outpatient “in-home” data from a predictive low-glucose suspend trial with over 1125 nights of data. A total of 178 sets were created by using different parameters for the PISA detection algorithm to illustrate its range of available performance. Results: The tracings were reviewed via a web-based analysis tool by an engineer with an extensive expertise on analyzing clinical datasets and ~3% of the CGM readings were marked as PISA events which were used as the gold standard. It is shown that 88.34% of the PISAs were successfully detected by the algorithm, and the percentage of false detections could be reduced to 1.70% by altering the algorithm parameters. Conclusions: Use of the proposed PISA detection method can result in a significant decrease in undesirable pump suspensions overnight, and may lead to lower overnight mean glucose levels while still achieving a low risk of hypoglycemia.

Continuous subcutaneous insulin infusion pumps and continuous glucose monitors enable individuals with type 1 diabetes to achieve tighter blood glucose control, and are critical components in a closed-loop artificial pancreas. Insulin infusion sets can fail and CGM sensor signals can suffer from a variety of anomalies. In this paper algorithms are developed to detect infusion set failures and sensor signal anomalies; both in-patient and out-patient studies are presented. A threshold-based method, based on high glucose concentrations, is shown to be adequate to detect infusion set failures. Pressure-induced sensor attenuation (PISA), which can occur when a subject rolls over and puts pressure on their sensor, is a particularly challenging problem. An algorithm based on non-physiological rates-of-change, coupled with a maximum attenuation time window, is developed to detect and compensate for PISAs. These algorithms can be used either in advisory mode for current open-loop technology, as well as an additional safety/fault detection layer as part of a fully closed-loop artificial pancreas.

OBJECTIVENocturnal hypoglycemia is a common problem with type 1 diabetes. In the home setting, we conducted a pilot study to evaluate the safety of a system consisting of an insulin pump and continuous glucose monitor communicating wirelessly with a bedside computer running an algorithm that temporarily suspends insulin delivery when hypoglycemia is predicted.RESEARCH DESIGN AND METHODSAfter the run-in phase, a 21-night randomized trial was conducted in which each night was randomly assigned 2:1 to have either the predictive low-glucose suspend (PLGS) system active (intervention night) or inactive (control night). Three predictive algorithm versions were studied sequentially during the study for a total of 252 intervention and 123 control nights. The trial included 19 participants 18-56 years old with type 1 diabetes (hemoglobin A1c level of 6.0-7.7%) who were current users of the MiniMed Paradigm® REAL-Time Revel™ System and Sof-sensor® glucose sensor (Medtronic Diabetes, Northridge, CA).RESULTSWith the final algorithm, pump suspension occurred on 53% of 77 intervention nights. Mean morning glucose level was 144±48 mg/dL on the 77 intervention nights versus 133±57 mg/dL on the 37 control nights, with morning blood ketones >0.6 mmol/L following one intervention night. Overnight hypoglycemia was lower on intervention than control nights, with at least one value ≤70 mg/dL occurring on 16% versus 30% of nights, respectively, with the final algorithm.CONCLUSIONSThis study demonstrated that the PLGS system in the home setting is safe and feasible. The preliminary efficacy data appear promising with the final algorithm reducing nocturnal hypoglycemia by almost 50%.

A habituating multiple model predictive controller (MMPC) is developed by combining the habituating model predictive control law and multiple-model Kalman filter predictor. The habituating control is applied to blood glucose regulation in the intensive care unit, and includes both IV glucose input and IV insulin infusion in order to provide nutrition supply and improve disturbance rejection. The performance of Habituating MMPC is compared with Habituating MPC, SISO MMPC and SISO MPC on their ability to regulate BG for the fifteen in silico patients and to reject insulin sensitivity variation. The simulation results indicate that the Habituating MMPC strategy outperforms the other three control strategies by providing the tightest glucose control for a patient population, and producing the least amount of glucose variability while rejecting disturbances in insulin sensitivity.

In the intensive care unit patients benefit from being fed and from having well controlled glucose levels. Insulin and glucose infusion serves as manipulated inputs to regulate blood glucose, while glucose infusion serves as a sole nutritional input. In this paper, a model predictive control strategy, based on simultaneously manipulating glucose and insulin infusion, is developed to improve blood glucose regulation in intensive care unit patients. In the short term, glucose infusion is used for tighter glucose control, particularly for disturbance rejection, while, in the long-term (24h period), glucose infusion is used to meet nutritional needs. The “habituating control” algorithm is proposed and tested against a model predictive control (MPC) strategy that only manipulates insulin. The simulation results indicate that the Habituating MPC strategy outperforms the single input–single output MPC by providing faster setpoint tracking and tighter glucose control for a patient population, and producing less glucose variability while rejecting disturbances in insulin infusion and insulin sensitivity.

BACKGROUND:An insulin pump shutoff system can prevent nocturnal hypoglycemia and is a first step on the pathway toward a closed-loop artificial pancreas. In previous pump shutoff studies using a voting algorithm and a 1 min continuous glucose monitor (CGM), 80% of induced hypoglycemic events were prevented.METHODS:The pump shutoff algorithm used in previous studies was revised to a single Kalman filter to reduce complexity, incorporate CGMs with different sample times, handle sensor signal dropouts, and enforce safety constraints on the allowable pump shutoff time.RESULTS:Retrospective testing of the new algorithm on previous clinical data sets indicated that, for the four cases where the previous algorithm failed (minimum reference glucose less than 60 mg/dl), the mean suspension start time was 30 min earlier than the previous algorithm. Inpatient studies of the new algorithm have been conducted on 16 subjects. The algorithm prevented hypoglycemia in 73% of subjects. Suspension-induced hyperglycemia is not assessed, because this study forced excessive basal insulin infusion rates.CONCLUSIONS:The new algorithm functioned well and is flexible enough to handle variable sensor sample times and sensor dropouts. It also provides a framework for handling sensor signal attenuations, which can be challenging, particularly when they occur overnight.

By construction, model predictive control (MPC) relies heavily on predictive capabilities. Good control simultaneously requires predictions that provide consistent, strong filtering of sensor noise, as well as fast adaptation for disturbances. For example, controllers seeking to regulate the blood glucose levels in persons with Type 1 Diabetes should filter noise in the continuous glucose monitor (CGM) readings, while also adapting instantly to meals that trigger an extended upsurge in those same readings. One way to do this is to switch between multiple models with distinct dynamics. When the data suggest that there is a disturbance then the relevant model is given more influence on the predictions. When there is no evidence of the disturbance the non-disturbance model is given precedence. To reduce the effect of sensor noise we include prior information about the likely timing of the meal disturbances. Specifically, we model the system as making discrete transitions to new disturbances, allowing us to include the prior information as the prior probability of those transitions. Since each transition engenders a new disturbance case, we present a method to combine the cases that minimizes error and computational load. Here we develop a set of prior probabilities for meals that encode knowledge of the time of day, the timing of the last meal, sleep announcement, and meal announcement. We use this to detect and estimate current or past meals as well as anticipating future meals. Additionally, since this application can have asymmetric actuation and costs, violating the certainty equivalence principle, we also provide estimates of the prediction uncertainty. This method reduces 2h prediction error by 45% relative to an algorithm without meal detection and 18% relative to one with meal detection. For 3h prediction these improvements jump to 66% and 30% respectively. This algorithm improves the accuracy of prediction uncertainty estimates.

Automatic regulation of blood glucose in patients with Type 1 Diabetes is challenged by unknown or unannounced food consumption. Yet we know most food is ingested in discrete meals, spaced by brief periods of fasting. Treating meals as discrete events, this paper proposes a set of prior probabilities relating distinct meals independently of daily patterns. Using these prior probabilities, closed-loop blood glucose control can attain anticipatory behavior, implicitly lowering glucose levels when patients should be hungry and meals are most likely to occur. This improves overall performance and individual meal responses. The benefits occur because blood glucose prediction, which anticipates future food intake, can achieve near zero mean errors even over a long prediction horizon. As a side effect, such predictors can easily be tuned or tested against unrestricted out-patient data which, by definition, contains unknown meals. We validate our approach in both prediction and control of blood glucose levels. We see improved prediction accuracy over 1-4 hour horizons and a significant reduction in the blood glucose risk index for simulated closed-loop control.