Use of the Interoperable Artificial Pancreas System for Type 1 Diabetes Management During Psychological Stress

Automated insulin delivery (AID) systems have not been designed to account for psychological and physiological stress in individuals with type 1 diabetes (T1D). Prior studies have noted psychological stress can cause an increase in catecholamine and cortisol levels and can impair insulin sensitivity,^1,2 with acute stress affecting postprandial glucose levels. ³ We have reported that psychological stress can also cause changes in glucose variability, with potential increases in the number of low blood glucose readings and altered behavior, such as reduced carbohydrate intake. ⁴ We enhanced our zone-model predictive control algorithm running on the Harvard University interoperable Artificial Pancreas System ⁵ using a continuous function of glucose velocity and insulin-on-board to gradually increase insulin infusion under conditions of sustained hyperglycemia, while retaining the previous design to protect from controller-induced hypoglycemia, as a potential improvement for treatment of stress-induced hyperglycemia.

We piloted at-home use of the system in two adults with T1D (64-year-old male, A1c 5.3% and a 31-year-old female, A1c 6.0%) over a two-week period in a randomized crossover comparison to sensor-augmented pump (SAP) (NCT04142229) at Sansum Diabetes Research Institute in California and Mayo Clinic in Minnesota during March and April of 2020. Although neither subject was diagnosed with COVID-19, both subjects encountered multiple psychological stressors related to the pandemic, including job loss, closing of the study clinical centers, cancellation of the planned formal psychological and physiological stress induction sessions, challenges in purchasing food for themselves and their families, and adhering to strict isolation rules. Electrodermal activity (EDA) measurements were collected throughout each day using the Empatica E4 wristband (Empatica Srl, Milano, Italy) as a biomarker of stress.

Both subjects were randomized to the AID arm first. Comparing both arms, subjects had good glycemic control during SAP use, with Subject 1 having 90.3% sensor glucose time-in-range (TIR) 70-180 mg/dL overall and 87.1% overnight, and Subject 2 having 71.2% TIR overall and 71% overnight. This high TIR with SAP use also came at the cost of significant amount of hypoglycemia, in particular for Subject 1, who had 6.7% time <70 mg/dL overall, and 11% time <70 mg/dL overnight. During the two weeks of AID use, for Subject 1, TIR 70-180 mg/dL increased to 92% overall and to 89.4% overnight, with a decrease in time <70 mg/dL to 3.1% overall and to 1% overnight. For Subject 2, TIR 70-180 mg/dL increased to 84.6% overall and to 94% overnight, with a decrease in time <70 mg/dL to 0.5% overall and to 0.1% overnight (Figure 1).

Figure 1.

Paired comparison of 24 hours day and night sensor glucose time in range 70-180 mg/dL during the SAP and AID arms for each two-week period. The solid lines connect individual subjects, with the bubble size proportional to time in hypoglycemia below 70 mg/dL. Time in range improved for both subjects, while at the same time reducing hypoglycemia (time <70 mg/dL).

Continuous decomposition analysis of EDA, ⁶ defining a stress event as a response amplitude of at least 0.5 micro-Siemens, showed a greater number of events per day in the SAP arm for Subject 1 (median [interquartile range {IQR}]: 82.0 [167.0] vs 23.0 [20.0]) as well as for Subject 2 (median [IQR]: 122.5 [62.0] vs 56.0 [121.5]).

These results support the benefit of AID over SAP under stressful conditions. Our goal is to further explore improvements to AID technology to improve handling of stress-induced glucose variability.