Automated Insulin Delivery with Remote Real-Time Continuous Glucose Monitoring for Hospitalized Patients with Diabetes: A Multicenter, Single-Arm, Feasibility Trial

Abstract

Introduction: Multiple daily injection insulin therapy frequently fails to meet hospital glycemic goals and is prone to hypoglycemia. Automated insulin delivery (AID) with remote glucose monitoring offers a solution to these shortcomings. Research Design and Methods: In a single-arm multicenter pilot trial, we tested the feasibility, safety, and effectiveness of the Omnipod 5 AID System with real-time continuous glucose monitoring (CGM) for up to 10 days in hospitalized patients with insulin-requiring diabetes on nonintensive care unit medical-surgical units. Primary endpoints included the proportion of time in automated mode and percent time-in-range (TIR 70-180 mg/dL) among participants with >48 h of CGM data. Safety endpoints included incidence of severe hypoglycemia and diabetes-related ketoacidosis (DKA). Additional glycemic endpoints, CGM accuracy, and patient satisfaction were also explored. Results: Twenty-two participants were enrolled; 18 used the system for a total of 96 days (mean 5.3 ± 3.1 days per patient), and 16 had sufficient CGM data required for analysis. Median percent time in automated mode was 95% (interquartile range 92%-98%) for the 18 system users, and the 16 participants with >48 h of CGM data achieved an overall TIR of 68% ± 16%, with 0.17% ± 0.3% time <70 mg/dL and 0.06% ± 0.2% time <54 mg/dL. Sensor mean glucose was 167 ± 21 mg/dL. There were no DKA or severe hypoglycemic events. All participants reported satisfaction with the system at study end. Conclusions: The use of AID with a disposable tubeless patch-pump along with remote real-time CGM is feasible in the hospital setting. These results warrant further investigation in randomized trials.

Keywords: Automated insulin delivery; Hybrid closed-loop; Inpatient diabetes.

FIG. 1.

Automated insulin delivery system set up in the hospital for the AIDING study. (A) The devices all communicated via Bluetooth in the patient room, primarily managed by the bedside nurse. However, we did test the ability to manage interaction with the system using the phone-based Controller from outside the rooms. (B) The data wirelessly streams to a secure cloud via cell phone signal. (C) The cloud connected wirelessly to the Dexcom Follow application on an iPad at the nursing station, which provided real-time glucose telemetry at the nursing unit. This was programmed to alarm for any glucose drop below 80 mg/dL (4.4 mmol/L) and for prolonged glucose level above 300 mg/dL (16.7 mmol/L) for more than 2 h. CGM values below 80 mg/dL were assessed using finger-stick CBG and hypoglycemia was treated according to each institutional protocol, which was indicated for glucoses <70 mg/dL (3.9 mmol/L). Guidance for interventions to prevent hypoglycemia (e.g., small amount of juice for glucose just above lower limit but accompanied by a downward trend arrow) was neither provided by the protocol nor was it prohibited. Prolonged hyperglycemia necessitated reaching out to study for advice on correction bolusing, system troubleshooting, or setting adjustments, as well as increased frequency of CBG assessments until glucose dropped back below 300 mg/dL. (D). Data were also available continuously to the study via the cloud in the Dexcom Follow application, the Dexcom Clarity application, and Insulet's investigational data management software. a. The figure depicts CGM connection to the Controller because this was used on the study devices; however, it should be noted that the commercial release of the product does not connect the Controller (instead it connects to the patient's personal cell phone or CGM receiver). AIDING, Automated Insulin Delivery for INpatients with DysGlycemia; CBG, capillary blood glucose; CGM, continuous glucose monitoring.

FIG. 2.

Study flow diagram. Enrollment was performed in two phases. In phase 1, finger-stick CBG was assessed and compared with simultaneous CGM value (using validation process described in the Research Design and Methods section) six times daily, before breakfast, before lunch, before dinner, before bedtime, and two times overnight. After data safety monitoring board review, the study moved into phase 2, which reduced the daily number of CBG checks and validations to four by removing the two overnight assessments. In each phase, 11 patients were enrolled, 1 patient was discharged within 48 h of enrollment, and 1 patient was withdrawn from the study. In phase 1, one participant was withdrawn due to initiation of high-dose (≥4 g daily) acetaminophen. Nine participants from each phase completed the study, one of whom from each phase did not have the prespecified 48 h of CGM data required for inclusion in glycemia analysis. Therefore, 16 participants in total were included in the final glycemia analysis.

FIG. 3.

Average participant-aggregated glycemic control by day of study enrollment. Values are expressed as the average of the proportions of time each participant spent in each range on the specified day of enrollment. Each “day” represents a full 24-h period, and day 1 starts from the time of completion of initial continuous glucose monitor warm-up (i.e., days are not midnight-midnight). Day 8 includes data from one participant who had a sensor that was erroneously low, ranging 75–175 mg/dL below finger-stick CBG level on two consecutive scheduled checks. “Initial pod” and “adaptation” are behind the bars and highlight a specific action of the Omnipod 5 system initial start-up, in which the algorithmic basal dosing is approximately half as aggressive with the initial Pod. The algorithm subsequently increases its automated basal calculations with the first Pod change that occurs after 48 h on the system. Correspondingly, the bars in the figure show a notable improvement in glycemic control after this period.

FIG. 4.

Hourly overview of glycemic control and insulin delivery with AID. (A) This panel illustrates the compiled 24-h glucose profiles from 16 selected patient hospitalizations that were eligible for glycemic analysis. The median glucose level is represented by a solid line, while the interquartile range is highlighted by a darker shaded region. The 5–95 percentile range is shown as a lightly shaded area. For contextual reference, the target glucose range (70–180 mg/dL; 3.9–10 mmol/L) is displayed within a light green band. (B) This panel visualizes the aggregated hourly data on basal insulin (blue) and total insulin (orange) administration from the same 16 patient hospitalizations over a 24-h period. Median insulin delivery rates are indicated by solid lines, with the interquartile range represented by darker shaded regions. The 5–95 percentile range is outlined by the lighter shaded areas. AID, automated insulin delivery.