NCT05876273

Neural-net Artificial Pancreas (NAP)

Adaptive Motif-Based Control (AMBC): Pilot 1 - Neural Net Implementation of Automated Insulin Delivery

• Status: Completed
• Phase: N/A
• Study type: Interventional
• Enrollment: 15 (actual)
• Sponsor: University of Virginia · Academic / Other
• Sex: All
• Age: 18 Years
• Healthy volunteers: Not accepted

Summary

This study is intended to assess a Neural-net Artificial Pancreas (NAP) implementation of an established AP controller - the University of Virginia Model Predictive Control Algorithm (UMPC). The health outcomes achieved on NAP will be compared to the health outcomes achieved on UMPC in a randomized crossover design. The investigators will consent up to 20 participants, ages ≥18.0, with a goal of completing 15 participants.

Detailed description

The study will follow a randomized cross-over design assessing glycemic control on a Neural-net Artificial Pancreas (NAP), compared to the previously tested University of Virginia Model Predictive Control (UMPC) algorithm, in a supervised hotel setting: The study will involve Tandem t:slim X2 Control-IQ (CIQ) users who will continue to use their CIQ systems, except during the hotel sessions, which will use the DiAs prototyping platform, connected to a Tandem t:AP research pump and a Dexcom G6 sensor, and implementing NAP or UMPC. The study sensor will be the same sensor used by CIQ - it will be disconnected from CIQ and connected to DiAs. Following enrollment, one week of automated insulin delivery (AID) data will be downloaded from the participants' pumps or t:connect accounts and will be used to establish a baseline and initialize the control algorithms. Participants will be then studied at a local hotel for 20 hours, including an 18-hour experiment, randomly receiving either NAP or UMPC. Participants will then receive the opposite intervention either sequentially during the same hotel stay, or in a second hotel stay up to 28 days following the first hotel stay. During these 18-hour hotel sessions participants will be followed to compare blood glucose control on NAP vs. UMPC. The study meals and activities will be kept the same between study sessions. The investigators will analyze non-inferiority of NAP compared to UMPC, but this pilot feasibility study is not powered to formally test noninferiority. The primary outcome is percent time in range (TIR) (70 to 180 mg/dL) on NAP vs UMPC. Secondary outcomes include frequency of hypoglycemia (time below range = TBR) and hyperglycemia (time above range = TAR), as well as other safety and control metrics.

Conditions

• Type1 Diabetes

Interventions

Type	Name	Description
DEVICE	Neural-net Artificial Pancreas	NAP is a neural-net implementation of the previously tested UMPC algorithm (below).
DEVICE	University of Virginia Model Predictive Control	A previously tested artificial pancreas control algorithm, based on a differential-equation model of the human metabolic system in diabetes.

Timeline

• Start date: 2023-05-30
• Primary completion: 2023-09-08
• Completion: 2023-09-10
• First posted: 2023-05-25
• Last updated: 2024-07-31
• Results posted: 2024-07-31

Locations

1 site across 1 country: United States

Regulatory

• FDA-regulated device study