• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

Researchers Propose Model to Replicate Patient Experience in Artificial Pancreas System

Article

A methodology to monitor patient-in-the-loop modes and faults was examined using a hybrid automata model that was able to replicate patient’s real-life operation modes.

A study examining automated insulin delivery systems found that a system built around an hybrid automata (HA) model was able to detect mode and “patient-in-the-loop” faults in real time, according to findings published in Sensors (Basel).

Automated insulin delivery systems require that their users, namely patients with type 1 diabetes (T1D), are in the control loop by providing information that affects the performance and safety of the system. The current paper discusses the development of fault monitoring strategies to ensure safety and performance and diagnose potential patient modes and faults.

When compared to open-loop therapy, research has found that artificial pancreas (AP), a closed-loop (CL) system in which insulin is delivered automatically through a pump at an insulin infusion rate dependent on continuous glucose monitor (CGM) readings, is safe, robust, and efficacious for patients with T1D. Users of APs must provide inputs, maintain the system, and undertake any training before using the technology, as they are tasked with wearing and operating them.

The patient-in-the-loop design phase for AP technology makes the patient an integral part of its operation, therefore any developed AP must establish fault tolerant control (FTC) strategies to reduce the impact of patient behaviors and decisions and maintain stability, performance, and safety for the user. The current study discusses the implementation of a model-based fault detection (FD) system compatible with any kind of hybrid or fully CL control architecture that announces meals and/or/uses insulin and/or CHO as a control action.

The hybrid AP system used in this work allows the patient to take on fundamental roles, whether it be a sensor role when announcing meals and exercise or an actuator role when the CHO controller suggests rescue CHO. Investigators determined patient-in-the-loop faults to be poorly estimated meal boluses, when patients do not follow the CHO controller recommendations, or when the dynamical plant is behaving unexpectedly.

The normal operation for the system includes 3 automata modes, which include input patient information about meals and exercise, and events triggered by the consistency of residuals. Faulty modes account for structural faults, such as not eating rescue CHO or injecting inappropriate boluses for meals. By using an HA, investigators aimed to mimic real patient operational modes and transitions.

An online diagnoser was involved with the system and responsible for detecting mode changes within the HA. Patient-in-the-loop faults were identified when a triggered transition led to one of the faulty modes.

If several signatures have the same binary combination and are not isolable, using a binary transition and fault detection system might not be enough to elicit detection. Transitions may be supplemented with external information like input events, the sign of the residuals and by the value and trend of the measured variables.

Investigators used a reduced version of the nonlinear Hovorka model in their study, a compartmental model containing 4 main sub-systems: carbohydrate absorption, subcutaneous insulin absorption, insulin action on glucose uptake and removal, and the BG dynamics. The model is individualized by the patient’s weight, but the insulin sensitivity parameters are adapted based on the fasting patient glucose and basal infusion rate.

A bank of interval zonotopic observers, including a meal observer, resting and fasting observer, altered insulin sensitivity observer, and rescue carbohydrates observer, helped generate appropriate residual signals. Once residuals were generated for each observer, they were fed to the signature analysis module.

Across 4 complete simulations, 480 meals were consumed and 20 exercise sessions were performed, and faults were introduced in a total of 160 meals, with over- and underestimations of ±60% and feed-foward rescue CHO were also simulated with patient-in-the-loop faults on 10 occasions. Meals were correctly detected in 461 meals out of the total 480 meals, yielding a high sensitivity of 96.0%. All exercise sessions were correctly detected.

The information in this study could be used as a tool to reconfigure CL controllers, monitor the system continuously and classify patient-in-the-loop behaviors, authors concluded.

Reference

Beneyto A, Puig V, Bequette BW, Vehi J. A hybrid automata approach for monitoring the patient in the loop in artificial pancreas systems. Sensors (Basel). 2021;21(21):7117. doi:10.3390/s21217117

Related Videos
dr jennifer green
dr jennifer green
dr ken cohen
dr ian neeland
Yael Mauer, MD, MPH
Pregnant Patient | image credit: pressmaster - stock.adobe.com
Diana Isaacs, PharmD
Beau Raymond, MD
Robert Zimmerman, MD
Beau Raymond, MD
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.