Assessing Detection Algorithms for Insulin Pump Malfunctions in Type 1 Diabetes

The goal of this clinical trial is to test the effectiveness of fault-detection algorithms in detecting malfunctioning of the insulin infusion system in an artificial pancreas (also known as Automated Insulin Delivery system) for type 1 diabetes. The main questions it aims to answer is: "Are the proposed algorithms effective in detecting insulin suspension?" Effectiveness accounts for both high sensitivity (i.e. the fraction of suspension correctly detected) and low false alarm rate. The study has three phases: * free-living artificial pancreas data collection, * in-patient induction of hyperglycemia (mimicking an insulin pump malfunction), * retrospective analysis of the collected data to evaluate the effectiveness of the proposed algorithms in detecting insulin suspension.

In individuals with type 1 diabetes, adjusting insulin doses to accommodate the ever-changing conditions of daily life is crucial for achieving satisfactory metabolic control. To address this challenge, researchers have developed an Automated Insulin Delivery (AID) system, commonly known as an artificial pancreas. This system comprises of an insulin pump, a continuous glucose monitoring (CGM) sensor, and a sophisticated control algorithm. The algorithm uses CGM data to calculate the insulin dose required to maintain good glycemic control, and it automatically commands the insulin infusion. However, artificial pancreas systems can experience malfunctions, some of which are highly risky. The most dangerous malfunctions include insulin pump failures and infusion set occlusions, which lead to prolonged interruptions in insulin delivery. This exposes the patient to the risk of hyperglycemia and, even more dangerously, ketoacidosis, a severe complication that can result in hospitalization and, in severe cases, death. Unfortunately, patients do not always notice these issues in a timely manner. This study aims to test new algorithms for detecting pump/infusion set malfunctions that result in reduced or interrupted insulin delivery. The study consists of three phases: * Phase 1: Preliminary Data Collection (Free-living Data) In this phase, data related to glycemic trends and insulin administration in free-living conditions are collected. This data is obtained from a download form the patient's artificial pancreas. The one-month session is designed to gather a substantial amount of patient-specific data to enable the algorithms to learn how insulin and meals impact the patient's glycemia as recorded by the CGM sensor. During this phase, the patient continues to use their artificial pancreas in their daily life. * Phase 2: Induction of Hyperglycemia The second phase involves the patient visiting the clinic, where, according to a specific protocol and a defined schedule, insulin infusion is temporarily suspended to simulate a pump malfunction. The resulting episode of hyperglycemia is closely monitored under medical supervision. At the end of the experiment, the study team assists the patient in restoring euglycemia before returning home. * Phase 3: Retrospective Data Analysis In this phase, the collected data is retrospectively analyzed to evaluate the effectiveness of the proposed algorithms in detecting insulin suspension, simulating a pump malfunction. The sensitivity of the tested methods is assessed as the fraction of insulin suspensions (simulating a malfunction) correctly detected. The uniqueness of this dataset lies in the controlled induction of malfunction, achieved by disconnecting the insulin pump and monitoring the resulting hyperglycemic episode. The presence of malfunctions in this data is certain and precisely characterized in terms of the start time and duration. The dataset resulting from this experimentation will be a valuable tool for the scientific community, enabling the retrospective testing of fault detection algorithms.