Enabling Fully Automated Closed Loop Control in Type 1 Diabetes Through an Artificial Intelligence Meal Detection Algorithm and Pramlintide

点击翻译按钮获取中文摘要

Summary People with type 1 diabetes (T1D) have autoimmune destruction of beta cells resulting in insufficient insulin to maintain normal glucose levels, thereby requiring exogenous insulin treatment, typically delivered subcutaneously either continuously infused through an insulin pump or by injection multiple times throughout the day. There are now commercial closed loop systems available that enable automated delivery of fast- acting insulin in response to wireless continuous glucose monitoring (CGM) sensors that inform an automated control algorithm to calculate and deliver insulin via a pump. Unfortunately, current commercial closed loop systems are hybrid systems that require users to announce meals to the system, and these hybrid systems do not control glucose well following meals. People oftentimes forget to announce their meals or misestimate their carbohydrate intake, thereby leading to poor overall postprandial glucose control. The primary benefit of hybrid closed loop systems has been during the overnight period when meals are not consumed. In a normal working beta cell, the hormone amylin is co-secreted with insulin in response to meals to help suppress glucagon production and delay gastric emptying, thereby reducing postprandial glucose increases. Pramlintide is an analog of the endogenous hormone amylin. Our commercial partner, Adocia, has developed a coformulation of insulin and pramlintide. In this project, we will develop a dual hormone (insulin and pramlintide), fully automated closed loop system with a meal detection algorithm that will enable substantial improvements in postprandial glycemic control for people with T1D while minimizing patient burden by not requiring a meal announcement to the system. Our group has previously developed a multi hormone closed loop system (insulin and glucagon) and we have developed models of insulin, pramlintide, glucagon, and carbohydrate kinetics and dynamics that will be integrated into a new insulin+pramlintide closed loop model predictive control (MPC) system (Aim 1). We have also developed a meal detection algorithm that will be integrated with this MPC control algorithm to dose insulin and pramlintide shortly after a meal is detected, thereby eliminating the need for the user to announce this meal to the system (Aim 1). We will use our in silico simulator of glucose metabolism to identify the optimal dosing amount and timing when insulin and pramlintide are delivered in response to the meal detection algorithm (Aim 1). Next, we will evaluate the optimal insulin and pramlintide dosing therapies identified in Aim 1 and evaluate the top 4 strategies during an in-clinic meal test in a 4-arm randomized crossover study in 14 people with T1D (Study 1a, Aim 2). We will then evaluate the optimal insulin and pramlintide dosing therapy identified in Study 1a and compare with various insulin-only hybrid and automated therapies in an in-clinic randomized crossover study (Study 1b, Aim2). Finally, in Aim 3, we will evaluate the optimal insulin and pramlintide dosing therapy determined in Aim 2 and compare it with the Tandem Diabetes Control-IQ commercial hybrid closed loop system.