Improving Glycemic Management in Patients with Type 1 Diabetes Using a Context-aware Automated Insulin Delivery System

项目摘要 自动胰岛素输送(AID)系统为帮助1型糖尿病患者提供了大量机会 (T1D)改善血糖控制，降低HbA1c。然而，AID仅在 夜间，当用餐、锻炼和压力对艾滋病没有明显的挑战时。此外，低血糖 (&lt；70 mg/dL)在1型糖尿病患者中仍然很常见，即使在 爱滋病的设置，特别是与运动。将情景感知集成到AID中可能会 改善白天的血糖升高时间(70-180毫克/分升)，减少甚至可能消除 低血糖症。上下文信息可以包括推断的食物摄入量、胰岛素剂量、推断的运动类型 和持续时间，以及运动模式。AID可以设计为识别上下文模式，该模式 与进餐血糖反应差和低血糖有关，然后调整胰岛素剂量以应对 这些模式可以提前实现，有助于缓解这些问题。在这笔赠款中，我们将探索如何根据上下文 可以在AID中使用信息来帮助(1)避免低血糖和(2)减少餐后 血糖紊乱。我们将首先进行数据收集研究，从人们那里收集丰富的数据集 对于T1D患者，他们将使用传感器增强泵疗法来管理他们的血糖。数据将从以下地点收集 这30名患者超过28天；数据将包括多变量上下文信息，包括连续 血糖监测(CGM)数据、胰岛素数据、食物数据、体力活动数据(心率和加速度计)、 以及使用新的基于信标的上下文感知来收集的室内/室外上下文移动模式 我们的团队与行业合作伙伴MotioSens合作开发了名为MotioWear的传感系统。 接下来，我们将利用该上下文数据集来构建贝叶斯血糖预测算法。这将是 包括将对彼此相似上下文序列进行分组的聚类算法 导致类似的血糖升高结果。这一上下文感知的血糖预测算法将集成到 适应性、个性化、基于智能手表的情景感知AID(CA-AID)系统。具有以下特征的上下文模式 导致低血糖或餐后血糖紊乱的高可能性将提示胰岛素剂量。 针对将来观察到的类似上下文序列(即CA-AID)调整攻击性因子 对于低血糖可能性高的上下文序列，如有氧运动，会减少胰岛素)。我们 预计将情景感知集成到AID中将显著缩短目标时间 在白天的范围内，将有助于减少低血糖的时间。将对CA-AID进行安全性评估 小规模的先导性研究。然后，我们将在6周内对40名1型成人患者进行CA-AID评估 糖尿病患者接受胰岛素泵治疗。20人将接受CA-AID，而其他20人将获得标准 (非情景感知)AID。这项研究的主要结果衡量标准是在范围内的时间百分比。我们 假设与非上下文感知AID相比，CA-AID将在范围内增加10%的时间。

Project Summary Automated insulin delivery (AID) systems offer substantial opportunities for helping people with type 1 diabetes (T1D) to improve glucose control and lower HbA1c. However, the AID has only shown a benefit during the nighttime when meals, exercise, and stress do not significantly challenge the AID. Furthermore, hypoglycemia (<70 mg/dL) remains a common occurrence in people with type 1 diabetes and continues to occur even in the setting of AID, particularly with exercise. Integrating context awareness into an AID has the potential to improve glycemic time in range (70-180 mg/dL) during the daytime and reduce and possibly eliminate hypoglycemia. Contextual information can include inferred food intake, insulin dosing, inferred exercise type and duration, as well as movement patterns. An AID can be designed to recognize contextual patterns that relate to poor glycemic responses to meals and hypoglycemia and then adjust insulin dosing in response to these patterns in advance and help mitigate these problems. In this grant, we will explore how contextual information may be used within an AID to help (1) avoid hypoglycemia and (2) reduce postprandial dysglycemia. We will first conduct a data gathering study whereby we will collect a rich data set from people with T1D who will use sensor augmented pump therapy to manage their glucose. Data will be collected from these 30 patients over 28 days; data will include multivariable contextual information including continuous glucose monitoring (CGM) data, insulin data, food data, physical activity data (heart rate and accelerometry), as well as indoor/outdoor contextual movement patterns gathered using a novel beacon-based context-aware sensing system called MotioWear developed by our group in collaboration with our industry partner MotioSens. Next, we will utilize this contextual data set to construct a Bayesian glucose prediction algorithm. This will include a clustering algorithm that will group contextual sequences that are similar with each other and which lead to similar glycemic outcomes. This context-aware glucose prediction algorithm will be integrated into an adaptive, personalized, smartwatch-based context-aware AID (CA-AID) system. Contextual patterns that have a high likelihood of leading to hypoglycemia or postprandial dysglycemia will inform an insulin dosing aggressiveness factor to be adjusted for similar contextual sequences observed in the future (i.e. the CA-AID will reduce insulin for contextual sequences with high likelihood of hypoglycemia such as aerobic exercise). We expect that integrating context awareness into an AID will lead to significant improvements in time in target range during the day and will help reduce time in hypoglycemia. The CA-AID will be evaluated for safety in a small pilot study. We will then evaluate the CA-AID within a 6 week clinical study in 40 adults with type 1 diabetes on insulin pump therapy. Twenty will receive the CA-AID while the other 20 will receive a standard (non-context-aware) AID. The primary outcome measures of this study is the percent time in range. We hypothesize that the CA-AID will increase time in range by 10% as compared with a non-context aware AID.