EAGER: Insulin Delivery for Diabetes Management in the Intensive Care Unit as a Feedback Control Problem

糖尿病影响了超过30%的住院患者。目前，在医院监测血糖的管理方案过于简单，往往无效。一个由一名糖尿病学家/内分泌学家和一名控制工程师组成的跨学科团队，旨在改善重症监护患者的血糖水平调节，特别是那些术后患者。目前的胰岛素剂量方案实际上是针对血糖水平的反馈控制算法，但很少有研究对其进行分析。本建议的中心主题是，反馈控制技术可以导致更好的执行和更安全的胰岛素给药方案。pi将建立并验证血糖对静脉胰岛素输注率反应的动态模型，该模型适用于ICU中不同类别的患者。该模型将以患者数据为基础，并将考虑重要的干扰因素，如葡萄糖输注速率、营养（口服和肠外）、胰岛素抵抗或致敏药物、患者体温和进入ICU时的临床严重程度。他们将使用获得的动态模型在闭环中对现有胰岛素输注方案进行广泛的控制分析，重点研究它们的名义响应以及它们对大范围模型参数变化的响应。在上述模型建立和闭环控制分析的基础上，pi将提出改进现有协议的建议，包括世界各地医院所采用的所谓书面协议，这些协议反映了更优的控制法。智力优势：目前还没有公布的各种方案对达到目标的速度和一致性或低血糖频率等因素的比较。目前也没有系统研究现有方案对患者营养（尤其是碳水化合物）、血管加压剂、类固醇和其他药物变化的稳健性。已公布的方案声称，46%至69%的血糖水平达到了目标。在一些研究中，低血糖率可以从0.3%到15%不等。如果一个方案能够迅速使血糖水平达到目标，对胰岛素抵抗/敏感性的变化做出适当和及时的反应，并解决饮食和药物制度问题，将对患者护理产生重大影响。通过自动记录胰岛素敏感性的变化来预防低血糖，可以消除对低血糖水平的恐惧。对于无反应的镇静患者，这将有助于减少对低血糖的担忧，而低血糖是使用胰岛素的最大障碍。最后，除了解决上述所有问题外，理想的方案还应允许医疗保健提供者为不同ICU环境（内科、外科、心脏或创伤烧伤）的患者滴定不同的血糖范围和敏感性，通过膳食或输注含有葡萄糖的方法减少血糖波动，并防止低血糖。这里提出的工作将使我们朝着这个目标迈出重要的一步。更广泛的影响：本研究制定的一般原则应适用于糖尿病的门诊管理，其范围是巨大的和各种其他设置，包括麻醉和疼痛管理。

0938288-GrizzleDiabetes impacts more than 30% of hospitalized patients. Currently, management protocols for monitoring blood glucose in the hospital setting are overly simplistic and often ineffective. An interdisciplinary team consisting of a diabetologist/endocrinologist and a control engineer seeks to improve the regulation of blood glucose levels in patients in the intensive care setting, especially those who are there postoperatively. The current insulin dosing protocols are actually feedback control algorithms for blood glucose levels, but little work has been done to analyze them as such. The central theme of this proposal is that feedback control techniques can lead to better performing and safer insulin dosing protocols. The PIs will build and validate a dynamic model of blood glucose response to intravenous insulin infusion rate that is adapted to various classes of patients seen in the ICU. The model will be based on patient data and will account for important perturbing factors such as dextrose infusion rates, nutrition (oral and parenteral), insulin resistant or sensitizing drugs, patient body temperature and clinical severity when entering the ICU. They will perform an extensive control analysis of existing insulin infusion protocols in closed-loop with the obtained dynamic model, with an emphasis on studying their nominal response as well as their response to a wide range of model parameter variations. On the basis of the model building and closed-loop control analysis performed above, the PIs will propose improvements to the existing protocols, including the so-called paper protocols that are employed in hospitals worldwide, which reflect a superior control law.Intellectual Merit: There have been no published comparisons of various protocols for such factors as rapidity and consistency of reaching targets or frequency of hypoglycemia. Neither has there been a systematic study of robustness of the existing protocols to patient variations such as nutrition (especially carbohydrates), changing vasopressors, steroids and other drugs. Published protocols claim between 46 to 69 % of blood glucose levels reach goal. Hypoglycemia rates can range from 0.3 to 15 % in several studies. A protocol that could bring blood glucose levels to goal quickly, respond appropriately and in a timely manner to changing insulin resistance/sensitivity, and address meals and drug regimes would have a significant impact on patient care. Preempting hypoglycemia by automatically noting the changing insulin sensitivity could address the fear of low blood glucose levels. In an unresponsive sedated patient, this would help minimize the concern for hypoglycemia which is the biggest barrier to the use of insulin. Lastly the ideal protocol in addition to addressing all the above issues would allow the health care provider to titrate a different blood glucose range and sensitivity for patients in different ICU settings (medical, surgical, cardiac, or trauma burn), minimize blood glucose swings with meals or infusions containing dextrose, and prevent hypoglycemia. The work proposed here will take us a significant step toward this goal.Broader Impacts: The general principles developed by this research should be transferable to the outpatient management of diabetes where the scope is of an enormous magnitude and a variety of other settings, including anesthesia and pain management.