Global Sensitivity Analysis enabled Uncertainty Domain Reduction for Euglycemic Control of Type 1 Diabetics

糖尿病的祸害困扰着超过10%的美国人口。2017年，糖尿病护理成本估计超过3000亿美元。在确诊的人中，有5%到10%是1型糖尿病患者，他们不断监测自己的血糖，用胰岛素调节血糖，以求生存。在过去的几十年里，连续血糖监测仪和胰岛素泵有了深刻的改进，这使得部署人工胰腺成为可能。该设备旨在模拟健康胰腺的功能，从而减轻糖尿病患者管理疾病的负担。糖尿病患者体内对胰岛素反应的巨大变异性，以及对极低血糖(可能是致命的)的恐惧，迫使开发强大的控制器来解释这种变异性。这笔赠款支持基础研究，以模拟公认的胰岛素反应特征的每日变化，并合成自动进餐检测算法。这与全面的全球敏感性分析相结合，将使可部署的血糖调节实时控制器的开发成为可能。这项研究的结果将有助于降低数百万糖尿病患者的医疗成本，提高患者的生产力，帮助美国经济和社会福祉。为了实现人工胰腺的全部潜力，这项由这笔拨款支持的研究将对包括黎明和黄昏现象在内的每日胰岛素敏感性进行建模，并开发一种膳食检测算法，以对未公布的膳食做出反应，或纠正公布的膳食中错误计算的碳水化合物。这两个目标将加强现有的葡萄糖-胰岛素模型。然而，随着不确定参数的增加，所提出的模型的复杂性将增加。这要求开发一个框架，用于基于非矩的全球敏感度指标来对排序参数进行排序，这些参数对预测血糖浓度的不确定性最大。由此产生的不确定空间的降维将允许准确的不确定性量化，并有助于开发健壮的控制器，使血糖调节的范围内时间性能指标最大化。美国食品和药物管理局批准了用于验证开发的控制器的软件，300名患者的数据集将支持研究结果。虽然这项研究的直接受益者是1型糖尿病患者，但一旦成熟，这项研究也有可能造福于2型糖尿病患者。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The scourge of diabetes afflicts over ten percent of the US population. In 2017, the cost of diabetes care was estimated to be over $300 billion. Five to ten percent of those diagnosed are Type 1 diabetic who constantly monitor their blood glucose to regulate it with insulin for survival. There have been profound improvements over the past few decades in continuous glucose monitors and insulin pumps, which now make it feasible to deploy an artificial pancreas. This device aims to emulate the function of a healthy pancreas and thus reduce the burden of managing the ailment for diabetics. The tremendous variability across diabetics in how the body responds to insulin and the fear of extremely low blood glucose (which can be fatal) mandates the development of robust controllers to account for this variability. This grant supports fundamental research to model well-recognized daily variation in insulin response characteristics and synthesize automated meal detection algorithms. This, in conjunction with comprehensive global sensitivity analysis, will enable the development of deployable real-time controllers for blood-glucose regulation. Results from this research will help reduce the cost of health care for millions of diabetics and improve the patient’s productivity, helping the US economy and well-being of society.To realize the full potential of an artificial pancreas, the research supported by this grant will model diurnal insulin sensitivity, which includes the dawn and dusk phenomena, and develop a meal detection algorithm to respond to unannounced meals, or rectify incorrectly counted carbohydrates in announced meals. These two goals will enhance the existing glucose-insulin models. However, the increased complexity of the proposed model will be burdened with increase in the dimension of the uncertain parameters. This mandates the development of a framework for non-moment based global sensitivity metrics to rank order parameters contributing the most to uncertainty in forecasting blood-glucose concentration. The resulting reduced dimension of the uncertain space will permit accurate uncertainty quantification and aid the development of robust controllers that maximize the time-in-range performance metric for blood-glucose regulation. A Food and Drug Administration approved software for validating the developed controllers and a 300-patient dataset will buttress the research outcomes. Although the immediate beneficiaries of the research are Type 1 diabetics, once matured, the research has the potential to also benefit people with Type 2 diabetes.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.