RAPID: Data-Driven Models to Optimize Ventilator Therapy in ICU COVID Patients

新型冠状病毒（COVID-19）是由SARS-CoV-2病毒引起的四种传染病之一。尽管这种复杂疾病的临床体征和患者症状在表现和严重程度上各不相同，但临床医生和研究人员报告了体质症状（咳嗽和发烧）、上呼吸道和下呼吸道症状以及胃肠道症状。其中最令人担忧的是危及生命的急性呼吸窘迫综合征（ARDS）患者。严重急性呼吸窘迫综合征的病理生理是由于肺功能的快速下降，危重患者需要插管进行有创机械通气，以对抗肺功能恢复，降低外周血管血氧饱和度（SpO2），降低器官衰竭和死亡的风险。通过正呼气末压（PEEP）实现呼吸机设置以增加SpO2和氧气输送。然而，长时间在高PEEP下控制通气会显著增加呼吸机相关肺损伤（VALI）的风险。RAPID项目将开发新的工程策略，优化呼吸机控制，在最短的时间内最大限度地提高SpO2，同时最小化PEEP和呼吸机使用时间，以最大限度地减少VALI和随后的并发症，并改善患者的良好预后。在COVID-19患者的管理中，这些策略对优化供氧、微创呼吸机使用和机械肺损伤具有重要意义。此外，对呼吸机要求和操作环境的了解强调了对可用呼吸机的需求。严重急性呼吸窘迫综合征的治疗是复杂的，迫切需要策略和方案。为了实现这一目标，我们将开发数据驱动的线性参数变化（LPV）动力系统模型，将患者临床状态和呼吸机输入与输出变量患者SpO2联系起来。将使用来自电子健康记录（EHR）的数据和从患者监测中获得的每分钟生理时间序列（PTS）数据（例如，心率、呼吸频率、SpO2）来表征患者状态。我们将首先利用使用呼吸机的非covid -19患者的回顾性数据开发LPV模型，以帮助治疗肺炎和ARDS等疾病。然后，我们将测试LPV模型对COVID-19患者使用呼吸机的预测能力。最后，我们将基于LPV模型制定COVID-19患者呼吸机优化控制策略，以调节ICU患者SpO2水平。试图使用基于机制模型的控制策略来控制一个复杂的生物系统通常是棘手的。然而，LPV框架允许实现复杂的最优策略，不仅允许比其他经典方法更好的性能，而且还提供稳定性和性能保证。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The novel Coronavirus (COVID-19) is one of four infectious diseases caused by the SARS-CoV-2 virus. Although the clinical signs and patient symptoms of this complicated disease vary in presentation and severity, clinicians and investigators have reported constitutional symptoms (cough and fever), upper and lower respiratory tract symptoms, as well as gastrointestinal symptoms. Among the most concerning is the life threatening acute respiratory distress syndrome (ARDS) in patients. The pathophysiology of severe ARDS results from a rapid decline in pulmonary function and requires intubation of patients in critical condition for invasive mechanical ventilation to combat lung recruitability, reduced peripheral capillary oxygen saturation (SpO2) and risks of organ failure and death. Ventilator settings to increase SpO2 and oxygen delivery is achieved with positive end-expiratory pressure (PEEP). However, controlling ventilation at a high PEEP for extended periods of time significantly increases risk for ventilator-associated lung injury (VALI). This RAPID project will develop novel engineering strategies for optimal ventilator control to maximize SpO2 in minimal time, while minimizing PEEP and the duration of ventilator use are needed to minimize VALI and subsequent complications, and to improve favorable patient outcomes. In the management of patients with COVID-19, these strategies are significant to optimize oxygen delivery, minimal invasive ventilator use and mechanical lung injury. Further, the understanding of ventilator requirements and operative settings highlights the need for available ventilators. The management of severe ARDS is complicated and strategies and protocols are desperately needed.To achieve this goal, we will develop data-driven linear parameter-varying (LPV) dynamical systems models that relate patient clinical state and ventilator inputs to the output variable patient SpO2. Patient state will be characterized using data from the electronic health record (EHR) and minute-by-minute physiological time-series (PTS) data (e.g., heart rate, respiratory rate, SpO2) acquired from patient monitoring. We will first develop the LPV model using retrospective data from non-COVID-19 patients who are on ventilators to help treat conditions such as pneumonia and ARDS. Then, we will test the predictive capabilities of the LPV model in COVID-19 patients who are placed on ventilators. Finally, we will develop an optimal ventilator control strategy for COVID-19 patients to regulate SpO2 levels in ICU patients based on the LPV model. Attempting to control a complex biological system using control strategies based on mechanistic models is generally intractable. However, the LPV framework allows for sophisticated optimal strategies to be implemented that not only allow for better performance than other classical methods, but also provides stability and performance guarantees.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.