Continuous-Signal Driven Predictive Models in Neurological Intensive Care Units

描述(由申请人提供)：大多数医疗干预措施主要是为了在传统的患者管理方案下对临床症状或异常生理值做出反应。相比之下，如果得到对关键临床事件的可靠和可操作的预测的支持，预防性或前瞻性方案可以显著改善结果并降低不断上升的医疗成本。然而，在大多数临床环境中，卫生保健专业人员只能访问最多到目前为止的数据。我们研究的长期目标是通过对不同的医疗、生理和生物数据进行预测性数据挖掘，为个体患者的关键参数和事件提供可靠的临床预测。作为神经科重症监护环境中典型的临床预测应用，本项目的主要目的是展示在一个新的分类器融合框架下，基于我们小组最近在处理连续的颅内压信号中发现的机器学习算法和新的量化指标，对两种常见的颅内二次损伤(包括急性脑室增大(脑室增大)和急性颅内压升高(ICP))进行早期识别的有效性。 在神经科重症监护病房(NICU)中，预测脑室增大和颅内压升高的需求尤其相关，在NICU中，使用一系列连续监测的信号来支持复杂和严重神经疾病患者在急性期的治疗。这些危重病人容易受到多种形式的延迟性但可治疗的继发性损伤。因此，在临床症状出现之前及早发现二次胰岛素抵抗是支持采用前瞻性患者管理的直接相关因素，其有效性可在后续的随机临床试验中得到证明。 我们提出的两个目标是首先建立一个通用框架，支持将连续的生理信号合并到由一组分类器组成的预测模型中。这些分类器相对于感兴趣的时间以不同的时间间隔间隔，并且它们的结果被融合以提供改进的预测。第二个目标包括两个子目标，以建立两个在神经危重护理环境中有用的典型预测。第一个预测涉及脑室扩大的早期发现，第二个预测涉及急性颅内压升高的预测，这两种预测都是颅脑损伤和出血性脑损伤后常见的继发性损害。 如果成功，对颅内压升高和脑室扩大的预测可能会导致潜在的范例从传统的反应性患者管理转变为更积极的管理方案，并改善患者的预后，提高神经危重护理的效率。 公共卫生相关性：该项目的目标是首先开发一个通用框架，以支持将连续的生理信号纳入可用于提供临床预测的预测模型中。然后，我们将调查两个典型预测的表现，这两个预测在管理神经重症监护病房的脑损伤患者方面非常有用。第一个预测涉及脑室扩大的早期发现，第二个预测涉及急性颅内压升高的预测，这两种预测都是颅脑损伤和出血性脑损伤后常见的继发性损害。因此，他们的成功预测可以导致从传统的反应性患者管理向更主动的管理方案的范例转变，并改善患者结局并提高神经危重护理的效率。

DESCRIPTION (provided by applicant): Most medical interventions are primarily prescribed to respond to clinical symptoms or abnormal physiological values under conventional patient management protocols. In contrast, prophylactic or proactive protocols can substantially improve outcomes and decrease escalating costs of healthcare if supported by reliable and actionable forecasts of critical clinical events. However, in most clinical environments, health care professionals are only able to access data that are, at best, up to the present time. Long-term goal of our research is to provide reliable clinical forecasts of individual patient's critical parameters and events by predictive data mining of heterogeneous medical, physiological, and biological data. As a prototypical clinical forecast application in a neurological intensive care environment, the main objective of the proposed project is to demonstrate the efficacy of earlier recognition of two common intracranial secondary insults including acute ventriculomegaly (enlargement of the brain ventricles) and acute elevation of intracranial pressure (ICP) based on integrating, under a novel classifier fusion framework, machine learning algorithms and novel quantitative metrics that were recently discovered by our group in processing continuous ICP signals. The need for forecasting ventriculomegaly and elevated ICP is particularly relevant in a neurological intensive care unit (NICU) where an array of continuously monitored signals are used to support management of patients of complex and severe neurological disorders at their acute phase. These critically ill patients are susceptible to many forms of delayed but treatable secondary injuries. Therefore, an early detection of developing secondary insults prior to clinical symptoms is directly relevant to support the adoption of proactive patient management whose efficacy can be demonstrated in a follow-up randomized clinical trial. We propose two aims to first build a general framework supporting incorporation of continuous physiological signals into a predictive model that comprise of a set of classifiers. These classifiers are spaced at different time intervals relative to the time of interest and their results are fused to provide an improved forecast. The second aim includes two sub aims to build two protypical forecasts useful in a neurocritical care environment. The first forecast concerns early detection of ventriculomegaly and the second forecast concerns prediction of acute ICP elevation, both of which are common forms of secondary insult after traumatic and hemorrhagic brain injury. If successful, the forecast of ICP elevation and ventriculomegaly can lead to a potential paradigam shift from conventional reactive patient management to a more proactive management protocol and improve patient outcome and enhance the efficiency in neurocritical care. PUBLIC HEALTH RELEVANCE: The objectives of this project are to first develop a generic framework for supporting incorporation of continuous physiological signals into a predictive model that can be used for providing clinical forecasts. Then we will investigate the performance of two protypical forecasts that are very useful in managing brain injury patients in a neurocritical care unit. The first forecast concerns early detection of ventriculomegaly and the second forecast concerns prediction of acute ICP elevation, both of which are common forms of secondary insult after traumatic and hemorrhagic brain injury. Therefore, their successful forecast can lead to a paradigam shift from conventional reactive patient management to a more proactive management protocol and improve patient outcome and enhance the efficiency in neurocritical care.