Predicting Patient Instability Noninvasively for Nursing Care (PPINNC)

描述（由申请人提供）：降压单元（SDU）上的患者接受连续无创生命体征（VS）监测，以便于护士检测心肺不稳定（CRI），但我们的数据显示，即使当单个VS监测器发出警报时，护士也无法快速检测CRI发作，也无法在80%的时间内尽早寻求帮助。我们最近证明，通过使用基于复杂性建模的算法，我们可以提高CRI的检测。该提案旨在进一步应用基于复杂性建模的算法，在明显不稳定之前预测CRI，具有足够的前置时间和准确性，以支持护理决策 进行先发制人的治疗临床决策支持系统（CDSS）持续处理来自不同来源的复杂数据，并应用预测算法及早提醒临床医生即将发生的事件。一种数据驱动的CDSS方法使用称为“机器学习”的人工智能类型来评估移动时间序列数据并学习导致事件的数据模式。然而，应用CDSS预测CRI事件受到缺乏适当详细的数据集学习支持的限制。我们最近证明，静态VS数据的人工神经网络（ANN）机器学习在任何连续监测的VS报警之前检测到CRI长达9.5分钟。通过应用机器学习来评估较长移动时间段内的高频VS数据，并包括人口统计和临床数据，将出现更敏感和更具体的CRI预测，并具有更长的交货期。我们建议：1）从我们现有的数据源中组装一个复杂的多维数据集作为学习平台，2）开发和验证用于SDU患者动态CRI预测的机器学习模型，并测试哪些模型在预测能力，模型简约性和事件提前时间方面最有效，以及3）使用这些发现为护士开发一个原型CRI预测CDSS工具。我们已经证明，我们有能力将大型高频数据集与定义的CRI事件组合在一起，从而进行机器学习，这使得我们的团队（护理，医学，数学，计算生物学，工程，统计学）具有进行这项工作的独特资格。研究结果可以促进CRI护理从反应性护理转变为先发制人的护理方法。开发一种灵敏、特异、简约和临床实用的方法来预测患者不稳定性，对于降低可预防的发病率和死亡率、改善患者安全、护理（监测频率、病例负荷和混合、工作人员分配）和护理提供系统（分诊、床位分配、预防不良事件）具有重要意义。

DESCRIPTION (provided by applicant): Patients on step-down units (SDU) undergo continuous noninvasive vital sign (VS) monitoring to facilitate nurse detection of cardiorespiratory instability (CRI), yet our data show nurses do not quickly detect CRI onset nor seek help early 80% of the time even when individual VS monitors are alarming. We recently demonstrated that by using a complexity modeling-based algorithm we could improve detection of CRI. This proposal seeks to further apply complexity modeling-based algorithms to predict CRI prior to overt instability with sufficient lead-time and accuracy to support a nursing decision for preemptive therapy. Clinical decisional support systems (CDSS) continuously process complex data from disparate sources and apply predictive algorithms to alert clinicians early to impending events. One data-driven CDSS approach uses an artificial intelligence type called "machine learning" to evaluate moving-time series data and learn data patterns leading to an event. However, applying CDSS to predict CRI events has been limited by lack of suitably detailed datasets for learning support. We recently demonstrated that artificial neural network (ANN) machine learning of static VS data detected CRI up to 9.5 min before any continuously monitored VS alarmed. By applying machine learning to evaluate high-frequency VS data over longer moving time blocks and including demographic and clinical data a more sensitive and specific CRI prediction with longer lead-times will emerge. We propose to: 1) assemble a complex multidimensional dataset from our existing data sources to serve as the learning platform, 2) develop and validate machine-learned models for dynamic CRI prediction in SDU patients, and test which models are most effective relative to predictive ability, model parsimony and lead-time to event, and 3) use these findings to develop a prototype CRI prediction CDSS tool for nurses. Our demonstrated ability to assemble large high-frequency datasets with defined CRI events from which machine learning occurs makes our team (nursing, medicine, mathematics, computational biology, engineering, statistics) uniquely qualified to conduct this work. Study findings can foster a shift in CRI care from a reactive to preemptive nursing approach. Developing a sensitive, specific, parsimonious and clinically practical means to predict patient instability has important implications for reducing preventable morbidity and mortality, improving patient safety, nursing care (monitoring frequency, case load and mixture, staff allocation) and care delivery systems (triage, bed allocation, prevention of adverse events).