Predictive Informatics Monitoring in the Neonatal Intensive Care Unit

新生儿ICU中的早产儿需要住院，直到他们达到 生理成熟期，平均60天。然而，在医院里，他们正处于 亚急性潜在灾难性疾病的风险，如感染、呼吸道疾病 失代偿导致紧急计划外插管和颅内出血。 这些疾病很常见，而且是致命的。在每一种情况下，早期诊断都有希望 通过早期干预改善预后。 我们小组的长期目标是开发这种新的预测性监测策略 作为早期预警系统，通过高级数学和统计分析 来自床边监护仪的波形和其他信息学数据。这种做法 最近带领该小组及其在其他7个中心的同事完成了NICHD- 赞助对3000名早产儿进行随机临床试验，这是有史以来规模最大的一次 在这一人群中进行。结果非常重要--只是简单地展示了结果 临床医生使用预测性监护仪后，死亡率降低了20%以上。 然而，这种预测性工具需要使用胸导联进行ICU水平监测，以进行EKG和 呼吸信号。如果有策略，可以帮助更多的婴儿使用 仅仅是无处不在的脉搏血氧仪，它可以提供心率和氧气饱和度数据 每隔1或2秒。推导使用这个小数据流的预测算法 需要相关临床信息和监测数据的大型数据库，包括重要的 来自多个地点的多个婴儿的体征和波形。临床医生团队和 数学家-弗吉尼亚大学、华盛顿大学和圣彼得堡大学合作 Louis和哥伦比亚大学-将发现基于血氧测定仪的异常表型 并开发算法来检测它们。大型数据库和 这项工作的计算能力已经在日常使用，UVA和哥伦比亚大学的合作 在该奖项的头几年卓有成效，这项竞争性续签提案 将使他们能够进行随机临床试验，以测试 新的监控器。 这代表着患者护理的范式转变-报告以下趋势的监视器 健康和疾病的发展，而不是转瞬即逝的价值观。

Premature infants in the Neonatal ICU require hospitalization until they reach physiological maturity, an average of 60 days. While in the hospital, though, they are at risk of subacute potentially catastrophic illnesses such as infection, respiratory decompensation leading to urgent unplanned intubation, and intracranial bleeding. These illnesses are common and deadly. In each case, early diagnosis has the promise to improve outcome through early intervention. The long-term goal of our group is to develop such novel predictive monitoring strategies as early warning systems through advanced mathematical and statistical analysis of waveforms and other informatics data from the bedside monitor. This kind of approach recently led the group and its colleagues at 7 other centers to complete a NICHD- sponsored randomized clinical trial in 3000 premature infants, the largest ever conducted in this population. The result was very important - simply showing the results of a predictive monitor to clinicians reduced the death rate by more than 20%. This predictive tool, though, requires ICU level monitoring with chest leads for EKG and breathing signals. Many more infants could be helped if there were strategies for using just the ubiquitous pulse oximeter, which provides heart rate and O2 saturation data every 1 or 2 seconds. Deriving predictive algorithms that use this small data stream requires large databases of relevant clinical information and monitor data, including vital signs and waveforms, from many infants at multiple sites. The team of clinicians and mathematicians – a collaboration of University of Virginia, Washington University – St Louis, and Columbia University – will discover oximetry-based phenotypes of abnormal physiology and develop algorithms to detect them. The large-scale databases and computing capability for this work is in daily use, the UVa-Columbia collaboration has been productive in the first years of this award, and this competitive renewal proposal will leave them in the position to undertake randomized clinical trials to test the impact of the new monitoring. This represents a paradigm shift in patient care – monitors that report trends of development of health and illness rather than fleeting values.