基于重症监护室(ICU)大数据的创伤决策支持系统研究

Intensive care unit (ICU) is a key hospital ward for treatment of critical illnesses. ICU constantly and closely monitors body functions of severe patients. ICU clinicians need to make illness severity judgment and clinical intervention decision based on the monitored physiological data. Big data in the ICU is from multiple sources and it has the characteristics of complexity, temporality, and variability. Thus illness severity judgment and clinical intervention in the ICU should be accurate and prompt. How to use big data in the ICU to aid real-time dynamic clinical decision is an important research topic. In this project, using trauma ICU as a clinical area, first, we propose to build an ICU big data platform with distributed computational framework to integrate all ICU data of those involved hospitals in Beijing; second, based on Bayesian method, evidential reasoning rule and time-series analysis, we propose a model for clinical evidence mining from ICU big data and another two types of models for dynamic combination of pieces of evidence from multiple sources; third, based on the constructed ICU big data platform and the evidence mining and dynamic evidence combination models, we propose to develop a trauma ICU decision support system prototype; fourth, we propose to prospectively validate the developed system in all involved ICUs. The trauma ICU decision support system developed in this project will aid ICU clinicians to make online severity judgment together with dynamic outcome prediction for ICU trauma patients. It will facilitate ICU clinicians to make active clinical intervention for severe trauma patients. Consequently, it will help improve the quality of trauma care in ICU and reduce trauma mortality. The research outcome of this project will have reference significance for using ICU big data to aid clinical decision of other critical illnesses in the ICU.

重症监护室(ICU)是重症创伤救治的关键场所。ICU需要持续监测创伤患者生理功能，并根据实时监测数据进行病情判断和救治。ICU大数据多源复杂且随时变化。ICU伤情判断和临床干预必须准确及时。如何利用ICU大数据辅助临床决策是医学领域亟待解决的重要课题。本项目选择创伤ICU作为研究领域，基于分布式计算框架构建ICU大数据平台，集成北京市多家参研医院ICU数据；利用贝叶斯方法及证据推理进行临床证据挖掘与学习、结合证据推理规则和时间序列分析构建多源证据动态融合的模型；采用原型法开发一套基于ICU大数据的创伤决策支持系统；应用现场观察法开展ICU创伤病患临床调查、进行前瞻性系统验证研究。该系统将辅助ICU医护人员进行在线动态的创伤评估和结局预测，促进ICU医护人员进行主动临床干预。从而提升ICU创伤救治质量、降低创伤致死致残率。该项目的研究成果对基于ICU大数据的其它危重症临床决策亦具有借鉴意义。

如何利用医疗大数据辅助临床决策是医学领域亟待解决的重要课题。本研究基于真实世界创伤和ICU大数据，提出了一套基于证据推理规则、数据驱动的临床决策支持方法。在该方法中，贝叶斯推理范式下的似然度计算用于临床证据挖掘以构建证据库；证据推理规则用于多则证据融合以产生决策建议；基于真实世界观测结局与系统推理结局的误差最小化用于训练临床证据的权重和可信度。同时，本研究也应用多种机器学习方法进行不同疾病的预后预测建模研究，并利用数据分析及建模软件进行模型构建和系统实现。本项目研发的数据驱动的临床决策支持系统不仅在方法学上有创新，而且对辅助急诊与ICU医护人员进行在线的创伤评估与结局预测有很强的实践意义，将促进医护人员进行临床主动干预，帮助改善患者预后。