Innovative Methods for Real-time Risk Modeling of Postoperative Complications

项目总结 外科手术存在术后并发症的风险，这些并发症可能会很严重， 费用昂贵，并将患者的生命置于危险之中。围手术期的风险分层 决策支持可以帮助规划和缓解这些复杂情况。研究的目的是 支持了解这些并发症的风险因素和开发风险模型 通过高质量的注册数据，如国家外科质量改进项目 (NSQIP)注册表。越来越多的研究表明，术中风险因素会影响 并发症的风险，但即使在NSQIP中也很难捕捉到这些风险。 在这项工作中，我们建议开发和实现基于以下方面的高级风险模型 术前和实时传输高分辨率术中数据。这个系统将会有 为患者建立术前基线状态的能力，跟踪他的病情 手术的进展并提供了对患者不同疾病风险的最新估计 手术前、手术中和手术后的任何时间的自动并发症(无人 干预)。这将帮助我们理解术中数据在预测 术后并发症。 我们在两个地点开展我们的项目：在明尼苏达大学附属美景健康中心 服务部和梅奥诊所。我们将开发可以耐心等待的建模技术 考虑到异质性(例如健康差异)，但又能产生可移植的模型 这两个地点。我们在两个站点独立构建模型，并对模型进行交叉验证 从制度上并在临床决策支持系统中实施经过验证的模型 各自的站点。所实现的系统构成了未来交互实时的基础 围手术期决策支持系统。

PROJECT SUMMARY Surgical procedures carry the risk of post-operative complications, which can be severe, expensive and put patients' lives at risk. Risk stratification in the context of perioperative decision support can help plan for and mitigate these complications. Research aimed at understanding the risk factors and developing risk models for these complications is supported by high-quality registry data, such as the National Surgical Quality Improvement Project (NSQIP) registry. A growing body of research indicates that intraoperative risk factors influence the risk of complications, but they are poorly captured even in the NSQIP. In this work, we propose developing and implementing advanced risk models based on preoperative and real-time streaming high-resolution intraoperative data. This system will have the ability to establish a preoperative baseline state for a patient, track his condition as the surgery progresses and provide an up-to-date estimate of the patient's risk of different complications at any time before, during and after surgery automatically (without human intervention). It will help us understand the value of intraoperative data in predicting postoperative complications. We carry out our project at two sites: at the University of Minnesota affiliated Fairview Health Services and Mayo Clinic. We will develop modeling techniques that can take patient heterogeneity (e.g. health disparities) into account, yet produce models that are portable across the two sites. We construct models at the two sites independently, validate the models cross- institutionally and implement the validated models in the clinical decision support systems of the respective sites. The implemented system forms the foundation of a future interactive real-time perioperative decision support system.