Real-time Infection Prediction in Inpatient Postoperative Care

术后感染是最常见的外科并发症，每天影响40多名退伍军人 每一次感染的总护理费用估计增加了25,000美元。超过一半的术后SSI 在患者出院后被诊断为。当前的决策支持工具 术后感染的预测准确性很差。这使得识别患者身份变得特别困难。 出院后有发生SSI的风险，这使出院计划复杂化。准确地 识别出院时感染风险较高的患者有助于改进出院计划、识别 适当的出院后跟踪时间，以及更好的目标资源--昂贵的出院后监测。 体温或疼痛等生命体征一直被证明可以预测感染。他们照例是 在住院期间收集，但仍是感染风险预测的未开发信息来源 模特们。我们假设我们可以提高现有感染风险预测模型的准确性 通过包括这些实时生命体征数据。在计划我们的更大规模的研究时，我们确定了几个可行性 在我们着手开发该工具并在临床上测试它之前，应该解决的问题 环境。因此，我们正在寻求HSR&D试点资金，以最终敲定我们计划的HSR&D的协议 研究人员发起的研究性研究。 我们的研究目标如下： 1.进行发展性形成性评估，评估可行性、可接受性、潜在有用性 以及初步设计用于预测出院时感染风险的决策支持工具。 2.检查VA公司数据仓库中收集的住院患者生命体征数据的完整性。 我们使用了以人为中心的设计思维框架来计划我们的研究。对于目标1，我们将使用定性 方法分析来自两个退伍军人管理局的20个结构化观察和15个半结构化访谈的数据 分布在至少六个退伍军人管理局。这些定性分析将验证我们提出的工具可以被纳入 在退伍军人管理局的临床护理中，临床提供者将接受该工具，并且它将被认为是有用的。我们 还将开始开发该工具的用户界面的初始原型，并在退伍军人管理局临床护理中进行部署。 Aim 2解决了使用可能不完整的生命体征数据构建我们的模型的可行性问题。这个 弗吉尼亚州企业数据仓库(CDW)是访问从以下位置收集的住院患者生命体征的标准位置 整个退伍军人事务部。不幸的是，我们发现并不是所有的局部生命体征数据都能进入CDW和一些 生命体征数据仅在VISN特定的数据库中捕获。这让我们担心 当使用国家CDW数据来改进具有生命体征数据的现有模型时，缺少数据。对于目标2，我们将使用 用定量方法分析VISN 21中6个退伍军人医院入院的所有患者的现有数据库。至 确认开发我们的模型和工具原型的可行性，我们计划确定所有生命体征如何 在退伍军人事务部捕获，检查生命体征数据的完整性，并探索影响和潜在偏差 使用不完整的CDW生命体征数据。AIM 2的研究结果将决定开发该工具的策略 在弗吉尼亚州帕洛阿尔托医疗保健系统，并将该工具扩展到其他设施和VISN。 使用生命体征数据更新当前的感染风险预测分数可能会大大增加 术后感染风险预测的准确性。更准确地预测当时的高危患者 出院计划将改进出院计划，并帮助提供者识别将受益于 术后早期随访。它还可以帮助更好地分配其他资源密集型和昂贵的资源 预防性策略，如预防性抗生素或负压伤口治疗。这将导致 大大降低了成本、发病率和死亡率。此外，这项研究还将告知退伍军人事务部研究社区 关于储存在退伍军人管理局内的生命体征数据的可用性和使用情况。

Postoperative infections are the most common surgical complication, affecting upwards of 40 Veterans each day and adding an estimated $25,000 to the total cost of care per infection. More than half of all postoperative SSIs are diagnosed after the patient has been discharged from the hospital. Current decision support tools for postoperative infections suffer from poor predictive accuracy. This makes it particularly hard to identify patients at risk for developing an SSI after hospital discharge which complicates discharge planning. Accurately identifying patients with a high risk of infection at discharge could help to improve discharge planning, identify adequate timing for post-discharge follow up, and better target resource-expensive post-discharge surveillance. Vital signs, such as temperature or pain, have consistently been shown to predict infections. They are routinely collected during inpatient stays but have remained an untapped source of information for infection risk prediction models. We hypothesize that we can improve the accuracy of existing infection risk prediction models by including this real-time vital sign data. While planning our larger study, we have identified several feasibility concerns that should be addressed before we embark on developing the tool and testing it in a clinical environment. Thus, we are seeking HSR&D Pilot funding to finalize the protocol of our planned HSR&D Investigator-Initiated Research study. Our study aims are as follows: 1. Perform a developmental formative evaluation assessing the feasibility, acceptability, potential usefulness and initial design of a decision support tool for predicting infection risk at discharge. 2. Examine the completeness of inpatient vital sign data collected in the VA Corporate Data Warehouse. We have planned our study using a person-centric design thinking framework. For aim 1, we will use qualitative methods to analyze data from 20 structured observations at two VA facilities and 15 semi-structured interviews across at least six VA facilities. These qualitative analyses will verify that our proposed tool can be incorporated into clinical care at the VA, that clinical providers will accept the tool, and that it will be perceived as useful. We will also begin to develop an initial prototype for the tool’s user interface and deployment within VA clinical care. Aim 2 addresses a feasibility concern about building our model with potentially incomplete vital sign data. The VA Corporate Data Warehouse (CDW) is the standard location for accessing inpatient vital signs collected across the entire VA. Unfortunately, we have discovered that not all local vital sign data make it into the CDW and some vital sign data are only captured in VISN-specific databases. This leads us to concerns about the impact of missing data when using national CDW data to refine existing models with vital sign data. For aim 2, we will use quantitative methods to analyze an existing database of all patients admitted to six VA facilities in VISN 21. To confirm the feasibility of developing our model and tool prototype, we plan to determine how all vital signs are captured in the VA, examine the completeness of the vital sign data, and explore the impact and potential biases of using incomplete CDW vital sign data. The findings of aim 2 will determine strategies for developing the tool at VA Palo Alto Health Care System and expanding the tool across other facilities and VISNs. Updating current infection risk prediction scores with vital sign data has the potential to greatly increase the accuracy of risk prediction for postoperative infection. A more accurate prediction of high-risk patients at the time of discharge will improve discharge planning and help providers identify high-risk patients that would benefit from earlier postoperative follow-up. It could also help better allocate other resource-intensive and expensive preventative strategies such as prophylactic antibiotics or negative pressure wound therapy. This would result in substantial reductions in costs, morbidity, and mortality. Also, this study will inform the VA research community on the availability and use of vital sign data stored within the VA.