Automation and Multi-Site Validation of a Personalized Empiric Antibiotic Advisor

 描述(由申请人提供)：如果选择了不适当的抗生素，不良的抗生素处方做法可能会增加社区的细菌耐药性，并损害患者。这项工作的目标是自动化和验证一个工具，该工具旨在帮助供应商在获得培养结果之前更好地选择抗生素。该工具使用来自医院的微生物学数据 以及来自患者的临床信息，以预测哪种抗生素方案最有可能覆盖患者的感染。这项研究将在两个地点进行，以提高研究结果的概括性，并创建一个可以很容易地传播到其他地点的工具。这项工作将分三个目标完成。在第一个目标中，将对两个机构的微生物学数据进行格式化，以便将其用于预测模型。包括这两个机构的微生物学家、医生和药剂师在内的专家将创建规则，这些规则将用于创建用于格式化数据的自动算法。在第二个目标中，预测模型将在这两个机构的回溯性队列中开发和验证。接下来，由主题专家开发的处方规则将添加到模型输出中。这将允许该工具根据规则和预测建模输出推荐的抗生素方案。研究人员将通过将该建议与每个患者实际给予的初始抗生素进行比较来评估该建议的质量。覆盖的感染百分比 将对每种抗生素方案进行比较，并对每种方案的覆盖范围和成本进行比较。在第三个目标中，将在电子健康记录内创建一个用户友好的界面，以显示在目标二中创建的工具的输出。这项工作是一个更大项目的第一阶段，该项目旨在研究让提供者获得最新的、本地的、个性化的微生物学数据是否会改善处方做法，并最终改善患者的护理和结果。

 DESCRIPTION (provided by applicant): Poor antibiotic prescribing practices can increase bacterial resistance in the community and harm patients if an inappropriate antibiotic is chosen. The goal of this work is to automate and validate a tool intended to help providers better choose antibiotics before culture results are available. The tool uses microbiology data from the hospital and clinical information from the patient to predict which antibiotic regimen would be most likely to cover the patient's infection. The study will be done at two sites in order to improve the generalizability of the findings, and to create a tool that can be easily disseminated to other sites. The work will be accomplished in three aims. In the first aim, the microbiology data from two institutions will be formatted so that it can be used in predictive models. Experts including microbiologists, physicians, and pharmacists at both institutions will create rules that will be used to create an automated algorithm for formatting the data. In the second aim, predictive models will be developed and validated on a retrospective cohort at both institutions. Next, prescribing rules developed by subject matter experts will be added to the model output. This will allow the tool to output a recommended antibiotic regimen based on the rules and predictive modeling. The investigators will assess the quality of this recommendation by comparing it to the actual initial antibiotic given to each patient. The percentage of infections that were covered by each antibiotic regimen will be compared, as will the breadth of coverage and cost of each regimen. In the third aim, a user-friendly interface will be created within the electronic health record to display the outputs of the tool created in aim two. This work is the first phase of a larger project intended to study whether giving providers access to up-to-date, local, personalized microbiology data will improve prescribing practices and ultimately patient care and outcomes.