Evaluating diagnostic decision support systems for patients requiring urgent primary or emergency care or with stroke

医疗诊断是有效卫生保健的重要组成部分，但误诊、延误诊断和不正确的分诊 尤其在紧急或急诊护理环境中常见，并且是不良临床事件的主要促成因素。而 医疗保健的改善在解决误诊方面很重要，患者通过认识到潜在的 症状严重，及时就医。这些问题对于需要治疗的患者尤其重要。 对可能危及生命的疾病（如短暂性脑缺血发作（TIA）、中风或 心肌梗死他们往往不能认识到他们的症状的严重性，可能无法及时寻求治疗， 导致错过治疗和更差的结果。中风患者通常必须在4小时内接受治疗， 取得良好的反应，公众教育运动没有很大的帮助。医疗智能手机应用程序 在美国和世界范围内，被称为症状标志物（SC）的诊断广泛适用于患者。他们一直 显示可由患者使用，并且可以影响患者决策和寻求护理的行为。我们工作中的证据 其他研究表明，如果使用得当，SC可以实现接近医生的诊断和分诊准确性 （根据症状数据）。然而，大多数研究都是基于医生创建的病例摘要，SC应用程序不是 患者使用，缺乏证据表明患者特征的影响，或SC对患者决策的影响。 为了解决这些差距，我们将评估一个领先的症状检查器的安全性，可用性，诊断和分类准确性 在急诊科或紧急初级护理中由患者使用，以及SC输出对患者决策的影响 寻求照顾。在目标1中，我们将招募700名患者，当他们在急诊室就诊时使用Ada Health的SC应用程序。 护理或罗得岛医院（RIH）的急诊科，包括可能有TIA或中风症状的患者。 这建立在我们之前对Ada SC的研究基础上，在这些地点招募了241名患者。紧急程度 他们打算寻求的护理将在使用Ada之前和之后进行评估，沿着关于应用程序可用性的问卷调查。 诊断和分诊的准确性将与看过病人的医生的评估和Ada结果进行比较 都与独立医生查看Ada收集的症状数据进行了比较和批评。我们还将 评估不同的诊断和分诊数据对患者决策的影响。 在目标2中，我们将利用2个独特的数据集，其中包括在RIH的艾德中观察到的2300例可能患有TIA或卒中的患者，并使用 机器学习技术，以创建新的算法，以改善早期诊断和风险分层。性能将 与现有的算法和指南进行比较，以准确诊断和有效管理这些疾病。 我们还将分析一个社区中使用Ada应用程序的158，000名可能患有TIA或中风的患者的数据集 设置，并评估患者特征的影响，包括年龄，性别，种族，民族，国家和社会， 经济组的症状，比较这与RIH数据集的结果。这些研究的结果将会改善 我们对症状检查器安全性的理解，识别高危患者并指导他们寻求护理的能力， 对包括TIA或中风患者在内的广泛患者的卫生服务使用的潜在影响。

Medical diagnosis is a critical component of effective health care but misdiagnosis, delayed diagnosis and incorrect triage is common especially in urgent or emergency care settings, and a major contributor to adverse clinical events. While improvements in health care are important in addressing misdiagnosis, patients have a key role by recognizing potentially serious symptoms and seeking care in a timely manner. These concerns are of particular importance for patients requiring urgent or emergency care with potentially life-threatening diseases, such as transient ischemic attacks (TIA), stroke or myocardial infarction. They often fail to recognize the seriousness of their symptoms and may fail to seek care promptly, resulting in missed treatments and poorer outcomes. Patients with stroke typically must be treated within 4 hours to achieve a good response, and public education campaigns have not significantly helped. Smartphone apps for medical diagnosis termed Symptom Checkers (SCs) are widely available to patients in the US and worldwide. They have been shown to be usable by patients, and can affect patient decision making and care seeking behavior. Evidence from our work and others has shown that, if used correctly, SCs can achieve accuracy of diagnosis and triage close to that of physicians (relying on symptom data). However, most studies are based on case summaries created by physicians, SC apps are not used by patients, and lack evidence on the effects of patient characteristics, or SCs influence on patient decision making. To address these gaps we will evaluate the safety, usability, diagnostic and triage accuracy of a leading symptom checker in use by patients in an emergency department or urgent primary care, and the effect of SC outputs on patients’ decisions to seek care. In aim 1 we will recruit 700 patients to use a SC app from Ada Health when they are seen in urgent primary care or the emergency department at Rhode Island Hospital (RIH), including those with possible symptoms of TIA or stroke. This builds on our previous studies of the Ada SC with 241 patients recruited in these locations. The level of urgency of care they intend to seek will be assessed before and after use of Ada, along with a questionnaire on app usability. Diagnostic and triage accuracy will compared to the assessment of the physician who saw the patient, and the Ada results both compared with, and critiqued by, independent physicians viewing the symptom data collected by Ada. We will also evaluate the effects of different presentations of diagnosis and triage data on patient decision making. In aim 2 we will utilize 2 unique data sets of 2300 patients with possible TIA or stroke seen in the ED at RIH, and use machine learning techniques to create new algorithms to improve early diagnosis and risk stratification. Performance will be compared with existing algorithms and guidelines on accurate diagnosis and effective management of these conditions. We will also analyze a data set of 158,000 patients with possible TIA or stroke who used the Ada app in a community setting, and evaluate the influence of patient characteristics, including age, sex, race, ethnicity, country, and socio- economic group on their symptoms, comparing this to the RIH data set results. The results of these studies will improve our understanding of symptom checker safety, ability to recognize high risk patients and direct them to seek care, and the potential impact on health service use, for a broad range of patients including those with TIA or stroke.