SBIR Phase I: Software for Developing Consumer-Driven Health Care Solutions

小型企业创新研究(SBIR)第一阶段项目的更广泛影响/商业潜力是开发软件，自动识别和标记患者在接受治疗时遇到的问题。这些信息准确地告诉卫生保健提供者他们需要解决什么问题来改善病人的护理。该软件的应用包括：a)针对不同的医疗保健情况开发不同的软件版本，该技术可用于改善诊所、急诊科、初级保健机构、门诊中心等的患者体验；b)早期干预。医疗保健领导人要求建立一个早期预警系统。通过记录患者在医院期间与护士和医生的对话，这种反馈可以用来在患者出院前发现和解决问题；c)新的医疗保健提供模式。一家医院希望减少心脏病患者的住院时间；医生希望确保患者准备好回家。该软件正在分析表现良好的心脏病患者与提早出院时出现问题的患者的故事。他们的反馈将帮助患者远离医院，如果他们入院，请在他们出院前后改善他们的护理；d)听从医生的命令。当慢性护理患者没有按照医生建议的治疗计划进行治疗时，他们的健康问题往往会变得更糟。该软件可用于收集慢性病患者的反馈，并确定这些患者为何不遵守医嘱的模式；以及(E)可以实施激励和吸引患者的解决方案；以及e)其他行业。该软件可用于分析临床试验期间患者的反馈。由此产生的信息将使在临床试验中招募和留住患者以及改善临床结果变得更容易。最终，这些分析的结果可以用来为FDA在制药行业内的决策提供信息。拟议的项目开发了使用高级自然语言处理(NLP)技术来分析患者的软件。在访谈和开放式回答调查问题中提供关于他们的卫生保健经验的答复，以便向医院提供关于如何改善护理和患者结果的具体、可操作的信息。到目前为止，医院主要依靠调查来为他们改善患者体验的尝试提供信息。这项研究将开发一个NLP应用程序，用于在医疗保健环境中挖掘患者反馈。与对预先定义的问题进行评级的标准调查相比，患者自由讲述他们的“故事”的数据提供了对患者经历更清晰、更准确的看法。挑战在于，表达和体验的多样性需要复杂的技术来对信息进行分类，确定情绪，并提取必要的细节，以便向医院提供可行的建议。该团队提出了数据注释、模式匹配和机器学习技术的组合，用于分类和从非结构化患者反馈中提取核心概念，如问题根本原因。由于访谈包含了许多信息量最大的数据，我们还将评估哪些语音识别技术可以最好地将音频转换为文本，以便随后进行分类和信息提取。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is the development of software that automatically identifies and labels problems that patients encounter when they receive care. This information tells health care providers exactly what they need to fix to improve patient care. Applications of the software include: a) the developing different software versions for different health care situations, the technology can be used to improve the patient experience in clinics, emergency departments, primary care settings, outpatient centers, etc.; b) Early interventions. Healthcare leaders are asking for an early warning system. By recording patient conversations with nurses and doctors while they are still in the hospital, this feedback can be used to identify and resolve problems before the patient is discharged; c) New health care delivery models. One hospital wants to reduce the length of stay for heart patients; doctors want to be sure that patients are ready to go home. The software is analyzing the stories of heart patients who did well versus patients who had problems when they left the hospital early. Their feedback will help keep patients out of the hospital and if they are admitted, improve their care before and after they leave; d) Following the doctor's orders. When chronic care patients don't follow through on the treatment plan recommended by their doctor, their health problems often get worse. This software can be used to collect feedback from chronically ill patients and identify patterns in why these patients are not following doctor's orders solutions that motivate and engage patients can be implemented; and e) Other industries. This software can be used to analyze patient feedback during clinical trials. The resulting information will make it easier to recruit and retain patients in clinical trials as well as improve clinical outcomes. Ultimately, results from these analyses could be used to inform FDA decision making within the pharmaceutical industry. The proposed project develops software that uses advanced Natural Language Processing (NLP) techniques to analyze patients? responses about their health care experiences in interviews and open response survey questions in order to provide hospitals with concrete, actionable information on how to improve care and patient outcomes. To date, hospitals have relied primarily on surveys to inform their attempts to improve patient experiences.This research will develop an NLP application for mining patient feedback across health care settings. Data where patients freely tell their "stories" provides a clearer and more precise view of the patient's experience than a standard survey with ratings on predefined questions. The challenge is that the variability of expression and experiences requires sophisticated techniques to be able to classify the information, determine the sentiment, and extract essential details that can provide actionable recommendations to hospitals. The team proposes a combination of data annotation, pattern matching, and machine learning techniques for classification and information extraction of core concepts like problem root causes from unstructured patient feedback. Since interviews comprise much of the most informative data, we will also evaluate which speech recognition technologies can best convert audio to text, for subsequent classification and information extraction.