Extraction of Symptom Burden from Clinical Narratives of Cancer Patients using Natural Language Processing

使用自然语言处理从癌症患者的临床叙述中提取症状负担

• 批准号: 10179677
• 负责人: Meliha Yetisgen
• 金额: $ 44.45万
• 依托单位: UNIVERSITY OF WASHINGTON
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-05-19 至 2024-04-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10179677
• 关键词: Active Learning; Address; Adoption; Adverse effects; Affect; Anatomy; Anxiety; Cancer Burden; Cancer Patient; Caring; Cessation of life; Characteristics; Clinical; Communities; Data; Data Set; Decision Support Systems; Desire for food; Diagnosis; Disease; Documentation; Drowsiness; Electronic Health Record; Engineering; Evaluation; Frequencies; Funding; Future; Goals; Gold; Grain; Hospitals; Impairment; Information Retrieval; Institutes; Institution; Intervention; Location; Malignant Neoplasms; Malignant neoplasm of prostate; Methods; Modeling; Natural Language Processing; Nature; Nausea; Oncology; Outcome; Outpatients; Pain; Palliative Care; Patients; Performance; Physicians; Practice Guidelines; Prevention; Professional Organizations; Provider; Publications; Quality of life; Records; Reporting; Resources; Semantics; Services; Severities; Shortness of Breath; Specialist; Structure; Supervision; Supportive care; Symptoms; System; Technology; Testing; Time; Training; Utah; Visit; Washington; Well in self; Work; base; cancer care; cancer complication; cancer therapy; cancer type; care systems; cohort; cost effectiveness; deep learning; experience; field study; implementation barriers; improved; innovation; instrument; large cell Diffuse non-Hodgkin' s lymphoma; learning strategy; multidisciplinary; novel; oncology service; open source; patient health information; relating to nervous system; repository; symptom management; symptomatic improvement

Project Summary / Abstract Cancer patients frequently experience high levels of pain, tiredness, shortness of breath, decreased appetite, nausea, drowsiness, anxiety, and decreased sense of wellbeing, often related to the disease itself, its treatments, or both. This high symptom burden leads to significant impairment of cancer patients’ quality of life and may be associated with impaired survival. Optimal symptom management is required to minimize symptom burden and maximize quality of life for cancer patients throughout the course of their disease. Supportive care in cancer (SCC) teams are multidisciplinary teams that are focused on the prevention and management of the adverse effects of cancer and its treatments across the continuum of the cancer experience from diagnosis through treatment and beyond. These teams typically lack the resources to see all cancer patients and need to prioritize patients with the highest need, often relying on oncology physicians for referral. However, oncology physicians are often too focused on curing cancer than treating its symptoms. As a result, SCC services are often accessed by chance even when available, often later in the cancer trajectory. To improve recognition of SCC needs and to identify the symptom burden of cancer patients for better management and care, we propose to build natural language processing (NLP) approaches that can automatically extract symptom information from unstructured narratives. The proposed systems will utilize neural nets and build on the state of the art information extraction methods. To accomplish our goals, we will create a dataset of clinical notes for a large cohort of prostate cancer and Diffuse Large B Cell Lymphoma (DLBCL) patients treated in Seattle Cancer Care Alliance (SCCA) and Huntsman Cancer Institute (HCI) between 1.1.2015 and 1.1.2020. We focus on these two types of cancer as examples of two very different and prevalent cancer types. We propose to represent symptom burden documented in clinical narratives with a generalizable frame representation that captures fine-grained details including presence/absence, change-of- state, severity, characteristics, duration, frequency, and anatomy information related to patient symptoms. We will use active learning to create a diverse and representative gold standard annotated with symptom frames to train and test the proposed neural-based NLP approaches. All models and their implementations produced during the execution of this project will be shared with the community as open source resources. After successful completion of the project, the developed NLP methods will be integrated into the information access methods of SCCA and HCI clinical repositories.

项目总结/摘要 癌症患者经常经历高水平的疼痛、疲劳、呼吸急促、食欲下降， 恶心、嗜睡、焦虑和幸福感下降，通常与疾病本身有关， 治疗，或两者。这种高症状负担导致癌症患者的生活质量显著受损 并且可能与受损的存活有关。需要最佳的症状管理， 在癌症患者的整个疾病过程中，减轻症状负担并最大限度地提高生活质量。 癌症支持性治疗（SCC）团队是多学科团队，专注于预防和治疗癌症。 管理癌症的不利影响及其在癌症连续体中的治疗 从诊断到治疗的经验。这些团队通常缺乏资源来查看所有 癌症患者，需要优先考虑最需要的患者，通常依赖肿瘤医生， 转介。然而，肿瘤科医生往往过于专注于治愈癌症，而不是治疗其症状。作为 因此，SCC服务通常是偶然获得的，即使在可用的情况下，通常在癌症轨迹的后期。到 提高对SCC需求的认识，并确定癌症患者的症状负担， 管理和护理，我们建议建立自然语言处理（NLP）方法，可以 自动从非结构化叙述中提取症状信息。拟议的系统将利用 神经网络和建立在最先进的信息提取方法。为了实现我们的目标，我们将 创建前列腺癌和弥漫性大B细胞淋巴瘤大型队列的临床记录数据集 在西雅图癌症护理联盟（SCCA）和亨斯迈癌症研究所（HCI）接受治疗的DLBCL患者 2015年1月1日至2020年1月1日之间。我们把这两种类型的癌症作为两种非常不同的例子， 常见的癌症类型我们建议用临床叙述中记录的症状负担来表示， 可概括的帧表示，捕获细粒度的细节，包括存在/不存在， 与患者症状相关的状态、严重程度、特征、持续时间、频率和解剖结构信息。我们 将使用主动学习来创建一个具有症状框架注释的多样化和代表性的黄金标准， 训练和测试所提出的基于神经的NLP方法。生成的所有模型及其实现 在这个项目的执行过程中，将作为开源资源与社区共享。后 项目的成功完成，开发的NLP方法将被集成到信息访问中 SCCA和HCI临床储存库的方法。