Comparing Rule-Based and Deep Learning Models for Patient Phenotyping
MIT Critical Data, Laboratory for Computational Physiology（麻省理工学院关键数据与计算生理学实验室）；Harvard John A. Paulson School of Engineering and Applied Sciences（哈佛大学约翰·A·保尔森工程与应用科学学院）；Massachusetts Institute of Technology（麻省理工学院）；Harvard T.H. Chan School of Public Health（哈佛大学陈曾熙公共卫生学院）；Philips Research North America（飞利浦北美研究院）；Beth Israel Deaconess Medical Center（贝斯以色列女执事医疗中心）；Massachusetts General Hospital（麻省总医院）；Tufts University School of Medicine（塔夫茨大学医学院）；University of Massachusetts（马萨诸塞大学）；Washington University School of Medicine（华盛顿大学医学院）
Sebastian Gehrmann, Franck Dernoncourt, Yeran Li, Eric T. Carlson, Joy T. Wu, Jonathan Welt, John Foote, Edward T. Moseley, David W. Grant, Patrick D. Tyler, Leo Anthony Celi
Objective: We investigate whether deep learning techniques for natural language processing (NLP) can be used efficiently for patient phenotyping. Patient phenotyping is a classification task for determining whether a patient has a medical condition, and is a crucial part of secondary analysis of healthcare data. We assess the performance of deep learning algorithms and compare them with classical NLP approaches.
Materials and Methods: We compare convolutional neural networks (CNNs), n-gram models, and approaches based on cTAKES that extract pre-defined medical concepts from clinical notes and use them to predict patient phenotypes. The performance is tested on 10 different phenotyping tasks using 1,610 discharge summaries extracted from the MIMIC-III database.
Results: CNNs outperform other phenotyping algorithms in all 10 tasks. The average F1-score of our model is 76 (PPV of 83, and sensitivity of 71) with our model having an F1-score up to 37 points higher than alternative approaches. We additionally assess the interpretability of our model by presenting a method that extracts the most salient phrases for a particular prediction.
Conclusion: We show that NLP methods based on deep learning improve the performance of patient phenotyping. Our CNN-based algorithm automatically learns the phrases associated with each patient phenotype. As such, it reduces the annotation complexity for clinical domain experts, who are normally required to develop task-specific annotation rules and identify relevant phrases. Our method performs well in terms of both performance and interpretability, which indicates that deep learning is an effective approach to patient phenotyping based on clinicians' notes.
LATTE: Label-efficient Incident Phenotyping from Longitudinal Electronic Health Records
Harvard Medical School（哈佛医学院）；VA Boston Healthcare System（VA波士顿医疗系统）；University of Minnesota（明尼苏达大学）；Harvard University（哈佛大学）；Mass General Brigham（麻省总医院布莱根医疗系统）；Brigham and Women’s Hospital（布莱根妇女医院）；Duke University（杜克大学）；Harvard T.H. Chan School of Public Health（哈佛大学陈曾熙公共卫生学院）
A Self-Supervised Learning-based Approach to Clustering Multivariate Time-Series Data with Missing Values (SLAC-Time): An Application to TBI Phenotyping
University of Arizona（亚利桑那大学）；University of Cincinnati（辛辛那提大学）；Virginia Tech（弗吉尼亚理工大学）
Generalized and Transferable Patient Language Representation for Phenotyping with Limited Data
The University of Texas Health Science Center at Houston（德克萨斯大学休斯顿健康科学中心）；McGovern Medical School, The University of Texas Health Science Center at Houston（德克萨斯大学休斯顿健康科学中心麦戈文医学院）
Interpretable phenotyping of Heart Failure patients with Dutch discharge letters
MOX - Modelling and Scientific Computing Lab, Department of Mathematics, Politecnico di Milano, Milano, Italy（数学系模拟能力与科学计算实验室，米兰理工大学，意大利）；Department of Cardiology, Amsterdam University Medical Center, University of Amsterdam, Amsterdam, The Netherlands（心脏病学部，阿姆斯特丹大学医学中心，阿姆斯特丹，荷兰）；Department of Medical Informatics, Amsterdam University Medical Center, University of Amsterdam, Amsterdam, The Netherlands（医学信息学部，阿姆斯特丹大学医学中心，阿姆斯特丹，荷兰）；Amsterdam Public Health, Digital Health, Personalized Medicine, Amsterdam, The Netherlands（阿姆斯特丹公共卫生，数字健康，个性化医学，荷兰阿姆斯特丹）；Amsterdam Cardiovascular Sciences, Atherosclerosis and Aortic Disease, Cardiomyopathy and Arrhythmia, Amsterdam, The Netherlands（阿姆斯特丹心血管科学，动脉粥样硬化和主动脉疾病，心肌病和心律失常，荷兰阿姆斯特丹）；Amsterdam Reproduction & Development, Amsterdam, The Netherlands（阿姆斯特丹生殖与发育，荷兰阿姆斯特丹）；Department of Epidemiology and Data Science, Amsterdam University Medical Center, University of Amsterdam, Amsterdam, The Netherlands（流行病学与数据科学部，阿姆斯特丹大学医学中心，阿姆斯特丹，荷兰）；Department of Clinical Science and Education, Södersjukhuset, Karolinska Institutet, Stockholm, Sweden（临床科学与教育部，索德斯医院，卡罗林斯卡研究所，斯德哥尔摩，瑞典）
Contextual Phenotyping of Pediatric Sepsis Cohort Using Large Language Models
Department of Surgery, Duke university school of medicine（杜克大学医学学院外科部）；Department of Anesthesiology, Duke university school of medicine（杜克大学医学学院麻醉科）；Indian Institute of Technology, Goa（印度去邦理工学院）；Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre（全球健康研究所，BC儿童医院和BC妇女医院及健康中心）；Department of Anesthesia, Pharmacology & Therapeutics, University of British Columbia（不列颠哥伦比亚大学麻醉学、药理学与治疗学系）；BC Children’s Hospital Research Institute, BC Children’s Hospital（BC儿童医院研究学院，BC儿童医院）；Department of Pediatrics, University of British Columbia（不列颠哥伦比亚大学儿科系）；Walimu, Kampala, Uganda（乌干达坎帕拉沃尔imu）；Department of Community Health, Mbarara University of Science and Technology（马巴拉拉大学科学与技术学院社区健康系）
A Cross-institutional Evaluation on Breast Cancer Phenotyping NLP Algorithms on Electronic Health Records
University of Minnesota（明尼苏达大学）；Mayo Clinic（梅奥诊所）
目的：临床大语言模型的通用性在模型开发过程中往往被忽视。本研究通过乳腺癌表型提取任务评估了基于BERT的临床NLP模型在不同临床环境中的通用性。材料与方法：从明尼苏达大学和梅奥诊所的电子健康记录中收集了两个乳腺癌患者的临床语料库，并按照相同的指南进行注释。我们开发了三种类型的NLP模型（即，条件随机场、双向长短期记忆和CancerBERT）从临床文本中提取癌症表型。评估了模型在不同测试集上的泛化能力，这些测试集具有不同的学习策略（模型转移与本地训练）。实体覆盖率得分与模型性能的相关性进行了评估。结果：我们手动注释200和161临床文件在UMN和MC，分别。研究发现，两个研究所的语料库中目标实体之间的相似度高于整体语料库。CancerBERT模型在两个临床机构的独立测试集和排列测试集上获得了最好的性能。在一个研究所开发并在另一个研究所进一步微调的CancerBERT模型与基于本地数据开发的模型相比，实现了合理的性能（微F1：0.925 vs 0.932）。结论：结果表明，在三种临床NLP模型中，CancerBERT模型具有最好的学习能力和泛化能力。模型的通用性与语料库中目标实体的相似性有关。
A Corpus for Detecting High-Context Medical Conditions in Intensive Care Patient Notes Focusing on Frequently Readmitted Patients
Massachusetts Institute of Technology（麻省理工学院）；University of Massachusetts（马萨诸塞大学）；Harvard T.H. Chan School of Public Health（哈佛大学陈曾熙公共卫生学院）；Massachusetts General Hospital（麻省总医院）；Tufts University School of Medicine（塔夫茨大学医学院）；Beth Israel Deaconess Medical Center（贝斯以色列女执事医疗中心）；Washington University School of Medicine（华盛顿大学医学院）；Philips Research North America（飞利浦北美研究院）；Harvard University（哈佛大学）；Adobe Research（Adobe研究院）
Comparing representations of long clinical texts for the task of patient note-identification
Inria（法国国家信息与自动化研究所）；Inserm（法国国家健康与医学研究院）；Data Science Platform, Imagine Institute（伊马金研究所数据科学平台）；Hôpital Universitaire Necker-Enfants Malades（内克尔-恩范茨-马拉德大学医院）；Assistance Publique - Hôpitaux de Paris (APHP)（巴黎公共医疗援助医院集团）；Université de Lorraine（洛林大学）；CNRS（法国国家科学研究中心）；LORIA（洛林计算机科学与应用实验室）；INESC-ID（里斯本大学学院计算机与系统工程研究中心）；Instituto Superior Técnico（里斯本高等技术学院）；Universidade de Lisboa（里斯本大学）
Conversational Assistants to support Heart Failure Patients: comparing a Neurosymbolic Architecture with ChatGPT
Department of Computer Science（计算机科学系）；Department of Biomedical and Health Information Sciences（生物医学与健康信息科学系）；Department of Medicine（医学系）；Department of Communications（通讯系）
An evaluation of GPT models for phenotype concept recognition
Rare Care Centre（罕见病护理中心）；Perth Children's Hospital（珀斯儿童医院）；Telethon Kids Institute（Telethon儿童研究所）；School of Electrical Engineering, Computing and Mathematical Sciences, Curtin University（科廷大学电气工程、计算与数学科学学院）；SingHealth Duke-NUS Institute of Precision Medicine（新加坡保健服务集团杜克-新加坡国立大学精准医学研究所）；Lawrence Berkeley National Laboratory（劳伦斯伯克利国家实验室）；Western Australian Register of Developmental Anomalies（西澳大利亚发育异常登记处）；King Edward Memorial Hospital（爱德华国王纪念医院）；Faculty of Health and Medical Sciences, University of Western Australia（西澳大利亚大学健康与医学科学学院）；University of Colorado Anschutz Medical Campus（科罗拉多大学安舒茨医学校区）；The Jackson Laboratory for Genomic Medicine（杰克逊基因组医学实验室）；Institute for Systems Genomics, University of Connecticut（康涅狄格大学系统基因组学研究所）
