LATTE: Label-efficient incident phenotyping from longitudinal electronic health records.
LATTE: Label-efficient incident phenotyping from longitudinal electronic health records.
复制标题
DOI：
10.1016/j.patter.2023.100906
复制
发表时间：
2024-01-12
期刊：
影响因子：
6.5
通讯作者：
Cai, Tianxi
中科院分区：
文献类型：
--
作者：
Wen, Jun;Hou, Jue;Bonzel, Clara -Lea;Zhao, Yihan;Castro, Victor M.;Gainer, Vivian S.;Weisenfeld, Dana;Cai, Tianrun;Ho, Yuk-Lam;Panickan, Vidul A.;Costa, Lauren;Hong, Chuan;Gaziano, J. Michael;Liao, Katherine P.;Lu, Junwei;Cho, Kelly;Cai, Tianxi
电子健康记录（EHR）数据越来越多地用于支持真实世界的证据研究，但由于缺乏临床事件的精确时间而受到限制。在这里，我们提出了一个标签有效的事件表型（LATTE）算法，以准确地注释纵向EHR数据的临床事件的时间。通过利用预先训练的语义嵌入，LATTE选择预测特征，并通过访问注意力学习将其信息压缩到纵向访问嵌入中。LATTE对目标事件和访问嵌入之间的顺序依赖性进行建模，以获得计时。为了提高标记效率，LATTE从未标记的患者中构建纵向银标准标签来进行半监督训练。LATTE用于评估2型糖尿病、心力衰竭和多发性硬化症复发的发病情况。LATTE在提供高预测可解释性的同时，始终实现了对基准方法的实质性改进。事件发生的时间有助于发现类风湿性关节炎患者心力衰竭的危险因素。提出了一种事件表型分析方法来识别临床事件的时间，我们通过利用EHR嵌入和预测替代物来实现标签效率。结果有助于评估类风湿关节炎患者的心脏风险电子健康记录（EHR）在常规临床护理期间收集的数据越来越多地被转化和临床研究人员用于解决各种问题，例如识别疾病或表型之间的关联，预测疾病风险或预后，或支持治疗的安全性和有效性。这些研究的可行性依赖于从EHR数据中精确推断临床事件的时间和顺序，以定义基线资格和患者结局。基于规则的提取方法可能不准确，现有的机器学习方法通常需要大规模的标签进行训练。识别临床事件发生时间的更好方法可以帮助扩大EHR数据的使用，以解决重要的医学问题，并可以提高分析结果的质量。一个标签效率的方法，LATTE，提出了从纵向电子健康记录中识别临床事件的时间。它在识别2型糖尿病、心力衰竭和多发性硬化症复发的发病方面实现了显著改善的性能。LATTE具有很强的跨站点可移植性，并且通过指示驱动预测的重要功能和访问量来高度解释。
Electronic health record (EHR) data are increasingly used to support real-world evidence studies but are limited by the lack of precise timings of clinical events. Here, we propose a label-efficient incident phenotyping (LATTE) algorithm to accurately annotate the timing of clinical events from longitudinal EHR data. By leveraging the pre-trained semantic embeddings, LATTE selects predictive features and compresses their information into longitudinal visit embeddings through visit attention learning. LATTE models the sequential dependency between the target event and visit embeddings to derive the timings. To improve label efficiency, LATTE constructs longitudinal silver-standard labels from unlabeled patients to perform semi-supervised training. LATTE is evaluated on the onset of type 2 diabetes, heart failure, and relapses of multiple sclerosis. LATTE consistently achieves substantial improvements over benchmark methods while providing high prediction interpretability. The event timings are shown to help discover risk factors of heart failure among patients with rheumatoid arthritis. An incident phenotyping method is proposed to identify timings of clinical events We achieve label efficiency by exploiting EHR embeddings and predictive surrogates Model is validated on incident type 2 diabetes, heart failure, and multiple sclerosis Results facilitate assessing cardiac risks among patients with rheumatoid arthritis Electronic health record (EHR) data collected during routine clinical care are increasingly used by translational and clinical researchers to address a variety of questions, such as identifying associations between diseases or phenotypes, predicting disease risk or prognosis, or supporting the safety and efficacy of treatments. The feasibility of these studies relies on precisely inferring the timing and ordering of clinical events from EHR data to define baseline eligibility and patient outcomes. Rule-based extraction methods can be inaccurate, and existing machine-learning approaches generally require large-scale labels for training. Better methods for identifying the timing of clinical events could help expand the use of EHR data to address important medical questions and could improve the quality of the resulting analyses. A label-efficient method, LATTE, is proposed to identify the timings of clinical events from longitudinal electronic health records. It achieves significantly improved performance in identifying the onset of type 2 diabetes, heart failure, and relapses of multiple sclerosis. LATTE has strong cross-site portability and is highly interpretable by indicating the important features and visits that drive the predictions.
登录
查看更多内容
影响因子：
8.4
作者：
Eisenhauer, E. A.;Therasse, P.;Verweij, J.
通讯作者：
Verweij, J.
通过腹腔镜辅助结肠切除术与开放式结肠切除术管理，早期结肠癌患者电子健康记录中5年生存的临床证据的时间趋势。
影响因子：
13.8
作者：
通讯作者：
--
影响因子：
1.9
作者：
Hou, Jue;Chan, Stephanie F.;Wang, Xuan;Cai, Tianxi
通讯作者：
Cai, Tianxi
DOI：
10.1093/jamia/ocaa079
发表时间：
2020-08-01
影响因子：
6.4
作者：
Ahuja, Yuri;Zhou, Doudou;Cai, Tianxi
通讯作者：
Cai, Tianxi
使用结构化临床/行政数据检测肺和大肠癌复发，以实现结果和人群健康管理。
影响因子：
3
作者：
Hassett MJ;Uno H;Cronin AM;Carroll NM;Hornbrook MC;Ritzwoller D
通讯作者：
Ritzwoller D