从电子健康记录中准确识别痴呆症的检索增强型大型语言模型框架
medRxiv : the preprint server for health sciences
|February 6, 2026
概括
使用大型语言模型 (LLM) 的新搜索增强生成 (RAG) 框架显著改善了从电子健康记录 (EHR) 中识别痴呆症. 这种先进的方法优于用于准确的痴呆症表型化的传统方法.
科学领域:
- 医疗信息学 医疗信息学
- 医疗保健中的人工智能
- 临床数据科学 临床数据科学
背景情况:
- 从电子健康记录 (EHR) 中准确地确定痴呆症表型对于研究和临床干预至关重要.
- 现有的基于规则和基于关键字的方法与不一致的文档和临床细微差别作斗争.
- 大型语言模型 (LLM) 提供了改善复杂临床数据分析的潜力.
研究的目的:
- 开发和评估一个检索增强生成 (RAG) 框架,用于从EHR中增强痴呆症识别.
- 为了比较RAG-LLM与传统基于规则和关键字过的LLM方法的性能.
- 克服当前方法在捕捉临床细微差别和处理文档变化的局限性.
主要方法:
- 从Mass General Brigham EHR数据组建了一个队列,识别了潜在的痴呆病例.
- 开发并比较了三个痴呆症确定方法:基于规则的,关键字过的LLM和基于RAG的LLM.
- 优化了嵌入模型,检索方法,LLM和每个方法的提示工程的配置.
主要成果:
- 基于RAG的LLM分类器在与基于规则 (F1=0.823) 和关键字过的LLM (F1=0.903) 相比,实现了更高的性能 (F1=0.933).
- 从RAG-LLM管道中排除结构化的ICD代码改善了正预测值 (PPV) 和F1得分.
- 错误分析表明,结构化代码的依赖导致了假阳性,而错过的上下文线索导致了假阴性.
结论:
- 一个基于RAG的LLM管道,特别是没有结构化的ICD代码,显著提高了从EHR数据中确定痴呆症.
- 这种方法超越了传统的基于ICD的规则和关键字过,以改善痴呆症的识别.
- RAG-LLM框架支持痴呆症研究中的先进患者护理,预测建模和风险分析.
相关概念视频
Purpose of Health Records I
1.8K
The vital purpose of health records is to provide a complete and accurate account of a patient's medical history, including communication, diagnostic and therapeutic orders, care planning, research, and quality review.
Here's a breakdown of how health records serve these purposes:
Here's a breakdown of how health records serve these purposes:
1.8K
Purpose of Health Records II
1.4K
Health records serve various essential purposes in the healthcare system. Here are some key purposes:
1.4K
Dementia
580
Dementia is a collective term for cognitive disorders primarily affecting memory, thinking, and reasoning. It is not a specific disease but a syndrome, with Alzheimer's disease being the most common cause, accounting for approximately 60-80% of cases. Other types include vascular dementia, Lewy body dementia, and frontotemporal dementia. Dementia affects millions worldwide, particularly older adults, though it is not a normal part of aging.
The progression of dementia is generally gradual....
The progression of dementia is generally gradual....
580
Electron Orbital Model
72.3K
Orbitals are the areas outside of the atomic nucleus where electrons are most likely to reside. They are characterized by different energy levels, shapes, and three-dimensional orientations. The location of electrons is described most generally by a shell or principal energy level, then by a subshell within each shell, and finally, by individual orbitals found within the subshells.
The first shell is closest to the nucleus, and it has only one subshell with a single spherical orbital called the...
The first shell is closest to the nucleus, and it has only one subshell with a single spherical orbital called the...
72.3K
Retrieval
447
Retrieval is the process of getting information out of memory storage and back into conscious awareness. This ability is essential for daily tasks like brushing hair and teeth, driving to work, and performing job duties. Retrieval occurs in three ways: recall, recognition, and relearning.
Recall involves accessing information without cues, such as during an essay test, where individuals must retrieve facts and concepts from memory unaided. Another example is remembering the name of a colleague...
Recall involves accessing information without cues, such as during an essay test, where individuals must retrieve facts and concepts from memory unaided. Another example is remembering the name of a colleague...
447
Data Reporting and Recording
5.5K
Reporting and recording are crucial in data documentation. The timely, thorough, and accurate documentation of facts is essential when recording patient data. Failure to record findings during an assessment or interpretation of a problem will result in loss of information and make the patient document unreliable. The reader is left with general impressions if the information is not specific. A recording is documenting data of the individual's health information in a traceable, secure, and...
5.5K


