减少临床笔记冗余以高效使用大语言模型
Clinical Note Bloat Reduction for Efficient LLM Use
Jordan L. Cahoon · Chloe Stanwyck · Asad Aali · Rachel Madding · Sulaiman S. Somani · Emma Sun · et al.
中文摘要
临床笔记常因模板、复制粘贴及自动填充产生大量重复文本(笔记膨胀),稀释临床信号并增加大语言模型(LLM)成本。TRACE 利用电子健康记录(EHR)元数据识别模板化及复制内容,无元数据时采用基于频率的去重。在涵盖肝移植、产科及住院人群的四个队列(530 万份笔记)中,通过盲法医生审查及金标准标注评估 TRACE。对比原始与 TRACE 处理笔记在 20 项信息抽取任务及 5 年生存、产后出血、30 天再入院预测中的零样本 LLM 与嵌入分类器表现。结果显示,仅 0.3-6.6% 的移除文本被标记为作者生成,TRACE 捕获了 86% 的标注模板字符。信息抽取 F1 差异在 -0.009 至 +0.004 之间,预测 F1 差异在 -0.011 至 +0.018 之间。在斯坦福健康护理 1,000 名随机患者中,TRACE 减少 47.3% 的图表文本(7.427 亿字符),平均每位患者减少 220,167 个 token。基于 2024 年就诊量,预计三年净节省 100 万至 1358 万美元。TRACE 在保留性能的同时显著减少冗余,利用未充分利用的 EHR 元数据可降低 LLM 推理成本并扩展纵向上下文。
关键要点
- 01临床笔记中的模板化重复文本(笔记膨胀)稀释临床信号并增加 LLM 处理成本。
- 02TRACE 利用 EHR 元数据识别并移除模板化及复制内容,无元数据时采用频率去重。
- 03在 530 万份笔记中,TRACE 捕获 86% 的模板字符,仅 0.3-6.6% 移除文本为作者生成。
- 04信息抽取与预测任务的 F1 分数差异极小,表明 TRACE 在去重的同时保留了关键临床信息。
- 05TRACE 减少 47.3% 的文本量,预计三年可为大型医疗中心节省 100 万至 1358 万美元。
解读
尚无解读。
原始英文摘要
arXiv:2604.16364v3 Announce Type: replace-cross Abstract: Background: Clinical notes contain extensive duplicated text from templates, copy-paste, and auto-populated fields ("note bloat"), diluting clinical signal, limiting longitudinal context, and increasing large language model (LLM) costs. Methods: TRACE removes note bloat using note-level EHR metadata to identify templated and copied content, with frequency-based de-duplication when metadata are unavailable. We evaluated TRACE using blinded physician span review and gold-standard templated-text annotations across four cohorts spanning liver transplant, obstetrics, and inpatient populations at multiple health systems (5.3M notes). We compared zero-shot LLMs and embedding-based classifiers using original and TRACE-processed notes for 20 information extraction tasks and prediction of 5-year survival, postpartum hemorrhage, and 30-day readmission. Results: Only 0.3-6.6% of removed text was flagged as author-generated; TRACE captured 86% of annotated templated characters. Information extraction F1 differences averaged by cohort ranged from -0.009 to +0.004; task-specific prediction F1 differences ranged from -0.011 to +0.018. Among 1,000 randomly sampled Stanford Health Care patients, TRACE reduced chart text by 47.3% (742.7M characters), averaging 220,167 fewer tokens per patient. Using 2024 encounter volumes at a large tertiary academic center and one query per encounter, projected three-year net savings ranged from $1.00M to $13.58M across evaluated model pricing schemes, including initial and annual TRACE processing costs. Conclusion: TRACE substantially reduces clinical note redundancy while preserving information extraction and prediction performance. Underused EHR metadata can reduce LLM inference costs, expand usable longitudinal context, and support scalable clinical AI.