Scholarlink

Purpose of Review <p>This scoping review focused on the use of large language models (LLMs) to extract clinically-relevant heart failure (HF) information from electronic health records (EHRs).</p> Recent Findings <p>We identified 10 studies that focused on the ability of LLMs to extract HF data from EHRs and to clinically use the data. While early demonstrations of extraction accuracy and clinical prediction endpoints were encouraging, some studies offered limited descriptions of baseline characteristics and most studies had clinical endpoints that lacked specificity relevant to heart failure phenotype, goal-directed medical therapy, medication titration, and other important evidence-based domains known to impact clinical outcomes in heart failure.</p> Summary <p>Even with the application of a comprehensive search strategy, relatively few studies were identified meeting the pre-specified criteria for inclusion in this review. All included studies that met criteria were retrospective with some case control designed studies. Additional LLM studies are needed to better define how heart failure diagnosis, prevention, management, and treatment can benefit from the use of large&#xa0;language models associated with electronic medical records. This is particularly true for areas of interest where LLM analytics may be applied (e.g., early diagnosis, HF phenotype and treatment variation, medication adherence, medication titration) to further advance the evidence-base for diagnosis and phenotype-specific treatment, deepen understanding barriers to treatment and management, and close adoption gaps widened by variations in clinician behavior.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Use of Large Language Models to Extract Heart Failure Information from Electronic Health Records: A Scoping Review

  • 期刊论文

  • 发表日期：

    2026年4月22日

查看全文
  • Jocelyn Carter,
  • Nancy Anoruo,
  • Narmeen Rehman,
  • Yu Otaki,
  • Melis Lydston,
  • Harry B. Burke

摘要

Purpose of Review

This scoping review focused on the use of large language models (LLMs) to extract clinically-relevant heart failure (HF) information from electronic health records (EHRs).

Recent Findings

We identified 10 studies that focused on the ability of LLMs to extract HF data from EHRs and to clinically use the data. While early demonstrations of extraction accuracy and clinical prediction endpoints were encouraging, some studies offered limited descriptions of baseline characteristics and most studies had clinical endpoints that lacked specificity relevant to heart failure phenotype, goal-directed medical therapy, medication titration, and other important evidence-based domains known to impact clinical outcomes in heart failure.

Summary

Even with the application of a comprehensive search strategy, relatively few studies were identified meeting the pre-specified criteria for inclusion in this review. All included studies that met criteria were retrospective with some case control designed studies. Additional LLM studies are needed to better define how heart failure diagnosis, prevention, management, and treatment can benefit from the use of large language models associated with electronic medical records. This is particularly true for areas of interest where LLM analytics may be applied (e.g., early diagnosis, HF phenotype and treatment variation, medication adherence, medication titration) to further advance the evidence-base for diagnosis and phenotype-specific treatment, deepen understanding barriers to treatment and management, and close adoption gaps widened by variations in clinician behavior.