Paper Type

Complete

Abstract

LLMs are increasingly embedded in EHR systems, transforming how healthcare organizations manage clinical information and support care delivery. Despite rapid technical progress, a systematic understanding of how LLMs are operationalized within healthcare information systems remains limited. This study presents a systematic literature review of 41 peer-reviewed studies published between 2020 and 2024 that examine LLM-enabled applications in EHR contexts. Following PRISMA guidelines, the review identifies six dominant application domains: Clinical Text Generation, Clinical Decision Support and Prediction, Disease Identification, Information Extraction, Message Prioritization, and Redundancy Analysis. In addition, five categories of LLM architectures and deployment approaches are analyzed, including encoder-only, decoder-only, encoder–decoder, clinical/domain-specific, and multilingual models. Building on this synthesis, the study proposes a conceptual framework linking LLM capabilities and model types to specific healthcare information system tasks and clinical workflow requirements. This framework offers structured guidance for future research and system design in AI-enabled healthcare environments.

Paper Number

1259

Comments

SIGHealth

Download

Share

COinS
 
Aug 15th, 12:00 AM

Exploring the Role of Large Language Models in Electronic Health Records: A Systematic Literature Review

LLMs are increasingly embedded in EHR systems, transforming how healthcare organizations manage clinical information and support care delivery. Despite rapid technical progress, a systematic understanding of how LLMs are operationalized within healthcare information systems remains limited. This study presents a systematic literature review of 41 peer-reviewed studies published between 2020 and 2024 that examine LLM-enabled applications in EHR contexts. Following PRISMA guidelines, the review identifies six dominant application domains: Clinical Text Generation, Clinical Decision Support and Prediction, Disease Identification, Information Extraction, Message Prioritization, and Redundancy Analysis. In addition, five categories of LLM architectures and deployment approaches are analyzed, including encoder-only, decoder-only, encoder–decoder, clinical/domain-specific, and multilingual models. Building on this synthesis, the study proposes a conceptual framework linking LLM capabilities and model types to specific healthcare information system tasks and clinical workflow requirements. This framework offers structured guidance for future research and system design in AI-enabled healthcare environments.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.