JMIR Publications

ABSTRACT

Background:

The global nursing shortage, which was exacerbated by heavy workloads and high turnover rates associated with the COVID-19 pandemic, continues to undermine care quality and nurse well-being. Although digital health technologies have enhanced coordination, improved communication, and reduced clinical errors in nursing, they have also increased nurses’ documentation burden. Advancements in large language models (LLMs) and other generative artificial intelligence (GenAI) tools currently facilitate the generation of precise reports from electronic medical records (EMRs), thus streamlining documentation workflows, saving time, and reducing nurses’ workloads. Accordingly, the integration of LLMs into electronic nursing documentation systems warrants exploration.

Objective:

This study explored the integration of an LLM into an in-house nursing information system (NIS) implemented across three hospitals in Taiwan in order to reduce the time and effort required for nursing handover documentation.

Methods:

A multidisciplinary team of nursing specialists and information technology experts at Taipei Medical University restructured the organization’s existing nursing handover documentation process to facilitate interactions with the LLM. Moreover, they developed prompt-based interfaces to generate section-specific content automatically for the nursing handover document. The LLM-integrated NIS was subsequently deployed across three hospitals in Taiwan: Taipei Medical University Hospital (TMUH), Wan Fang Hospital (WFH), and Shuang Ho Hospital (SHH). We then extracted and analyzed NIS log data to compare documentation times before and after LLM implementation, thus quantifying time savings.

Results:

The integration of the LLM into the workflow of nursing handover documentation markedly reduced documentation time. Under the existing workflow, nurses spend 25–35 min per shift completing handover documentation for 5–7 patients, with the documentation time per patient being approximately 5 min. However, with the tailored LLM prompts, the average documentation time per patient decreased to less than 20 s across all three hospitals, with approximately 4.7 min saved per case. Thus, monthly time savings of 2765–2949, 2140–2236, and 3383–3498 h were achieved at TMUH, WFH, and SHH, respectively.

Conclusions:

The integration of the LLM into the NIS enabled the rapid and efficient generation of nursing handover documentation through the use of tailored prompts and EMR data, substantially reducing documentation time. Although the developed GenAI-assisted approach enabled reliable extraction of abnormal radiology and examination values, it occasionally generated hallucinations when processing patient vital signs, highlighting the need for further model accuracy enhancement. Continuous comparative analysis of LLM-generated and nurse-finalized content can inform subsequent model refinements and the establishment of rules for retrieval-augmented generation.