Making nursing visible: AI-assisted standardization of electronic health record interventions using generative pre-trained transformer models and retrieval-augmented generation process

Abstract

Background

Nurses provide essential care that significantly impacts patient outcomes. However, nursing care is rarely examined due to a lack of standardized nurse-generated data suitable for large-scale analysis.

Purpose

To test the performance of retrieval-augmented generation (RAG) and generative pre-trained transformer (GPT) models in standardizing electronic health records nursing intervention terms.

Methods

The process involved: (a) manually mapping a subset of local terms to standardized terms to create a “gold standard” for comparison, (b) implementing and assessing the RAG pipeline with GPT models for standardization, and (c) incorporating the output into an AI-assisted manual mapping workflow.

Discussion

GPT models were comparable to human mappers but lacked sufficient accuracy for fully automated mapping.

Conclusion

Although GPT models are not yet reliable for full automation, they can reduce human workload and streamline the standardization process. This advancement represents an important step toward making nursing contributions visible through data-driven evaluation of nursing practices.