Using Prompt Engineering to Optimize a RAG Pipeline for EHR-Nursing Data Standardization

Standardizing nursing care plan data from electronic health records is critical for interoperability and large-scale research but is often hindered by the heterogeneity of local terminologies. This study evaluates an optimized Retrieval-Augmented Generation (RAG) pipeline using large language models to map local care plan problems, goals, and interventions from two health systems to standardized nursing terminologies. We compared multiple prompt engineering techniques. Structured Prompts combined with Few-Shot Learning achieved the highest performance, with moderate accuracy (45-70%). Although full automation remains out of reach, this approach reduced human workload and supports scalable, human-in-the-loop semantic interoperability of nursing data.