SPELL: A scalable NLP method using regular expressions and large language models for clinical information extraction.

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning (ML)
• Clinical Informatics

Background:

• Electronic Health Records (EHRs) contain valuable unstructured clinical narratives.
• Analyzing unstructured EHR data at scale is challenging for research and decision-making.
• SPELL (Snippet-Primed rEgex LLM Pipeline) was developed to address this challenge.

Purpose of the Study:

• To develop a scalable NLP workflow for extracting structured variables from large collections of clinical narratives.
• To combine regular-expression-based snippet retrieval with locally hosted large language model (LLM) inference.
• To enable accurate and computationally efficient extraction of information from unstructured EHR data.

Main Methods:

• SPELL retrieves text snippets using task-specific regular expressions and applies LLM inference on these snippets.
• All processing is conducted within institutional computing environments.
• Accuracy and efficiency were evaluated on obstetric tasks (blood loss, estimated due date, HELLP syndrome) and a generalizability task (ventricular tachycardia detection).

Main Results:

• SPELL processed 31 million clinical notes, demonstrating scalability.
• Snippet-based inference reduced processing time by 71-87% compared to full-document LLM inference.
• High accuracy was achieved: 98% for blood loss, 92% for estimated due date, and 94% F1 for HELLP syndrome.

Conclusions:

• A hybrid regex-snippet-LLM pipeline enables accurate and computationally efficient extraction from unstructured EHR narratives.
• SPELL offers a scalable solution for leveraging valuable information within clinical notes.
• This approach enhances the utility of EHR data for research and clinical decision support.