Efficient Standardization of Clinical Notes using Large Language Models

TL;DR

This study investigates standardizing clinical notes using a large language model (GPT-4) to address variability in writing style, abbreviations, and formatting. The method expands abbreviations, fixes spelling and grammar, replaces non-standard terms, and reorganizes notes into canonical sections, producing JSON-ready outputs suitable for ontology mapping and HL7 FHIR interoperability. On a dataset of 1618 de-identified neurology notes, the approach achieved quantitative improvements and a subset showed no loss of clinical content, enabling semi-structured data extraction of medications and signs/symptoms. The authors discuss limitations such as copy-paste practices, data carry-forward, privacy considerations, and scalability, highlighting that LLM-based standardization should be complemented by systemic process improvements.

Abstract

Clinician notes are a rich source of patient information but often contain inconsistencies due to varied writing styles, colloquialisms, abbreviations, medical jargon, grammatical errors, and non-standard formatting. These inconsistencies hinder the extraction of meaningful data from electronic health records (EHRs), posing challenges for quality improvement, population health, precision medicine, decision support, and research. We present a large language model approach to standardizing a corpus of 1,618 clinical notes. Standardization corrected an average of $4.9 +/- 1.8$ grammatical errors, $3.3 +/- 5.2$ spelling errors, converted $3.1 +/- 3.0$ non-standard terms to standard terminology, and expanded $15.8 +/- 9.1$ abbreviations and acronyms per note. Additionally, notes were re-organized into canonical sections with standardized headings. This process prepared notes for key concept extraction, mapping to medical ontologies, and conversion to interoperable data formats such as FHIR. Expert review of randomly sampled notes found no significant data loss after standardization. This proof-of-concept study demonstrates that standardization of clinical notes can improve their readability, consistency, and usability, while also facilitating their conversion into interoperable data formats.

Figures (7)

• Figure 1: Note Length. Mean note length in characters was 6420 $\pm$ 3691.
• Figure 2: Acronymns and Abbreviations. Normalization expanded a mean of 15.8 $\pm$ 9.1 acronyms and abbreviations per note.
• Figure 3: Grammatical Errors. Note normalization corrected a mean of 4.9 $\pm$ 1.8 grammatical errors per note.
• Figure 4: Slang, Jargon, and Non-Standard Terms. GPT-4 identified and corrected a mean of $3.1 \pm 3.0$ non-standard terms per note. An example of a non-standard term substitution is "feeling blue" $\rightarrow$ "symptoms of depression"
• Figure 5: Spelling Errors. GPT-4 identified and corrected a mean of $3.3 \pm 5.2$ spelling errors per note.