JMIR Publications

ABSTRACT

Background:

Clinical narratives are vital components of electronic health records (EHRs) in hospital information systems. The adoption of EHRs has led to increased documentation times for hospital staff, resulting in brevity and frequent use of short forms, i.e. all types of abbreviations and acronyms. This potentially hinders comprehension for professionals and patients alike.

Objective:

This review aims to establish an overview of types of short forms that are processed in clinical narratives, and of the natural language processing techniques applied for identification, expansion and disambiguation of short form content.

Methods:

In the databases Web of Science, Embase, MEDLINE, and EMBR, publications that matched the inclusion criteria were searched following PRISMA guidelines for a systematic scoping review. Original, peer-reviewed publications focused on short-from processing in human clinical narratives were included from January 2018 until February 2023. Types of short forms were extracted, in conjunction with assigning for each target objective (identification, expansion, disambiguation) multidimensional research methodologies. Natural language processing study recommendations and study characteristics were sequentially assigned with occurrence rates for evaluation.

Results:

From a total of 6,331 articles, only 14 publications were included in the final analysis. Rule-based approaches were predominantly used for the identification of short forms, string similarity, and vector representations for expansion, and embeddings and deep learning approaches were applied for disambiguation.

Conclusions:

Scope and types of what to classify as a clinical short form were often not explicitly defined in the papers. This is a challenge for reproducibility and for establishing if certain methodologies even qualify to be used for a given type of short form. Analysis of a subset of natural language processing recommendations for determining quality and reproducibility showed only partial fulfillment of those recommendations. Single-character abbreviations are underrepresented in investigations regarding processing of clinical narratives, as well as investigations in languages other than English. For future studies, research on these two topics would be of interest, while type of content descriptions should be included in each paper.