Page 32 - AIH-1-1

Artificial Intelligence in Health NLP in EHR

may be problematic when applied to other languages. based on the users’ purposes for information extraction.
The data utilized in these articles encompassed a variety If the user’s goal is to search for keywords only, then
of formats, including structured, unstructured, semi- the phonetic level suffices. When the search extends to
structured, clinical notes, pathological reports, radiology identifying similar terminologies to the entered keyword,
reports, visit or progress notes, EHR, and EMR. The the semantic level becomes necessary. For decision-making
focus of the articles often revolved around symptoms within and beyond the immediate context, the discourse
and individual diseases such as cancer, peripheral arterial and pragmatic levels are required, respectively. The
disease, geriatric syndrome, rheumatoid arthritis, and subsequent discussion delves into the operations carried
diabetes. However, there is a recognized need for a system out at each NLP level.
that thoroughly analyzes a patient’s health records,
empowering physicians to make decisions about their 4.1. Phonetic level
patients’ health. The phonetic level operates by searching for words based
on their sounds. In this process, the user enters the
4. Proposing a model for EHRs expected word for the search. If the sound of the entered
The seven levels of NLP encompass a spectrum from word matches with any word stored in the EHR, relevant
phonetic to pragmatic, progressing from general to more information related to that word is displayed. This level
specific aspects of information retrieval. At the phonetic becomes essential when the sole objective is to perform a
level, the focus is on identifying words based on their search operation.
sounds, while the morphological level involves searching 4.2. Sematic level
for suffixes, prefixes, or other additions to a word.
Advancing to the lexical, syntactic, and semantic levels, the The perspectives of two physicians may vary, and so too
aim is to discern grammar, meaning, and alternative words. can their choice of words. Consequently, a simple keyword
The discourse and pragmatic levels involve determining search may prove inadequate in such cases. In these
intra- and interrelationships among sentences, predicting cases, where the goal is to search for terminologies that
contextual elements both inside and outside the immediate are similar, the semantic level becomes important. For
context. Based on this understanding, we propose an NLP instance, if a physician is searching for a patient’s blood
model that collaborates with EHR. This model consists pressure, variations in terminologies such as “high BP” or
of four NLP levels: phonetic, semantic, discourse, and “hypertension” should be retrieved and displayed from the
pragmatic (Figure 2). The selection of these four levels is records.

Figure 2. The proposed model for EHR.
Abbreviations: EHR: Electronic health records; NLP: Natural language processing.

Volume 1 Issue 1 (2024) 26 https://doi.org/10.36922/aih.2147