Page 24 - AIH-1-1

Artificial Intelligence in Health NLP in EHR

for patient care or managerial decision-making. Denmark, with distinctive clinical characteristics in clinical care.
Israel, and the Netherlands are the only countries that use Moreover, it reduces methodological heterogeneity in
AI for natural language processing (NLP) to convert text- defining phenotypes, thereby mitigating the obscuration
based data into coded formats. The difficulty associated of biological heterogeneity in research related to allergy,
with manual searches through the health records of asthma, and immunology. The previous studies have
[16]
every patient prompts the research question: How can we advocated for the use of NLP in EHR .
automate the search through EHR? Therefore, we endeavored to investigate information
During emergencies, when physicians need to go retrieval from EHR using different NLP levels. In this
through multiple records, the cognitive load increases, review, we conducted a systematic literature review (SLR) to
delaying the decision-making process. According to the explore the integration of different NLP levels within EHR.
OECD report, Denmark, Israel, and the Netherlands are This article details the use of NLP to extract information
the countries using NLP to convert text data to coded from specific departments or diseases. However, working
data. However, their use of NLP is confined to automatic at a certain level with a particular approach does have
alerts and predictive analytics. This review explores limitations, highlighting the need for innovative ideas to
the application of NLP in keyword searches to display drive further improvement.
information relevant to the searched keyword.
2. Study method
1.1. NLP
We conducted SLR to examine the various levels of NLP
The usability of EHR is affected by the need for increased applicable to EHR. The SLR was conducted according to
navigation . A system is considered usable if it is easy guidelines outlined in Tranfield et al. in three stages.
[15]
[18]
to learn, easy to use, and error-tolerant. Usability metrics
include learnability, the number of trials to reach a certain 2.1. Review planning
performance level, the number of items and sequential The objective of this study was to examine the effects
steps to be memorized, efficiency, time on task, task steps, of various NLP levels on EHR. Therefore, the SLR was
task success, mental effort, error prevention and recovery, designed to include only research papers that addressed
error occurrence rate, and error recovery rate . Issues the use of NLP levels in EHR. The procedure for selecting
[16]
such as poor user design, lack of data security technology, articles for the final review is discussed below.
differences in semantics and data dictionaries, and a lack of
interoperability have impeded EHR implementation. The 2.2. Review procedures
framework for EHR usability, known as TURF (task, user, An illustration of the review procedure is presented
representation, and function) , includes: in Figure 1, with detailed discussions provided in the
[17]
(i) A theory that describes, explains, and predicts following sub-sections.
usability differences.
(ii) A method to objectively define, evaluate, and measure 2.2.1. Selection of papers
usability. We conducted a comprehensive search on PubMed using
(iii) A process designed to incorporate good usability. the keyword “Natural Language Processing” within
(iv) Once fully developed, a potential principle for the timeline of 2016 – 2023. In 2016, countries such as
developing EHR usability guidelines and standards. Australia and Sweden launched their EHR initiative,
TURF defines usability as the extent to which a and India published standards for implementing EHR.
system is useful, usable, and satisfying for intended users Therefore, 2016 was chosen as the base year for searching
to accomplish goals within the work domain through papers. We applied the following filters: “full text,” “-and-,”
specific sequences of tasks. TURF stands for task, and “-free full text-.”
user, representation, and function, which are the four The selection process involved three phases. In the
components that determine the usability of an EHR system. first phase, the author with a background in information
These components are further described as measurements technology assessed titles and abstracts, subsequently
of usability in several case studies. selecting articles based on the following inclusion criteria:
As a substantial amount of valuable clinical information (i) Reporting on NLP in EHR.
is locked in clinical narratives, NLP techniques, as an AI (ii) Using synonyms of EHR, including electronic medical
approach, have been leveraged to extract information records, clinical decision support systems, health
from clinical narratives in EHR. This NLP capability information systems, and clinical notes.
enables automated chart review for identifying patients (iii) Articles written in English.

Volume 1 Issue 1 (2024) 18 https://doi.org/10.36922/aih.2147