Page 33 - AIH-1-1

Artificial Intelligence in Health NLP in EHR

4.3. Discourse level

Suppose a patient is experiencing symptoms such as a cold
and headache; in this scenario, the system should diagnose
this condition as the flu.
4.4. Pragmatic level
Consider a scenario where a patient is experiencing acid
reflux, which is possibly attributed to previous medications.
In this scenario, the system should conduct a search for the
patient’s previous medication, identify recent medicines,
and assess whether the current symptoms result from the
side effects of any previous medicines or represent a totally
new situation. This NLP level is required to enhance the
decision support capabilities of the system.
5. Limitations and future scope

This study has several limitations:
(i) The literature review relied on articles collected from
the PubMed database. Potential omission of significant
research papers not indexed in the PubMed database Figure 3. Future scope of NLP in EHR.
may exist. Abbreviations: EHR: Electronic health records; NLP: Natural language
(ii) The review period spanned seven years (2016 – 2022). processing.
While not exhaustive, we consider it comprehensive, between two sentences. Therefore, interpretation is heavily
given its coverage of numerous academic journals,
including those focused on NLP, EHR, and related dependent on information retrieval. This review suggests
subjects. that different NLP levels interpret different aspects of
(iii) We adhered to a set of keywords for the literature health records. Hence, using the appropriate level for
review. Any disagreements regarding the inclusion of interpretation is crucial. While English is the naturally
specific articles or keywords were resolved through processed language, future considerations should include
discussion. foreign language translation for international coverage
of health records. In India, Ayurvedic medicine names
The proposed model employs different levels to
interpret different aspects of health records. Unlike are often mentioned in Sanskrit, and sometimes doctors
existing models operating on a single level, the proposed prescribe medications in regional languages. Therefore,
model combines different levels to enhance automation in electronic version translation from regional languages to
the decision-making process. The system offers substantial a unique form and vice versa is necessary. Figure 3 depicts
benefits to both physicians and patients. For doctors, it the combination of these four levels, which will be a topic
provides quick lookup, fast access, and a decision support for future study.
system. Patients benefit from reduced paperwork and
the convenience of not having to carry physical records. 6. Conclusion
Due to the availability of all related records, the patient The present study employed a SLR to investigate the current
is the obvious beneficiary of accurate treatment. This state of research in EHR and NLP. The SLR was conducted
opens avenues for researchers and database managers to on 47 articles, categorized into seven categories based on
explore different ways of managing such vast information, NLP levels. The results of the SLR indicated a significant
thereby improving aspects such as storage, security, and focus on NLP applications within the research community.
privacy. Our review suggests that the information retrieval Notably, this study reveals that a substantial proportion of
process affects the interpretation of information. When
two physicians read the same patient’s history, their the research is related to the semantic level.
perspectives and analyses may differ, forming part of an Acknowledgments
annual review process. Similarly, when NLP tools interpret
EHR, different tasks can be performed, ranging from I would like to thank Dr. Hegler Tissot for his valuable
simple keyword searches to more complex interrelations comments on this SLR.

Volume 1 Issue 1 (2024) 27 https://doi.org/10.36922/aih.2147