Leveraging large language models to mimic domain expert labeling in unstructured text-based electronic healthcare records in non-english languages

Table 1 Distribution of presenting complaints by data clusters

Presenting Complaints	Structured text data	Unstructured text data (“Others ()”)	Typographical errors
Total RTI Patient	98,904	3828	909
Fever	76,408	1131	246
Cough	53,866	890	143
Fatigue	19,926	33	19
Sore throat	10,897	302	47
Ear pain	9568	162	21
Respiratory Distress	4630	31	16
Non-cardiac chest pain	4077	78	8
URTI	3513	70	-
Crackles	718	249	2
Wheezing	246	5	34
COVID	9	289	423
Other RTI complaints^a	80	1449	117
Total words in the text^b	717,153	64,529	12,117
Total categorized label^c	183,938	4689	1076

This table summarizes the distribution of presenting complaints from patients admitted to the PED. The complaints are categorized into structured text data, unstructured text data (“Others ()”), and typographical errors. The data includes RTI-related complaints like fever, cough, and sore throat, as well as non-respiratory issues such as ear pain and non-cardiac chest pain. The table also presents the total word count and categorized labels extracted through standard filtering methods. The structured text data contains the highest number of RTI complaints, while the unstructured category reflects poorly labeled cases, many of which were identified as RTIs after further analysis. a: Other RTI labels in English are: Flu, Cold, Nasal congestion, Wheezing, Rhonchi, Asthma, Croup, Bronchiolitis, Pneumonia, Febrile convulsion, Lymphadenitis, Tonsillitis, Influenza, Laryngitis, Sputum. b: Total words in the text: The total counts of words within the text data, segmented by data clusters. c: Total categorized label: The number of categorical variables that can be extracted from the content of text data through standard filtering methods

ISSN: 1472-6947