JMIR Publications

ABSTRACT

Background:

Extracting accurate medication information from Thai hospital records presents challenges due to the narrative style of medical notes, which often combine Thai and English terminology. This study aimed to address these difficulties by leveraging ontology-based named entity recognition (NER) and pre-trained transformer models.

Objective:

The primary objective was to investigate the effectiveness of ontology-based NER and pre-trained transformer models in extracting medication information from unstructured Thai hospital records, thereby improving data standardization and interoperability in Thai healthcare.

Methods:

An annotated dataset comprising 90 discharge summaries was developed, based on SNOMED-CT and FHIR standards for medical terminology. Three deep learning models—BioClinicalBERT, ClinicalBERT, and Microsoft BiomedNLP—were trained and subsequently evaluated on this dataset to assess their performance in entity recognition.

Results:

Among the models tested, ClinicalBERT demonstrated the highest overall F1-score in entity recognition. It performed particularly well at identifying drug substances and dosage entity types.

Conclusions:

The findings suggest that ontology-based medication information extraction using transformer-based models holds significant promise for enhancing data standardization and interoperability within the Thai healthcare system. This approach offers a viable solution for overcoming the complexities of unstructured Thai medical notes.