JMIR Publications

ABSTRACT

Background:

In the field of obstetrics, electronic medical records capture crucial information about pregnant women, from pregnancy and delivery to postpartum recovery. This information is of vital importance for obstetrics-related research. However, the diagnostic descriptions used in electronic medical records exhibit diversity and lack of standardization, making data aggregation and analysis from multiple sources highly complex and presenting challenges for medical research. The recent advancements of ChatGPT showcase its proficiency in comprehending and generating human-like text, indicating significant potential for text extraction and standardization tasks in the medical domain.

Objective:

The study aims to utilize ChatGPT to mine and explore real-world obstetric data and to create a preliminary knowledge graph of obstetric diagnostic terminologies.

Methods:

To achieve our objective, we employed a three-step approach. Firstly, we extracted obstetric diagnostic descriptions from electronic medical records. Next, we implemented a composite strategy that integrated ChatGPT models and similarity-based methods for further processing. Furthermore, we explored and validated four different prompting techniques to identify the most effective approach for standardizing diagnostic description.

Results:

Upon conducting our experiments, we achieved promising results. The accuracy of our method was found to be competitive with the BERT model, as it achieved an impressive F1-score of 0.923. During the process, we successfully categorized 1100 diagnostic terms into 107 distinct subcategories using clustering algorithms. This categorization formed a preliminary knowledge graph.

Conclusions:

Our study demonstrates the effectiveness of utilizing ChatGPT for standardization of obstetric diagnostic descriptions from real-world data. Standardizing diagnostic descriptions enhances the precision and efficiency of obstetric diagnosis. By creating a preliminary knowledge graph with distinct subcategories, we contribute to the effective standardization and classification of large-scale real-world medical data. Overall, our research aims to offer valuable information for future obstetric research. Clinical Trial: The study was approved by the People’s Hospital of the Guangxi Zhuang Autonomous Region in China (Ref. No. KT-KJT-2021-67), and registered in ChiCTR under identifier ChiCTR2300072225.