Akademik Veri Yönetim Sistemi
6th International Conference on Data Science and Applications (ICONDATA'24), Priştine, Kosova, 2 - 06 Eylül 2024, ss.38-44, (Tam Metin Bildiri)
Large language models (LLMs) are artificial intelligence systems focusing on natural language processing using deep learning techniques. These models are often used in processes such as understanding, generating, and analyzing texts. Recently, these models have been used in the creation of decision support systems for issues such as medical diagnosis and analysis of clinical notes in healthcare. When patients apply to healthcare institutions, International Classification of Diseases (ICD) codes and summary epicrisis information for their complaints are manually entered into the system by doctors. ICD code is the general name given to the international classification system of patients. Doctors may select incomplete or generic codes while doing this. To improve this situation, we wanted to test whether the ICD codes of patient electronic health records (EHR) can be determined automatically. For this purpose, it was analyzed whether the ICD codes entered the system based on the complaints of patients aged 0-15 years who applied to the pediatrics polyclinic of a private hospital in Samsun/Turkey in the last year can be identified by artificial intelligence. 2422 data were used in the study, and since BDM contains information about ICD codes, all data was used as test data. In the study conducted using ChatGPT and Co-pilot, ICD codes entered the system by doctors in line with the summary epicrisis information and ICD codes obtained from artificial intelligence were compared. ChatGPT showed 36.3% accuracy in the accuracy matrix, and Co-pilot showed 46.7% accuracy. The results show that for AI-based systems to be a potential tool for determining ICD codes, physicians should write EHRs more comprehensively and indicate additional findings obtained from the examination in the records.