Presentation Information
[POS-24]Numerical representation of clinical diagnostic time series for modeling neurodegenerative disease progression
*Keita Nakao3,1, Yohei Kondo2, Honda Naoki1, Yuichiro Yada1 (1. Nagoya University Graduate School of Medicine (Japan), 2. Center for One Medicine Innovative Translational Rsearch (Japan), 3. Hiroshima University (Japan))
Keywords:
Electronic Health Records,Machine Learning
Recent years have seen a growing adoption of Electronic Health Records (EHRs) in the healthcare industry, which digitize paper-based medical records and allow for the centralized and continuous management of medical data such as patient histories and test results. The utilization of EHRs is expected not only to improve the efficiency of diagnoses and treatments but also to pave the way for analyses using artificial intelligence and machine learning. However, much of the data contained in EHRs is recorded in natural language, making it difficult to handle with conventional probabilistic models. Therefore, in this study, we developed a method to represent natural language clinical diagnoses as a time series of numerical vectors by applying large language models, hierarchical clustering, and support vector machines. We applied the proposed method to a large-scale EHR dataset for Parkinson’s disease and present the results of analyzing disease progression using a continuous-time hidden Markov model.