Application of HDBSCAN algorithm for clustering vector representations of text data in phenotypic group identification among rheumatoid arthritis patients receiving biological therapy

Object of the study - text data from electronic medical records (EMR) of patients suffering from rheumatoid arthritis. The aim of the work was to investigate the possibility of using the HDBSCAN algorithm for clustering generalized vector representations of text data obtained using the BERT transform model pre-trained on Russian-language texts in order to divide patients into phenotypic groups based on medical indicators, to assess the applicability of the algorithm for identifying a group with a difficult course of the disease (D2T-patients), and to investigate the possibility of identifying a group of patients with future efficacy/ineffectiveness of biological therapy. Problem Statement: 1) Analyze the data, construct vector representations, prepare for the application of the clustering algorithm. 2) Apply the clustering algorithm and analyze the results obtained. 3) Evaluate the possibility of applying the algorithm using the methods of mathematical statistics. 4) Analyze separately the possibility of selecting a group of patients with future efficacy/ineffectiveness of taking bDMARDs. The volume of the work is 18 pages, there are 7 illustrations, 5 tables. The list of used literature contains 25 titles. Department of Computer Science, National Research University Higher School of Economics, Moscow, 2024.