Psychosis Relapse Prediction Leveraging Electronic Health Records Data and Natural Language Processing Enrichment Methods

• DOI: 10.3389/fpsyt.2022.844442
• Publication Date: 2022-04-11T11:32:45Z
• Authors: Dong Yun Lee; Chungsoo Kim; Seongwon Lee; Sang Joon Son; Sun-Mi Cho; Yong Hyuk Cho; Jaegyun Lim; Rae Woong Park
• Keywords: Health records

Background Identifying patients at a high risk of psychosis relapse is crucial for early interventions. A relevant psychiatric clinical context often recorded in notes; however, the utilization unstructured data remains limited. This study aimed to develop psychosis-relapse prediction models using various types notes and structured data. Methods Clinical were extracted from electronic health records Ajou University Medical Center South Korea. The population included with psychotic disorders, outcome was within 1 year. Using only data, we developed an initial model, then three natural language processing (NLP)-enriched (psychological tests, admission notes, nursing assessment) one complete model. Latent Dirichlet Allocation used cluster into similar topics. All applied least absolute shrinkage selection operator logistic regression algorithm. We also performed external validation another hospital database. Results total 330 included, 62 (18.8%) experienced relapse. Six predictors model 10 additional topics added enriched models. derived all showed highest value area under receiver operating characteristic (AUROC = 0.946) internal validation, followed by based on psychological test assessments, (0.902, 0.855, 0.798, 0.784, respectively). assessment note, AUROC 0.616. Conclusions NLP-enrichment method. Models more effective than suggesting importance prediction.