Combining structured and unstructured data for predictive models: a deep learning approach

摘要: Background: The broad adoption of Electronic Health Records (EHRs) provides great opportunities to conduct health care research and solve various clinical problems in medicine. With recent advances success, methods based on machine learning deep have become increasingly popular medical informatics. However, while many studies utilize temporal structured data predictive modeling, they typically neglect potentially valuable information unstructured notes. Integrating heterogeneous types across EHRs through techniques may help improve the performance prediction models. Methods: In this research, we proposed 2 general-purpose multi-modal neural network architectures enhance patient representation by combining sequential notes with data. fusion models leverage document embeddings for long note documents either convolutional or short-term memory networks model signals, one-hot encoding static representation. concatenated is final which used make predictions. Results: We evaluate 3 risk tasks (i.e., in-hospital mortality, 30-day hospital readmission, length stay prediction) using derived from publicly available Medical Information Mart Intensive Care III dataset. Our results show that data, outperform other only. Conclusions: learn better helps reduce errors.