Abstract
Unstructured clinical medical text, as an important part of the electronic health records, is characterized by large quantities and can store substantial disease-related information of patients. Currently, the disease risk assessment model based on the analysis of clinical medical text designs relevant characteristics aiming at certain diseases, and different characteristics are identified from the text using different methods. In this way, changes of disease performance characteristics are difficult to adapt. Furthermore, it is hard to use the risk assessment model in other disease applications. As a result, this paper establishes the universal disease risk assessment model using the data of clinical medical text, conducts the independent study, and extracts disease characteristics from substantial historical data to avoid the limitations designing disease characteristics. First, this paper analyzes the medial clinical text to determine the contents related to the disease characteristics of patients. Second, learn the representation of clinical text with unsupervised learning methods, and study and extract the disease characteristics from the substantial historical data of patients in the convolutional neural network to assess disease risks. Finally, make a contrast experiment of disease risk assessment using the clinical text data of patients with cerebral infarction, patients with pulmonary infection, and patients with coronary atherosclerotic heart disease from the data of a second grade-A hospital in China and related methods. The experiments show that the approach proposed in this paper achieves the disease risk assessment for different diseases with the same structure.