Auxiliary Diagnosis of Type 2 Diabetes Complication Based on Text Mining

In recent years, the use of electronic medical records becomes more pervasive. In addition to structured data, part of the unstructured data is also of great value in medical research. How to analyze these electronic documents through automation methods has already become present research focuses. For patients, the information they can provide include basic information, history of present illness (HPI) and Chief Complaint. Our research has tried using such information to identify diabetics with different complications (neurological complications, eye disease, diabetic nephropathy, diabetic foot). Our datasets come from hospital, which include 1622 electronic medical records of patients with type 2 diabetes. Among them, 678 cases were diagnosed non-insulin-dependent diabetes with neurological complications, 359 cases as non-insulin-dependent diabetes with diabetic foot, 316 cases with diabetic nephropathy, the rest 269 cases have eye disease. Our experiment exploit Text-CNN in deep learning technology as our training model, apart from this, we extracted basic information, history of present illness (HPI) and Chief Complaint from these patients as the input of our model for classification, which achieved a relatively good performance (precision: 93.6%, recall: 88.4% and F1: 90.5
^%
). Therefore, the proposed method has been proved to be effective in the classification of common complications of diabetes patients from the EMR of type 2 diabetes patients, so as to carry out an early diagnosis and assessment for these patients.