Abstract: Disease prediction based on electronic medical record generally predicts the disease according to the patient's symptoms, and rarely studies on the time sequence relationship between the diseases. A new representation of electronic medical record is introduced, which considers the context-aware information of medical diseases with time series. Each disease is transformed into a digital vector similar to its "semantics" using Doc2Vec. Based on these vectors, the BiLSTM model is used to predict the potential diseases of elderly patients, which can play an early warning role in diseases of the elderly. Finally, real hospital diagnostic data is used in the experiments, and the results show that the model can effectively predict new diseases of the elderly, and it also has certain stability while ensuring the accuracy of prediction.

Key words:  , contextual, Doc2Vec, Bi-directional long short-term memory(BiLSTM), data mining, disease prediction