Analysis of Correlation between Medical Characteristics and Health Status of Breast Cancer Patients Based on Knowledge Graph

Nowadays much attention has been focused on the medical decision support system of breast cancer, as it is a very common malignant disease with a high recovery rate. At the same time, knowledge graph technology is widely applied in the medical field and contributes a lot for diagnosis and treatment. In this paper, the relationship between patients’ health conditions and other diagnostic information is studied for the purpose of assisting the medical support on the basis of knowledge graph (KG). We got a dataset with 100,000 electronic records of female patients and use different models to construct three graphs: the quadruplet model, the noisy OR model, and the traditional triplet model. The learning model Bi-LSTM is applied for model evaluation and the traditional model performs the best, with 0.45 accuracy. After the KG built by it is further studied, three closest features to health condition are found – “LymphNodeNumber”, “DistantMetastases” and “LymphNode”. We conclude that these three characteristics will to a large extent influence patients’ health status and should be considered into medical therapy. We believe our findings can help build the support system.