Graph theory, as an effective tool to analyze the structure of complex networks, provides new perspectives for health management based on medical health records. The study utilizes a dynamic hypergraph network to construct a disease prediction model that extracts patients’ symptom information from EHR data. Specific disease development patterns are obtained by constructing two sub-supergraphs, and the disease prediction performance is improved by finely differentiating the different effects of diseases on patients and the different patterns of disease emergence in the time series. Compared to five baseline models in the MIMIC-III dataset, the model in this paper achieves the best prediction performance. After practical application of the model in healthcare, the incidence of health emergencies was reduced to 1.9%. The health management strategy based on the disease prediction model proposed in this paper improves health management effectiveness.