In-hospital mortality prediction is significant for evaluating a patient's severity of illness ahead of the time. The outcome of the evaluation can help physicians to identify which patient is at risk and needs immediate care, it can further increase the efficiency of use of medical resources. In this study, I proposed a method that is similar with the one in our Electronic Health Records (EHRs) research at the CBL Lab and utilized the hyperbolic embedding of ICD-9 medical ontology for the prediction model. The results outperformed the benchmark prediction model and demonstrated that the hyperbolic embedding on ICD-9 is more effective than other graph embedding methods.