Accurate estimation of the physiologic limits of patients during their stay in the intensive care unit is critical for predicting mortality risk. However, patient populations in different ICUs may differ in age, severity of illness, and medication. Creating the models that are relevant for diverse populations poses a significant challenge. Existing models trained on data fromspecific ICUs may performpoorly in other settingsbecause of changes in the distribution of characteristics. To avoid those problems,we propose a deepsequence framework model topredictmortalityrisk based ontimeseries data received fromICUs.Weconductedexperimentsusingthepublic