Accurate prediction of medium-term survival after admission is necessary for identifying end-of-life patients who may benefit from earlier goals of care (GOC) discussions. While previous studies have leveraged admission data from electronic health records (EHRs) to predict the hospital one-year mortality risk (HOMR) score, they focused on single admissions, without considering longitudinal patient history and its impact on prognostication. To address this gap, we developed the Ensemble Long Short-Term Memory (ELSTM) neural network, which learns from multiple visits of the same patient to improve the accuracy of the HOMR score. Furthermore, in this work, we generated a synthetic dataset and made it publicly available to encourage further research in this area while safeguarding patient privacy.