Machine learning based early risk assessment and prognosis prediction of Heart Failure (HF) are beneficial for disease management, but challenging due to the limited availability of large survival datasets. In this paper, we study whether models originally trained to predict the risk of HF hospitalization can be repurposed to estimate mortality. We hypothesize a relationship between hospitalization and mortality risks based on the progressive nature of HF that could be leveraged to unlock data limitations. Using our previously developed models based on 30-second lead I ECG and basic patient information, we evaluate their performance in predicting mortality in multiple cohorts: HF patients in the MUSIC dataset with additional insights into the cause of death and left ventricular ejection fraction, as well as other risk groups in the SaMi-Trop and CODE-15% datasets. Our results demonstrate that our HF hospitalization models are capable of effectively stratifying mortality risk among different populations.