Background: Elderly hospitalized cancer patients face high risks of inpatient hospital mortality. Identifying patients at high risk of hospital mortality could help with risk stratification, and potentially help inform future interventions aimed at improving outcomes. We evaluated the predictive capacity of a high-dimensional machine-learning prediction tool to predict inpatient mortality, and compared the performance of this new tool to existing prediction indices. Methods: We identified cancer patients 75 and over who presented to an emergency department (ED) and were subsequently hospitalized from the National Emergency Department Sample (NEDS) between 2016 and 2018. We used an extreme gradient boosting approach to predict the risk of death during hospitalization. Model covariates included patient demographics, hospital characteristics, and International Classification of Diseases, version 10 (ICD-10) diagnosis codes recorded during the ED visit. The data were split 75%/25% into training/testing datasets. We constructed the model with training data and evaluated performance within the test data using area under the curve (AUC), with an AUC of 1.0 indicating perfect prediction. We compared the performance of this risk prediction model to standard prediction indices including the Hospital Frailty Risk Score, modified 5-item frailty index, and Charlson comorbidity index. Results: We identified 1,892,690 weighted-hospitalizations among elderly cancer patients, of which 133,379 (7.0%) who died in the inpatient setting. Our final predictive model included 238 features, which contained 5 demographic variables, 3 hospital characteristics, and 230 ICD-10 diagnosis codes. The predictive model achieved an AUC of 0.92. Our comparator models including the Hospital Frailty Risk Score, modified 5-item frailty index, and Charlson comorbidity index achieved AUCs of 0.67, 0.56, and 0.60, respectively. Conclusions: Using a high-dimensional machine-learning model enabled a high level of precision in predicting hospital mortality among elderly cancer patients, substantially out-performing existing prediction indices. High-dimensional prediction models show promise in helping to identify patients at risk of severe adverse outcomes, though additional validation is needed as well as research studying how to implement these tools into practice.