This study describes the development of a model to predict length of inpatient stay as the survival of daily discharge decisions made by attending providers. The dataset consisted of n=283,209 patient-days from 44,339 inpatient visits during 2019 at a multi-facility healthcare system in the southeastern United States. Clinical, operational, and financial electronic medical record data were used—including provider attending records, orders and results, procedures, diagnoses, vital signs, as well as patient movement and status updates. Patient visit history from the year preceding the study time period was also included. Time-varying predictors were aggregated for each patient at the time of admission and daily during the inpatient stay. Daily discharge likelihood, remaining days until discharge, and total length of hospital stay were predicted using a non-parametric discrete-time survival framework, with discharge rate as the hazard function. Feature importance was assessed empirically and results were interpreted in the context of potential hospital performance improvement initiatives. Background on the importance of hospital performance improvement and the central role of length of stay, as well as a thorough review of relevant literature, is also included.