| Hospitals are complex, patient throughput systems that experience undesired effects such as customer impatience and high utilizations common in any entity flow system. Non-stationary arrival and service rates, multiple patient types, and high-level clinical decision-making in a multi-node queuing network cause the hospital's throughput complexities. Hospitals desire to make resource decisions supported by scientific modeling but are inhibited by the throughput complexities. Existing operations research approaches require explicit knowledge of all service rate profiles, service capacity, and staffing profiles that are typically unknown in hospitals. Existing hospital research creates complex, empirical statistical models of patient flow that are not throughput theory based, are not generically applicable, and provides little insight to model quality-of-fit and inference.;This dissertation presents hospital patient flow approximations that are throughput theory based, only require limited data inputs, are generically applicable, and provide measures of quality-of-fit and inference. It derives queuing and throughput theory based Markov models as a foundation for regressions to describe and predict key patient throughput effects. Specifically, it creates a queuing based nonlinear regression to predict emergency department reneging, a discrete time Markov chain regression to predict inpatient inventories, and a Markov decision process for dynamically adjusting inpatient staffing. Statistical quality-of-fit measures and inference are derived unique to the Markovian models. Significant results include a queuing based prediction of long-term reneging, superior prediction of inpatient inventories, and a decrease in full probability and required staffing by the removal of inpatient discharge seasonality. |
