Improving decision making in healthcare operations

Date of Completion

January 2010


Business Administration, Management|Health Sciences, Health Care Management




This dissertation examines three distinct problems in the healthcare operations field and demonstrates how decision making can be improved in each setting. The first problem addresses how to make dynamic, real-time allocation decisions of testing time slots for a Cardiac Diagnostic Testing Center (CDTC) in the presence of multiple competing patient classes. Using historical data from the CDTC at a partner hospital, we explore the performance of a dynamic model which is based on a multi-commodity network flow formulation of the CDTC's scheduling problem. It incorporates the possibility of "bumping" different patient classes in order to accommodate requests from other patient classes. Our results indicate that this approach will result in drastically improved quality of service for inpatients with little or no reduction in the quality of service to the outpatients, relative to existing policy at our partner hospital. We then shift to exploring another important problem involving healthcare resources: during a Mass Casualty Incident (MCI), to which one of several area hospitals should each victim be sent? These decisions depend on resource availability (both transport and care) and the survival probabilities of patients. We focus on the critical time period immediately following the onset of an MCI and are concerned with how to effectively transport victims to the different area hospitals in order to provide the greatest good to the greatest number of patients while not overwhelming any single hospital. We explore two settings. First, we examine the deterministic version of the problem. We compare our proposed model with a model in the extant literature and also against several current policies commonly used by the so-called incident commander. Finally, we explore a dynamic and stochastic setting for the MCI problem, looking at how our decisions affect the expected number of survivors as events unfold over time. To address the infamous "curse of dimensionality", we develop a heuristic methodology that combines the use of a stochastic dynamic program (SDP) and Monte Carlo simulation to arrive at a final decision policy. We call this methodology "scenario iteration." Managerial insights are also discussed for each of the MCI models. ^