Reducing operation rooms labor costs: Capturing information on workload heterogeneity and dynamic adjustments of staffing level

I study in my dissertation the problem of setting nurse staffing levels in hospital operating rooms when there is uncertainty about the daily workload. Hospital operating rooms often experience uncertain workload that results in overtime and inefficiency. Operating Rooms (ORs) are one of the most critical resources in a hospital: they drive a significant portion of a hospital's revenue and costs. With growing expenses and reduced government subsidies, hospitals have been under pressure to reduce OR costs without sacrificing quality of care. Because workload varies substantially in days and specialties, and the lead time to arrange work shifts is long (typically 2--3 months), a hospital has to balance under-utilization of its capacity with the risk of having the ORs run late, which requires the hospital to pay the staff at overtime rates and induces employee frustration and frequent turnovers. To address these issues, I focus my research in the following directions: (1) identify the key factors that can reduce the workload uncertainty and develop effective methods to predict future workload; (2) explore how information, data and forecasting methods can be used to facilitate the staffing decision, and how these elements affect the cost performance; (3) improve the decision process, allow adjustments to be made to the allocated staffing levels as new information of the workload becomes available; (4) integrate forecasting methods with the dynamic staffing model to provide practitioners with an implementable solution.;In the first chapter I present an empirical investigation on how heterogeneity in demand types, choices of empirical models and data availability can affect a newsvendor's performance in this healthcare service setting. Using a newsvendor framework, I consider the problem of determining optimal staffing levels with different information sets available at the time of decision: no information, information on number of cases, and information on number and types of cases. I develop empirical models for the daily workload distribution and study how its mean and variance change with the information available. I use these models to derive optimal staffing rules based on historical data from a US teaching hospital and prospectively test the performance of these rules. The numerical results suggest that hospitals could potentially reduce their staffing costs by up to 39%--49% (depending on the absence or presence of add-on cases) by deferring the staffing decision until procedure-type information is available. The improvements are robust to changes in the assumptions about newsvendor costs.;However, hospitals may not be able to delay the staffing decision until all scheduled cases are known, because contractual and scheduling constraints often require OR managers to reserve staffed hours several months before the actual day of surgery. Hence, in the second chapter I study two decision models that allow an OR manager to adjust the staffing level with some adjustment costs as workload information is updated over time: I first consider a two-stage binomial-tree model that assumes the workload information sets to be nodes of a finite discrete tree, where the transitions between the nodes represent upward/downward corrections in workload forecast. Building on the intuition of the binomial-tree, I then consider a mixture-normal model, where workload is assumed to be normal with uncertainty about the mean and variance resolved in the first stage and the actual workload observed in the second stage. In both models, the optimal adjustment policy is of a control-limit type, where the control limits depend on the mixture or transition (in the binomial-tree case) probabilities and the cost parameters. I conduct numerical experiments to study how the value of having the flexibility to adjust staffing levels changes with the adjustment costs and workload uncertainty. I also demonstrate how to implement these decision models by applying the optimal policies to operating room data from a US teaching hospital.