Discrete-event Simulation and Optimization to Improve the Performance of a Healthcare System

Thesis (Ph.D.)--University of Washington, 2016-08

Healthcare systems have attracted the attention of management and analysis due to their high percentage of the gross domestic product (GDP) and increasing rate of growth of expenditures. Within the various types of healthcare problems, this dissertation focuses on resource allocation decisions because they can significantly improve the performance of care delivery systems. Since most healthcare systems have significant uncertainty and complexity, and mathematical closed form models may not exist, discrete-event simulation is considered here as a suitable approach to model a healthcare system. Also, multiple objectives are a natural consideration in healthcare problems where at least cost and health outcome are usually considered. This research has two contributions: developing decision support models for three healthcare systems; and developing algorithms for simulation optimization that can be used to provide insights into healthcare problems. The three healthcare resource allocation problems in this dissertation are: (i) design an occupational health campaign in which two departments in a hospital must coordinate their activities in order to provide a high level of service with intensive service demands in a compressed time period; (ii) investigate how to optimize hepatitis C screening and treatment allocation strategies; (iii) evaluate cost and health quality trade-offs when allocating portable and console type ultrasound instruments for orthopedic clinics. These healthcare systems are modeled with discrete-event simulation and optimization. The occupational health campaign was designed with limited resources from a clinic and a laboratory to meet high demands for influenza immunization and tuberculosis (TB) screening. All design configurations of the campaign were simulated to understand the system performance. Hepatitis C screening and treatment allocation strategies involving budget constraints were simulated to maximize the total discounted health utility gain of a cohort over its lifetime. The number of strategies was too large to be exhaustively investigated, so the first simulation optimization algorithm was applied to explore sensitivity of solutions. The portable ultrasound machine allocation problem considered portable ultrasound machines in orthopedic clinics as a viable alternative to a centralized MRI service, regarding the trade-off between cost and health utility loss. This problem also involved a large number of designs as well as a high degree of randomness and multiple objectives. The use of the second simulation optimization technique played an important role in this situation. The second contribution in this dissertation is to develop two simulation optimization algorithms that can be applied to a general black-box simulation with randomness, such as those used in the hepatitis C screening and treatment strategy and the portable ultrasound instrument allocation problem. The first algorithm focuses on single objective problems and approximates a target level set of solutions bounded by a quantile. The approximated level set provides insights through the shape of the level set and allows decision makers to consider issues out of the simulation model. The second algorithm is designed for multiple objective problems and provides an approximated Pareto optimal set and efficient frontier for decision makers to investigate the trade-offs between objective functions. The algorithms include several important characteristics. First, they can process mixed integer and continuous variables. Second, the objective function can be a black-box simulation with noise or uncertainty. Third, statistical qualities of solutions provided by the algorithms are derived to support decision making. The developed algorithms are evaluated with test functions and applied to the hepatitis C screening and treatment strategy and the ultrasound instrument allocation problem.