Integrated Simulation and Optimization for Decision-Making under Uncertainty with Application to Healthcare



Journal Title

Journal ISSN

Volume Title



Many real applications require decision-making under uncertainty. These decisions occur at discrete points in time, influence future decisions, and have uncertainties that evolve over time. Mean-risk stochastic integer programming (SIP) is one optimization tool for decision problems involving uncertainty. However, it may be challenging to develop a closed-form objective for some problems. Consequently, simulation of the system performance under a combination of conditions becomes necessary. Discrete event system specification (DEVS) is a useful tool for simulation and evaluation, but simulation models do not naturally include a decision-making component. This dissertation develops a novel approach whereby simulation and optimization models interact and exchange information leading to solutions that adapt to changes in system data.

The integrated simulation and optimization approach was applied to the scheduling of chemotherapy appointments in an outpatient oncology clinic. First, a simulation of oncology clinic operations, DEVS-CHEMO, was developed to evaluate system performance from the patient and managements perspectives. Four scheduling algorithms were developed for DEVS-CHEMO. Computational results showed that assigning patients to both chairs and nurses improved system performance by reducing appointment duration by 3%, reducing waiting time by 34%, and reducing nurse overtime by 4%.

Second, a set of mean-risk SIP models, SIP-CHEMO, was developed to determine the start date and resource assignments for each new patients appointment schedule. SIP-CHEMO considers uncertainty in appointment duration, acuity levels, and resource availability. The SIP-CHEMO models utilize the expected excess and absolute semideviation mean-risk measures. The SIP-CHEMO models increased throughput by 1%, decreased waiting time by 41%, and decreased nurse overtime by 25% when compared to DEVS-CHEMOs scheduling algorithms.

Finally, a new framework integrating DEVS and SIP, DEVS-SIP, was developed. The DEVS-CHEMO and SIP-CHEMO models were combined using the DEVS-SIP framework to create DEVS-SIP-CHEMO. Appointment schedules were determined using SIP-CHEMO and implemented in DEVS-CHEMO. If the system performance failed to meet predetermined stopping criteria, DEVS-CHEMO revised SIP-CHEMO and determined a new appointment schedule. Computational results showed that DEVS-SIP-CHEMO is preferred to using simulation or optimization alone. DEVSSIP-CHEMO held throughput within 1% and improved nurse overtime by 90% and waiting time by 36% when compared to SIP-CHEMO alone.