Article Preview
Top1. Introduction
This paper considers an inventory optimization problem commonly found at the pharmacy department of large hospitals in Thailand. Unlike health care systems in some countries, where physicians only give medicine prescriptions and patients then purchase medicines from pharmacy stores, in Thailand all health care systems are one-stop services that prepare medicines and medical supplies for their patients. This characteristic makes inventory management of medicines and medical supplies an essential task in Thailand's healthcare systems. In such a system, a central stockroom, managed by the pharmacy department, provides medicines and medical supplies (referred to as items in this paper) to dispensaries and other medical departments (e.g., operation rooms) of the hospital. When item inventory levels become depleted at dispensaries, replenishment orders are placed to the central stockroom. These orders are considered as demand for the central stockroom. Similarly, the central stockroom would place replenishment orders to the suppliers when its inventories become depleted. Effective inventory management at the central stockroom is crucial since it directly impacts the dispensaries and medical department's ability to serve outpatient and in-patient demands. At the same time, keeping the inventory management cost low is also important. The key to minimizing inventory management cost, while maintaining a service level, is the inventory policy implemented at the central stockroom.
A major hospital, which not only provides real data but also is the industrial user of this study, is typical and representative of the Thai health care system. Currently, the hospital central stockroom uses the 
policy, where a replenishment order is placed to bring the inventory position (IP) to the order-up-to level S when an item's IP falls on or below the reorder point s. Each item at the central stockroom is managed independently regarding how the values of s and S are set. Performance of the central stockroom, therefore, relies on how well these values are determined. Unlike the random demands from in-patients at medical departments and from outpatients at the dispensaries that arrive regularly, demands for an item at the central stockroom are much more variable and intermittent. In addition, the incoming lead times of replenishment items from the suppliers are relatively long and highly variable due to the suppliers' delivery schedules. Both sources of variability make the problem very challenging.
The shortage of items at the central stock room is handled differently depending on the patient type. In the case of medicine shortage to outpatients, the hospital provides delivery service free-of-charge to the patients as soon as the shortage item becomes available. For in-patients that need the items for their treatment, a shortage is not an option. In such situations, the hospital must arrange special pickup from another nearby health care system or an expedited delivery from the supplier. This type of shortage is considered as a backlog that either can wait (for out-patients) or is immediately fulfilled (for in-patients). Both cases incur backlog costs to the hospital. The backlog cost is considered as a part of the total inventory management cost of the central stockroom, in addition to ordering costs and item holding costs. The objective of the inventory optimization in this healthcare system is, therefore, to determine the inventory policy parameters 
that minimizes the total cost.
Due to the excessive amount of variability in the systems, a simulation-optimization approach to solving this problem is proposed. The approach features a search algorithm that is performed on a simulation model. The simulation component, which captures the system's random behavior, provides accurate estimates of the system measures of performance, while the optimization component searches for high-quality solutions. Contributions of the paper are two-fold: (1) an effective approach is developed for a typical health care inventory system in Thailand, where demands are highly variable and intermittent and supplier lead times are very long and highly variable, and (2) effectiveness of the proposed approach is demonstrated using a real case study of a large hospital in Thailand.