Improving Product Availability in Hospitals: the Role of Inventory Inaccuracies

All players in the healthcare industry face increasing public and political pressure to improve quality of care and control costs. Hospitals, on the frontline of this challenge, face nursing shortages and financial constraints. Survey data indicate that missing medication and supplies interrupt nurses more than twice per shift, increasing costs and putting patients at risk. These challenges persist even though over 72% of U.S. hospitals have deployed Automated Dispensing Machines (ADMs), electronic cabinets that automate inventory management processes and improve product availability. This research investigates the role of inventory inaccuracies, i.e., mismatches between book inventory and physical inventory on hand, as drivers of product availability in hospitals. The research objectives are three-fold: (1) characterize the sources of inventory inaccuracies prevalent in a hospital context; (2) quantify the impact of inventory inaccuracies on product availability and performance metrics; and (3) identify and evaluate practical strategies that hospitals can use to improve product availability by reducing and mitigating inventory inaccuracies. This thesis views the hospital supply chain as a socio-technical system and addresses the research questions using a multilevel, multi-method approach. The research is empirically grounded by the case study of Lambda, a New England area hospital that provided qualitative and high-frequency transactional data from its network of 108 ADMs that stock over 21,000 product-location combinations. First, by classifying sources of inventory inaccuracies this thesis identifies Imperfect Demand Recording as a hospital-specific source of such inaccuracies. Recording Accuracy is proposed as a metric of user behavior at product and location levels, and reveals that between five and thirty percent of product usage is not recorded. Then, a single-product Discrete-Event Simulation (DES) model shows that Imperfect Demand Recording causes large reductions in availability unless mitigated by frequent and consistent (i.e., equally-spaced) inventory counts, and that service level estimates provided by ADMs can have a large, optimistic bias. Assuming that count timing is independent of inventory state, an analytical model provides a closed-form generalization of the simulation results and shows that variability in cycle count has a nonlinear and substantial effect, causing 35% of counts performed at Lambda to be ineffective. Finally, a sequential and iterative framework integrating the managerial implications of these contributions is proposed.