Oakley, David and Worthington, David and Onggo, Bhakti Satyabuhdi Stephan (2019) Online simulation for the operational management of inpatient beds. PhD thesis, Lancaster University.
2019oakleyphd.pdf - Published Version
Available under License Creative Commons Attribution-NoDerivs.
Download (4MB)
Abstract
In many modern hospitals, resources such as beds, theatre time, medical equipment and staff are shared between patients who require immediate care and must be dealt with as they arrive (emergency patients), and those whose care requirements are known to the hospital some time in advance (elective patients). Caring for these two types of patients poses a logistical challenge, since some portion of each resource must be set aside for emergency patients when planning for the number and type of elective patients to admit. Failing to strike this balance can result in negative outcomes, such as patient-stays on non-ideal wards, or increased waiting time for elective procedures (in the case of public health services). The potential benefits of using discrete event simulation (DES) models in healthcare are well established, and they are often preferred to other modelling approaches because of their ability to emulate the randomness seen in real systems, at a level of detail which is necessary for models to be convincing. However, their use is often limited to strategic or tactical decision making, and few have attempted to produce models which can help hospitals with short-term (operational) decision making. This is where Online Discrete Event Simulation (ODES) can help. An ODES (also known as symbiotic simulation) takes all the components of a DES model, and adds the ability to load the state of the real system at run-time to make predictions about how the real system might evolve in the short-term. This thesis reports the development of a whole-hospital, proof-of-concept ODES to assess the impact of elective admissions decisions, on wards which are shared with emergency patients. The model is parameterised by analysing 18 months of patient administrative data from an Australian General Hospital. Since ODES is a relatively new method, this research focuses on formalising the model development process, resulting in a new “black-box” validation method for handling conditionally distributed simulation outputs. Additionally, a new probabilistic routing method is developed to better represent inter-ward dependencies during peaks in bed demand. A statistical analysis of the relationship between ward transfers and ward occupancy is conducted on real hospital data to parameterise so-called “Dynamic Transition Matrices” for this purpose. Finally, the ODES is used to demonstrate how additional patient-level information (which might only become available after admission) can affect the predicted bed census. Clinicians’ discharge date estimates fit this criterion, and the case is made for more scientific use of this type of information, as part of an operational ODES model.