Data-based optimisation of intra-hospital patient transport capacity planning

Abstract

Efficient and timely organisational healthcare processes are urgent for patient satisfaction and medical success in hospitals. Despite process analysis and problem identification, there are especially challenges in evaluating and implementing planning alternatives. This is also valid for the planning of resource capacities. There are currently few use cases that offer data-driven, automated solutions and typically significant effort in modeling complex processes and systems is involved. Therefore, we explore the use of a combination of neural networks and metaheuristic algorithms to optimise organisational capacity planning in healthcare. These techniques allow for autonomous learning and optimisation of processes. A Multilayer Perceptron (MLP) is developed in a use case utilising data from approximately 3.5 years of accompanied intra-hospital patient transport in a German hospital in order to be able to make accurate predictions about delayed transports on a day of the week basis. A data preprocessing was performed, aggregating case-wise transportation information into hourly information to serve as input and labelling data for the MLP training. Using a genetic algorithm (GA), hourly input variables such as the number of active transporters, the number of planned transports, or the automation rate of transport dispatching are adapted in order to reduce the model predicted number of delayed transports throughout a day. Through this approach, a theoretical reduction in delayed transports on a day of the week ranging from 27% to 42% could be achieved merely through resource reallocating, without adding additional resources. The performance of both MLP and GA are validated using various measures.