Demand-Responsive Windows Scheduling in Tertiary Hospital Leveraging Spatiotemporal Neural Networks
Long waiting queues have been a stressful problem in many tertiary public hospitals, which significantly impact the accessibility and quality of health care. One of the key challenges to solve this problem is to provide enough registration windows to serv
- PDF / 2,624,121 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 70 Downloads / 168 Views
Fujian Key Laboratory of Sensing and Computing for Smart City, School of Informatics, Xiamen University, Xiamen, China [email protected] 2 Xiang’an Hospital of Xiamen University, Xiamen, China
Abstract. Long waiting queues have been a stressful problem in many tertiary public hospitals, which significantly impact the accessibility and quality of health care. One of the key challenges to solve this problem is to provide enough registration windows to serve hospital visit demand under the limited medical and human resources. Traditional window shift scheduling methods are usually based on experiences and biased historical data, which may not accurately reflect the actual hospital visit demand. In this work, we propose a demand-responsive window scheduling framework by accurately modeling and forecasting the fine-grained hospital visit demand from real-world human mobility data. Specifically, in the first phase, we extract hospital visit demand from taxi drop-off events around hospitals, and build a graph model to capture their spatiotemporal patterns. In the second phase, we propose a spatiotemporal graph neural network (ST-GNN) to accurately forecast the hospital visit demand, which simultaneously captures the spatial correlation by graph convolutional networks (GCN) and the temporal dependency by gated recurrent units (GRU). Finally, we exploit a queuing theory model to achieve demand-responsive windows scheduling. Evaluation results using real-world data from Xiamen City show that our framework accurately forecasts hospital visit demand, and effectively schedules hospital registration windows, which consistently outperforms the baselines.
Keywords: Hospital management mobility data · Deep learning
1
· Graph neural network · Human
Introduction
The problem of long waiting queues in tertiary hospitals is troubling patients, which may affect the medical experience of patients and delay the treatment of emergency and critical patients [1]. For example, the average queuing time in Detroit VA Medical Center is as long as 42.3 min [2]. Due to the limitation of c Springer Nature Switzerland AG 2020 Z. Yu et al. (Eds.): GPC 2020, LNCS 12398, pp. 231–243, 2020. https://doi.org/10.1007/978-3-030-64243-3_18
232
Z. Wang et al.
medical and human resources, the hospitals’ administrations can not schedule enough service windows to satisfy the demand, causing it necessary to exploit effective registration windows scheduling strategies to reduce queuing time. The registration windows scheduling requires the forecast of the amount of the hospital visit demand in the next period of time to schedule the shift in advance, e.g., 6 h. The conventional methods are to forecast hospital visit demand based on experiences [3] and the data in the hospital visit registration system [4]. However, experience-based methods are not able to respond to real-time demand. While the methods based on the historical registration system data can not reflect the actual queuing situation since the registration system can not reflect the over-demand par
Data Loading...