Deep reinforcement learning in patient admission process to optimize bed occupancy and revenue

Abstract

This paper discusses the problem of bed allocation with the example of a Swiss rehabilitation center. The aim is to challenge the benchmark of manual patient management by utilizing deep reinforcement learning (DRL) coupled with discrete event simulation to maximize bed occupancy and revenue. The comparison between random allocation, queuing systems, and deep reinforcement learning indicates the opportunities to optimize patient management at admission. Findings show that DRL agents can outperform random allocation and queuing systems when waiting lists are not utilized automatically. The integration of waiting lists significantly enhances the random system and queues, yet DRL agents maintain the highest results.