Improving Emergency Department Efficiency by Patient Scheduling Using Deep Reinforcement Learning.

Area of Science:

• Artificial Intelligence
• Operations Research
• Healthcare Management

Background:

• Emergency departments (EDs) face challenges with overcrowding and long patient wait times.
• Patient flow complexity arises from diverse acuity levels, personalized treatments, and resource interdependencies.
• Dynamic factors like patient composition and resource availability further complicate ED scheduling.

Purpose of the Study:

• To develop and apply a deep reinforcement learning (RL) model for optimizing patient scheduling in emergency departments.
• To formulate a mathematical model and Markov decision process (MDP) for ED patient flow.
• To design a deep Q-network (DQN) algorithm for determining optimal patient scheduling policies.

Main Methods:

• Formulation of a mathematical model and Markov decision process (MDP) for the ED environment.
• Implementation of a deep reinforcement learning (RL) algorithm, specifically deep Q-networks (DQN).
• Comparison of the RL approach against traditional dispatching rules to evaluate performance.

Main Results:

• The deep RL model significantly outperformed existing dispatching rules in simulated scenarios.
• Key performance improvements included minimizing weighted patient waiting times.
• The RL approach also reduced penalties associated with emergent patient care.

Conclusions:

• Deep reinforcement learning (RL) offers a successful strategy for enhancing emergency department (ED) patient scheduling.
• The developed RL model provides effective decision support in dynamic ED environments.
• This approach demonstrates the potential to improve overall ED operational efficiency and patient outcomes.