Behavior-aware deep reinforcement learning for multi-objective outpatient scheduling optimization.

Area of Science:

• Operations Research
• Behavioral Economics
• Artificial Intelligence

Background:

• Outpatient scheduling faces inefficiencies like long waits, resource underutilization, and high no-show rates.
• Current methods often overlook patient behavioral heterogeneity, simplifying satisfaction to mere waiting time.

Purpose of the Study:

• To develop and evaluate MO-SAC-B, a multi-objective deep reinforcement learning framework integrating behavioral science for outpatient scheduling.
• To enhance scheduling efficiency and patient satisfaction by modeling psychological constructs of waiting and no-shows.

Main Methods:

• Constructed a behavior-driven discrete-event simulation environment incorporating prospect theory, patience decay, and calibrated no-show dynamics.
• Employed a multi-objective Soft Actor-Critic algorithm with satisfaction-aware reward shaping and adaptive weight adjustment.
• Validated the framework using real outpatient data from a tertiary hospital.

Main Results:

• MO-SAC-B reduced mean waiting time by 21.9% and patient satisfaction by 12.7 points.
• The framework decreased the no-show rate by 25.8% compared to the leading baseline.
• Behavioral components proved crucial, especially under high patient flow, with synergistic effects.

Conclusions:

• Integrating behavioral science into AI-driven scheduling significantly improves outpatient operational efficiency and patient experience.
• MO-SAC-B offers a robust and adaptable solution for optimizing hospital appointment systems, even amidst demand fluctuations.