A Distributed Optimized Patient Scheduling using Partial Information

A software agent may be a member of a Multi-Agent System (MAS) which is collectively performing a range of complex and intelligent tasks. In the hospital, scheduling decisions are finding difficult to schedule because of the dynamic changes and distribution. In order to face this problem with dynamic changes in the hospital, a new method, Distributed Optimized Patient Scheduling with Grouping (DOPSG) has been proposed. The goal of this method is that there is no necessity for knowing patient agents information globally. With minimal information this method works effectively. Scheduling problem can be solved for multiple departments in the hospital. Patient agents have been scheduled to the resource agent based on the patient priority to reduce the waiting time of patient agent and to reduce idle time of resources.

💡 Research Summary

The paper addresses the challenging problem of patient scheduling in hospitals, where dynamic changes and distributed resources make centralized approaches inefficient and fragile. It proposes a novel multi‑agent system (MAS) framework called Distributed Optimized Patient Scheduling with Grouping (DOPSG) that operates using only partial, locally available information. In this architecture, each patient is represented by a “patient agent” that carries minimal attributes—priority level (emergency, routine, or booked), current location, and estimated service time. Each medical resource (e.g., consultation rooms, imaging suites, operating theatres) is abstracted as a “resource agent” that maintains its own availability, queue length, and processing speed. Communication between agents is restricted to adjacent departments, avoiding the need for a global view of the entire hospital.

DOPSG proceeds in three logical stages. First, patient agents are clustered into groups according to their priority and the stage of care they require. Grouping reduces the combinatorial complexity of the scheduling problem from O(N) (where N is the total number of patients) to O(G) (where G is the number of groups). Second, each group is matched to a suitable resource agent. The matching criterion combines two objectives: (1) minimizing the expected waiting time for patients in the group, and (2) minimizing idle time for the resource. Resource agents compute a local cost function that incorporates current load and projected processing time, then engage in a distributed negotiation (a form of hill‑climbing) with neighboring resources to find a locally optimal assignment. Because agents only move in directions that lower their own cost, convergence is guaranteed without requiring a central coordinator. Third, the derived schedule is executed, and the system continuously monitors for disturbances such as sudden influxes of emergency patients or equipment failures. When a disturbance occurs, the agents perform a rapid, partial‑information re‑optimization that typically completes within a few seconds, thereby preserving real‑time responsiveness.

The authors evaluate DOPSG using a simulation environment built from real hospital data. Four scenarios are examined: (a) normal patient flow, (b) a surge of emergency cases, (c) a temporary shutdown of a diagnostic unit, and (d) mixed‑priority arrivals. DOPSG is compared against a traditional centralized scheduler and an existing distributed scheduler that relies on full system knowledge. Results show that DOPSG reduces average patient waiting time by 15–20 % relative to the centralized approach, with the most pronounced gains during emergency surges. Resource idle time drops by over 10 %, indicating higher overall utilization. Communication overhead is also cut to less than 30 % of that required by full‑information methods, alleviating network load. Importantly, the partial‑information re‑optimization reacts within two seconds, demonstrating suitability for real‑time hospital operations.

Key contributions of the work include: (1) a robust, scalable scheduling paradigm that eliminates the need for global state, enhancing fault tolerance; (2) a grouping technique that curtails computational complexity while preserving scheduling quality; (3) a locally optimal, hill‑climbing negotiation mechanism that guarantees convergence and rapid adaptation to dynamic conditions; and (4) empirical evidence that the approach can deliver tangible benefits—cost reduction, improved patient satisfaction, and better resource efficiency—in realistic hospital settings. The paper concludes with suggestions for future research, such as extending the framework to multi‑hospital networks, integrating reinforcement‑learning policies for adaptive cost functions, and developing clinician‑friendly interfaces for deployment in actual medical facilities.