View from Europe: Tracking tool can solve scheduling problems

Most hospitals experience difficulties optimizing the use of expensive resources such as CT and MR scanners and operating theaters. Ideally, these resources should be in use continuously, with no idle periods, but also remain readily available when required. Patient waiting times should be kept as short as possible.

Emergencies that simply cannot be planned for complicate scheduling. Administrators can either leave capacity deliberately unallocated, or they can work with fully allocated resources and accept delays when emergencies occur. The latter option usually means planning is disturbed for the rest of the day, with a possible domino effect on other resources. All scheduling heuristics must also consider delays from late-arriving patients, equipment failure, and the interdependency of procedures.

Our radiology department generally has advance schedules for the use of expensive resources, including CT, MR, and angiography (Figure 1). Relatively inexpensive resources such as chest x-ray need no advance reservation. We accept just a few patients in modality waiting rooms because these facilities are limited (Figure 2), and we cannot supervise critically ill patients. Long waiting times also cause stress for patients and radiology staff and reduce overall treatment efficiency by delaying other procedures.

A well-organized patient transportation service is essential for following pre-established planning in radiology and adhering to examination times. Patients may need guidance to the treatment room or transport in a wheelchair or gurney. Problems meeting the schedule may still occur from a breakdown in real-time communication between transportation staff and the transportation scheduler or the scheduler's lack of knowledge about the patient's immediate health. The scheduler may not know at any given moment where the patient to be transported is located, where the transportation staff are, or how long a procedure will take. Emergencies can also interfere with planned patient transport.

Most weaknesses in existing manual or semiautomatic transportation planning systems result from inflexible scheduling at modalities, ignorance of the status of radiology and transportation resources, and lack of feedback from the scheduler.

Elimination of these weaknesses requires a planning system that receives automatic input of new data when the status of the resources changes. This allows, for instance, for continuous localization of transportation staff. The ideal system would continually reschedule for modalities and transportation resources in real-time and automatically feed data back to all transportation and radiology staff.

REAL-TIME SCHEDULING

We are developing a system to tackle these issues.1 We hope to achieve continuous localization using industry-standard radio-frequency tags that store a unique ID (typically 128-bit). These identifiers may be very small (0.4 mm x 0.4 mm), may communicate on RF signals with a range of 125 kHz to 2.45 GHz, and can either be powered by a battery (active tags) or use the impulse emitted from antennae passed by the wearer (passive tags).

Every movable resource within the hospital, whether human (transportation staff and patients) or nonhuman (wheelchairs, beds, gurneys), would be tagged with a unique RFID (radio-frequency identification) chip. Displacement of resources, tracked by following displacement of the chips, would then feed automatically into the information system. This process keeps resource localization information up to date and permits calculations on the relative proximity of resources.

We envisage the use of personal digital assistants together with wireless local area network (WLAN) technology for real-time transmission of orders to transportation and radiology staff. All participants would learn about scheduling changes via automatic updates sent to their PDAs. Scheduling changes involving medical issues may require user interaction about whether a moved appointment is acceptable or not. If no doctor is available to respond to the message within a certain time interval, the information system will reject the proposed scheduling change.

Conflict can arise among three aspects of scheduling: predictive scheduling, dispatching, and reactive rescheduling. Centrally controlled schedules (predictive scheduling) cover most assignments of outpatients to staff and equipment in advance. Dispatching summons inpatients when the required resources are available. Optimized appointment sequences are amended in response to emergencies and/or cancellations (reactive rescheduling).

The optimization techniques used for predictive scheduling in radiology, mostly "constraint-based," do not allow for continuous, reactive rescheduling or dispatching. Reactive rescheduling inevitably alters the predictive schedules, disrupting the stability and robustness of the entire process. The lack of stability makes inpatient dispatching hard to coordinate. Inpatients find themselves competing with one another for resources during peak periods, instead of slotting into vacant positions. Solving the coordination problem involves resolving the goal conflicts among the three scheduling approaches.

TIME IS MONEY

General resource allocation problems in economics can be addressed through the introduction of price information and its inclusion in utility functions. Price information on the availability of a product or resource, interpreted in monetary units, constitutes a standardized "language." Relative shortages of resources and comparisons between resources of the same type become transparent.

Translation of basic hospital logistics information into monetary units makes it possible to engage different scheduling strategies in the same environment and build on their respective strong points using the introduction of a virtual monetary substitute in the form of time points. The time points constitute a direct connection to the optimization variable "time," enabling evaluation of the time required and spent in the hospital.

The next stage is to attach artificial intelligence technology known as autonomous software agents (ASA) to the tagged physical resources (Figure 3).2,3 Every entity-staff member, modality, or transportation device-is represented in the information system by an ASA. Each ASA negotiates its entity's interests with the other agents. An ASA will sense the environment's current state-for example, by comparing the tagged resource's location with the known schedule-and match that to the desirable goal state. If the two states differ, the ASA chooses an appropriate action. For instance, it may try to renegotiate the next appointment.

Appointment negotiation is carried out using price-based offers. The ASA network, which is continuously negotiating, essentially forms a virtual, electronic marketplace that trades resources for virtual money.

This abstract concept may become clearer through an example. An ASA representing an emergency patient in urgent need of a CT examination will pay sufficient virtual money to the ASA representing the CT scanner to obtain the examination immediately. This ASA will also pay compensation to the agents representing patients whose examinations are now delayed. The negotiation leads to an immediate rescheduling of all subsequent actions, which is communicated to all actors.

The transportation staff member who is in the best location to take the emergency patient to the CT scanner, or is simply less busy, is deemed "cheapest." He or she will receive the message on a PDA prompting the immediate transport. This staff member's ASA will also be compensated by the ASA of the emergency patient.

THEORY TO PRACTICE

We have developed a simplified, small-scale physical model to test this system. The model comprises a waiting room, transportation area, two examination rooms with one modality each, two transportation staff, two radiology staff, four patients, a fox, and a chicken. Transportation staff are each equipped with a PDA. Passive RFID chips and antennae make localization of all moving persons and objects possible.

Localization of objects in this mock-up functions robustly. The software autocorrects for detection errors, and the connection of the system with the PDAs also works. The ASA marketplace works as a simplified model with few participants and simple rules.

The model has certain limitations when compared with intended system realization. Detection distance for the RFID chips we are using is limited to about 30 cm. This range requires installation of a profusion of antennae in a hospital-wide system. Alternative options include switching to passive chips that work with longer antennae (occupying more space) or with longer frequencies (with possible electromagnetic interferences), or using active tags (requiring heavy batteries). Adoption of Bluetooth or WLAN technology for this function will invoke unreasonable costs.

We have defined only a basic set of action rules for the ASAs, and these must be refined before deployment in a real-life, critical environment. The coordination mechanism must be developed to consider a larger number of resources and rooms, nearing the complexity of the real environment. Clinical implementation will require an evaluation of how the existing coordination mechanism scales with an increasing number of resources, to produce stable conditions for appointment allocation.

Mobile IT makes it possible in principle to inform every user about all changes in the hospital environment, including real-time notification of scheduling decisions. This may improve the scheduling situation in general, but it could also create information overload for the human participants. Such a situation will create a demand for so-called reachability management, which describes the individual adjustment of type and frequency of information provisioning.4

Use of software agents requires demarcation between automatic and manual involvement in scheduling decisions. Agents that make changes of their own accord will notify users only when the changes have been made. This decreases the amount of information the user receives and the level of interaction needed. Users opting for a lower level of software autonomy must manually accept or reject every proposition the system makes.

Acceptance of this proposed system will doubtless depend on the ability of users to change autonomy levels as appropriate. Doctors may desire full autonomy on every decision, while transport staff may need only to react to changed scheduling decisions. We plan to use expert and user interviews to evaluate optimal levels of control.

The next step in this research project is transition of the small-scale model to a simulated real-time environment. Computer scientists, medical researchers, and economists will work together over the next two years to transfer the working technical prototype to a real-size, complex hospital. The team will also evaluate the negotiation protocols, strategy rules, and market properties to achieve a secure, flexible, and dependable information system that blends seamlessly into the demanding hospital environment.

References

1. Sackmann S, Eymann T, Muller G. EMIKA-Real-time controlled mobile information systems in health care applications. In Bludau HB, Koop A, eds. Mobile computing in medicine. Bonn: Kollen Druck und Verlag, 2002.

2. Wooldridge MJ. Intelligent Agents. In: Weiss G, ed. Multiagent systems. Cambridge, MA: MIT Press, 1999:27-78.

3. Paulussen TO, Jennings NR, Decker KS, Heinzl A. Distributed patient scheduling in hospitals. Presented at 18th International Joint Conference on Artificial Intelligence, Acapulco, MX; Aug. 2003:1224-1229.

4. Reichenbach M, Damker H, Federrath H, Rannenberg K. Individual management of personal reachability in mobile communication. In: Yngstrom L, Carlsen J, ed. Information security in research and business. London: Chapman & Hall, 1997:163-174.

DR. KOTTER is a senior staff radiologist and head of imaging IT, and PROF. DR. LANGER is head of diagnostic radiology at Freiburg University Hospital in Germany. DR. EYMANN is an assistant professor, and PROF. DR. MULLER is a professor of telematics, at the Institute of Computer Science and Social Research, Albert-Ludwigs-University in Freiburg.