Dynamic appointment scheduling in outpatient department with patient preferences


Student thesis: Doctoral Thesis

View graph of relations


  • Jin WANG


Awarding Institution
Award date16 Feb 2015


Traditional healthcare systems are proving to be inadequate as they come under pressure from aging populations. In response, appointment systems aimed at improving resource utilization and service efficiency are being installed. However, current systems still have many shortcomings in terms of satisfying patient preferences and stochastically optimizing service durations and patient arrival patterns. Of these, the issue of patient preferences is particularly serious because it is a matter that affects the patients directly. Real time appointment scheduling is essentially a problem of sequencing. Schedulers make decisions upon receiving a patient's request online. A well-developed appointment system should take into account future events (such as future patient requests). This turns the problem usually into a dynamic programming (DP) type. Since this is usually a complex and computationally intensive problem, its solution demands a methodical approach. In summary, the main challenge in appointment scheduling is to efficiently handle patients' appointments online with due consideration of patient preferences. This thesis tackles this challenge by modeling and optimizing patient appointments using a DP model. The thesis begins with an examination of the dynamic appointment scheduling problems associated with single patient preference scenarios in outpatient departments. A two-dimensional choice model that takes into account preferences with regard to physicians as well as time slots is developed. A DP model then optimizes the scheduling of sequential appointments constrained by patient preferences. An improvement over existing models is the evaluation of booking decisions on the basis of the degree to which patient preferences are satisfied. Next, the characteristics of the model are analyzed to help formulate booking policies. Based on the characteristics, two kinds of approximate dynamic programming algorithms are developed in order to avoid the curse of dimensionality typically associated with DP models. Experimental results have pointed to the need for further fine-tuning of the model and improving the efficiencies of the two proposed algorithms. The thesis then proceeds to consider the dynamic appointment problem with multiple preferences. Unlike in the case of single preferences, patient decisions after receiving the scheduler's assignments are also considered. Next, a DP model is proposed for scheduling sequential appointments in a manner maximizing patient satisfaction levels. Certain adaptive dynamic programming algorithms are formulated to avoid the curse of dimensionality. The algorithms are capable of dynamically capturing patient preferences, updating the values of being in a state, and thus raising the quality of decision-making. Numerical experiments are conducted next to evaluate the performance of the algorithms. Evaluation considerations include the convergence behaviors under different settings, i.e., the number of iterations needed before convergence and the accuracy of results. In some cases, the patient might want to choose a specific appointment time slot on his/her own rather than being assigned one by the scheduler unilaterally. A mechanism for facilitating negotiations between the patient and the scheduler is therefore included in the proposed model. The scheduler does not just assign a patient to a particular physician and time slot, instead a set of options is offered to the patient. A separate DP model is used to deal with this requirement. A decomposition method and a column generation algorithm are used while performing the computations needed in solving the model. Numerical studies have shown that these are acceptably efficient and accurate. The effects of booking horizon and patient arrival rate are also studied. A policy concerning how one can make use of the solutions proposed by the model is introduced next along with a quasi-optimal policy aimed at mitigating computational complexity. Since no-shows are common in advance appointment systems, appointment scheduling with no-shows is also examined. Finally, an overbooking policy is introduced to reduce losses resulting from no-shows.

    Research areas

  • Hospitals, Medical appointments and schedules, Waiting lists, Dynamic programming