CORS / Optimization Days 

HEC Montréal, May 29-31, 2023


HEC Montreal, 29 — 31 May 2023

Schedule Authors My Schedule

QMHOII Queueing Models in Healthcare Operations II

May 31, 2023 10:30 AM – 12:10 PM

Location: Rona (blue)

Chaired by Marco Bijvank

4 Presentations

  • 10:30 AM - 10:55 AM

    An oncology clinic increases efficient operations

    • Alexandru Popp, presenter, University of Saskatchewan
    • Chris Zhang, University of Saskatchewan

    In this research there is an what-if scenario (WIS) that is compared to a baseline simulation model (BAS). The environment is a Canadian ambulatory oncology clinic. The WIS was created to test a very specific approach that the Saskatoon Cancer Clinic (SCC) can implement and become more efficient.
    The only administrative change that is made (the WIS) is changing the assignment of 2 rooms for one oncologist to having 2 oncologists sharing 3 examination rooms. The intent of this scenario is to increase the number of patients served as well as to reduce the waiting period for the first appointment of patients and the time waiting in the clinic
    While the BAS was verified and validated, so was the WIS validated statistically via multiple tests. There are 14 Key Performance Indicators (KPIs) that are considered. Having various KPIs, each type of stakeholder is able to understand the results in their own context.
    The average number of patients that are serviced per day increased in the scenario. This is one characteristic that the SCC management desires because it results in a higher average utilization of the examination rooms. Out of the 14 KPIs measured, there is only one that is worse for the WIS compared with the BAS.

  • 10:55 AM - 11:20 AM

    Partial Flexibility in the Interface of Emergency Department and Hospital Wards

    • Mahdi Shakeri, presenter,
    • Haji Babak, Sharif University of Technology
    • Kamali Farrokhvar Leily, California State University, Northridge

    Emergency department (ED) crowding remains a significant challenge for healthcare systems worldwide. Boarding time, the time ED patients wait for admission to inpatient wards, is recognized as a critical contributor to this problem. We propose a partially flexible approach for admitting ED patients to hospital wards with the goal of minimizing board time while avoiding negative impacts on the quality of care and staff satisfaction. To achieve this, we consider a two-class queueing system with two heterogeneous server pools as our base model and analyze it by adopting a Matrix Analytic approach. By utilizing probability-generating functions (PGFs), we derive closed-form expressions to approximate the queue length and waiting time for the corresponding queueing system. We further introduce a partially flexible routing policy, designed based on process flexibility principles and the findings from the base model. To assess the effectiveness of our proposed approach, we conduct a simulation study using a fully flexible structure as a benchmark. The fully flexible structure allows all patients to be assigned to any inpatient ward. Our approach achieves similar performance in terms of boarding time as the fully flexible structure while reducing complexity and avoiding the negative impacts of high flexibility levels.

  • 11:20 AM - 11:45 AM

    Data-driven capacity estimation

    • Yawo Kobara, presenter, Rotman School of Management, University of Toronto
    • Opher Baron, Rotman School of Management, University of Toronto
    • Dmitry Krass, Rotman School of Management, University of Toronto

    Data-driven capacity estimation is one of the most significant issues in practical queuing theory.
    While several models have been put forth in the literature, they typically assume the presence of complete information, i.e., time-stamps for arrivals, service starts and departures. In many practical settings only a subset of these timestamps is available (often consisting of system arrivals and departures).

    This paper presents two simple algorithmic approaches to estimate the capacity of a partially observable G/G/c queueing system. The observer can only see the arrival and departure times of the customers and wishes to estimate the number of servers.

    One algorithm is based on Krivulin's recursion, while the other one is based on the number of overtakes. Using both, synthetic data, and real-life data from the emergency department of a larger urban hospital data, an comparative evaluation of the two algorithms is conducted. Our results show for a stationary G/G/c system, our algorithms produce extremely accurate estimates of the number of servers based on rather limited data.

  • 11:45 AM - 12:10 PM

    Managing Capacity Reservation for Low-Priority Strategic Patients

    • Marco Bijvank, presenter, University of Calgary
    • Guanlian Xiao, Cape Breton University

    Emergency department crowding prevents timely care delivered to patients in urgent need of treatment. Delay announcements can be an approach to divert low-acuity patients. We study a healthcare system that operates two parallel tracks, i.e., a shared track and a dedicated track, to serve high-priority and low-priority patients. High-priority patients are assigned to the dedicated track. However, when the dedicated track is relatively busy, high-priority patients are diverted to the shared track with a non-preemptive priority over low-priority patients. Low-priority patients are strategic, and they choose to join the waiting queue on the shared track, or to balk from the system and seek service elsewhere. Their join-or-balk decision is made based on the utility of joining after obtaining delay information. In our study, we consider two types of expected waiting time information to be revealed to low-priority customers: long-term expected waiting time, and real-time expected waiting time. Under both strategies, we evaluate how much capacity to reserve for the dedicated track when the total number of servers over both tracks is a fixed constant, and what impact this has on the expected wait times for both types of patients as well as the throughput rate.