Seyed Iravani--Research

We address the benefits of virtual appointments in chronic care by considering the clinical and operational aspects of the problem setting. We used a finite-horizon stochastic dynamic programming model to study the following questions: (i) Given limited capacity in office and virtual appointments, what should be the optimal appointment scheduling policy? (ii) Observing that shorter office revisit intervals result in increased interaction between patients and physicians in chronic care, what is the impact of virtual appointments on the revisit intervals for office visits? and (iii) What is the value of virtual appointments in the overall health status of the patients? The contribution of the project is that it considers the clinical aspects of the problem as well by incorporating the disease progression and the differences in the treatment effectiveness of office and virtual appointments. While the optimization problem has a complex structure, we develop some easy-to-implement scheduling policies that perform very well in most settings. (Joint work with A. Bayram, S. Deo, and K. Smilowitz)

In this project, we develop a knowledge-based queueing model to study such teletriage systems. We also develop a novel approach to model knowledge of agents in similar service systems that allows us to compute the fraction of misclassification of patients. Teletriage systems are one of what we call hierarchical knowledge-based service systems (HKBSS) in which hierarchically organized agents with different knowledge levels assess cases and must either issue a decision or refer the case to a higher level. Other HKBSS examples include the U.S. Department of State Bureau of Consular Affairs, in which agents must decide whether or not to grant visa applications and can refer cases to a supervisor, and the mortgage department of a bank, in which loan officers must decide whether or not to approve mortgages and can refer cases to a manager.
While speed-versus-quality trade-offs are common in operations management, those of an HKBSS present a unique modeling challenge of how to incorporate the presentation of agent knowledge and decisions into a queueing model. Therefore, we develop a novel approach to model agents' knowledge and using partially observable Markov Decision Process (MDP) and queueing theory, we characterized the structure of the optimal decisions in each hierarchy of HKBSS systems. (Joint work with S. Saghafian, W.J. Hopp, Y. Cheng, and D. Diermeire)

Virtual appointments, including e-mail, phone, and online consultations, can offer a cost-effective alternative to traditional office appointments for managing chronic conditions. They also permit physicians to see more patients and increase patients’ access to care. Demand for virtual appointments is increasing, especially with the Covid-19 pandemic. Despite the increased usage of virtual appointments and their observed benefits, especially in chronic care, the integration of virtual appointments with office appointments limits large scale adoption. In this project, we studied a chronic care setting in which patients are scheduled for routine follow-up visits through virtual or office appointments.

Triage is the process of assessing patient urgency in the Emergency Departments (ED) of hospitals, traditionally through a short interview by a triage nurse. In addition to speed, accuracy is vital in triage because errors in patient classification can cause dangerous delays in treating urgent patients. In order to improve triage decisions, some hospitals have begun experimenting with telemedical physician triage (TPT). In TPT, after examining a patient, a triage nurse has the option to refer that patient to a telemedicine booth through which a remote physician conducts a video conference and renders a triage decision. The telemedical physician typically services multiple hospitals; hence, patients referred to the TP may have to wait in a queue. Therefore, when considering referring a patient to the TP, a triage nurse must balance the queueing delay with the benefit from a review by the more knowledgeable physician.

In 2000, due to economic stress and low performance, the Board of Directors of Virginia Mason Medical Center issued a mandate for change with the goal of improving organizational culture and becoming the quality leader in healthcare. Seattle’s Virginia Mason Medical Center is an integrated healthcare system with 400 physicians, 5000 employees, 9 locations, and a 336-bed hospital. The leadership had the vision of Virginia Mason Production System (VMPS), which was modeled based on the lean concepts in Toyota Production System. In 2002, to have their senior leaders learn lean principles and to see how lean manufacturing works in action, Virginia Mason sent all its senior executives to Japan. Their executives worked on the production line at the air conditioning plant of Hitachi. They mapped the process, measured flow times, inventory and throughput. The lesson they learned, according to senior leaders, was that healthcare has many features in common with manufacturing. Virginia Mason therefore required all its 5000 employees to attend an “Introduction to Lean Operations” course. Using several lean operations tools, Virginia Mason was able to significantly improve its operational and economical performance. Inventory was reduced by 53 percent, floor space was reduced by 41 percent, lead times were reduced by 65 percent, and setup times were reduced by 82 percent. These waste reductions resulted in significant savings in capital expenses, including $1 million for an additional hyperbaric chamber, $1 to $3 million for endoscopy suites, and $6 million for new surgery suites.
Virginia Mason was not the only healthcare institution that benefited from lean operations. Using operations management principles, the University of Pennsylvania Medical Center was able to reduce hospital-acquired infections, saving 57 lives and reducing costs by over $5 million over 2 years. Avera McKennan Hospital in South Dakota was able to reduce the length of stay of emergency patients by 29 percent and thus avoided $1.25 million in new emergency department (ED) construction. By eliminating waste, Denver Health, Colorado, was able to achieve cost saving of $200 million over 7 years, while achieving “the lowest observed-to-expected mortality among the academic health center members of the University Health System Consortium in 2011.”
My research utilizes operations research techniques such as queueing theory, Markov decision process, stochastic dynamic programing and game theory to improve the performance of healthcare systems. Below are two of my past projects.