Access Type

Open Access Dissertation

Date of Award

January 2020

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Industrial and Manufacturing Engineering

First Advisor

Ratna B. Chinnam

Second Advisor

Evrim Dalkiran

Abstract

Healthcare systems face difficult challenges such as increasing complexity of processes, inefficient utilization of resources, high pressure to enhance the quality of care and services, and the need to balance and coordinate the staff workload. Therefore, the need for effective and efficient processes of delivering healthcare services increases. Data-driven approaches, including operations research and predictive modeling, can help overcome these challenges and improve the performance of health systems in terms of quality, cost, patient health outcomes and satisfaction.

Hospitals are a key component of healthcare systems with many scarce resources such as caregivers (nurses, physicians) and expensive facilities/equipment. Most hospital systems in the developed world have employed some form of an Electronic Health Record (EHR) system in recent years to improve information flow, health outcomes, and reduce costs. While EHR systems form a critical data backbone, there is a need for platforms that can allow coordinated orchestration of the relatively complex healthcare operations. Information available in EHR systems can play a significant role in providing better operational coordination between different departments/services in the hospital through optimized task/resource allocation.

In this research, we propose a dynamic real-time coordination framework for resource and task assignment to improve patient flow and resource utilization across the emergency department (ED) and inpatient unit (IU) network within hospitals. The scope of patient flow coordination includes ED, IUs, environmental services responsible for room/bed cleaning/turnaround, and patient transport services. EDs across the U.S. routinely suffer from extended patient waiting times during admission from the ED to the hospital's inpatient units, also known as ED patient `boarding'. This ED patient boarding not only compromises patient health outcomes but also blocks access to ED care for new patients from increased bed occupancy. There are also significant cost implications as well as increased stress and hazards to staff. We carry out this research with the goal of enabling two different modes of coordination implementation across the ED-to-IU network to reduce ED patient boarding: Reactive and Proactive. The proposed `reactive' coordination approach is relatively easy to implement in the presence of modern EHR and hospital IT management systems for it relies only on real-time information readily available in most hospitals. This approach focuses on managing the flow of patients at the end of their ED care and being admitted to specific inpatient units. We developed a deterministic dynamic real-time coordination model for resource and task assignment across the ED-to-IU network using mixed-integer programming.

The proposed 'proactive' coordination approach relies on the power of predictive analytics that anticipate ED patient admissions into the hospital as they are still undergoing ED care. The proactive approach potentially allows additional lead-time for coordinating downstream resources, however, it requires the ability to accurately predict ED patient admissions, target IU for admission, as well as the remaining length-of-stay (care) within the ED. Numerous other studies have demonstrated that modern EHR systems combined with advances in data mining and machine learning methods can indeed facilitate such predictions, with reasonable accuracy. The proposed proactive coordination optimization model extends the reactive deterministic MIP model to account for uncertainties associated with ED patient admission predictions, leading to an effective and efficient proactive stochastic MIP model.

Both the reactive and proactive coordination methods have been developed to account for numerous real-world operational requirements (e.g., rolling planning horizon, event-based optimization and task assignments, schedule stability management, patient overflow management, gender matching requirements for IU rooms with double occupancy, patient isolation requirements, equity in staff utilization and equity in reducing ED patient waiting times) and computational efficiency (e.g., through model decomposition and efficient construction of scenarios for proactive coordination). We demonstrate the effectiveness of the proposed models using data from a leading healthcare facility in SE-Michigan, U.S. Results suggest that even the highly practical optimization enabled reactive coordination can lead to dramatic reduction in ED patient boarding times. Results also suggest that signification additional reductions in patient boarding are possible through the proposed proactive approach in the presence of reliable analytics models for prediction ED patient admissions and remaining ED length-of-stay. Future research can focus on further extending the scope of coordination to include admissions management (including any necessary approvals from insurance), coordination needs for admissions that stem from outside the ED (e.g., elective surgeries), as well as ambulance diversions to manage patient flows across the region and hospital networks.

Download

Share

COinS