Our work environment is intricately linked with productivity and well-being. Given the motivation and investment in time and resources, it is perhaps possible today to change almost any aspect of our work environment.
In reality, however, our scope is limited by several resource constraints. To work within these constraints, we need to understand how they are interrelated.
Improving productivity, quality and safety are some of the key motivating factors driving the need to change the work environment. This post will deal with how computational models can augment techniques to improve the ED environment.
Building Effective Models
The organization of objects within the layout in an ED can be quite complex. Imagine walking into the department for the very first time. Every room, its relation to other rooms, equipment at bedside and elsewhere, has a purpose. It is hard to grasp the complexity in the first visit. When building a computational model for an ED, this is where one would start.
The rules in the work environment will need to be meticulously defined and coded into the model. Using approaches in machine learning, these rules can also be learned from data without needing to be explicitly coded1,2.
Irrespective of the approach, the fidelity of data used (i.e., the degree to which it approximates real-world conditions) determines the utility of the model. Choosing the right fidelity ensures that we do not overly simplify or complicate the model.
In the plot above, patient-related events in an ED are shown to be distributed across time and space. Depending upon the modeling goals or problem statement, these events may have to be represented at different levels of fidelity3. To illustrate this, let's consider two hypothetical problem statements.
1.What is the optimum space requirement for the waiting room in a given emergency department (ED)?
2.How can we reduce medication errors in the ED?
The following table gives examples of the fidelity required in terms of the data for each event during the patient visit in an ED to build a model for the given problem statements.
For a model to determine the optimum waiting space, variation in terms of patient arrival and start of triage should suffice. The dynamics in the treatment and discharge can be modelled together as one event. Thus, the granularity of the triage, treatment and discharge events are reduced.
For a model to reduce medication errors, all events may need to be either of the same or higher fidelity as the model for optimum waiting space. In particular, higher fidelity is required for the treatment event. Here, details of diagnoses, procedures, medication orders and administration should be considered. In addition, activities of providers including their background activities and the pharmacy as well as several spatial attributes such as travel distances, location of equipment and visual cues may be required.
Benefits and Limitations of Models
Defining the problem statement, creation of rules and selection of fidelity are steps in the planning and building phase of the model. After the model is built, it needs to be validated. The entire process is often iterative4.
Given the investment in the effort upfront, a computational model may not be appropriate for simple problems that can be solved using common sense. Problems tackled during a kaizen event may be of this nature. However, if a customizable model already exists, it can be reused to test ideas arising out of such events.
It is often the case that insights obtained from experiments within an emergency department are not easily disseminated and implemented at other emergency departments5. While the model itself is not universally applicable, the rules used to build the model can function as a common platform to share this intelligence.
In process improvement efforts such as lean, reduction of waste and bottlenecks in the flow are the focus. Releasing one bottleneck can potentially create others downstream in the process flow. In a complex and non-linear system, it may not be possible to detect and solve these latent bottlenecks before implementing the solution. A computational model such as a discrete event simulation model (one in which the process steps execute one at a time) can assist by predicting such bottlenecks prior to implementation.
In the design of spaces, use of building information models (BIM) is becoming universal. These models capture the physical and functional aspects of spaces and are pretty information intensive. Integrating relevant information from BIM into a computational model can allow the exploration of the subtle connections between spaces and processes. For example, using computational fluid dynamics (a branch of fluid mechanics that uses algorithms to analyze the flow of liquids and gases) within a space defined by the BIM model, one can explore the impact of air flow on infection control6. This can be done, for example, while evaluating before-and-after scenarios of placing new equipment within a patient room.
Computational models can also be deployed in real time to monitor both spatial and process attributes. Such models, for example, can be devised to find the most efficient way to perform tasks by a provider while reducing travel time and increasing the time with patients. There is potential to improve patient outcomes and reduce staff burnout by looking at spatial attributes in conjunction with processes.
To conclude, computational models are useful as decision support tools in a complex multidimensional and nonlinear environment such as the ED. Models should be at the right fidelity and complexity to be meaningful. With the application of machine learning and advances in cloud computing, computational models can be built and operated faster than ever before.
Computational models can be a valuable tool to explore the relationship between interrelated spaces and processes to help EDs realize new opportunities to improve.
Additional reading: Aditazz White Paper: “Operational Modeling: A Key to Addressing Emergency Department Overcrowding”
1. Taranath N, Patil SB, Patil P, Subbaraya CK. Medical decision support system for the missing data using data mining - A survey. In: Contemporary Computing and Informatics (IC3I), 2014 International Conference on. ; 2014:192-196. doi:10.1109/IC3I.2014.7019697.
2. Bisbal J, Engelbrecht G, Villa-Uriol MC, Frangi AF. Prediction of cerebral aneurysm rupture using hemodynamic, morphologic and clinical features: A data mining approach. Lect Notes Comput Sci (including Subser Lect Notes Artif Intell Lect Notes Bioinformatics). 2011;6861 LNCS(PART 2):59-73. doi:10.1007/978-3-642-23091-2_6.
3. Gunal MM. A guide for building hospital simulation models. Heal Syst. 2012;1(1):17-25. doi:10.1057/hs.2012.8.
4. Marshall DA, Burgos-liz L, Eng I, et al. Selecting a Dynamic Simulation Modeling Method for Health Care Delivery Research — Part 2 : Report of the ISPOR Dynamic Simulation Modeling Emerging Good Practices Task Force. Value Heal. 2015;18:147-160. doi:10.1016/j.jval.2015.01.006.
5. Welch SJ. Using Data to Drive Emergency Department Design: A Metasynthesis. HERD Heal Environ Res Des J. 2012;5(3):26-45. doi:10.1177/193758671200500305.
6. Tang JW, Noakes CJ, Nielsen P V, et al. Observing and quantifying airflows in the infection control of aerosol- and airborne-transmitted diseases: an overview of approaches. J Hosp Infect. 2011;77(3):213-222. doi:10.1016/j.jhin.2010.09.037.