Building and Implementing a Predictive Decision Support System Based on a Proactive Full Capacity Protocol to Mitigate Emergency Department Overcrowding Problems

Emergency departments are often overcrowded, leading to patient safety risks, poorer patient outcomes, and treatment delays with increased morbidity and mortality. Overcrowding is a complex multifactorial problem, including usage by those without insurance with needs that are not acute, complexity of complaints and injuries, large patient volumes, limited resources, and poor patient flow. Per the American College of Emergency Physicians, the key to improving patient flow and avoiding overcrowding is a full capacity protocol (FCP), an approach used internationally that focuses on communication between the ED and inpatient floors. FCP uses patient flow measures (PFMs) that set criteria to trigger escalating levels of interventions. PFMs include metrics such as the number of boarding patients in the ED while waiting for an inpatient bed. Currently, these systems are reactive since they use real-time values of PFMs, when mitigating strategies may be implemented too late.

This research will use deep learning models to move a reactive FCP for ED overcrowding interventions into a proactive FCP by predicting patient flow measures so that interventions may be activated to avoid overcrowding.

The specific aims of the research are as follows:

• Develop deep learning models to predict PFM values and incorporate them in a proactive FCP.
• Develop a discrete event simulation (DES) model to evaluate the effectiveness of the proactive FCP.
• Design, evaluate, and implement a decision support system (DSS) based on the proactive FCP.
• Expand and generalize the DSS by standardizing data input and output interfaces.

Multiple deep learning models will be built to predict PFM values which will be included in the proactive version of the FCP. A DES model will be used to measure the effectiveness of the proactive versus reactive FCP to compare outcomes generated, such as average length of stay, waiting time, and staff satisfaction. The proactive patient flow decision support system (PPF-DSS) dashboard will display both real-time and predicted values of PFMs and the PPF-DSS will use the proactive FCP to activate the FCP interventions and automate parts of it. From there, the researchers will standardize the data input and output interfaces of the PPF-DSS using FHIR so that the tool may be shared in an electronic health record vendor-agnostic way. The finalized tool will allow sites to configure their workflows and policies, incorporate the ability to train the underlying model, allow for evaluation, and have streamlined processes to allow for customized implementation.

This research will result in a shareable vendor-agnostic proactive FCP DSS tool that will allow the early identification of potential ED overcrowding situations with resultant triggering of mitigating actions to avoid these situations. With less overcrowding, it would be expected that there would be resultant improvements in patient safety, outcomes, and mortality.