Analytics approach aims to cut overcrowded ERs

Using data analytics to understand hospital emergency department overcrowding and wait times, two researchers have developed a methodology to predict future ER demand.

Hospitals that use their analysis could use the results to reduce wait times for patients by as much as 15 percent, the researchers contend.

The methodology uses machine learning technology to assess data on known patterns of ER activity, say Carri Chan, associate professor of business at Columbia Business School, and Kuang Xu, an assistant professor at Stanford Graduate School of Business.

Carri Chan

Their approach takes into account factors such as time of day, general level of severity, holidays, weather patterns, bad air quality, flu season and special events, to predict how many walk-in patients will come during a certain time period. That data then can help providers determine when to begin diverting them to their primary care physician, an urgent care facility or another hospital, as well as when to start diverting ambulances to other facilities.

The methodology enables hospitals to establish thresholds for when decisions on diversion start to be made to have the largest impact on reducing congestion, Chan says.

Such a data-driven approach may be able to help hospitals make decisions before patients start to show up and begin waiting for treatment. For instance, a typical approach for a hospital may be to wait until 10 patients are seeking non-emergency care and to divert patients from that point forward, but if a hospital diverts the first or second non-emergency patients coming in during a certain time, they free up resources to better treat patients who truly need emergency care.

Before using the methodology, hospitals first must collect and analyze their own data to understand their current patterns, according to Chan. She suggests first testing the algorithm and methodology on a small sample.

“Try the methodology in a small case study for a week and compare delays with the new algorithm, compared with the status quo,” she counsels.

Now, the authors are working with start-up analytics companies to have the proposed algorithm and methodology incorporated in a software package. The study is available here.

For reprint and licensing requests for this article, click here.