The urgency of the situation is clear from the name: If you’re headed to the emergency room, you’re hoping to receive medical treatment quickly and efficiently.
Unfortunately, as emergency room utilization has increased in recent years, wait times have ballooned, creating a situation that sometimes denies patients timely treatment when they need it most. One study indicated that 47% of U.S. hospitals’ emergency departments are either at or over capacity. Another study found that patients who arrived during times of high crowding had a 5% higher chance of in-patient death.
When Stanford Graduate School of Business professor Kuang Xu and Columbia Business School professor Carri Chan teamed up to investigate this phenomenon, they vowed to create the groundwork for a theoretical cure to the system’s congestive ills. Their answer, as published in a recent issue of Manufacturing & Service Operations Management: a model that focuses on translating predictions about future ER arrivals into better decisions to minimize patient waits.
Xu says that when applied efficiently this model can reduce delays by up to 15% and might actually save lives. “Above all else we’re motivated by understanding the value of information,” says Xu, who specializes in data-driven decision-making within dynamic environments. “The more you know, the more you can slash the waiting time.”
In recent years, one way to deal with long wait times at ERs in many states has been to divert patients to receive care at other facilities. Once congestion in a particular ER reaches a tipping point — in other words, once a delay already exists — the hospital puts out the equivalent of a red alert and tells ambulance services to transport incoming patients elsewhere.
Current diversion models rely exclusively on congestion data, but Xu’s model incorporates predictive data into the mix — even if that data is unreliable, or “noisy” — to determine when congestion will build in the future. If this predictive data suggests patient arrivals may spike, in what Xu calls “bursty episodes,” the algorithm calls for hospitals to divert patients immediately, before congestion can occur and trigger ongoing delays.
Essentially, the new model calls for preemptive action instead of reactive response, reducing ER wait times for all involved.
“These delays can have significant, life-changing ramifications,” Xu says. “These are the kinds of changes that potentially could affect all of us.”
Challenges to the Approach
The new approach to predictive modeling of ER traffic is noteworthy because, to a certain extent, it is error-proof — that is, the model is set up to reduce delays even when the predictive information is noisy.
Xu says the approach still is impacted by unreliable data in the sense that the more noise that exists, the worse the performance will be. Still, he adds, the main novelty of the approach is to push people to think in terms of future information.
“If you’re prepared to handle congestion, you’re never really in a bad spot,” he says. “Any degree of predictive information added to the equation is going to be better than none of it.”
Xu notes that the model is not capable of handling unpredictable “shocks” to the system — unforeseen events such as terror attacks or train crashes that cause an increase in the magnitude of arrivals. Because these occurrences are exceptional, Xu and Chan did not consider them at all.
“We did not invent new ways of predicting future arrivals to the ER,” says Xu, highlighting that the focus of the work was not on making predictions, but instead on using predictions to make better decisions. He adds that there are other complicated equations that take into account weather and other factors that could create spikes in future patient arrivals.
Looking forward, Xu and Chan plan to advance their research by analyzing the impact of predictive analytics on ER waiting times, sharpening the model to make it more useful for a general audience.
Xu adds they also hope their research drives policy change and inspires hospital administrators to come up with their own ways of managing “bursty episodes” proactively.
“We’re hoping to educate practitioners, not just hand them a disk of software and say, ‘Use this,’” he explains. “Imagine a situation where a hospital gauges its own level of business and puts a notice on the website that says, ‘We’re really busy right now; if you’re not that sick, don’t come in.’ That simple change could improve delays significantly.”
He adds: “It matters less whether you’re full at the moment and more who will come at another time.”
However medical health facilities embrace this new approach, eliminating the chronic overcrowding and long wait times that plague ERs around the world likely would require an end-to-end transformation of the way medical care is delivered. Until then, Xu’s model facilitates predictive analytics to make meaningful incremental improvements within the existing system and to help patients receive the care they need when they need it.