Reducing Waiting Times in the Emergency Department via “Diversion”

In the past 25 years, demand for emergency healthcare in the United States has exploded, pushing capacity-strained emergency departments across the country to the limit and prompting organizations from the National Academy of Medicine to the General Accounting Office to label the American emergency care system “in crisis.”

Between 1992 and 2012, emergency room utilization, defined as the number of visits per 1,000 people, grew by approximately 18% in the United States, according to a 2012 survey by the American Hospital Association. For many of those patients—the majority of whom require immediate care—even just a few minutes can make a significant difference in their health. In some instances, patients will be diverted to other hospitals, or encouraged to receive care at urgent care facilities when congestion is very high. Adopting a proactive approach to patient diversions could help hospitals minimize patient wait times and raise the odds of achieving the best outcomes.

The use of predictive analytics in healthcare has been growing. Much effort has been spent developing models to predict patient risk of bad outcomes, such as admissions (and/or readmissions). These models have largely been developed with the intent to guide operational decision making and allow hospital administrators and clinicians to better utilize limited healthcare resources. While there has been substantial attention paid to the development of these models, too little has been done to demonstrate whether—and how—they can best be utilized to improve system performance.

In our recent paper, we examine how predictions of patient arrivals to the emergency department can be used to make better diversion decisions, resulting in less waiting for patients who are treated.

As emergency departments have faced growing numbers of patients over the past several decades, they have increasingly been forced to “divert” patients to receive care at other hospitals or encourage them to visit urgent care facilities or their primary care physicians. In practice, such strategies are employed in an online fashion where only information about the current congestion is utilized. Once the congestion in the emergency department reaches a maximum threshold, patients will be diverted.

Rather than relying exclusively on present congestion, however, emergency departments can use future information to make better diversion decisions, allowing them to treat the same number of patients, while incurring much lower waiting time.

Utilizing predictions of patient arrivals to determine when congestion is going to build in the future, we propose a policy to strategically divert patients in order to minimize the delay experienced by treated patients. Intuitively, our proposed policy diverts patients at the beginning of a “bursty episode” of patient arrivals. Because these patients, if admitted, tend to delay all subsequent arrivals during the bursty episode, their diversions are beneficial for reducing the overall delay. In contrast, an online policy can start diverting only after a bursty period materializes, by which point long wait times have already become inevitable.

Graph of time-variation in arrivals

The figure above depicts the time-variation in arrivals to the emergency department and shows how the proposed proactive diversion policy begins diverting patients much earlier in the day than the optimal, online diversion policy.

One of the challenges with using predictive models to guide operational decisions is that such models can only provide noisy versions of actual demand realizations. A natural question is whether using such noisy information can still improve delay or if it is better to simply ignore future information and resort to the current information used by online diversion policies.

We find that even diversion decisions based on noisy predictions will never increase delays—in most cases delay will even be reduced. These gains, however, are potentially achieved at the expense of over-diversion. As such, each emergency department is associated with a quantifiable “noise tolerance” such that our proposed methodology still provides improved delay guarantees as long as the prediction model’s accuracy is within the designated noise tolerance.

By carefully using future information to make strategic diversion decisions, the proposed policy can treat the same number of patients while simultaneously reducing the wait time of patients by up to 15% compared to the current best practice. Even when using a predictive model with very high noise and minimally improved predictive capabilities over average arrival rates, these savings can be up to 8%.

With the increase in the development of predictive analytics in healthcare, it is of growing importance to understand how such models can be used to improve delivery of care and operational decision making across the field. This work takes a first step toward understanding how to use predictions of patient arrivals to reduce wait times, thereby providing timelier access to care and, potentially, better patient outcomes.


Xu K, Chan CW (2016) Using future information to reduce waiting times in the emergency department via diversion. Manufacturing Service Operations Management, ePub ahead of print March 22,