A growing number of hospitals are turning to predictive analytics to anticipate and manage volume better. The approach, which involves using sophisticated simulation and modeling techniques, enables administrators to get ahead of patient surges and to focus on pressure points. For example, Johns Hopkins Hospital in Baltimore has made significant progress on a range of measures, using a centralized command center to monitor the hospital’s data streams. The approach enables the hospital to accelerate decision-making and optimize hospital resources. Investigators at Columbia University believe similar modeling techniques can be used to avert ED congestion when used in conjunction with proactive diversion strategies.

  • The 5,000-square-foot command center at Johns Hopkins Hospital monitors 14 IT systems on a 24/7 basis so that all relevant inputs can factor into decision-making about beds, transfers, consults, admissions, discharges, and other aspects of care. Administrators say they have been able to achieve 96% accuracy in their predictions.
  • In just 10 months of operation, the data-driven command center has achieved dramatic improvements, including a 30% reduction in the number of emergency patients who must wait for an inpatient bed and a one-hour reduction in the time it takes to get out the door to retrieve a patient identified for transfer to the Hopkins facility. In addition, the hospital has all but eliminated procedure cancellations due to OR holds.
  • Investigators at Columbia University contend that by using predictive analytics to guide proactive diversion strategies, ED delays can be reduced by as much as 15%.

Hospitals have been relying on data to manage staffing and resources for decades, but EDs still become overwhelmed at times, and unanticipated surges often lead to long waits, boarding, and other negative effects. However, some pioneering medical centers are taking data-driven operations to a whole new level, leveraging predictive analytics to boost performance on a range of measures while also maximizing resources. Further, some are even experimenting with these newer modeling techniques to see if anticipated patient surges can be averted through the use of diversion strategies.

Leverage High-tech Tools

Jim Scheulen, PA, MBA, the chief administrative officer for the Johns Hopkins Department of Emergency Medicine in Baltimore, manages the operations of five busy EDs, which altogether handle roughly 250,000 patients a year. He also oversees the hospital system’s transfer center and a new command center at Johns Hopkins Hospital that monitors 14 IT systems on a 24/7 basis so that all the relevant inputs can factor into decision-making about beds, transfers, consults, admissions, discharges, and other aspects of care.

The command center, which was developed in concert with GE Healthcare Partners and began operations in February 2016, is seeking to make use of the same kinds of sophisticated, mathematical tools that many other industries have long since integrated into their practices.

“We recognized that running a huge hospital like Hopkins is as complex as running any other complex industry,” Scheulen explains.

For example, he notes that aerospace companies and automobile manufacturers use systems engineering tools to maximize efficiency and boost productivity.

“What that means is that besides the usual process improvement work, they also use all sorts of simulation models, deep data analytics, statistical modeling, and predictive modeling ... as a routine part of what they do,” he says. “The other thing they do: To operate in these complex environments, they bring together everybody who needs to be together to make the place run on a daily basis, they give them all the information they need to run it, and they give it to them in real time.”

Centralize Decision-making

For Scheulen, the concept of a centralized command center that houses all the relevant information as well as all the key decision-makers in one place makes perfect sense.

“It used to be that when we got a referral in from another hospital or we wanted to admit another patient through the ED, all of these calls would have to go through admitting, our transport group, and our bed managers,” he explains.

Given that none of these players were in the same place, they would have to communicate through phone calls, faxes, emails, or texts, an approach that would delay the whole process, Scheulen observes.

“We thought if we were just able to bring all these people together, we would take all of that away,” he says.

Further, with improved efficiency, the thinking was that the hospital also could make headway on issues such as ED boarding and OR holds, and hospital administrators could expedite the process of accepting patients referred to Hopkins from other hospitals for the tertiary and quaternary care that Hopkins can provide.

In fact, in just 10 months of operation, the data-driven command center has achieved dramatic improvements, including a 30% reduction in the number of emergency patients who must wait for an inpatient bed.

“We have reduced the time from when an order [for admission is completed] and the bed is assigned by at least 45 minutes,” Scheulen explains. “We have also seen a very significant reduction in the amount of time it takes for us to go get a patient who might be having an emergency at an outside hospital and needs to come in. We have reduced the time it takes for us to get out the door to get that patient by one hour.”

Even more dramatic has been the effect on OR holds.

“We used to routinely have several OR holds a day, which would lead to case cancellations, but over the last several months we have had essentially zero cancellations due to OR holds,” Scheulen notes, adding that many other improvement efforts are in process. “We are still developing a lot of tools, and learning them as we go.”

Apply Predictive Analytics

The command center is housed in a 5,000-square-foot facility that is positioned in the center of the hospital campus. “It is a very cool, awesome-looking, NASA-like place,” Scheulen says.

There are 22 information screens or “tiles” that retrieve and display a constant flow of information that feeds into the center from the hospital’s IT systems. At any time of the day or night, command center staff can see real-time information about incoming ambulances, pending admissions or discharges, backups in the OR, and many other aspects of care documented in the hospital’s electronic medical record system.

Four groups of personnel monitor all this information throughout the day: the people who take calls from physicians who want to transfer patients to the hospital, bed management nurses, the paramedics who arrange transportation for patients arriving at the hospital from other facilities, and admitting staff who handle all the financial clearances and registrations for patients being admitted, Scheulen explains. Guiding all the decisions taking place in the command center are complex calculations that enable the hospital to anticipate and manage patient volume.

“The first thing we did was build a simulation model that replicated the operations of Johns Hopkins Hospital,” Scheulen explains. “It was unbelievably complex and very difficult to do, but once we had it done, then we were able to see where we should target our process improvement efforts.”

On a daily basis, administrators use the mathematically based tools to look two or three days ahead to see what the occupancy of beds will be and put appropriate plans in place. For example, perhaps the surgical schedule needs an adjustment to manage the anticipated capacity better, or perhaps certain patients can be discharged earlier to clear beds on days when demand is expected to spike. The information enables decision-makers to act proactively to avert bottlenecks.

“We use that information when we have our bed huddles,” Scheulen explains. “We are also developing a tool ... that will enable us to predict on an individual patient basis who will be able to go home three days from now, two days from now, and tomorrow.”

With this type of data, staff can begin the discharge process sooner so patients exit the hospital earlier in the day, creating needed capacity, Scheulen notes.

“People will say that they already do this, but they actually don’t because even if they are in a huddle or a multidisciplinary setting, they might say that a patient is going to go home in two days, but they aren’t sure ... and then all of a sudden the patient is ready to go home, but they haven’t written any of the [necessary] orders,” he explains. “So if we can use this [predictive] information, and then change the conversation during these rounds [or huddles] to say that this predictive model says a patient is almost certainly going to go home tomorrow, then we can have the residents tonight prepare their prescriptions and whatever else the patient needs to be [ready for discharge] tomorrow.”

Fine-tune Predictions

Scheulen acknowledges that when it comes to emergency medicine, there are events that you cannot predict, but the effect of these events on patient utilization predictions is surprisingly small.

“Hopkins in particular has a lot of data,” he says. “We are within 96% accuracy in our predictions.”

Further, the comprehensive view provided by the predictive, analytical modeling used in concert with the command center has helped hospital administrators focus on the very specific, targeted things they need to do to make better progress, Scheulen observes.

For example, administrators have been able to determine that it is not always a lack of beds that slows throughput, but rather provider availability. “So our department of medicine is planning to restructure the way its residency program operates ... so that providers will be available all the time and there won’t be any gaps,” Scheulen explains. “That is going to make a major difference.”

Another alteration in the works has to do with the way emergency providers arrange for consults from specialty providers, a process that has been plagued with bureaucracy and delays, leading to boarding in the ED.

“We are changing the way the consult process works, putting it through the command center and having time expectations for when [a specialty provider] is going to respond,” Scheulen says.

As processes change and improve, administrators continue using the simulation model to determine what the next process improvement efforts should be, Scheulen notes.

“The people at Hopkins recognize that we have an incredible demand for our service, and they do everything in their power to take care of it,” he says. “We are literally down now to looking at how many minutes a bed is open between patients. That is how tightly we are beginning to be able to run the ship.”

Scheulen is not at liberty to report what the cost was to build and implement the command center, although he acknowledges that it was substantial. However, he notes that hospitals and EDs do not necessarily need a project of this size and scope to take advantage of the kind of predictive analytics that Hopkins leverages.

“We recognize that not everybody needs all of this; nor are people likely to be able to afford it,” he adds. Some hospitals, for example, may want to focus on one problem area or “tile,” Scheulen observes.

“It is possible that someone may just want the predictive discharge tile, or the set of things that [factor into] predictive discharge ... or someone may just want to know the status of every bed in the hospital all the time,” he says. “That is exactly what we are thinking about now — how to do this at places that aren’t as complex as Hopkins.”

The command center has been a three-year journey thus far, but Scheulen notes that the hospital system is just getting started.

“We are not done by any stretch of the imagination, and that is the point,” he says. “We are trying to make people understand that this is a new way of doing business.”

Act to Avert Congestion

While much of the work around predictive analytics is used to create capacity and boost efficiency in busy hospitals, investigators are looking at leveraging this type of data to ease the demand for service in busy EDs that are destined to become overwhelmed with patients.

“What we want to do is take these predictions of when you expect demand is going to be high, and instead of waiting for the congestion to build up, leaving many patients waiting for hours and hours to get care ... we want to proactively start making decisions to try to reduce the overall congestion,” explains Carri Chan, PhD, an associate professor of business in the division of decision, risk, and operations at Columbia University in New York City.

Specifically, Chan proposes proactive decision-making and diversion strategies so that certain patients who would otherwise access care in a particular hospital’s ED actually go elsewhere to receive care. Chan notes that she uses the term “diversion” in the broadest sense to encompass everything from ambulance diversions to simply educating certain patients who arrive in the ED about an alternative site where they can receive care much faster.

While such diversion tactics are not novel, Chan proposes implementing them much earlier, essentially heading off periods of predicted congestion.

“Typically, what hospitals do nowadays is they wait until that buildup of patients occurs ... before they start diverting patients,” she explains. “What we are actually proposing is that you actually know ahead of time that this may be occurring, so you make diversion decisions earlier in a proactive manner.”

For example, when a low-acuity patient arrives in the ED, even though the ED is not yet packed, this patient can be given the option of receiving care at an adjacent urgent care facility to ease delays for the 50 patients administrators know will be arriving in the next hour, Chan observes.

Chan specifies that such conversations would be carried out during triage, so clinicians already would have determined that the patient is dealing with a low-acuity issue. She also stresses that the patients would be given the choice of whether to wait for care in the ED or to access care at the alternative site. Using simulation techniques that are based on published data regarding average arrival rates and service times, investigators have determined that such proactive diversion tactics can reduce delays by up to 15%.

“We looked at how arrivals are varying by time of day and day of the week, and we used that for these predictions for the future,” Chan explains. “Even if we say we will always take patients by ambulance and that we will only divert the lowest acuity patients, we still find that we get substantial gains.”

Of course, it should be noted that to be in compliance with EMTALA, patients also must be advised about receiving care at an alternative site only after they have received a medical screening exam. Also, the alternative care site must be on the hospital campus, as defined by EMTALA regulations, and the process of moving patients must occur in accordance with CMS regulations.

Pair Predictions with Diversion

Chan notes that she is in communication with several EDs that are interested in the approach. She also notes that each facility must create a predictive model built with data that is specific to its own operations.

“Each facility has different types of patients who are coming in, so part of the process is to first collect enough historical data so that we feel confident that these predictions are somewhat accurate,” she explains.

“The more detailed information that is available, the better the predictive power of the algorithm, and the better our approach will do,” Chan offers. For example, she notes that information about weather patterns has big predictive power regarding ED demand. “If it is a particularly icy day, you will see more car accidents. Similarly, flu season can create surges in demand.”

The approach involves taking these data inputs and putting them together with how the ED will act on the predictions.

“Let’s say we are only able to predict 80% of the actual arrivals to the ED,” Chan says. “We will still have improvements in reducing delays, but it certainly won’t be as great as if we had 90% of arrivals predicted, so that is a tradeoff that one needs to consider.”

One of the reasons why investigators focus on using diversion interventions to address anticipated surges in demand is because these mathematical models become much less accurate the farther out you try to predict.

“If I want to predict how many patients are going to come into the ED in three weeks, there is going to be much less accuracy than if I want to predict how many are likely to come in the next two hours,” Chan explains. “Diversion decisions are being made on a minute or hourly time scale, whereas staffing decisions are often made weeks in advance, so we wanted to account for that reality.”

Going forward, predictive models are only going to become more commonplace in healthcare planning, and people will become more adept at using such data, Chan predicts.

“If we know the variability, we can modify things appropriately,” she says.


  1. Xu K, Chan C. Using future information to reduce waiting times in the emergency department via diversion. Manufacturing & Service Operations Management 2016;18:314-331.


  • Carri Chan, PhD, Associate Professor of Business, Division of Decision, Risk and Operations, Columbia University, New York. Email: cwchan@columbia.edu.
  • Jim Scheulen, PA, MBA, Chief Administrative Officer, Department of Emergency Medicine, Johns Hopkins Medicine, Baltimore. Email: jscheule@jhmi.edu.