Can you predict how busy your emergency department is likely to be tomorrow by looking at Google searches today? If that sounds a little like Wimpy gladly promising to pay for his hamburger on Tuesday, no one could fault your skepticism. After all, could it really be that easy?
A new study published in Annals of Emergency Medicine by researchers in Stockholm, Sweden, suggests using real-time Internet search data just might work as a forecasting tool for ED ttraffic. (Ann Emerg Med 2014 Dec 2 [ePub ahead of print].)
Using visits from 6 p.m. to midnight to the Stockholm Health Care Guide, a regional medical website, Martin Nordberg, MD, Andreas Ekström, M Ed, and colleagues attempted to predict the next day's traffic at seven Stockholm-area emergency departments as well as the overall visits for the region. The model was more accurate in whole-region forecasting, with a 4.8 percent mean absolute percentage error.
“Previously, Google Flu Trends has used Internet data to detect how the seasonal flu spreads over the world,” said Dr. Ekström. “Internet data has also been used in other areas, like predicting the stock market and book and movie sales. This inspired us to test if it could be used to predict the number of ED presentations.”
The Stockholm Health Care Guide, Sweden's largest site for medical information, was a natural starting point, even though that also represents a limitation of the study, he pointed out. “Since the model relies on just one website, if the popularity of that site decreases, then so does the accuracy of the predictions.”
The study, which tracked all ED visits in the region (including adult, pediatric, and gynecologic), found that the website proved as good as or better than previous models using other types of data to predict the number of ED visits, said Dr. Ekström. “The best predicting factor has been calendar data, like weekday and holidays. We also use calendar data in our model, but when we add the number of visits to the website, we get an even better result. One of the biggest benefits with Internet data compared with calendar data is that it has the ability to reflect sudden changes in people's behavior before this change in behavior shows at the ED.”
Mathias Wargon, MD, PhD, the chief of the emergency department at Hôpital Saint-Camille in Bry-sur-Marne, France, found that day-of-the-week models were the most reliable, according to his systematic review of ED forecasting models. (Emerg Med J 2009;26:395.) Overall, he found an error rate of between 4.2 and 14.4 percent.
“I think it's a very good idea, what they've done in this article. I love this model,” Dr. Wargon said. “The fact that people looking at the website the day before predicts ED traffic for the day after, that shows that people who come to the ED don't always come in because they have an immediate emergency, but because they are in need of care. What they need is not only a doctor but a technologic platform. There's a doctor, radiology, blood exams, everything.”
His region does not have a central health information website like Stockholm, but he said Google analytics might be used in the same way. “I think it's brilliant to put ED requests into this model that we know very well and find out that it works,” Dr. Wargon said.
But a challenge remains: putting the forecasting data to work. “When you find out this information, what can you do with it? Can you have more nurses and doctors on days when you expect higher traffic, and have fewer doctors and nurses when you don't?” Dr. Wargon asked. “But people will say, you don't know for sure. It's like weather forecasting variability. If you can predict the number of patients who need beds, the others can wait but will wait less because overcrowding is linked to boarding patients.”
The more important forecasting figure, he said, is not the number of visits but the number of hospitalizations. “If you have a condition where you can wait in the ED, it's not a good thing, but nobody will die,” he said. “But if you are truly emergent or you are 80 and it's difficult to lie on a stretcher or [sit] in a chair, it's better for you to have a bed with a nurse. An important use of forecasting data, whether it's a model like this or another, would be to let you know how many beds you need to have free. If you can predict the number of patients who need beds, the others can wait.”
Access the linksin EMN by reading this on our website or in our free iPad app, both available atwww.EM-News.com. Comments?Write to us email@example.com.