Mar 7, 2007 (CIDRAP News) – Some professors at the University of Iowa think that a kind of betting system may help predict the timing and extent of the next influenza pandemic, and a large health foundation is betting they are right.
The group last week unveiled the "Avian Flu Market" (AFM), a system designed to pool expert opinions from around the world about the probability of a human influenza pandemic sparked by avian flu. Participants in essence will bet on the probability of various pandemic-related events, and the "price" of a given prediction will reflect the number of participants favoring it.
Prediction markets give experts a financial incentive to participate, facilitating the quick collection of views from widely dispersed specialists and translating them into probability estimates. The AFM doesn't require participants to bet their own money, but those who make the most accurate predictions can win small educational grants.
The market, now in a testing phase, is supported by a $245,685 grant from the Robert Wood Johnson Foundation (RWJF), according to a Mar 1 news release from the foundation.
"It's an experiment" of a kind that that has worked well in other fields, said Philip M. Polgreen, MD, an assistant professor of medicine at the University of Iowa's Carver College of Medicine, who is co-principal investigator on the project and also directs the Infectious Disease Society of America's Emerging Infections Network.
The AFM is patterned after the Iowa Flu Activity Market, which was set up in 2004 to help predict the behavior of seasonal flu in Iowa.
"Using a small group of healthcare workers, less than 50, we've been able to predict when the [annual flu] season would start, when it would peak, and when it would end, 2 to 4 weeks in advance," Polgreen told CIDRAP News.
"Each one of those people [participating] has a different slice, different experiences, different amounts of information," he said. "Unlike surveys, prediction markets allow people to register how strongly they believe" in a given outcome. "Every time someone makes a trade, we get a new price—it changes the probability."
Polgreen and colleagues described the Iowa Flu Activity Market in an article in the Jan 15 issue of Clinical Infectious Diseases (CID). The market tries to predict what level of flu activity (widespread, regional, sporadic, or local outbreaks—or none) will be measured in Iowa in any given week by the Centers for Disease Control and Prevention.
In 2004-05, the market's forecast by the end of the target week was correct 71% of the time (the CDC figures are released a week later), and the market forecast 1 week in advance was correct 50% of the time, according to the report. For comparison, forecasts based strictly on historical averages would have been right about 36% of the time.
Participants in the AFM will be recruited experts, starting with the editorial board of ProMED, the Program for Monitoring Emerging Diseases, the electronic disease-reporting system of the International Society for Infectious Diseases, Polgreen said. The ProMED board consists of about 20 people of various backgrounds in various places.
After that, the plan is to add other ProMED participants. "We have funding to expand to about 100 volunteers," Polgreen said. The aim is to make the group reflect the makeup of ProMED's membership and eventually expand it to as many as 200 participants. They are expected to include physicians, veterinarians, epidemiologists, nurses, and public health workers.
Polgreen said he and ProMED editor Lawrence Madoff, MD, of Harvard Medical School will evaluate the applications from potential participants. Madoff is a co-principal investigator on the AFM; the principal investigator is Forrest Nelson, PhD, a University of Iowa economics professor.
Both the flu activity market and the AFM are spinoffs of the Iowa Electronic Markets (IEM), which were set up in 1988 and have focused mainly on elections. The markets have generally outperformed conventional opinion polls, according to Polgreen and associates.
AFM participants will be asked to answer specific questions about avian and pandemic flu possibilities, with each question being the basis for a "market," according to the RWJF news release. For example, one question is whether the World Health Organization (WHO) will raise its pandemic alert level to stage 4 by Jul 1 of this year, while another deals with how soon H5N1 avian flu will spread to the Americas.
Another question, according to an Associated Press report published last week, is whether a human case of H5N1 flu will occur in Hong Kong before Jul 1. "Yes" contracts on that question were trading at 43 cents, meaning participants put the probability of the event at 43%, the story said.
But Polgreen cautioned, "That's not really representative of what ProMED thinks right now." The market is in a beta (testing) phase now, with a small group of participants, he said. "Those prices are just from the beta market. Just to make sure the software worked, we've been running markets for the past 2 weeks just for this group."
When the market is up and running in about 2 more weeks, "anybody who wants to can go to the Web site and see the prices," Polgreen said.
Rather than putting up their own money, AFM participants will trade with virtual currency, and any payoffs will be in the form of grants for things like journal subscriptions, conference fees, and books, according to the RWJF. The worst participants can do will be to win no grants.
Not everyone thinks the AFM is a good idea. "It really compounds uncertainty," said Michael T. Osterholm, PhD, MPH, director of the University of Minnesota Center for Infectious Disease Research and Policy, publisher of CIDRAP News.
"Right now the world's best experts can't tell you what's going to happen, so having 100 people predict what they don't know doesn't add any more precision or intelligence to the issue," Osterholm said. "It's maybe not quite a gimmick, but it doesn't give me any more reason to think we'll be more informed about the next pandemic."
He added, "Whether you have a market or not, you still need good disease surveillance. This just seems like a glitzy way of doing disease surveillance."
David A. Halvorson, DVM, a veterinarian and avian health specialist at the University of Minnesota in St. Paul, said he was "intrigued" by the AFM but had some reservations.
He suggested that AFM participants might make predictions about questions on which they lack expertise. For example, veterinarians are not in a good position to predict when the H5N1 virus might start spreading from human to human, while physicians "would have no way to evaluate when it would spread from a bird to a human," he said.
In addition, he asked, "Does that make it a good study when you're betting someone else's money? I don't know."
The architects of the AFM say it is not meant to replace disease surveillance. "Prediction markets will never replace traditional infectious disease surveillance; we view them instead as a potential supplement," they said in their CID report. "If prediction market information is used in tandem with existing surveillance systems, it can guide the deployment of resources and increase the efficiency of professionals who are monitoring emerging infectious diseases."
The key advantage of prediction markets, proponents say, is that they collect the disparate information held by many participants. "People on the front lines have information they're usually not asked about," said Polgreen.
RWJF news release
University of Iowa news release
Polgreen PM, Nelson FD, Neumann GR. Use of prediction markets to forecast infectious disease activity. Clin Infect Dis 2007 Jan 15;44(2):272-9 [Full text]