SAN FRANCISCO — The same principles that keep commodities traders in business were used for the first time to predict disease, accurately forecasting influenza activity up to 6 weeks in advance, Philip M. Polgreen, M.D., said.
Instead of buying corn or soybean futures, a group of 61 health care providers in Iowa used an Internet site to trade information about influenza activity during last year's flu season. In this innovative pilot study, the investigators aggregated and analyzed the trades and compared them with later surveillance data on actual flu activity from the Centers for Disease Control and Prevention.
Results showed that the influenza “prediction market” provided useful forecasts of disease activity up to 6 weeks in advance, with accuracy increasing as the target week grew nearer, Dr. Polgreen said at the annual meeting of the Infectious Diseases Society of America.
Market forecasts made 2–3 weeks in advance matched the level of influenza activity later reported by the CDC 80% of the time, reported Dr. Polgreen of the University of Iowa, Iowa City, and his associates.
“Our data suggest that prediction markets can forecast influenza activity quickly, accurately, and inexpensively by aggregating the expert opinion of health care providers,” he said.
Even a 2-week warning of increased influenza activity would be helpful because it would allow time to vaccinate high-risk patients and health care providers and also would allow planning for increased hospital admissions, Dr. Polgreen said.
Historically, no accurate predictor of flu activity has been available. Although yearly influenza outbreaks are certain, the timing, severity, and strains are not known.
Outbreaks typically peak over 2–3 weeks and last for about 2–3 months. Vaccines must be administered early enough to provide protection (at least 2 weeks before exposure) but not too early, because their effectiveness wanes over time.
Prediction markets were used first in 1988 to forecast the winner of Iowa's presidential caucus, and have proved more accurate than Gallup polls. The method since has been used by Hewlett-Packard to forecast sales of its printers and by Eli Lilly & Co. to predict which drugs are most likely to advance through clinical trials.
About 30–60 active traders are needed for a prediction market to function. More are better, but the most important goal is to use traders who have solid information, such as primary care physicians, infectious disease specialists, pediatricians, emergency department physicians and nurses, epidemiologists, clinical microbiologists, public health officials, and pharmacists.
This year, the investigators will run an influenza prediction market for the full state of Iowa with about 100 traders.
How the Flu Market Worked
Each trader was given 100 “flu dollars” for trading on an interactive Web site that was accessible around the clock. Every even-numbered week during the flu season, the system introduced a new set of contracts, and traders bought or sold predictions of the level of flu activity that would occur 8 weeks after the contracts opened.
They chose from six levels of flu activity, which ranged from “no activity” through “widespread activity.” Trading could continue until the last day of the target week.
A clinical microbiologist who suddenly began seeing more respiratory cultures that tested positive for influenza might log on to the Web site and purchase shares for “widespread influenza activity” while also trying to sell shares for “little or no influenza activity,” Dr. Polgreen said.
The electronic interaction of all the traders led to a set of market prices that reflected the consensus probability for the future level of influenza activity.
When the CDC released its influenza surveillance report after the target week, traders who bought contracts correctly predicting the level of flu activity received 1 flu dollar per contract. All other contracts converted to a zero value.
At the end of the flu season, traders received $1 in educational grant money for every flu dollar they owned. The most successful trader amassed $213.