Researchers devise model to predict flu outbreaks

Using data from health records, researchers in Boston correctly predicted flu activity and the peak of the season in a recent study.

By Stephen Feller

BOSTON, May 11 (UPI) -- Data on influenza outbreaks often lag one to two weeks behind the actual spread of the virus, making it difficult to predict where it will hit and methods of prevention.

Researchers at Boston Children's Hospital combined electronic health records, historical patterns of the flu and a machine-learning algorithm to accurately predict national and local flu activity, they report in a new study.


Better tracking of the virus could help reduce the number of people affected by influenza each year, researchers say.

Although other real-time tracking systems have worked well, such as Google's Flu Trends tool that was shut down last August, the new system was able to predict the timing and magnitude of the flu's peak in the country -- a significant advance for using data to track infection.

"Our study shows the true value of considering multiple data streams in disease surveillance," Dr. John Brownstein, chief innovation officer at Boston Children's Hospital, said in a press release. "While Google data provide incredible real-time population wide information, clinical data add a more accurate and precise assessment of disease state. As EHR data become more ubiquitously available, we will see major leaps in our ability to monitor and track disease outbreaks."


For the study, published in the journal Scientific Reports, researchers pulled together information from Athenahealth -- a database containing insurance claims for 64 million people and medical records for 23 million patients seen by 72,000 healthcare providers -- and historical patterns of the flu collected between 2009 and 2012. The researchers used the data to predict flu activity for the next three years, finding the system was accurate.

The researchers used total weekly counts of doctor's visits, flu vaccine visits, flu visits, influenza-like illness visit counts and other unspecified doctors visits for other viral infections to make their predictions, finding their model was generally more than 90 percent accurate at matching actual flu records from 2013 to 2015 as monitored by the Centers for Disease Control and Prevention.

"Having access to near-real-time aggregated EHR information has enabled us to significantly improve our flu tracking and forecasting systems," said Dr. Mauricio Santillana, a professor at Boston Children's Hospital and the Harvard Medical School. "Real-time tracking will enable local public health officials to better prepare for unusual flu activity and potentially save lives."

Latest Headlines