Boston Children’s Hospital Used Data Synthesis to Make Real-Time Flu Predictions
A team of epidemiologists from Boston Children’s Hospital and Harvard Medical School has developed a real-time predictive model for flu forecasts, based on a method of combining sources it calls “ensemble modeling.”
“Weather forecasting is an established discipline and has become engrained in society,” says Dr. John Brownstein, the study’s lead author and chief innovation officer at Boston Children’s, in a statement. “We think the time is ripe for the same to happen with disease forecasting.”
While the Center for Disease Control (CDC) regularly monitors seasonal flu-like symptoms, it often uses only individual data sources to track the disease. As a result, reports often fall short of robust, timely snapshots of activity, leaving doctors and public health officials one to two weeks behind.
Ensemble modeling, by contrast, looks at a variety of non-traditional sources: national electronic health records (EHRs), Google searches, Twitter analytics, and crowd-sourced reports via Flu Near You. The team then synthesizes the data from these sites, as well as established models like Google Flu Trends, into a singular predictor similar to that used for hurricane trajectories.
The result? Instant “now-casts” showing flu patterns with a high degree of accuracy.
As reported in the journal PLoS Computational Biology, ensemble modeling offers more inclusive, far-reaching data than any single-source method before it. When compared with historical CDC flu data, ensemble modeling was able to more accurately predict both duration and magnitude of the disease.
Next, according to the statement, the Boston Children’s team hopes to apply this technique to specific geographical areas—right now, it can only offer national forecasts—as well as other diseases.
