January 11, 2019 (Infection Control Today)
Trying to keep track and predict highly contagious, highly susceptible influenza is incredibly difficult as people are constantly on the move in the United States. The CDC works to monitor influenza by tracking patient visits to HCPs for flu-like illness, however, gathering information by this method can become delayed by up to two weeks.
The Computational Health Informatics Program (CHIP) led a new study that sought to combine two forecasting methods with machine learning (AI) to better estimate local flu activity.
Results of this study are now published in Nature Communications.
“When applied to flu seasons from 2014 to 2017, this new approach, called ARGONet, was more accurate in predictions than an earlier, high-performing forecasting approach, called ARGO, in over 75 percent of the states studied.”
To date, ARGONet is the most accurate flu forecasting method – about a week ahead of former healthcare forecasting models. The more timely and accurate these methods become, the better public health officials can respond to epidemic outbreaks and improve health outcomes.
The new approach, ARGONet, uses artificial intelligence (AI) and two highly-accurate flu detection models, one of them ARGO, which uses information from electric health records, flu-related searches and historical flu activity in certain locations.
To take accuracy a step further, ARGONet also incorporates a model that can detect the spread of flu in neighboring areas, which may increase the risk of disease outbreak.
This AI technology was “fed” flu predictions from both models as well as real flu data, creating the most accurate predictions possible.
This is an exciting time of innovation which “will set a foundation for ‘precision public health’ in infectious diseases.”