How can you accurately track flu activity when it’s so contagious and spreads very easy while people travel and move about? The answer is you can’t. The CDC monitors patients who get to the hospital in cases of flu or flu-like illnesses, but the information can have a delay of almost two weeks.
However, the Boston Children’s Hospital’s Computational Health Informatics Program (CHIP) has found a way to estimate the flu activity by using machine learning and forecasting methods together.
The study has been published in Nature Communications this week. The program is called ARGONet, and it was tested on three years of flu seasons that began with September 2014 and ended in May 2017. Compared to the high-performing forecasting approach, ARGO made more accurate predictions in over 75% of the cases. This means that ARGONet will be the most accurate program to estimate influenza activity and that it can report all the U.S. flu activity a week ahead of the traditional reports.
The senior author of the paper and a CHIP faculty member, Ph.D. Mauricio Santillana (also an assistant professor at Harvard Medical School) explained why this program is vital:
“Timely and reliable methodologies for tracking influenza activity across locations can help public health officials mitigate epidemic outbreaks and may improve communication with the public to raise awareness of potential risks.”
How ARGONet Works?
The program uses two flu detection models and machine learning.
One of the models – ARGO (AutoRegression with General Online information), gets information from electronic health records but also all the Google searches on flu and the historical flu activity in a certain location. Compared to Google Flu Trends who worked from 2009 to 2015, ARGO alone was much better.
The second model gets the spatial and temporal patterns of flu in the areas, explained Santillana:
“It exploits the fact that the presence of flu in nearby locations may increase the risk of experiencing a disease outbreak at a given location.”
The machine learning was fed data from the two models and actual flu data to make the predictions as accurate as possible. Santillana added that the program continues to evaluate the “predictive power of each independent method and recalibrates how this information should be used to produce improved flu estimates.”
The first author on the paper and a CHIP investigator, Fred Lu, concluded that the models would be more accurate over time as it collects more online search volumes and as it gathers more electronic health records from healthcare providers.
Doris’s passion for writing started to take shape in college where she was editor-in-chief of the college newspaper. Even though she ended up working in IT for more than 7 years, she’s now back to what he always enjoyed doing. With a true passion for technology, Doris mostly covers tech-related topics.