Can analytics predict epidemics?

Angrew Ng published an interesting article “Stopping Coronavirus” in his Jan 29 issue of The Batch. It talks about a Canadian company that analyzes online information to predict epidemics spotted the upsurge in coronavirus at least a week ahead of public-health authorities. Link

What’s new: Canadian startup BlueDot alerted customers to the outbreak in the Chinese city of Wuhan on New Year’s Eve, Wired reported. The U.S. Centers for Disease Control and Prevention issued its warning on January 6, and the World Health Organization followed suit three days later. The respiratory illness as of this writing has infected more than 6,000 people and killed more than 130, mostly in China.
How it works: Founded in 2014, BlueDot aims to stop the spread of infectious diseases by giving healthcare workers early warning, so they can identify and treat people who become infected.

  • The company’s natural language processing model ingests 100,000 articles in 65 languages daily to track more than 100 infectious diseases. It ignores social media but scans news reports, government information, blogs, and forums related to human, plant, and animal diseases, as well as travel ticketing and local weather data.
  • Human analysts vet the model’s predictions. They issue reports to customers in business, government, and nongovernmental organizations, ultimately reaching healthcare facilities and public health officials in a dozen countries.

Behind the news: In 2008, Google undertook a similar effort to forecast influenza outbreaks based on search terms entered by users. In initial research, Google Flu Trends tracked the number of cases two weeks faster than the CDC. However, it dramatically underestimated the peak of the 2013 flu season and was shuttered soon afterward. Subsequent analysis concluded that the algorithm overfit seasonal search terms unrelated to flu.

Why it matters: Rapid detection of new diseases is crucial to avoid global pandemics. Virulent diseases often can be contained if they’re caught early enough, but every hour compounds the number of people exposed and thus the number of cases. An epidemic can quickly overwhelm healthcare systems, leaving people even more exposed.
We’re thinking: It’s hard to know how well today’s techniques will play out tomorrow. But the ability to catch potential pandemics before they explode is too valuable not to try.

This entry was posted in Machine Learning. Bookmark the permalink.

1 Response to Can analytics predict epidemics?

  1. jamet123 says:

    Like all predictions, such a prediction would only be useful if acted on. The question then is what would it take for a health authority to do something different. If the authorities and hospitals won’t change their behavior based on a prediction – and they likely won’t – then the prediction has absolutely no value. The myth that any prediction must be valuable is causing companies to waste vast amounts of money ….

Leave a Reply to jamet123 Cancel reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s