Does This Tweet Make Me Look PHAT? Tracking Obesity Trends with Social Media

December 2014
Topics: Health Services Administration, Public Health (General), Probability and Statistics
To provide public health agencies with real-time information on the effectiveness of their obesity-prevention campaigns, MITRE designed a model that supplements traditional disease surveillance by monitoring Twitter.
Meredith Keybl

We all know the numbers. We've heard them recited by grim news anchors, we've seen them broken down in charts and tables in magazine cover stories. Yet, when rattled off by Meredith Keybl, a MITRE epidemiologist specializing in healthcare systems, they lose none of their impact. "Two-thirds of American adults are obese or overweight. One in four young adults is too overweight to join the military. The medical cost of adult obesity in the U.S. is $150 billion per year."

Of course, health experts are not ignoring these numbers. Public health programs across the country, like Get Healthy Philly and Mass in Motion, are striving to get people to eat better and exercise more. To determine whether their efforts are making an impact, public health officials study the results of annual surveys like the Behavioral Risk Factor Surveillance System.

But the lag in time between when such surveys are conducted and when their results are published can leave public health programs operating blindly for a year or more. These programs could conduct their own surveys to get more immediate feedback, but many do not have the budget or manpower to do so.

MITRE sought to provide these programs with a method for taking the pulse of the public health at any given moment. Keybl and her team, inspired by a previous MITRE project that sought to determine the gender, age, and location of people based simply on their Twitter posts, designed a tool for tracking local health trends through community tweets.

Tweeting About Treats and Obliques

The Persistent Health Assessment Tools (PHAT) research project is designed to supplement traditional health survey data by making real-time measurements of social media to infer behavioral trends. "People tweet all the time about all sorts of things," says Keybl. "And even if they're not tweeting 'Wow, I just had four cronuts and don't I feel large,' maybe they're tweeting other things that will give us an indication of their health behaviors."

For example, tweets like the following could give public health officials insights into how the local community is pursuing or eschewing a healthy lifestyle:

To test how accurately PHAT could glean public health data from Twitter, Keybl and her team isolated tweets from each state. They built models from those tweets and used them to predict whether a particular state would rank among the heaviest 25 states. PHAT made the right determination in 49 out of the 50 experiments.

Massachusetts on the Dashboard

PHAT having proved itself in theory, Keybl and her team sought to prove it in practice. They partnered with the Massachusetts Department of Public Health to test PHAT in a specific locality. Their goal was to sharpen their focus from proving that Twitter was an effective signal of community health practices to proving that you could tune into that signal for specific communities.

As part of that effort, the PHAT team designed a dashboard that will locate and import tweets of interest. Using the dashboard, policy makers will be able to browse, cluster, and search through messages to observe specific instances of where a program's outreach was successful and to identify ways that outreach could be improved. "The dashboard can filter tweets based on keywords such as 'yoga,' 'fruits and vegetables,' 'farmer markets,' etc.," says Keybl. "This will allow agencies to see what people are saying about behaviors the agency is trying to encourage or, in a case like #NachosforBreakfast, discourage."

An Elite Tweet Team

The team's future goals for PHAT or follow-up projects are to develop the ability to cross-reference location-based health data gathered on social media with age and gender identification and the ability to investigate other health topics besides obesity, such as veteran homelessness or substance abuse. PHAT could be a crucial weapon in the battle against obesity. And it’s an example of the ground-breaking tools MITRE is developing for social and behavioral science research.

It's also a good example of MITRE's ability to gather together diverse subject experts into collaborative teams. "The PHAT team includes me as the epidemiologist, plus two computer scientists and a user interface developer," says Keybl. "The interface developer has never worked in public health before, one of the computer scientists had to look up the definition of 'epidemiologist,' and the last time I coded anything was when I was 14. But together we designed an innovative solution that will help public health agencies improve their program evaluation efforts."

—by Christopher Lockheardt


Publication Search