That was the headline of a UPI.com story that began:
“Twitter’s fast pace and knack for promoting public spats can surely raise heart rates and get the proverbial blood boiling, but the platform known for hashtags and half-formed thoughts can also predict heart attacks — or at least rates of heart disease.”
It’s a story of the type we see increasingly – apparently based entirely on a news release with no independent vetting of claims, and misinterpretation of those claims that were made by the researchers.
Thousands of results turn up on Google. I only looked at a handful.
The Telegraph of the UK published, “Angry tweeting ‘could increase your risk of heart disease’.”
Mother Jones headlined it, “Are You at Risk for a Heart Attack? The Answer May Lie in Your Twitter Stream.”
The Washington Post published “Tweets can better predict heart disease rates than income, smoking and diabetes, study finds.” No, it didn’t. The authors wrote:
“Our study had three major findings. First, language expressed on Twitter revealed several community-level psychological characteristics that were significantly associated with heart-disease mortality risk. Second, use of negative-emotion (especially anger), disengagement, and negative-relationship language was associated with increased risk, whereas positive-emotion and engagement language was protective. Third, our predictive results suggest that the information contained in Twitter language fully accounts for—and adds to—the AHD-relevant information in 10 representatively assessed demographic, socioeconomic, and health variables. Taken together, our results suggest that language on Twitter can provide plausible indicators of community-level psychosocial health that may complement other methods of studying the impact of place on health used in epidemiology (cf. Auchincloss et al., 2012) and that these indicators are associated with risk for cardiovascular mortality.”
“Associated with” is really important here. You can’t make that mean something it doesn’t mean. It doesn’t mean “can better predict.” It means there’s a statistical association. Not proof of predictive power.
But then again, one of the authors said this in an interview: “So predictions from Twitter can actually be more accurate than using a set of traditional variables.” That’s not the same as what they wrote. This is a problem: researchers who write manuscripts very carefully because they know they might be booted under peer review, but then speak in a more freewheeling manner when interviewed afterwards.
Ultimately, though, I hold journalists responsible for vetting claims. If you can’t do it yourself, turn to independent experts for guidance. Much of this silliness could have been avoided if more journalists had turned to independent sources. (We offer a list that is likely to be expanded considerably in 2015.)
ABCNews.com at least provided this independent perspective:
“Cardiologist Dr. Sahil Parikh, at UH Case Medical Center in Cleveland, Ohio, said he applauds the researchers’ creativity but said readers should take the results with a “very large grain of salt.” He said it’s “reasonable” to say that negative emotions related to stress can predict heart disease events because there’s a significant body of research to back that up. But the age difference between social media users and people having heart attacks doesn’t match up, he noted.”
Bethesda psychiatrist Susan Molchan, MD, who will be a continuing contributor to this site, had some advice for journalists:
“Journalists should note that scientists have begun to use the data available in social media to supplement traditional methods used in epidemiology to study and track disease. Generating these correlations between numbers of tweets and heart disease is a tool. It looks like it may be a useful one in studies of heart disease risk, although this is only one study. They should probably emphasize the point that this is based on huge pools of people—much larger than usual epidemiological studies – a big advantage in that sense— and has nothing to do with any one individual’s risk.”
Dr. Zack Berger of Johns Hopkins wrote to me:
“The idea is an intriguing one. But there are limitations to the paper, which the authors describe in detail in their discussion section. In short, “the people tweeting are not the people dying.” Moreover, the claim that the Tweet-based model predicts mortality from heart disease better than traditional risk factors is weakened when one looks at the actual results. Finally, and most important, the reports in the lay press overestimate, flatten out and misrepresent an intriguing hypothesis-generating study into something that these stories imply is relevant to health today. Not so.”
Dr. Steve Miles of the University of Minnesota Center for Bioethics, another of our new contributors, wrote:
“Sigh. Any study of associations can attain significance if the data set is large enough. This study does not try to show any correlations between what a person tweets and that person’s risk of a heart attack. At best the level of hostile tweets is associated with a 6% risk increased risk of heart attack; heart attacks are very uncommon in the age group that tweets. Maybe the risk is from the parents who see the tweet rather than the sender.”
Follow us on Twitter: