More than 16 million American adults suffer at least one major depressive disorder every year, according to the National Institute of Mental Health. Most people with depression don’t have access to quality psychiatric care, says Chris Danforth, associate professor of mathematics and statistics at the University of Vermont. “But almost all of them have phones,” Danforth says, and most of those are used for social media. Danforth wanted to know if there might be a way to mine the data on social media to predict if the user has depression.
Specifically, Danforth and his colleague, Harvard University graduate student Andrew Reece, recruited 166 people to share their Instagram accounts. Danforth and Reece asked the participants if they already had a clinical diagnosis of depression –71 did –and they gave each subject the CES-D, or the Center for Epidemiologic Studies Depression Scale, questionnaire to see who might have undiagnosed depression. Danforth and Reece then took nearly 44,000 photos from the subjects’ Instagram pages and ran them through computer algorithms that identify markers of depression, such as facial characteristics and color choices.
[Read: Are Depression Naps a Harmless Internet Meme or Concerning Sign?]
The analysis found that those who were depressed were more likely to post bluer, darker and grayer photos; post photos with human faces but show less of their own face; not use Instagram filters to adjust the photo’s brightness and coloring; use the Inkwell black and white filter when they did use filters; and not use the Valencia filter that lightens the tint of the photo. “They literally see the world through a darker lens,” Danforth says.
They also posted more frequently and had more comments on their posts but also had fewer “likes.” These predictive results were better than general practitioners’ diagnostic success rates, and they held up even when the analysis was done only on posts made before depressed individuals were first diagnosed.
Danforth says that computational analyses like these have shown remarkable utility in other areas. Using applied math in weather prediction, for instance, “is one of the great success stories in science,” he says. “The social sciences are going through a similar transition now, and it will help us understand how people behave. We are going from data-poor to data-rich.”
Limitations to the Study
This study has some limitations that must be noted. The sample was fairly small, and the subjects were essentially “professional test takers,” says Dr. John Torous, co-director of the digital psychiatry program at Beth Israel Deaconess Medical Center, which is affiliated with Harvard Medical School. “Is this reflective of the general population?” The depression diagnosis was based on a short survey, not a clinical diagnosis. Danforth acknowledges these questions about the study: “We don’t know if this works on the typical [Instagram] user.”
There is also the matter of whether social media in itself is a risk for depression. A 2016 study out of the University of Pittsburgh, for instance, found that the more time people spend on social media, the more likely they are to be depressed. Those who spent more than two hours a day on social media platforms had twice the risk of feeling socially isolated–a known risk factor for depression — compared to those who spent a half-hour or less online.
Despite its limitations, Torous still finds the study interesting. “Social media may offer clues into the risk for depression, and it also raises issues on how we conduct studies on this,” he says. Studying technology and its relationship to depression screening and treatment is of keen interest to Torous, who chairs an American Psychiatric Association work group on this topic. Technology is playing an ever larger role in just about all areas of medical care, and mental health is no exception. Indeed, Danforth published a more recent study, in October 2017, showing that computational models could predict the emergence of depression and p ost-traumatic stress disorder in Twitter users.
And that’s the tip of the iceberg. Scientists are working to develop apps that can use a person’s cellphone to track GPS, Bluetooth, Wi-Fi, fitness monitoring, phone and text data and even biological data such as heart rate, galvanic skin response and body temperature to give a remarkably accurate picture of the user’s mood and state of mind.
Danforth is working on a suicide risk prediction tool that hospitals could use to screen patients who visit an emergency room. “We will be trying to get thousands to hundreds of thousands of subjects to give us access to their social media feeds, to see what predictive information is available,” he says. “This sort of data may offer something that the [clinical] interview isn’t able to offer. I would never say these tools will replace a diagnosis from a trained physician. But doctors use all kinds of tools to help in their diagnosis. This is building tools to help them make their assessment and get their patients into treatment sooner.”
[See: Am I Just Sad — or Actually Depressed?]
Torous preaches caution, but remains optimistic. The Instagram study “was a very cursory look, and there is a risk in over-reading this,” he says. “But there are some data out there that may be relevant, and we should look more. The fact that it has gained so much press reflects how desperate we are to help people with mental illness.”
More from U.S. News
Am I Just Sad — or Actually Depressed?
Are Depression Naps a Harmless Internet Meme or Concerning Sign?
Can Instagram Photos Suggest Who Might Have Depression? originally appeared on usnews.com
