Doctors could one day use our social media posts as diagnostic tools.
People who are clinically depressed are more likely to post grey or muted photos on Instagram, or to choose filters that create these effects, says a new study, while happier people tend to publish more colorful snaps. The findings are preliminary, but researchers think they could one day help identify mental illness at earlier stages.
The new report—which has been posted online but not yet been peer-reviewed or published in an academic journal—analyzed more than 40,000 Instagram photos from a group of 166 volunteers, about 70 of whom reported being clinically depressed.
The researchers looked at different qualities of the photos, such as hue, brightness, color saturation, and contrast. They also counted the number of faces in each image, and the number of likes and comments they’d received. Then they plugged all of that information into a computer algorithm designed to compare results for depressed versus non-depressed individuals.
There were indeed some significant differences. Depressed ’grammers tended to post bluer, greyer, and darker photos than non-depressed people. The two groups’ favorite filters were different, too: Depressed people used “Inkwell” most often (which turns photos black and white), while happier posters largely preferred “Valencia” (which has a lightening effect).
In fact, the algorithm correctly predicted which participants suffered from depression at a better rate than general practitioners typically do during in-person patient assessments. And it did so even when the analysis was limited to photos posted before people’s diagnoses.
“The people who were posting these darker, greyer pictures didn’t necessarily know that they were depressed at the time,” says study author Christopher Danforth, Ph.D., associate professor of mathematics and statistics at the University of Vermont. (Danforth’s co-author, Andrew Reece, is a doctoral student in psychology and computational engineering at Harvard University.)
Depressed people also got fewer likes, and tended to post more pics with faces. However, their photos had fewer faces per photo than those of non-depressed people. The researchers speculate that many of these posts might be “sad selfies” versus group pictures with friends, although they didn’t parse out the data enough to know if this is true.
Until this research is replicated on a larger scale, Danforth cautions that it shouldn’t be put into any type of official practice. And, he says, medical ethicists and legislators would also need to weigh in on the implications for privacy rights, insurance coverage, and other important considerations.
But are the findings really legit? Ben Michealis, Ph.D., thinks they could be. After all, it’s known that depressed people experience brain changes that alter their perception of the world, says the author of Your Next Big Thing: 10 Small Steps to Get Moving and Get Happy.
“We know that people who are depressed get less enjoyment out of activities that they tend to like, they may move and think more slowly, have problems initiating activities and even perceive the passage of time differently,” says Michaelis, who reviewed the study but was not involved in the research. “The idea that people who are depressed would be attracted to certain modes of self-expression that involve less colors or certain types of filters makes intuitive sense in this regard.”
He thinks that with further research, computer programs like this one could be useful for doctors and patients—especially for people who are less conscious of their own depressive symptoms. “The use of algorithms like these may help to increase their self-awareness, or the awareness of the people around them who might try to help,” he says.
But, he adds, this research should still be considered a pilot test, and no one should be using it to make diagnoses (of themselves or others) just yet.
“We need to be responsible with the data,” says Michaelis. “My hope is that [this study] leads to additional research to validate this as a measure of depression.”
Danforth agrees. “I think that in the future, 10 years from now, doctors will have all manner of tools at their disposal to understand how patients are feeling,” he says. “Maybe one of those signals could be the types of words people sneak into their phone conversations, or the types of pictures they post to social media—but these would just be single signals contributing to a doctor’s assessment, not a diagnosis by themselves.”