The color palette and the number of faces that appear in photos that individuals post on their social media feeds could more accurately indicate whether they are suffering from depression than an in-person discussion with a physician, a machine-learning study suggests.
Mining data from the Instagram feeds of depressed individuals and healthy individuals, the researchers found that computational algorithms analyzing color and content of posted photos could successfully identify depressed patients in more than two thirds of cases, compared with fewer than half of cases in general practice consultations.
The results also showed that individuals with depression were more likely than healthy people to post blue, dark, and gray photos and to use filters to turn their pictures black and white.
Furthermore, although depressed users were more likely than healthy people to post photos showing faces, on average, the photos the depressed users posted contained fewer faces per photo.
The study was published online August 8 in EPJ Data Science.
Not Ready for Prime Time
Study investigator Andrew G. Reece, PhD, Department of Psychology, Harvard University, Cambridge, Massachusetts, told Medscape Medical News that the analytical technique employed in their study is not at the diagnostic stage, and "one big reason for that is that we don't know if the people that volunteered to be in our study are a good representation of the broader public.
"There may be something special about people who choose to divulge their personal mental health history and social media profiles to teams of researchers they don't know.... So we really have to be careful before we can say: 'Yes, these findings are something that can be applied to everybody.' "
Nevertheless, he believes that, if the findings were to be validated and are found to be generalizable, "then I see something like an opt-in program being available to patients."
This kind of analytic tool, said Dr Reece, could, for example, alert a physician via an electronic device to further screen for emerging depression.
"But, certainly, data privacy and making sure that the control of data is in the hands of the patients would be of upmost importance here," he added.
Although previous studies have shown that social media may be used for to screen for a range of conditions, these studies have relied on textual analysis. Visual media has not yet been explored.
The researchers note that there are almost 100 million new posts to Instagram every day and that it is one of the fastest growing online networks. They invited depressed and healthy Instagram users to complete separate series of questionnaires.
Both depressed individuals and healthy individuals were screened to determine their eligibility for the study. The former also completed the Center for Epidemiologic Studies Depression Scale, as well as other depression scales, such as the Beck Depression Inventory and the Kellner Symptom Questionnaire.
Individuals who met all the inclusion criteria were asked to share their Instagram usernames and history. The researchers then made a one-time collection of individuals' entire posting history.
Users of Amazon's Mechanical Turk (MTurk) crowdwork platform were then asked to rate a random selection of 20 photos, marking how interesting, likable, happy, and sad each photo seemed, on a continuous scale of 1 to 5. Each photo was rated by at least three raters. Following this, 100 posts were rated per study participant. For depressed individuals, the ratings covered the year before and year after diagnosis.
Data on user activity and community reaction, assessed on the basis of the number of comments and likes, were also collected, and face detection and pixel-level color analyses were performed on posts to identify any correlations with depression. In addition, use of Instagram image filters was assessed.
Privacy Concerns
From the original 509 individuals who expressed interest in taking part in the study, 221 (43%) refused to share their Instagram data.
Dr Reece believes this reflects a general mistrust about sharing social media data.
"In fact, they were much more likely and willing to share their personal, private, mental health history than they were their social media record," he said. He emphasized that respect for data privacy should form a central part of future studies.
Of the remaining 288 individuals, complete data were collected on 166 Instagram users, of whom 71 had depression.
This yielded a total of 43,950 posted photographs. The mean number of posts per user was 264.8. For depressed users, the mean number of posts was 349.5; for healthy uses, the mean number of posts was 201.5.
Underlining the fact that the data were skewed because of the relatively small number of participants who posted photos frequently, the median number of posts per user was 122.5 (for depressed users, 196.0; for health users, 100.0).
The mean age was 28.8 years for depressed individuals and 30.7 years for healthy participants.
The researchers found that Instagram posts by depressed users tended to be bluer, darker, and grayer than those from healthy participants. Moreover, posts by depressed individuals received more comments but fewer likes than those from healthy users.
Interestingly, depressed individuals were more likely than healthy users to post photos with faces, but on average, their posts had fewer faces per photo.
Depressed users were also more likely to use the Inkwell filter, which converts photos to black and white, whereas healthy individuals were more likely to use the Valencia filter, which lightens the tint.
The Hidden and the Obvious
The team also found that with respect to identifying individuals with depression, analysis of social media posts had lower specificity but was more accurate than assessment by general practitioners (GPs). The positive predictive value was 70% for analysis of social media posts vs 42% for assessment by GPs, as calculated on the basis of a meta-analysis of 118 studies by Alex J. Mitchell, MRCPsych.
Of the four ratings applied by the MTurk users, only sadness and happiness were significant predictors of depression, but the correlation was extremely low with the results from using computational features.
The computational analysis was able to predict depression on the basis of prediagnosis posts in only a third of cases. Despite having lower predictive power, the degree of specificity was higher for computational analysis than for GP assessments, using Dr Mitchell's and colleagues' analysis.
"Our findings establish that visual social media data are amenable to analysis of affect using scalable, computational methods. One avenue for future research might integrate textual analysis of Instagram posts' comments, captions, and tags," the investigators write.
"Considering the early success of textual analysis in detecting various health and psychological signals on social media, the modeling of textual and visual features together could well prove superior to either medium on its own," they add.
They also note that this research may serve as a "blueprint for effective mental health screening in an increasingly digitalized society.
"More generally, these findings support the notion that major changes in individual psychology are transmitted in social media use, and can be identified via computational methods," they add.
Dr Reece believes the study shows that "social media is just another form of how people communicate with each other" and that, just as it would not be surprising for a coworker to notice a dark or grim expression on a person's face when talking to them, the same can be true of a person's social feed.
"I think it's important for people to recognize that it's not just the content of what they post on social media, but there's the equivalent of having a grim look on your face that also may come across in how you choose to convey your experiences and your thoughts and feelings online," he said.
"Somehow, I think that that doesn't quite seem intuitive to people yet, and so both the hidden and the obvious are ways that people pick up on information about other people, and that seems to happen just as much in social media as it does in real life."
Interpret With Caution
However, Shari Harding, PMHNP-BC, assistant professor and program director of the psychiatric-mental health nurse practitioner track at Regis College, Weston, Massachusetts, who was not involved in the study, disagreed with that interpretation.
Although she believes the study was well done and shows how technology can be used for mental health screening, she told Medscape Medical News that "there's something very different about social media than being in person with somebody, and I think that you see that when you look at the whole body of literature about social media and depression.
"People are selective about what they're posting on social media, whereas if you're in person, having a conversation with somebody in a daily interaction, you're not necessarily in control of your microexpressions and lots of other facets of your interaction with people," she added.
Noting that she does not use social media, Harding said that one of her concerns regarding social media is that "people are giving you the highlight reel" of their lives, complete with perfectly posed selfies.
She believes that in contrast with these "highlight reels," when individuals look at their own lives in full, they see "the good, the bad and the ugly, so I think there's a tendency to be more unhappy the more that you spend time looking at this skewed reality on social media."
The relationship between social media and depression could be bidirectional, she said, inasmuch as "depressed people are going to use social media more, because they don't have other options, like maintaining face-to-face relationships."
Lori Russell-Chapin, PhD, professor and codirector for the Center for Collaborative Brain Research, Bradley University, Peoria, Illinois, is a self-confessed social media addict.
She agreed with Harding that studies such as these should be interpreted with caution. Because of the depersonalized nature of social media, users are able to present just one aspect of their lives, she noted.
For Dr Russell-Chapin, a greater concern, particularly given that 12% of the US population is said to be addicted to their smartphones, is that social media appears to be changing the function and structure of the brain. Among these are changes to alpha wave patterns that resemble those seen in patients with attention-deficit disorder and epilepsy.
"I find social media quite scary, and I'm probably as addicted to it as the next person, but I really think that we have to understand what it's doing to our brain," she said.
Study coauthor Christopher M. Danforth, PhD, received funds from the National Science Foundation. Dr Reece received support from the Sackler Scholar Program in Psychobiology.
EPJ Data Sci. Full text
Medscape Medical News © 2017
Cite this: Instagram Pics May Flag Depression - Medscape - Aug 21, 2017.
Comments