Facebook activity can be used to identify people with mood disorders and schizophrenia spectrum disorders more than a year prior to their first psychiatric hospitalization, according to research published in npj Schizophrenia.
Although these findings are not meant to substitute for clinical psychiatric assessments, the results suggest that social media data can be used in addition to clinical assessment to support decision-making.
“Much like an X-ray or blood test is used to inform health status, Facebook data, and the insights we gather, could one day serve to provide additional collateral, clinically meaningful patient information,” the researchers said in the report.
TOP-LINE DATA
The machine-learning algorithms used in the study correctly identified participants with schizophrenia with an accuracy of 52%, an accuracy of 57% for participants with mood disorders and an accuracy of 56% for participants without a psychiatric disorder.
The study analyzed the participants’ activity on Facebook, including the images they posted and the type of language they used in Facebook Messenger, the site's instant messaging service.
Compared to the participants without psychiatric disorders, both the mood disorder and the schizophrenic groups tended to post images with a smaller height and width. Those with mood disorders also tended to post images with more blue and less yellow colors, compared with those without psychiatric disorders.
There were significant differences in linguistic features between each group.
Schizophrenic participants were the most likely to use words related to perception, such as “see, hear or feel.” They were more likely to express negative emotions, use second-person pronouns, type using informal language like “btw, lol" and "thx,” and were less likely to use punctuation.
Meanwhile, words related to biological processes such as “blood and pain” and the use of first-person pronouns were associated more with the mood disorder group.
Overall, participants with mood disorders and schizophrenia were more likely to use swear words and anger-related language compared to those without a psychiatric disorder.
The study also examined how Facebook activity changed leading up to the hospitalization date that could suggest escalating symptoms. Some linguistic features existed well in advance of hospitalization, suggesting that certain activity could help clinicians identify people at high risk of developing schizophrenia or a mood disorder before the emergence of clinically significant symptoms.
Further, some language categories became increasingly different closer to the date of hospitalization, which could reflect changes in anxiety, mood, preoccupations, perceptions, social functioning, and other symptoms known to accompany the onset of illness, the authors said.
The categories that changed the most leading up to the date of hospitalization were the use of informal language, punctuation, words related to biological processes, words expressing negative emotions, swear words and anger-oriented language.
METHODS
Researchers recruited participants between the ages of 15 and 35 years old between March 2016 and December 2018. Participants were eligible if they had a primary psychotic disorder or mood disorder listed in their medical records. “Healthy volunteers” were also recruited based on their lack of clinically significant psychiatric symptoms in a screening process.
The average age was 23.7 years old, and 41.7% of the participants were male. The study had participation from a range of races, including Caucasian (42.2%), Black (29.1%), Asian (17.5%) and Hispanic (11.7%).
Participation involved a single visit during which all historical social media data were downloaded and collected. A total of 3,404,959 Facebook messages and 142,390 Facebook images were collected across 223 participants.
The study analyzed data stretching back 18 months before the first psychiatric hospitalization to minimize the potential confounds of medications, hospitalizations, relapses and receiving a formal psychiatric diagnosis on social media activity.
THE LARGER TREND
Past studies have looked into how a person’s posts on Facebook can be tied to a range of medical conditions. Researchers found that Facebook language was able to predict 21 medical conditions “beyond chance.”
Because of social media’s often negative impact on users’ mental health, some platforms have taken steps to help users avoid aggravating symptoms of mental illness.
For example, last year Facebook unveiled a "Let’s Talk" photo filter for Stories that invites peers struggling with mental health issues to privately reach out and confide in users with an open ear. Pinterest teamed up with researchers from Stanford Lab for Mental Health Innovation to create a tool aimed at giving users more resources. The tool offers short, evidence-based exercises for users to complete whenever they search for a trigger word, such as stress.
ON THE RECORD
“Furthermore, although Facebook data alone cannot yet be used to make a diagnosis, the integration of this information with clinical data could help to improve the accuracy and reliability of clinical diagnoses,” the researchers said in the study.
“Alternatively, the classifier could serve as a low burden screening tool for youth at risk of developing psychiatric disorders, providing objective collateral information, and identifying those in need of further evaluation.”
