What You Write on Facebook Could Be Used to Diagnose Conditions From Depression to Drug Abuse and STIs

Scientists have used people's Facebook posts to identify conditions from depression to STIs in a study. The researchers wanted to see if looking at a person's Facebook profile was a more accurate way of predicting and identifying conditions they had than their demographic information, such as age, sex, and race.

For the paper published in the journal PLOS One, the team recruited 999 participants and studied the 20 million words they had collectively posted in 949,530 Facebook status updates on the social media platform. Their posts included at least 500 words across all their posts in total.

A total of 21 conditions were featured, including digestive problems; injuries; pregnancy; skin disorders; anxiety; obesity; drug and alcohol abuse; and STIs.

Every condition could be predicted using a patient's Facebook data. But 18 categories were best-predicted using demographic and Facebook data; while 10 were best predicted using Facebook language versus demographic data.

Alcohol abuse, for instance, was linked with a participant using terms like "drink," "drunk," and "bottle." Hostile terms like "dumb," "bullsh*t," and "b*tches" were associated with drug abuse and psychoses.

Terms like "stomach," "head," and "hurt," which suggested an individual was suffering physical symptoms of a psychological disorder could predict they had depression.

And participants who used words like "god," "family," and "pray," meanwhile were more likely to have diabetes than those who mentioned them the least.

To see if they could find a pattern between a participants medical history and their Facebook page, the team also looked at their electronic medical records with the participants' consent.

The method was best at picking up conditions such as diabetes and mental health disorders like anxiety, depression and psychosis.

The authors argued that like the genome, which can reveal the genetics of an individual and their risk for certain conditions, their language use on social media could constitute a "social mediome." Social media is also easier to access than a person's genetic information, and could be used to personalize healthcare.

"Social media data are a quantifiable link into the otherwise elusive daily lives of patients, providing an avenue for study and assessment of behavioral and environmental disease risk factors," the authors wrote.

In the past, researchers have used social media posts to look at illness on a population-wide level, for instance during outbreaks.

Dr. Raina Merchant, the lead author on the study and director of the Penn Medicine Center for Digital Health, told Newsweek she didn't expect social media to be so predictive of health conditions "even when the language wasn't always explicitly about health."

However, she acknowledged the sample of patients were receiving care in one location, so the results might not be generalizable. In the study, the authors highlighted the approach is somewhat limited because the words only indicate a correlation with a condition, and can't explain why an individual has an illness.

"This is early work but it suggests we need to do further research about the opportunities for benefits in data mining and early detection and the potential for harm in data mining and predictions about health," said Merchant.

Experts in the field not involved in the study regard the research as an interesting addition to the burgeoning field of digital epidemiology, and praised the authors for making sure consent was central to their work. But they asked how privacy might be protected in a real-world scenario.

Adam Miner, licensed clinical psychologist of the Stanford University Department of Psychiatry and Behavioral Sciences, told Newsweek: "An open and important question is how social media behavior relates to health. This relationship opens the door to exploring earlier disease detection and intervention.

"Importantly, patient consent was obtained. However, it is unclear what additional expectations patients will have if they share social media data with clinicians. For example, will insurance companies or the government also have access to these data? Are there cases in which someone would be referred to the police for their online behavior? Clear policies and expectations will be crucial for all stakeholders," argued Miner.

Jason Colditz, researcher and program coordinator at the University of Pittsburgh Center for Research on Media, Technology and Health told Newsweek he was not surprised that Facebook text performed better than basic demographic indicators of age, sex, and race.

"It would be interesting to know how these data fare against more complete demographics such as education, socio-economic status, or residing in particular communities," he said. "These additional factors can be highly predictive of health conditions."

He continued: "If these procedures were replicated on a broader population of Facebook users without their informed consent, it would raise serious ethical concerns. This type of surveillance is particularly concerning around sensitive topics such as mental and behavioral health conditions or sexually transmitted infections. The risk lies not only in identifying these conditions but also in mis-identifying individuals who do not have these conditions."

Colditz also urged the public not to be inspired by the study to try to diagnose themselves or other Facebook users in a similar way.

"Diagnosing specific health conditions is best left to physicians who are trained to do so," he said. "Facebook users should be aware that the content they share and even the individual words that they use can provide information that can be used to identify aspects of their health that they might not want to be public knowledge."