Facebook posts might be better at predicting some health conditions than traditional demographic information, a new study suggests.
Researchers at Stony Brook University and The University of Pennsylvania Health System have found that the language people use on Facebook could help clinicians predict whether or not they suffer from illnesses such as depression and diabetes in a similar way to physical symptoms.
The research team used an automated data collection technique to gather and analyse the entire Facebook post history of almost 1,000 patients, who had also linked their electronic medical records to their social media profiles.
After acquiring and examining information on the language used in their posts and their demographics, such as age and sex, the researchers found they were able to identify 21 different conditions using data collected from participants’ Facebook posts alone. These included psychosis, anxiety and alcoholism.
In 10 of the conditions identified, researchers concluded that Facebook data was a better predictor than demographic information.
Some words had obvious links to the conditions. For example, the team found that those who used words such as “drink” and “bottle” in their posts were more likely to abuse alcohol.
Less obviously, those who used religious language in their posts, e.g. words such as “God” and “pray”, were 15 times more likely to have diabetes compared to those who rarely used those words.
Lead author Raina Merchant, director of Penn Medicine's Center for Digital Health, said: "This work is early, but our hope is that the insights gleaned from these posts could be used to better inform patients and providers about their health.
"As social media posts are often about someone's lifestyle choices and experiences or how they're feeling, this information could provide additional information about disease management and exacerbation."
Senior author Andrew Schwartz, assistant professor of computer science at Stony Brook University, added that our “digital language” reflects a different aspect of our lives to traditional medical data and could provide medical professionals with key diagnostic insights.
"Many studies have now shown a link between language patterns and specific disease, such as language predictive of depression or language that gives insights into whether someone is living with cancer,” he added. “However, by looking across many medical conditions, we get a view of how conditions relate to each other, which can enable new applications of AI for medicine."
In light of the findings, the research team suggests there may be potential for developing systems that allow patients to hand over their social media data to medical officials should they choose to so that they can provide clinicians with additional data that may improve how effectively their condition is treated.
"For instance, if someone is trying to lose weight and needs help understanding their food choices and exercise regimens, having a healthcare provider review their social media record might give them more insight into their usual patterns in order to help improve them," Merchant explained.
Merchant is conducting a trial later this year to test this theory.