According to the Substance Abuse and Mental Health Administration, about one in five Americans over 18 suffer from mental illness. Yet more than half of those 51.5 million adults haven’t received any treatment for their condition. A significant barrier to accessing treatment is the lack of a diagnosis. Researchers at The Feinstein Institutes for Medical Research believe that analyzing Facebook activity could help individuals who suffer from schizophrenia and mood disorders obtain an accurate diagnosis.

The study included 223 individuals who had at least one psychiatric hospitalization. In total, the participants consented to the use of 142,390 images and more than 3.4 million personal Facebook messages from the 18 months before the first hospitalization. The research team used novel machine-learning algorithms to identify characteristics exhibited by participants with schizophrenia spectrum disorders (SSD) and mood disorders (MD) that weren’t present in a healthy control group.

As might be expected, patients with MD show an increased use of certain words, such as negative emotion words and perception terms including “like,” “see,” and “feel.” However, more detailed findings seem to offer a window into the lived experience of mental illnesses, while ultimately helping researchers detect the presence of such a condition.

Individuals diagnosed with MD and SSD used more swear words than the control group, and those with MD used more words related to the body, such as “pain,” and “blood.” As participants with SSD approached their first hospitalization, their use of punctuation increased. Smaller size was associated with images posted by individuals with either diagnosis, and photos shared by those with MD featured more blue tones and fewer yellows than the control group.

The use of Facebook data for diagnosis naturally requires the patient’s informed consent. Ethical questions about the misuse of digital markers of mental illness will likely arise as these data tools move toward the market. Such technology could still support — and potentially replace — subjective screening tools currently used by mental health providers; these tools have a high error rate. Because early intervention can improve treatment response, the use of Facebook data might have a significant effect on long-term treatment outcomes. 

A paper on the Facebook study was recently published in Nature Partner Journals: Schizophrenia. The team’s leader, Dr. Michael Birnbaum, recently published similar findings regarding the use of internet search activity to predict diagnoses.