Inferring Psychiatric Illness Based on Digital Activity Crosses Milestone

Facebook activity shows promise to help infer mental illness diagnosis.

Source: Matthew Libassi/The Feinstein Insitutes for Medical Research, used with permission

In a sense, the field of psychiatry is stuck in the prehistoric age; as clinicians, we are reliant on self-reporting, and observations from family and friends, to identify and treat signs of psychiatric illness. This often contributes to long delays in in treatment and effective care. We need to break free of our antiquated ways and begin to pay more attention to the digital era we are currently living in.

My research career focuses on harnessing digital technology to learn how to improve the ways we screen, diagnose, and care for individuals with mental illness. A significant portion of my work and my colleagues' efforts through Northwell Health’s Early Treatment Program has been trying to understand how we can use Internet search and social media activity as clues into an individual’s experience of psychosis.

Now, for the first time, in two separate research studies, we have been able to accurately predict a patient’s first psychiatric hospitalization and diagnosis, more than a year before it happened.

In the first study, published in njp Schizophrenia, working in close collaboration with IBM Research, we developed novel machine learning (ML) algorithms utilizing users’ Facebook activity capable of predicting those who go on to develop schizophrenia spectrum disorders (SSD) and mood disorders (MD) with high degrees of accuracy up to 18 months in advance. The study analyzed 3,404,959 Facebook messages and 142,390 images across 223 consented participants. We observed statistically significant differences in online characteristics among the populations, like the words used, and the images posted, eighteen months before their first psychiatric hospitalization. For example, participants with SSD used more words related to perception (hear, see, feel), and participants with MD were more likely to upload photos with more blues and less yellows. Integrating Facebook data with clinical information could one day serve to inform and improve clinical decision making

In collaboration with Cornell Tech and Georgia Tech, the second paper published in Public Library of Science (PLOS) One analyzed more than 400,000 Internet search queries, utilizing one year’s worth of data before the first psychiatric hospitalization. Again, statistically significant differences in the timing, frequency, and content of searches among those with SSD and MD were observed. For example, participants with SSD and MD were more likely to search in the middle of the night.

Why does it matter?

While search and social media activity alone should never be used to diagnose mental illness, these studies highlight their potential for providing clinically meaningful information and demonstrate the promise of machine learning, and digital technology for psychiatry and better patient care.

We were intrigued and excited by these results; to be able to observe, in an objective way, online activity associated with someone's mental state, well before diagnosis/hospitalization is important. We were also able to track a patient’s change in behavior over time as their symptoms escalate. The dramatic rise in social media and internet use could provide an opportunity to inform clinical care raging from early relapse identification, hospitalization, and diagnosis.

We are on the cusp of change. This growing field of research will help us explore what’s possible and ask the tough questions, both ethically and practically. I hope that one-day internet use, social media (whatever it might be at that time) will be one of many methods of understanding and care for individuals with mental illness.