Search engines store so much of our online activity that it's possible our daily search patterns hold clues to our mental health.
"Google searches can really provide a lens into the way people are thinking and feeling," Michael Birnbaum, M.D., a psychiatrist and assistant professor at Feinstein Institutes for Medical Research, told ABC News.
More and more, online searches are becoming a primary resource for those seeking health-related information. This is especially true when it comes to stigmatized topics such as mental health, and new research suggests this digital data could prove to be an important tool for monitoring mental health monitoring.
Birnbaum, also the director for Northwell Health's Early Treatment Program, a clinical and research initiative for adolescents and young adults in the early stages of psychosis, partnered with researchers at Georgia Tech and Cornell Tech to "try to understand how we can use the internet to revamp psychiatry." His team set out to learn how mental illness manifests itself online and how experts can use that knowledge to improve ways of identifying patients with mental illness -- and delivering appropriate care.
A comparison of Google searches between healthy volunteers and patients with Schizophrenia Spectrum Disorders -- or SSD, psychiatric conditions characterized by symptoms of psychosis -- looked at 32,733 time-stamped searches, analyzing search activity timing, frequency and content. Researchers then developed machine learning algorithms that use search activity data to identify individuals with SSD and to predict psychotic relapse.
The study identified several important patterns. People with SSD searched less frequently and used fewer words in searches. During a relapse, people with SSD were more likely to use words related to hearing, perception and anger, and were less likely to use words related to health.
This isn't the first time online search activity has been linked to a person's medical information: Prior studies have shown that patterns and search content have been useful in predicting lung cancer, Parkinson's disease and pancreatic cancer up to a year in advance of the diagnosis.
Other online platforms have been studied in this context as well. For example, Birnbaum's group previously looked at how Facebook activity could be used to predict episodes of psychosis. Other researchers have identified patterns in Twitter and Instagram content as predictors of depression.
"The ability to reveal critically important changes from a person's baseline functioning in real time has major implications in how quickly people can be diagnosed or receive the help they need," Dr. Neha Chaudhary, child and adolescent psychiatrist at Massachusetts General Hospital and Harvard Medical School and Co-founder of Stanford Brainstorm, said in an email to ABC News.
Unlike many other medical fields, psychiatry relies entirely on subjective clinical impressions and self-reporting from patients or their families. A lack of objective data often results in significant delays in making a psychiatric diagnosis and initiating proper treatment.
In patients with mental illness, those delays may contribute to bad outcomes -- social isolation, unemployment or substance abuse. Any under-recognized psychiatric condition or relapse also comes with a risk of emergencies -- suicide, violence or the need for psychiatric hospitalization.
Digital data could augment self-reported data and help physicians recognize what patients need more quickly.
"If we can find a way to make sense of this data the right way, we may be able to create more accurate, robust tools to assess and identify those at risk for suicide and other public health crises," Chaudhary added.
Jack Rozel, M.D., M.S.L., medical director of resolve Crisis Services and president of the American Association for Emergency Psychiatry, told ABC News: "The goal is to use machine learning and large databases to figure out who is genuinely at risk of engaging in violence and suicide, for example."
But, Rozel, added, "It's tricky because we're talking about rare events, which makes it difficult to identify patterns, even with machine learning. If we are ever going to have a chance to predict and manage these high-impact events, we need to figure out how to use algorithms in a useful and evidence-based way."
"The power of the data," Chaudary added, "comes from its use in collaboration with professional help. That's where we'll really start to see a positive impact."
But a number of significant challenges remain.
"Unfortunately, the use of internet data today comes with major privacy and ethical concerns that must be navigated before this research can be implemented in the real world, in a way that's helpful without being harmful," she said. "In an ideal world, the benefit is clear: The data follows the person's behaviors, and the changes in behavior may follow changes to a person's mental health that may have otherwise gone unnoticed for months -- or even years."
Leah Croll, M.D., a neurology resident at NYU Langone Health, is a contributor to the ABC News Medical Unit.