Data passively collected from cell phone sensors can identify behaviors associated with a host of mental health disorders, from agoraphobia to generalized anxiety disorder to narcissistic personality disorder. New findings show that the same data can identify behaviors associated with a wider array of mental disorder symptoms.
Colin E. Vize , assistant professor in the Department of Psychology in Pitt's Kenneth P. Dietrich School of Arts and Sciences, is co-PI on this research, which broadens the scope of how clinicians might one day use this data to treat their patients.
The work was led by first author Whitney Ringwald (SOC WK '18G, A&S '21G), professor at the University of Minnesota who completed her graduate training at Pitt. Also on their team were former Pitt Professor Aiden Wright, now at the University of Michigan, and Grant King, one of Wright's graduate students.
"This is an important step in the right direction," Vize said, "but there is a lot of work to be done before we can potentially realize any of the clinical promises of using sensors on smartphones to help inform assessment and treatment."
In theory, an app that could make use of such data would give clinicians access to substantially more, and more reliable, data about their patients' lives between visits.
"We're not always the best reporters, we often forget things," Vize said of filling out self-assessments. "But with passive sensing, we might be able to collect data unobtrusively, as people are going about their daily lives, without having to ask a lot of questions."
As the first steps to realizing such a tool, researchers investigated whether they could infer if people were behaving in ways associated with certain mental health conditions. Previous research has connected passive sensor readings with behaviors that point to specific illnesses, including depression and post-traumatic stress disorder. This new work, published July 3 in the journal JAMA Network Open , expands upon that research, showing that it can be linked to symptoms that are not specific to any one mental health condition.
This is important, Vize said, because many behaviors are associated with more than one disorder, and different people with the same disorder can look, act and feel very differently.
"The disorder categories tend to not carve nature at its joints," he said. "We can think more transdiagnostically, and that gives us a little more accurate picture of some of the symptoms that people are experiencing."
For this study, Vize and a team of researchers used a statistical analysis tool called Mplus to find correlations between sensor data and mental health symptoms reported at baseline. The scientists then had to determine whether sensor data correlated with a set of broad, evidence-based symptom dimensions: internalizing, detachment, disinhibition, antagonism, thought disorder and somatoform, or unexplained physical symptoms.
In addition to the six dimensions, they also looked at what has been called the p-factor. This is not a specific behavior or symptom, rather it represents an ineffable, shared feature that runs across all kinds of mental health symptoms.
"You can think about it sort of like a Venn diagram," Vize said. If all the symptoms associated with all mental health issues were circles, the p-factor is the space where they all overlap. It is not a behavior in and of itself. "It's essentially what's shared across all dimensions."
The researchers made use of the Intensive Longitudinal Investigation of Alternative Diagnostic Dimensions study (ILIADD), which was conducted in Pittsburgh in the spring of 2023. From ILIADD, they analyzed the data of 557 people who had filled out self-assessments and shared data from their cell phones, including (but not limited to):
● GPS data that indicated how long people stayed home and the maximum distance they traveled from home
● Time spent walking, running and stationary
● How long their screens were on
● How many calls they received and made
● Battery status
● Sleep time
Using an app developed by researchers at the University of Oregon, the team was able to relate the sensor data to various mental health symptoms. Comparing the app's findings to questionnaires filled out by participants, Vize and team determined that the six dimensions of mental health symptoms, which reflect symptoms represented among many disorders, did correlate to the sensor data.
Interestingly, they also found sensor data correlated to the p-factor, a general marker of mental health problems. The implications of these findings are several-fold — ultimately, it may one day be possible to use this kind of technology to better understand symptoms in a patient whose presentation doesn't fit the category of any single disorder.
But for now, these data do not say anything about individuals' mental health; they deal in averages. Mental health is complex. Behavior varies wildly. "These sensor analyses may more accurately describe some people than others."
That's one of the reasons Vize doesn't see this kind of technology ever replacing a human clinician. "A lot of work in this area is focused on getting to the point where we can talk about, 'How does this potentially enhance or supplement existing clinical care?'
"Because I definitely don't think it can replace treatment. It would be more of an additional tool in the clinician's toolbox."
