Can Technology Help Us More Accurately Diagnose Mental Illnesses? (2024)

Whoever said the eyes are the windows to the soul probably didn’t imagine them being a key to diagnosing severe mental health conditions. But some research shows that what a person focuses on and how their pupils respond, among other eye movements, can reveal the presence of depression, PTSD or schizophrenia, as well as developmental disabilities like autism.

Now, Senseye, a Texas-based mental health platform, is aiming to bring that research out of the laboratory and into therapists’ offices and patients’ homes. The company is testing a phone app that measures how people’s eyes respond to various visual tasks as a way to detect PTSD.

The current diagnostic process is not as effective as it needs be, and mental health practitioners say the generally relied-upon surveys are too subjective and aren’t adaptable to various identities, leaving many patients struggling for long periods.

“Unfortunately, a lot of it ends up being kind of an educated guess situation,” says Jessica Jackson, a psychologist and chair of the American Psychological Association’s Mental Health Technology Advisory Committee. “If there was a test that could say their blood work shows they have PTSD, for example, that would speed things along, but that is not a current way to diagnose.”

Still, Berkowitz believes some of the nascent technological solutions have enormous potential. Current diagnostic tests “do not at all address the complexity of human beings,” he says. “We’ll now be able to have some sense of this is really what is going on for this particular person.”

As it stands, patients may have to wait weeks or years to receive an accurate mental health diagnosis. A 2009 study found that, on average, 3.5 years pass between a person's first major episode and a confirmed bipolar I disorder diagnosis. Another, from 2012, showed that PTSD is often misdiagnosed as depression. “These individuals are, symptomatically, kind of stagnant or even getting worse, and they’re not hitting an effective point of care until they’re in the active crisis,” says Aubrey Reider, senior manager of clinical operations at Senseye.

The misdiagnoses are a burden not only on the individual who is suffering and paying for medications that aren’t working or actively causing their conditions to worsen, but also on an already inefficient, overworked and understaffed mental health system.

More than half of the country already lives in what are called “mental health professional shortage areas,” where there’s an extreme lack of access to skilled providers (more specifically, the population-to-provider ratio is at least 30,000 to 1). The Health Resources and Services Administration (HRSA) is projecting a shortage of at least 69,610 mental health counselors in 2036.

The need for skilled mental health clinicians is great: One in five U.S. adults experience mental illness. And the demand for their services is only growing, with 38 percent more people receiving mental health care now than before the Covid-19 pandemic. But HRSA notes that close to half of those with mental illness don’t receive treatment.

Some researchers, academics and mental health experts believe updating our current assessments to incorporate more objective diagnostic tools could help shorten how long it takes to receive an accurate diagnosis, which would help current patients get better faster and free up clinicians to bring new patients into the system more quickly.

“In the case of mental health, you’re talking about wait times that can be five months. People potentially die while they’re on those wait lists, or they lose their jobs, or they lose their marriages, or they lose friendships,” says Matt Vogl, CEO of VXVY Mental Health, a digital mental health consulting company that grew out of the University of Colorado’s National Mental Health Innovation Center, which Vogl also led. “The wait list is a very bad place for a patient to be, because they’re people that obviously need care, expressed the interest in getting that care, but they just have to wait.”

The next generation of tests

The most common mental health assessments being used today are short, self-reported, written or verbal tests—a handful of questions that can help clinicians determine if, for example, anxiety or depression are present. These are screeners like the Patient Health Questionnaire-2 (PHQ-2), which is now commonplace during primary care office visits: Over the last two weeks, how often have you had little interest or pleasure in doing things? Over the last two weeks, how often have you felt down, depressed or hopeless?

Though these tests are free and easy to administer, they may not capture the full extent of what’s going on. They rely on subjective reporting, which can lead to false positives, false negatives or misdiagnosis. “It’s especially tricky in mental health because, very often, you have people who have multiple mental health conditions,” Vogl says. “And the symptoms often mirror each other.” As an example, he says, depression with co-occurring ADHD can present like bipolar disorder.

Primary care physicians are particularly bad at diagnosing mental health conditions. According to a 2011 study in Canada, more than 65 percent of the time, what they diagnose as major depressive disorder is something else. This has significant repercussions, because these doctors are also responsible for prescribing about 80 percent of antidepressants. And the misdiagnosis rates are even higher when talking about generalized anxiety disorder (71 percent) and bipolar disorder (92.7 percent).

Some companies are using algorithms and augmented machine learning in an effort to create more exacting—and faster—diagnostic assessments.

Adaptive Testing Technologies, which was started by Robert Gibbons, a statistician at the University of Chicago, developed the CAT-MH and K-CAT screening systems, available in English and Spanish. The former (Computer Adaptive Tests-Mental Health) offers specific modules for assessing depression, anxiety, substance use disorder, psychosis, PTSD, mania/hypomania, adult ADHD and suicidality. K-CAT is a version specific to youth ages 7 to 17 and their caregivers. Clinicians choose which modules to administer.

While the tests still require self-reporting by patients, rather than the standard 9- or 20-question evaluation, they don’t rely on a rote list of inquiries. Instead, the questions are tailored based on how the patient answers. This adaptive format improves accuracy while making the test less burdensome on patients, Gibbons says. The assessments take just a few minutes to administer and require clients to answer an average of ten questions.

At the University of Chicago Medical Center, 999 emergency patients were screened for depression and suicide risk using CAT-MH. The 30 patients the tool identified as having a severe suicide warning had not been flagged by emergency department staff, but the findings resulted in them being connected with the psychiatric team.

The University of California, Los Angeles, is currently using CAT-MH to screen incoming freshmen for depression, anxiety and suicide risk as part of a large-scale study. A paper outlining early results, published last July, noted that 516 students received care as a result of the assessment and they “on average, experienced significant reductions in their anxiety and depression symptoms.” While uptake when care was offered still remained relatively low (at 14.4 percent), data showed that students across ethnic and racial groups were equally likely to engage in treatment.

Similarly, Clinicom is an adaptive digital assessment tool sent by practitioners to patients that can screen for more than 80 mental health conditions in five languages. The questionnaire—which can be completed on a phone or a computer—requires a more in-depth process than CAT-MH, taking an average of 28 minutes to complete depending on the complexity of a patient’s history and needs. The reports are submitted directly to clinicians, and the system generates a suggested diagnosis (or more than one), which can be accepted, rejected or ruled out. That patented “clinical feedback loop” is what sets the survey apart, says CEO Ignacio Handal, whose father, Nelson Handal, a distinguished fellow of the American Psychiatric Association, developed the assessment.

The program uses augmented intelligence, so it learns from how the clinicians respond to the recommended diagnoses. In an open letter, Christopher Lucas, a psychiatrist at Upstate Medical University, said, “Clinicom can efficiently and accurately assess the full picture of a patient’s presentation, while allowing our clinicians to retain full diagnostic control—reducing false positives and negatives.”

“With Clinicom, I can screen an entire school district or the entire state of New York in a day,” Handal says. “It [the test] feels like it’s listening. And if it listens, people want to trust it more.”

The company spent 15 years performing clinical tests at two dozen facilities before releasing Clinicom as a commercial product. Its 2018 validation study, published in the Journal of Psychiatry Research, said that the tool had “a relatively high level of reliability” at around 78 percent. Disorders such as ADHD and major depressive disorder were “more consistently diagnosed.”

According to Handal, Clinicom is currently being used in nine countries, including on 58 State University of New York campuses.

New tools on the horizon

The “next wave” of this technology, per Vogl, is analyzing physiological mechanisms or biomarkers—eye movements, blood tests, voice analytics, heart rate and the like—to help diagnose mental health conditions in ways similar to how we do for physical health.

Senseye, which is in the midst of clinical trials with around 1,000 patients across the country, is part of this wave. The company is working toward Food and Drug Administration approval, with the hopes of coming to market in the next couple of years and eventually applying eye tracking to other mental health concerns, such as major depressive disorder and generalized anxiety disorder.

Another app is Ellipsis Health, which relies on a person’s voice sample to help diagnose anxiety and depression earlier. “We look at the semantic and acoustic aspects of speech,” says Mike Aratow, the company’s chief medical officer and co-founder. “Previous researchers had discovered, for example, if you’re depressed, you use more personal pronouns. We also look at the acoustic aspects—how you enunciate, your rhythm, spectral characteristics.”

Using machine learning, Ellipsis has trained its model to evaluate the severity of anxiety or depression. Frontline workers record the patient’s natural dialogue (with consent) during conversation. The program only needs to capture 60 to 90 seconds of data.

“We are the most helpful at the top of the funnel, which is where most people go—primary care physician checks, nurse checks, case management checks,” says Mainul Mondal, Ellipsis co-founder and CEO. “That’s where gap starts in big way.”

The platform, which took six years to build and validate, came online for commercial use earlier this year, and the team plans to eventually expand to additional disorders, such as bipolar and schizophrenia. Once connected with a clinician, patients can download an app that will allow them to capture 45-second voice journals to track progress.

Like most of these platforms, Ellipsis does not diagnose. Rather, it tells physicians the severity of symptoms; they then rely on their own expertise to make decisions or recommendations. All of these tools are designed to aid clinicians in evaluating their patients, not take over that role.

“The idea of this technology was never to get rid of the clinician,” says Adaptive Testing’s Gibbons. “It was to allow the clinicians to have the time to be able to provide treatment and pick appropriate triage based on high-quality psychological and psychiatric measurement.”

These approaches have upsides, proponents say. For one, they allow clinicians to home in on problems quicker. Patients may also be more comfortable revealing information via a phone or a computer rather than face-to-face. And because of the ease of use, people can be retested at regular intervals to determine if treatment is working or if adjustments need to be made.

According to the American Psychological Association’s Jackson, they’re also an opportunity to improve equity in mental health assessments. “For a field that’s been around a long time, this is a new way of being able to be more inclusive in thinking about how you diagnose.” She notes that many of the current diagnostic tools, like the leading Diagnostic and Statistical Manual of Mental Disorders (DSM-5-TR), rely on data from adult, heterosexual men and women who are of a certain age and often white. “One of the reasons that it could take longer to diagnose is because we’re holding criteria to somebody who doesn’t meet that criteria,” she says. These new technologies, she continues, are “a great opportunity to get it right.”

However, Jackson worries that there’s still not enough science supporting some of these efforts. She’s also concerned that there’s not enough transparency surrounding the data behind these databases, which generally start their algorithms by pulling from those traditional diagnostic tools, like the DSM-5 and PHQ screeners. Technology, she adds, can’t pick up on nuance in a way people—clinicians—can.

“Anytime that you are building something like Senseye, I don’t know of enough data that would say that you can test somebody’s eyes and see PTSD across demographics,” Jackson says. How, she wonders, could a voice assessment pick up the difference between major depressive disorder and grief?

Other challenges remain. Some are commonplace tech-related worries, like the fact that an internet connection is necessary and concerns about data security. Others are more esoteric: Will physicians be liable for mistakes resulting from artificial-intelligence-connected tools? How much time will clinicians actually save?

“There’s a healthy dose of skepticism to a degree among providers,” Jackson says. Still, she’s excited about the future potential.

“We’ve been doing things the same way since Freud and them: People get on a couch, we get information, we think about what it means, we do rule-outs. This will change how clinicians are trained, how we are able to do our work,” she says. “I’m very excited about the possibilities. And I temper that with caution. … I want to see companies get it right.”