“Carol,” a 57-year-old divorced female, was in my office for a fourth opinion regarding medication treatment for her nightmares and flashbacks of childhood abuse. It took several years before she was diagnosed with post-traumatic stress disorder (PTSD) because of her reluctance to disclose the abuse, leading to a decline in functioning. She now takes five different psychotropic medications and has tried a dozen others, but her symptoms continue unabated. Unfortunately, this scenario of delayed diagnosis and multiple failed treatments is all too common with PTSD.

But here is where we may be headed: Carol’s PTSD could be diagnosed by blood tests, voice recordings, smartphone data, and history. Tools using large data sets and Carol’s genetic makeup could swiftly guide personalized treatment toward a successful outcome. This is the promise of precision psychiatry, or the ability to tailor diagnosis and treatment to a specific individual.

PTSD Is a Puzzle

 

Source: Cottonbro/Pexels

PTSD is still shrouded in confusion and stigma. Its wide variety of possible symptoms—most visibly flashbacks, hyperarousal, mood dysregulation, nightmares, and avoidance of reminders of the trauma—overlap heavily with those of other psychiatric disorders, making accurate diagnosis very difficult. Patients can also be reluctant to reveal their trauma to clinicians, at times making diagnosis by history alone virtually impossible. And while there are specific treatments like cognitive behavioral therapy and antidepressants, large numbers of patients fail to respond to any given treatment.

Psychiatry is the only medical discipline that doesn’t depend on biomarkers (objective evidence capturing what is happening in a cell or organism) for diagnosis and/or treatment. But researchers across the country—in many newly-created centers for precision psychiatry—are trying to change that by starting to develop blood and brain tests and specific digital tools to detect PTSD.

Machine Learning Tools

One promising approach uses the concept of digital phenotyping, or data collected from (often personal) devices like voice recordings, smartphones, or watches. Individual data sets can then be bundled together and analyzed by various forms of artificial intelligence to create algorithms of possible PTSD symptoms This line of research—known as machine learning—promises earlier and better detection of PTSD.

It’s been nearly impossible to foresee who—after exposure to trauma—will actually develop the many and varied symptoms of PTSD. Yet the ability to predict who will develop the disorder could potentially save untold years of suffering. With the backdrop of the emergency room as a laboratory and fueled by rapid advances in artificial intelligence, researchers at NYU have begun to untangle this puzzle.

 

Source: Pixabay/Pexels

Isaac Galatzer-Levy, Ph.D., adjunct assistant professor of psychiatry at NYU and senior staff research scientist at Google, sums up the current state of research on PTSD prediction: “the original idea with emergency room prediction research was that we had many individual measures of stress that each conferred some small amount of risk. There’s an idea in machine learning that many weak predictions can build a strong prediction, and that matters a lot in the context of PTSD because there’s heterogeneous reasons that people develop it both socially and biologically.”

In a groundbreaking study published in Nature Medicine, Katharina Schultebraucks, Ph.D., associate professor of psychiatry at NYU, was able to develop a PTSD prediction algorithm for emergency department patients using data from existing electronic health records. She and her colleagues accurately identified those who were at risk of developing PTSD in the ER and over the 12 months following trauma exposure. “This development is very unique because we have the opportunity to implement a digital tool directly in the emergency department in a real-world setting and identify patients at risk when they are still in contact with the healthcare system. This can reduce barriers to successful treatment,” she said.

What Lies Ahead

Although this science is still in the early stages, researchers envision a future when multiple data-driven methods can be used to prevent, diagnose, and treat mental disorders like PTSD. For example, a 2019 study demonstrated that a speech-based algorithm could accurately distinguish war-zone PTSD sufferers from people in a control group, providing initial evidence that speech can be a diagnostic marker of PTSD.

 

Source: Divinetechygirl/Pexels

A 2020 study using “deep learning” (a form of artificial intelligence in which algorithms are modeled to work like the human brain) of video- and audio-based markers of mood and arousal showed promise in correctly identifying PTSD and depression one month after emergency room admission. According to author Dr. Schultebraucks, this study is “proposing an innovative alternative to self-reports or clinical assessments using digital phenotyping. It’s unique because it integrates multiple sources of information such as facial expressions, voice prosody, speech content and movement patterns in an automated way to support clinical decision making.”

And a just-published 2024 Nature Genetics study of more than 1 million people—the largest study ever attempted—identified 95 PTSD genetic risk blood markers, potentially paving the way for eventual PTSD-specific diagnostic tools and therapies.

Diagnostic Conundrum

One of the problems with PTSD is the extremely large number of ways that it can present. Drawing from a long list of potential symptoms, the ultimate presentation can be like a patchwork that evades detection by even the most astute of clinicians. “I think that a lot of what’s happening in research is really aimed at reinventing the diagnosis to make it more specific and more actionable for more precise treatment,” says Galatzer-Levy. The ultimate goal is to develop many subtypes of PTSD based primarily on genetic and data-driven profiles. “Fifteen years from now, an ideal outcome is where there really is not a PTSD diagnosis or treatment, but a much more nuanced understanding of how different types of stress affect us,” he adds.

The Human Touch

Among the criticisms of precision psychiatry is the fear that the “person” (both patient and clinician) might be disregarded. When it comes to the patient, a diagnosis cannot be made without taking cultural and social context into consideration—something that machine-generated algorithms might have a more difficult time accomplishing. And the AI-derived models currently under study couldn’t be utilized without a clinician’s expertise in terms of teasing out the fine points of the patient’s story. Yet when used judiciously by an experienced clinician, the precision models currently under study would add an immense amount of value. For patients like Carol, that could be a game-changer.