In a study led by psychiatrist Dr. Charles R. Marmar, machine learning software was able to pick up on vocal characteristics that help distinguish between people with and without PTSD.
Photo: gorodenkoff/Getty
One of the most difficult mental health problems to diagnose is post-traumatic stress disorder (PTSD), in part because patients often present inaccurate information in self-report assessments and clinical interviews. Researchers at ٺƵ Health are harnessing artificial intelligence (AI) to find objective methods for identifying the disorder in military veterans—and in 2019, they published two studies that significantly advanced the field.
Decoding Telltale Vocal Patterns of PTSD
The first study, , searched for evidence of PTSD in veterans’ voices. “Clinicians have long observed that individuals with psychiatric disorders display changes in voice quality, such as ‘pressured’ speech in bipolar disorder or monotone speech in depression,” explains first author Charles R. Marmar, MD, the Lucius N. Littauer Professor of Psychiatry and chair of the . “We wanted to determine whether such changes could serve as a biomarker for PTSD.”
Working with SRI International, the nonprofit research institute that developed the smartphone assistant Siri, the ٺƵ team created an algorithm designed to identify vocal characteristics associated with the disorder. After recording standard Clinician-Administered PTSD Scale (CAPS) interviews of 52 Iraq and Afghanistan War veterans with service-related PTSD, as well as 77 veterans without the disorder, researchers fed the recordings into speech analysis software, which detected 40,526 different variables. The algorithm then searched repeatedly through these features until it found 18—involving factors such as tone, pacing, and enunciation—linked to PTSD. Using these markers, the program correctly identified patients with an existing diagnosis 89 percent of the time.
“The advantage of speech is that it can be measured objectively, inexpensively, remotely, and noninvasively,” says , adjunct assistant professor of psychiatry, who co-led the study. “Although further research will be required before this technology can be deployed in the field, we believe it could eventually enable clinicians to screen for PTSD using a smartphone app, anywhere in the world.”
A Promising Blood Test Accurately Screens for PTSD
The gold standard for identifying any illness is to observe its physical traces using a standardized lab test. In September, used AI to discover bloodstream markers for PTSD, creating the possibility for a screening blood test—the first of its kind for PTSD and one of the first such tests for any type of psychiatric disorder. The research was led by Dr. Marmar and colleagues from ٺƵ, Harvard John A. Paulson School of Engineering and Applied Sciences, and the U.S. Army Medical Research and Development Command.
An experimental blood test may accurately screen for PTSD.
Photo: Andrew Brookes/Getty
Previous studies had detected changes in gene expression among PTSD patients, but findings were limited by small sample sizes and a narrow molecular focus. The new study, launched in 2012, took a cue from cancer research, in which integrated multiomic studies have generated novel insights into diagnostic markers. The researchers recruited a cohort of male, war zone–exposed veterans of the Iraq and Afghanistan conflicts, 83 with PTSD and 82 without. Blood samples were assayed for genetics, methylomics, proteomics, metabolomics, immune cell count, cell aging, endocrine markers, microRNAs, cytokines, and more.
Using machine learning to analyze nearly 1 million molecular candidates, the team narrowed the field first to 343, then to 77, and finally to 27 chemical signatures that (together with heart rate) outperformed the larger groups in identifying PTSD. The researchers then tested these markers on a separate pool of 52 veterans, half of whom had been diagnosed with PTSD. The markers achieved 81 percent accuracy in distinguishing between patients and controls, detecting those with PTSD 85 percent of the time, and correctly identifying those without the condition 77 percent of the time.
Hopeful Signs for Veterans and Other People with PTSD
For both studies, more research is needed (including with women and non-veterans) before the technique can be submitted to the U.S. Food and Drug Administration for approval. Within a few years, however, such tools could fill an urgent need. “PTSD affects up to 25 percent of combat veterans who served in Iraq or Afghanistan, and 7 percent of the general population,” notes Dr. Marmar.
“This illness is associated with functional impairments, including relationship conflicts, unemployment, reduced academic attainment, substance abuse, and adverse health outcomes. By gaining a deeper understanding of the disorder’s physiology, we hope not only to improve screening and diagnosis, but to develop better ways of treating or preventing it.”
The speech biomarker study was supported by a grant from the U.S. Army Medical Research Acquisition Activity (USAMRAA) and Telemedicine and Advanced Technology Research Center (TATRC), as well as by the Steven and Alexandra Cohen Foundation and Cohen Veterans Bioscience.
The blood test study was supported by grants from the U.S. Army Research Laboratory’s Army Research Office and the U.S. Department of Defense.
