This article is more than 5 years old. Information may no longer be current.
Speech-based technologies could detect PTSD in veterans
Click Here to Manage Email Alerts
Using a speech-based algorithm, researchers could objectively differentiate male warzone-exposed veterans with and without PTSD, according to a study published in Depression & Anxiety.
“Several studies in recent years have attempted to identify biological markers that distinguish individuals with PTSD, with candidate markers including changes in brain cell networks, genetics, neurochemistry, immune functioning and psychophysiology,” Charles R. Marmar, MD, Lucius N. Littauer Professor of Psychiatry, chair of the department of psychiatry and director of the PTSD Research Program at New York University School of Medicine, told Healio Psychiatry. “Despite such advances, the use of biomarkers for diagnosing PTSD remained elusive going into the current study, and no physical marker was applied in the clinic.
Marmar and colleagues developed technology to identify features of speech that differentiate veterans with PTSD from controls and tested it in 52 male warzone-exposed Iraq and Afghanistan veterans with the disorder and 77 age- and gender-matched controls. They excluded participants with major depressive disorder.
“Because measuring voice qualities is noninvasive, inexpensive and might be done over the phone, many labs have sought to design speech-based diagnostic tools,” Marmar said.
Using audio recordings of clinical interviews, the researchers obtained more than 40,000 speech features that were input to a random forest algorithm. They used this algorithm to build a classifier function via speech markers to predict PTSD.
The final random forest algorithm selected were based on 18 voice markers. The receiver operating characteristic curve had an area under the curve of 0.954. The PTSD probability cut point was 0.423 and the overall correct classification rate was 89.1%, according to the results.
The researchers found that among veterans with PTSD, occurrences of slow speech production reflected speech sections containing articulators that move more slowly or contain hesitations compared with controls. Additionally, veterans with PTSD had speech features that were more monotonous than controls and were more likely to generate flat speech features, the study showed.
“This is more about combining a Siri-like tool that can recognize voice features with additional formulas, perhaps someday wrapped into a cell phone app, which can sift through voice features for the 18 linked by the current study PTSD,” Marmar told Healio Psychiatry.
“Families can often tell broadly when a loved one’s voice changes with distress, but the tool aims to be more accurate than that, and specific to PTSD, such as those with the condition experiencing a strong, persistent distress when reminded of a triggering event,” he continued. “Veterans with the condition also often do not like crowds or going to clinics. Remote, noninvasive, accurate tools could relieve a great deal of suffering.” – by Savannah Demko
Disclosure: The authors report no relevant financial disclosures.
Perspective
Back to Top
Nils C. Westfall, MD
This study is what an early step toward successfully developing a method of diagnosing a mental illness, in this case DSM-IV PTSD, using voice analysis might look like.
The researchers showed that if you select a group of warzone-exposed veterans with PTSD (and limited comorbidity — for instance, no MDD) and a matched control group of warzone-exposed veterans without PTSD and know which group each subject belongs to, you can use automated analysis of voice markers and a random forest algorithm method to identify a panel of voice markers that best differentiates the cases from controls. In this study, the optimum random forest algorithm included 18 unique voice markers and yielded an overall correct classification rate of 89.1%.
It will be interesting to see how well this panel of voice markers differentiates PTSD cases from non-cases in blinded trials involving different populations and subjects with greater comorbidity. Whether this particular effort is ultimately successful or not, it advances the state-of-the-art in the struggle to identify clinically useful biomarkers for psychiatric disorders.
Click Here to Manage Email Alerts