A 2023 Gallup poll reported that 29% of adults in the US have experienced depression at some point in their life, and that 17.8% are currently experiencing the condition. Despite its high prevalence, depression screening is relatively rare in the outpatient setting, occurring in less than 4% of primary care encounters according to some estimates. There is thus a clear need to improve primary care screening for the condition.
In the current study, researchers examined the efficacy of machine learning technology to detect and analyze voice biomarkers of moderate and severe depression. To do so, they trained an AI model with 10, 442 unique voice samples from adults in the US and Canada, and validated it with an additional 4, 456.
Participants recorded a voice response to the prompt “How was your day?” in English. Voice clips ranged from 25 to 74.9 seconds in length, with a median length of 57.9 seconds. The researchers compared AI analyses of voice data with patients' results from a self-reported health questionnaire for depression.
Ultimately, the tool correctly identified depression in 71% of people with the condition. It also correctly ruled out 74% of people who did not have the condition. The researchers noted that the tool exhibited less precision for men relative to the full population due to the smaller population of men in their sample.
“Machine learning has potential utility in helping clinicians screen patients for moderate to severe depression. Further research is needed to measure the effectiveness of machine learning vocal detection and analysis technology in clinical deployment,” concluded the researchers in their paper.
Sources: EurekAlert, The Annals of Family Medicine