In the Journals

Voice-analysis application assesses well-being of patients with serious mental illness

Armen C. Arevian

An interactive voice response system that uses artificial intelligence effectively monitored the well-being of patients with serious mental illness, according to study findings published in PLOS One.

“How patients talk and what words they choose enables us to predict their mental health state and track changes over time,” Armen C. Arevian, MD, PhD, director of the Innovation Lab at the Jane and Terry Semel Institute for Neuroscience and Human Behavior, told Healio Psychiatry. “Mental health care can be more proactive, addressing needs early to help avoid things like hospitalization, homelessness and suicide attempts. This application offers an opportunity to make mental health care personalized as we identified changes in speech over time that were specific to individuals.”

According to Arevian and colleagues, serious mental illnesses include schizophrenia, bipolar disorder and major depression and affect nearly 10 million individuals in the United States. The clinical states of these individuals change over time, and these changes are difficult to assess and result in increased care utilization and disease burden, they noted.

To determine whether features derived from speech can act as a transdiagnostic marker of these clinical states, the researchers collected speech samples from 47 patients from a community-based mental health clinic who were diagnosed with schizophrenia, schizoaffective disorder, bipolar disorder or major depressive disorder. Patients called a toll-free number once or twice a week for 4 months and answered three open-ended questions:

  • How have you been over the past few days?
  • What's been troubling or challenging over the past few days?
  • What's been particularly good or positive?

These responses were then analyzed by an application called MyCoachConnect — an artificial intelligence program that used an individual's words to offer a personalized analysis for each patient. It primarily focused on the choice of words used in patient responses, how responses changed over time and, to a lesser extent, on tone of voice and other audio features. A total of 13 clinic providers reviewed patient responses and provided global assessment ratings.

The researchers reported that the system recorded 1,101 phone calls and 117 hours of speech. They noted that 92% of patients agreed that it was easy to use. Individually trained models demonstrated the highest correlation with provider ratings. Population-level models demonstrated statistically significant correlations with provider global assessment rating, future provider ratings, BASIS-24 summary score, depression sub score and self-harm sub score, as well as with the SF-12 mental health sub score; however, they did not correlate with other BASIS-24 or SF-12 sub scores.

“These approaches offer exciting future directions for scaling and integrating with clinics to transform care to be more personalized and proactive,” Arevian said. – by Joe Gramigna

Disclosures: Arevian reports a financial interest in Insight Health Systems Inc. and Arevian Technologies Inc., as well as a family relationship to Memorial Psychiatric Health Services, which provides psychiatric services for the R.O.A.D.S. Foundation clinic where participants were recruited. Please see the study for all other authors’ relevant financial disclosures.

Armen C. Arevian

An interactive voice response system that uses artificial intelligence effectively monitored the well-being of patients with serious mental illness, according to study findings published in PLOS One.

“How patients talk and what words they choose enables us to predict their mental health state and track changes over time,” Armen C. Arevian, MD, PhD, director of the Innovation Lab at the Jane and Terry Semel Institute for Neuroscience and Human Behavior, told Healio Psychiatry. “Mental health care can be more proactive, addressing needs early to help avoid things like hospitalization, homelessness and suicide attempts. This application offers an opportunity to make mental health care personalized as we identified changes in speech over time that were specific to individuals.”

According to Arevian and colleagues, serious mental illnesses include schizophrenia, bipolar disorder and major depression and affect nearly 10 million individuals in the United States. The clinical states of these individuals change over time, and these changes are difficult to assess and result in increased care utilization and disease burden, they noted.

To determine whether features derived from speech can act as a transdiagnostic marker of these clinical states, the researchers collected speech samples from 47 patients from a community-based mental health clinic who were diagnosed with schizophrenia, schizoaffective disorder, bipolar disorder or major depressive disorder. Patients called a toll-free number once or twice a week for 4 months and answered three open-ended questions:

  • How have you been over the past few days?
  • What's been troubling or challenging over the past few days?
  • What's been particularly good or positive?

These responses were then analyzed by an application called MyCoachConnect — an artificial intelligence program that used an individual's words to offer a personalized analysis for each patient. It primarily focused on the choice of words used in patient responses, how responses changed over time and, to a lesser extent, on tone of voice and other audio features. A total of 13 clinic providers reviewed patient responses and provided global assessment ratings.

The researchers reported that the system recorded 1,101 phone calls and 117 hours of speech. They noted that 92% of patients agreed that it was easy to use. Individually trained models demonstrated the highest correlation with provider ratings. Population-level models demonstrated statistically significant correlations with provider global assessment rating, future provider ratings, BASIS-24 summary score, depression sub score and self-harm sub score, as well as with the SF-12 mental health sub score; however, they did not correlate with other BASIS-24 or SF-12 sub scores.

“These approaches offer exciting future directions for scaling and integrating with clinics to transform care to be more personalized and proactive,” Arevian said. – by Joe Gramigna

Disclosures: Arevian reports a financial interest in Insight Health Systems Inc. and Arevian Technologies Inc., as well as a family relationship to Memorial Psychiatric Health Services, which provides psychiatric services for the R.O.A.D.S. Foundation clinic where participants were recruited. Please see the study for all other authors’ relevant financial disclosures.