A new study out of UCLA has determined that an AI-driven interactive voice application can be an effective way to monitor the wellbeing of patients being treated for serious mental illness.

For the study, UCLA researchers monitored 47 patients – all of whom were being treated by physicians for serious mental illnesses, including bipolar disorder, schizophrenia and major depressive disorder – for up to 14 months using an application called MyCoachConnect.

For the study, published recently in PLOS ONE, participants called a toll-free number one or two times a week and answered three open-ended questions when prompted by a computer-generated voice. The questions included: How have you been over the past few days?; What's been troubling or challenging over the past few days?; and What's been particularly good or positive?

The MyCoachConnect app was designed to collect personalized patient responses, explained Dr Armen Arevian, director of the Innovation Lab at UCLA’s Jane and Terry Semel Institute for Neuroscience and Human Behavior. More specifically, the AI was trained to use an individual’s own words to provide researchers with a personalized analysis for each patient, with the application focusing primarily on the choice of words the patients used in their responses, how their responses changed over time, with a smaller emphasis on audio features like tone of voice.

“Language provides contextual information relating to an individual’s life experiences and is sensitive to underlying neuro-psychiatric states,” the report explained. “For example, increased concrete word use in delirious states and reduced word fluency after sleep disruption has been observed. In the mental health domain, use of specific words including negative or positive emotion and first-person singular words have been associated with depressive states and exposure to traumatic events.”

The analysis of the data, conducted in collaboration with researchers from USC's Signal Analysis and Interpretation Laboratory (SAIL), found that the application's analysis was as accurate at monitoring patients' mental states as their treating physicians.

According to the report, “(t)hese results may have important clinical implications. . . Because the personalized model requires a period of individual-level training to achieve this degree of correlation, it may . . . inform how care is delivered. For example, health organizations may recommend more frequent assessments for individuals with (Serious Mental Illness) early in their care to set this baseline and to train their personalized model. This could then be followed by routine in-person assessments paired with remote interactive voice response engagement and monitoring.”

MyCoachConnect was developed and hosted on the Chorus platform, which was developed by Arevian at UCLA and allows people to visually create mobile and other computer applications without computer programming in as little as a few minutes.