Behavioral Health Could See Major Benefits from Voice Tech and AI, But Skeptics Wonder At What Cost

Yared Alemu faced high stakes when treating 4,000 low-income youth at a community health clinic in Atlanta.

In his mind, getting care right could mean helping patients stay in school. Not getting it right left open the possibility of entanglements with the law.

He also knew that he was fighting blind without an objective way to measure the trauma causing patients problems. He turned to speech emotion recognition (SER), a form of voice tech and artificial intelligence that seeks to measure the emotional content of speech regardless of what is said. 

“Trauma is primarily a physiological experience,” Alemu told Behavioral Health Business. “We’re developing these biomarkers from speech to identify the severity of mental health distress related to trauma. … We are able to leverage and develop these algorithms, detecting these problems objectively.”

Alemu and others look to advances in voice-based technologies and AI to solve deep and varied issues in behavioral health, such as lessening the administrative burdens of a limited workforce or bringing genuine objectivity to a highly subjective field. But these nascent tools bring a number of privacy and ethical questions with them as well.

Voice tech and objectivity

Several startups, including TQIntelligence, use different forms of speech recognition and AI technology.

Alemu left his job at the clinic to join a business accelerator at the Georgia Institute of Technology (Georgia Tech) and founded the startup TQIntelligence in 2016 to develop and commercialize pediatric trauma-focused SER.

“I literally walked out of my job and went to Georgia Tech as a first-time founder — I have no background in technology and no background when it comes to finance,” Alemu said. “But that frustration had been brewing for quite a while.”

Alemu contrasted SER with other forms of AI-backed voice tech that the company doesn’t use such as sentiment analysis, which focuses on what a person means, and natural language processing, which focuses on capturing literally the words a person says.

SER and voice biomarkers are emerging science. TQIntelligence is developing a voice-based biomarker algorithm and a clinical voice sample database and repository that represents marginalized youth — specifically African Americans, Latinos and Caucasian individuals within rural communities. The National Science Foundation has awarded TQIntelligence about $1.2 million in grants to develop these features beginning in 2019. 

When fully realized, the TQIntelligence platform, called Clarity AI, can use a 60-second voice sample to almost instantly determine the level of emotional distress of a child. In turn, clinicians would be able to measure progress over time and also pick up on previously undiagnosed issues.

“We’re in a bind as mental health providers,” Alemu said. “We ask the person that is really struggling and not knowing what the hell is going on to provide us with an insight so we can make some consequential decisions.”

This conundrum is even worse in children, who may be developmentally unable or unwilling to accurately describe their inner state or other issues.

“Children even under the best circumstances — and I have a couple of them at home — are clueless,” Alemu said.

Regardless of a child’s mental condition, many come from and are expressions of trauma, which TQIntelligence helps clinicians assess and address.

Low-income and Black children are more likely to experience traumatic events: 48% of white children, compared to 64% of Black children, have experienced one or more traumatic events. 

Bringing greater objectivity to behavioral health is an undergirding aspect of the rise of value-based care in the industry.

Beacon Health Options, a major managed behavioral health organization owned by Anthem Inc. (NYSE: ANTM), aims to use voice tech analysis and transcription of therapists’ sessions as a way to measure the impact of certain interventions and track new potential measures for value-based care models. 

Automating transcription and analysis is possible to do at scale with present technology and could be the basis for also gleaning insights into the skills of specific clinicians, Beacon Health Chief Medical Officer Neil Leibowitz said at Behavioral Health Business’ event VALUE.

Finding more time

Voice technologies and AI have applications as efficiency tools as well.

Many in the behavioral health space are caught between major trends that make maximizing clinician time a priority. These include increases in demand for services, a lack of reimbursement parity, and low rates and frustrations with administrative burdens often associated with taking insurance.  

In a poll by industry group the National Council on Mental Wellbeing (NCMW) and Morningstar, nearly all (98%) respondents said it was difficult to hire staff. Most (62%) said they were seeing waitlists grow, with 78% reporting increases in demand.

“When asked what short-term policy would help organizations meet increased demand for services, members reference additional funding, continuing funding for telehealth services, updating reimbursement rates, and reducing paperwork requirements,” a report based on the poll states. 

Psychiatrists spend as much as 20% of their time on administrative tasks, according to a 2014 study.

Eleos Health Inc., a voice-based care operations automation startup focused on behavioral health, allows clinicians to use a form of AI called natural language understanding during their sessions. The company was founded in March 2020 and is headquartered in Boston and Tel Aviv, Israel.

The company’s prime sales pitch is that using its voice tech tools can reduce time spent on administrative tasks by as much as 40%. Assuming that a clinician spends 15 minutes on notes for each session and holds a weekly workload of 40 to 50 patients, using Eleos could translate into freeing up four to five hours, according to a spokesperson for the company. 

“In behavioral health, the conversation is the treatment in a lot of ways,” Eleos Health CEO and co-founder Alon Joffe told BHB. “Despite that fact, we don’t really know what’s working in these conversations, and we are not leveraging this data to automate a lot of the workflow processes that clinicians need to do.”

Eleos Health helps clinicians in three ways, according to Joffe: generating session intelligence by quantifying the interventions used and topics discussed; automating session documentation by producing 300- to 400-word session summaries; and compliance automation that also suggests billing codes.

Joffe said that his product could also be an asset in talent attraction and retention.

“Providers are trying to really hone in [on] how can they be the best place to work for the clinicians,” Joffe said. “Documentation is definitely something that clinicians don’t love. If you can provide them with tools that help them focus on care, it’s a big asset.”

Suki AI Inc. offers a voice-based assistant that focuses on allowing clinicians to interface with their other devices in a more natural and efficient way. Across all specialties, Suki AI provides an average time savings of 72% during documentation.

One pilot with the American Academy of Family Physicians found that clinicians using Suki AI saw documentation time decrease 62% during clinical hours and 76% during after-hours. The report also found users reported major shifts in feelings of burnout being more manageable. 

While Suki AI doesn’t use SER or other technology to pull deeper meaning from voices, the product is able to understand the intent of the voice commands it is given so clinicians can speak naturally and accomplish administrative tasks. 

“This is obviously the extreme end of positive examples. … We’ve had folks [say] they went out on a date with [their] spouse for the first time in years because I didn’t have all this stuff hanging over my head when I got home,” Dr. Erin Palm, vice president of product management at Suki AI, told BHB, adding that they have helped older clinicians who felt burdened by using technology practice longer and new mothers continue practicing part-time while raising their children.

The worry

Recording the sessions of behavioral health clinicians worries Carrie Singer. She is a licensed clinical psychologist, the owner of a Rockville, Maryland-based group practice called Quince Orchard Psychotherapy, and the founder TheraHub, a tool that helps connect patients to in-network therapists who use telehealth platforms.

Privacy concerns top her list. While proponents of voice-based tools say that patient consent and opt-in practices are vital for their protection, Singer says clinicians don’t have that option if a payer makes the practice a requirement of reimbursement. Similarly, patients may be pressured against their judgment to participate lest insurance not cover the service.

“If a health insurance company says, ‘We need you to sign this form,’ you’re probably going to feel some pressure to do that. But is that really in your best interest?” Singer said.

Voice tech and AI tools could be helpful in helping train clinicians, and Singer knows of university programs that use them to educate prospective therapists.

But using tech in clinical settings puts “performative pressure” on clinicians that could limit their ability to practice according to their preferred style and clinical judgment. It also assumes that all clinicians are up on and proficient in skills that may not already be part of a clinician’s formal training.

Attempts to simplify allowed therapies and interventions based exclusively on outcomes and session data severely discount the role of human connection and intuition, in Singer’s estimation.“Why even bother having a therapist” she said. “Why don’t you program a chatbot to deliver that service?”

Behavioral health is inherently unlike physical care where symptoms can be addressed by manualized decision trees; one symptom may have different causes across patients. It also has to deal with issues where no specific medical intervention exists or could be conceived from looking at session data.

“I was once working with a little girl who was afraid to go to school because she was afraid she was going to choke on something — because she had choked on a hot dog at a party,” Singer said. “How the heck do you program a manualized treatment for that?”