A recent study out of Dartmouth found that the use of a generative artificial intelligence mental health chatbot decreased symptoms of depression, anxiety and eating disorders in the first randomized controlled trial of its kind.
Thestudy demonstrates potential for the gen AI chatbots as a safe therapeutic intervention for mental health conditions, the study authors said. It also shows promise to enhance digital therapeutics, which leverage software to provide treatment but have struggled with patient engagement.
The Dartmouth study contained 210 participants, and 106 people with clinically significant symptoms of major depressive disorder, generalized anxiety disorder or at high risk for feeding and eating disorders received access to Therabot, a mental health AI chatbot developed by clinicians at Dartmouth in 2019.
Users of Therabot can initiate a session in the chat interface or respond to scheduled notifications. The chatbot provides targeted interventions, question prompting, empathetic responses and supportive affirmations to address the user's needs. The continuous interaction by the chatbot leads to a deeper therapeutic experience, the company says.
Intervention arm participants received prompts to interact with Therabot every day for 30 days. Then, they could access Therabot for an additional four weeks, but were not prompted to use the chatbot. Participants completed testing pre-intervention and at the four- and eight-week marks.
While highly engaging for users, gen AI chatbots have not been studied for delivering clinical intervention, the study authors said.
Related
AI can crack open the 'black box' of effective mental health therapy at scale, study finds
“To date, conversational agents using Gen-AI have fallen under general purpose, wellness, or companion applications, rather than software intended for the diagnosis and treatment of mental health disorders,” the study says.
The Dartmouth researchers found that Therabot users saw a clinically significant decrease in symptoms for their conditions. Adherence to the application was strong: On average, participants used Therabot during 24 days of the month and engaged with the chatbot for 6.18 hours. Participants sent 260 messages on average.
Users with major depressive disorder experienced a 51% reduction in symptoms at the four- and eight-week marks. For generalized anxiety disorder, participants experienced a 31% decrease in symptoms. Users with eating disorders saw a 19% decrease in symptoms. None of the participants were taking medication for the disorders at the time.
Related
AI in action: Enhancing suicide risk detection in behavioral health
The researchers also studied therapeutic alliance, engagement and satisfaction. Participants said Therabot was easy to learn to use and intuitive, and they found interactions helpful and felt better after interactions.
Participants also reported therapeutic alliance similar to that of a human therapist. The study says therapeutic alliance, empathy and shared goals are a part of psychosocial therapies, which are difficult or impossible to emulate with non-generative AI technologies and digital health technologies to date have lacked.
Clinicians involved in the study monitored the responses of Therabot for safety and appropriateness. In 15 cases, clinicians had to contact participants for safety concerns like suicidal ideation and had to reach out to participants 13 times to correct Therabot.
With gen AI, patients can form a stronger therapeutic relationship similar to that of a human therapist, which can deliver more personalized treatment and the potential to scale broadly, the study found. Some rule-based AI chatbots have demonstrated clinical effectiveness to date, though they’ve used a more simplistic form of AI that limits the outputs to those preset by programmers.
If more widely validated, mental health chatbots could help relieve the waitlist backlogs of behavioral health providers around the country.
“Although empirically validated psychosocial treatments exist, they are resource intensive, and limited in scalability and accessibility, leading to fewer than half of the people with a mental health disorder receiving care,” the study says.
The control arm participants did not receive access to the app until the study concluded.
