How Artificial Intelligence Technologies Can Augment Mental Health Treatment
By Kal Patel, MD, MBA, CEO and Co-Founder, BrightInsight
The rise of technologies such as machine learning and artificial intelligence (AI) is spurring great interest in how they might be applied to healthcare, especially in the wake of the advent of generative AI applications such as ChatGPT.
One interesting use case is mental health. Recent research on the use of digital health technologies has shown that they can have a powerful and lasting effect on mental health conditions, which are a leading comorbidity for many physical diseases.
"A pervasive loneliness epidemic is spreading across the US,” says Dr. Nimita Limaye, Research VP, Life Sciences R&D Strategy and Technology. “According to the AHA, loneliness can increase the risk of heart disease, stroke, or death by about 30%. It is a silent killer, significantly impacting mental health. In a world seeking empathy, conversational AI chatbots powered by Gen AI can drive deeply personalized conversations and make a huge difference. Thirty percent of the healthcare and life sciences industries see conversational AI chatbots as their most promising Gen AI use case, as noted in IDC’s Future Enterprise Resiliency & Spending Survey Wave 6, July 2023.”
But what are we talking about when we talk about AI? Here’s a look at the distinguishing characteristics of these technologies.
Machine Learning vs. AI vs. Large Language Models: Defining the Terms
Terms like machine learning, AI, generative AI, and large language models are often used interchangeably (and incorrectly). With machine learning, which is a subset of AI, the program is told, “Each time input ‘X’ is received, create output ‘Y.’” The technology also allows for exponentially more inputs than a simple algorithm, making it ideal for incredibly complex tasks.
AI comes into play when multiple machine learning programs are combined, creating a network capable of gathering a huge variety of specified inputs.
Large language models, a type of generative AI, take it a step farther, expanding the scope of inputs beyond a specific task, to include enormous bodies of information from across the internet, vast libraries of literature, and reams of programming and code.
In short, machine learning and AI are specific to certain tasks, and are bound by a finite set of inputs, while the potential inputs for large language models are virtually unlimited.
Mental Health and AI
Mental health disease is the most common comorbidity for almost every physical malady, with a global economic burden of roughly $5 trillion – to say nothing of the suffering that figure implies.
Interestingly, many therapeutic modalities – for example, cognitive behavioral therapy (CBT), dialectical behavioral therapy (DBT), or interpersonal therapy (IPT) – lend themselves well to the algorithmic processes of machine learning, in terms of methodically obtaining a desired output for a given input, enhancing and improving the standard of care.
Woebot Health, a startup launched in 2017, has built a mental health app called Woebot that offers reliable and compassionate support to help reduce symptoms of stress, depression and anxiety. Woebot is designed by humans but powered by AI, and can integrate with health systems and payors to provide evidenced-based behavioral health solutions for adults, adolescents and new mothers. Woebot has been shown to establish a lasting working alliance with users comparable to the bond formed between humans.
How Woebot Works
At Woebot Health, the focus is on building empathetic relationships with users, before bringing them to a therapeutic exercise; studies have shown that this approach delivers better outcomes than diving straight into treatment, even when using the same digital tool. AI allows Woebot to create that empathetic environment with great specificity to both patient and situation. As a rules-based conversational agent, Woebot does not generate its own sentences, but rather selects an appropriate evidence-based response crafted by the company’s team of writers and clinicians.
Say, for example, that a patient has had a difficult argument with a family member. Rather than delivering a formulaic output – “That sounds like a relationship problem” – Woebot responds with true empathy: “Family situations can be hard – especially at this time of year.”
That lets the patient know that they’ve been heard and cared for, building motivation for interacting with the actual therapeutic tool, which in turn is structured to lead patients through a CBT-based exercise. For example, the tool could help a patient reframe a negative thought pattern, guided by insights gleaned from prior interactions.
Another important aspect involves accessibility. In addition to being available around the clock (some 77% of interactions with Woebot occur outside of traditional clinical office hours), digital mental health applications allow patients to feel comfortable revealing things they wouldn’t ever tell another person.
The Evidence: Clinical Trials
The Woebot approach has been validated in numerous studies, including a bond study of over 36,000 users which examined whether patients using the tool would report therapeutic bond levels similar to those found in other CBT modalities, including face-to-face therapy, group therapy and other digital interventions that don’t use a conversational agent.
The cross-sectional, retrospective study showed that Woebot users established working alliance scores comparable to traditional, outpatient CBT – both individual and group – within three to five days, compared with two to six weeks for traditional therapeutic modalities. This suggests that forming bonds is not the exclusive domain of human therapeutic relationships.
Another trial, the first of its kind, is under way to explore user satisfaction with a version of Woebot infused with LLM technology, a form of generative AI. The primary endpoint is user satisfaction at two weeks, as measured by a questionnaire, with a secondary endpoint of user satisfaction among a sub-sample of patients with at least mild symptoms of depression and/or anxiety. Results from this study will help Woebot Health understand how LLMs can be applied to accelerate the delivery of safe, engaging and potent digital solutions.
Woebot and the BrightInsight Platform
BrightInsight’s clients address a number of physical disease states for which mental health is one of the most prevalent comorbidities. Working with the BrightInsight Platform gives Woebot Health the opportunity to introduce conversational AI tools for those patient populations to improve quality of life, aiding them through and then off of a therapy as their condition improves.
Woebot is an ideal tool to augment healthcare outcomes by establishing a sympathetic and understanding relationship as a prelude to a procedure or treatment. As an example, consider needle phobia, which affects virtually all medical procedures and an estimated 63.2% of patients, leading nearly half to avoid blood draws and donating blood, and one-third to avoid vaccinations.
Woebot is also a prime example of how AI-based technologies like machine learning and large language models are being harnessed to improve quality of life and outcomes for some of the most difficult-to-treat conditions.
Kal Patel is the CEO and Co-Founder of BrightInsight, the leading global regulated Digital Health platform for biopharma and medical device companies.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.