Depression is a leading mental disorder impacting about 16 million Americans. According to the World Health Organization, the annual global economic impact of depression is estimated at $1 trillion and is projected to be the leading cause of disability by 2020.
As researchers aim to better predict, diagnose and treat depression, artificial intelligence is being explored as a potential solution.
Some of the questions that need answering to better understand the role of artificial intelligence in efforts to diagnose and treat depression:
- What types of AI applications are currently in use to manage depression?
- How has the market responded to these AI applications?
- Are there any common trends among these innovation efforts – and how could these trends possibly contribute to reducing the rates of people living with depression?
Depression AI Applications Overview
The majority of AI use-cases for managing depression appear to fall into three major categories:
- Virtual Counseling: Companies are developing software using machine learning to recognize episodes of depression and to provide support using natural language processing.
- Patient Monitoring: Machine learning is employed to monitor patients and to predict and prevent the onset of a mental health crisis.
- Precision Therapy: Firms are using machine learning analytics to track and correlate cognitive function, clinical symptoms and brain activity.
Virtual Counseling Using Natural Language Processing
San Francisco-based startup Woebot said its chatbot uses machine learning and natural language processing to help users manage their mood and mitigate depression.
Accessed through the Facebook Messenger platform, Woebot prompts users with questions to assess their mood. Over time, the algorithm, which is trained on cognitive behavior therapy (CBT) methods, learns the emotional profile of each user and recommends activities to help maintain a more balanced mood.
The early-stage startup bases its clinical significance on the results of a clinical trial conducted in partnership with Stanford University. The study demonstrated that Woebot users aged 18-28 experienced “significant reductions in anxiety and depression” compared with the control group that used an e-book published by the National Institutes of Health (NIH).
Woebot was used every day or almost daily by 85 percent of participants for two weeks. Effectiveness was measured using a standard patient health questionnaire for depression called PHQ-9 with scores ranging from 0 (no symptoms) to more than 20 for severe depression.
The chatbot hit 50,000 users in its first week and receives roughly 1 million messages each week and has secured Series A funding. However, the amount of funding is not specified. Woebot, which was launched in June 2017, is currently available for free but a paid version may be in the works as the startup moves toward a sustainable business model.
Wysa, a rival of Woebot, said it uses machine learning algorithms to learn user emotions and recommend interventions to help users maintain an emotional balance. For example, using natural language processing, the app responds to texts and suggests interventions including cognitive-behavioural techniques (CBT), meditation and breathing techniques.
In the demo below, Wysa is shown communicating with a user, demonstrating how it tracks activity and monitors progress.
Currently, Wysa has more than 200,000 users from 30 countries, the company said. However, no clinical data or case studies are available on the website. Wysa is offered for free, but access to human Wysa coaches (mental health professionals) costs $29 per month.
Patient Monitoring – AI and Predictive Analytics
Emerging from Johns Hopkins Technology Ventures (an incubator for startups), Baltimore-based Sunrise Health said it uses AI and predictive analytics to monitor patient activity and prevent the onset of a mental health crisis using support group texting (five to seven users per group).
The algorithms that are integrated in the app are trained with data generated from user texts. Predictive analytics is used to recognize patterns in patient activity, measure progress and to identify changes that may indicate a potential issue. Human moderators, such as therapists, are immediately alerted if the system predicts that an issue with a patient could occur.
The two-minute demo below offers a walkthrough of the app and provides a step-by-step explanation of the features available.
Beyond anonymous support group texting, users are also provided with the option of voice calls or face-to-face meetings with a therapist. Case studies on how the startup is reducing instances of depression are not yet available.
Founded in 2011, California-based startup Ginger.io uses machine learning to provide users with mental health support. Algorithms were trained using data from over 1 million consumers and through partnerships with over 40 healthcare organizations, the company said.
In contrast to similar applications, Ginger.io appears to provide users with access to a team of human mental health professionals and machine learning is used to learn from patient data and to align care with patient goals and objectives. Members of the care team include emotional support coaches, licensed therapists and board-certified psychiatrists.
In the interview below, members of the startup team discuss the strategy behind the app and how it may help fill gaps in traditional models of mental health therapy.
The company has so far raised $28.2 million and lead investors include Kaiser Permanente Investors and Khosla Ventures. Subscription packages range from $129-$349 per month. Case studies on the impact of Ginger.io on patients and healthcare systems are not currently available on the company’s website.
Precision Therapy with Machine Learning
Founded in 2014, California-based startup Mindstrong Health said it uses machine learning to help diagnose and treat behavioral health disorders by interpreting data generated from using smartphone technology.
The company’s method of data mining is primarily focused on “digital phenotyping.” The term was introduced by Harvard researchers in 2016 and is defined as a method of quantifying individual characteristics by analyzing data generated from an individual’s use of smartphones and other personal digital devices.
Mindstrong has trained its machine learning algorithms on an equivalent of 200 person-years of cognitive data – the combined measurement of individuals and their time contribution – from three clinical studies, the company said. Through analytics, researchers interpret the data and suggest correlations between specific digital activities and brain activity.
For example, in a clinical trial funded by the NIH and conducted in partnership with Stanford University, researchers said they aim to “define signals correlated with cognition, brain imaging and mood in patients with depression.”
Mindstrong appears to be heavily focused on research and proving the viability of digital phenotyping as evidenced by its series of clinical trials currently in progress. To date, the company has reportedly raised $14 million.
Potential for AI to Change Mental Health Care
According to the Centers for Disease Control and Prevention (CDC), the economic burden of depression is equivalent to 200 million lost workdays or up to $44 billion in annual costs to employers. Therefore, the potential for AI to deliver value for healthcare and the economy in general is promising.
AI applications in the mental health space are appearing to rely heavily on mobile technology, specifically smartphones. While this is quite normal for apps, there seems to be a greater emphasis on using the smartphone as an actual data source and not just as the mode of delivery as in the example of digital phenotyping.
Digital phenotyping appears to be gaining some traction in the research space both within and beyond the mental health sector. This growing trend may also lay the groundwork for more AI and IoT applications in the future.
As the majority of companies are still in the early stages, clinical data on the impact of these applications will be critical to making the case for their use. There are also many factors that should be considered such as:
- How long will a user need to use this technology?
- How does the duration of use impact the business model?
- Is there a risk of dependency and could it outweigh the benefits?
It’s clear that trials and tests will need to be conducted to understand the long-term impact of these applications.
Within the next 5 years, I see essentially no hope for applications that aim to replace the calibrated care of a mental health professional. The role of companion or confidante seems vastly out of reach for machines. The idea of today’s ‘chatbots’ helping to talk someone through a panic attack or challenging life circumstance seems absurd and irresponsible.
That being said, applications like that facilitate patient-therapist conversations and learn from trends with natural language processing (see Ginger.io) or learn from patient usage data to identify patterns correlated with depression – these seem both viable and potentially useful (for diagnostic and treatment purposes) in the near term.
Time will tell how the public responds the these applications in the years ahead – and how well they impact the wellbeing of users.
Header image credit: KQED