tl;dr: digital tools can help with intervention, compliance, and measurement of progress, but it will be difficult to build the necessary datasets to power these applications. Measurement is key, and is the primary focus of this post. Feel free to jump around to the headings you find most interesting.
Mental health is a massive problem facing our society. Globally, 322M people are afflicted with depression and 264M with anxiety). In the U.S., 1 in 5 adults experienced mental illness, over half of whom did not receive treatment.
COVID-19 has added fuel to the fire. During the pandemic, 40% of U.S. adults (n=5470) have been struggling with mental health or substance abuse, and 25.5% of young adults aged 18-24 have seriously considered suicide in the past 30 days (ref).
The root causes of mental health are complex: as with most diseases, it's a combination of genetics and environment. During COVID, common environmental factors include prolonged social isolation, stress, unemployment, stress about childcare, lack of structure for children, and many others. We've all felt down and stressed this year, and it's sadly unsurprising seeing the toll this has taken on society as a whole.
COVID has made delivery of care more difficult as well. Two-thirds of countries in the WHO reported disruptions to counseling and psychotherapy. The US has managed to adapt relatively swiftly, with delivery of remote health care increasing from from 7% to 85%, but many people aren't aware that these services exist or that they should be using them, and (as with the whole healthcare industry) these services are reactive rather than proactive; you’ll likely be very sick before every seeing a doctor.
On the bright side, this shift to remote delivery provides a great opportunity for technology to help. It can connect us with help as soon as it's needed. It can be used to gather, process, and find patterns in data better than any human. It's always with us, and can help guide us towards better, healthier behaviors. And at the limit, it can do all of this at an extremely low cost, so the benefits will be available to all.
This post will focus on depression and anxiety, the two most prevalent mental health issues. We'll discuss what detection and treatment look like today, and highlight how technology might be able to impact each area.
Screening tools tell you if something might be wrong; either you're likely okay, or there's something worth investigating in greater detail. This step is often done with a questionnaire at the doctor's office, for example the PHQ-2 survey. With so many cases of undiagnosed mental health, an important goal is to have as many people screened as possible. Emailing everyone a survey is not feasible. Increasing awareness, decreasing stigma, and passively incorporating screening tools into our daily life could help.
The next step is to provide a diagnosis. It must be high accuracy and is currently done by a highly trained clinician. The status quo involved gathering evidence from lab tests, a physical exam, and an interview, followed by a clinician synthesizing the information and inferring a diagnosis. Depression and other mental health disorders are diagnosed based on the symptoms found in DSM-5, including fatigue, diminished interest or pleasure in most activities, observable reduction of physical movement, and others, weighted by both frequency and severity. Once a diagnosis has been determined, it's time to create a care plan to intervene.
Lifestyle interventions are things you can change in your daily life. For mental health management, this includes avoiding alcohol and drugs, staying active, eating well, meditating, spending more time with loved ones, and keeping a gratitude journal. These will not only help your mental state, but help you live a flourishing life. Everyone acknowledges that these are important, but few succeed in making these stick as long-term habits.
Talk therapy is currently our best way of rewiring our brain circuits to break the old, bad patterns and create new ones. This means sitting down and talking with a trained expert. Depending on the specifics of the condition, this can include cognitive behavioral therapy (CBT), dialectic behavior therapy, or others.
This is an area in which technology can help. A given treatment course can be six to 24 one-hour sessions, which can incur a bill of $300 - $3000 in total (before insurance kicks in). Not only is it possible for natural language processing to deliver effective CBT sessions (see Woebot as one example), but people report being more comfortable opening up to a bot than a human.
From Science, The Synthetic Therapist
Medications is the third major component. We all want to rely on willpower and lifestyle changes as much as possible, but when there’s something out of balance at the chemical or biological level, drugs are a critical component of care.
Gene therapy isn't a viable treatment for mental health today, but there is evidence of a genetic basis for mental health. As CRISPR and related technologies mature, they might be a viable method to cure mental health issues before they even begin.
It's only through repeated action that we can change ourselves and improve. Even the most cutting-edge care plan is useless if the person receiving the care doesn't comply. This is a massive weak spot in our existing medical system - at the end of a visit your doctor will tell you to eat better and exercise more, and send you on your way. How many actually do, even if it's a matter of life or death? One JAMA article found that 1 in 4 men don't make any lifestyle changes after a heart attack or stroke (interestingly, the rate for women is closer to 1 in 13). Even something as simple as taking one's medications is a huge problem.
"Nonadherence to medical recommendations is a leading cause of morbidity and mortality in a wide array of disease processes in all age groups." [Monitoring Drug Adherence]
There are many levers available to improve compliance with a given care plan. A person can be motivated intrinsically; anyone can change their lives if they have the right "why", a good recipe for developing new habits, enough willpower, and a community that supports and encourages them. There are also extrinsic motivators, like social accountability or even monetary incentives.
For mental health treatment, accountability and support often come in the form of team-based care, which can be comprised of a primary care doctor, a psychiatrist, a therapist, a pharmacist, a social worker, and family member(s).
Compliance is an area that technology can add value. For example, software can help:
Provide tools to create and track habits. Maybe we can learn from Instagram and other popular apps how to get people engaged in their health?
Provide educational resources and references for care plan
Connect with a community
Provide coaching and easy check-ins with your care team for accountability, encouragement, and help
The existing medical industry hasn't addressed this need, because creating fast, intuitive software is hard and expensive, and the people with the skills to do so are generally not working at big healthcare companies.
If you're training for a marathon, it's very motivating to look at your mileage and pace improving over time. You're putting in effort and seeing results! A similar pattern exists for tracking progress for weight loss (via a scale), diabetes management (via a continuous glucose monitor), and blood pressure management (via an at-home blood pressure cuff). However, there's nothing that fills this need for mental health. Mood tracking apps and questionnaires are notoriously subjective and high variance, and it's too expensive for a clinician to give each patient a daily or weekly assessment. Ideally this measurement would happen passively, and in real time.
Why is measurement important?
The patient is more engaged and motivated to improve when they can track progress
The clinician can remotely monitor their patient, and frequently update their care as needed
The system as a whole can learn what's working, and improve over time
Mental Health Biomarkers
Unfortunately, there isn't a single, easily measurable biomarker that correlates highly with a person's mental health state. But there is hope; there are many subtle signals that indicate that someone could be depressed:
Patterns in socializing, exercise, and other common activities
Sleep quality and duration
Patterns in movement, speech, facial expression
How they're using their computers and phone
Of course, machine learning seems well suited to extract and classify these complex and subtle signals, but data availability and data quality are major issues.
Show Me The Data
As with many areas of the healthcare industry, the data you'd need to train such a model is exceptionally hard to obtain. Basic behavior data (steps, sleep etc.) is mostly available via Apple Healthkit or app APIs, but outcomes data is locked away unstructured in various EMRs. And in many cases we don't capture or expose other important sources of input data. For example we could get useful signal from typing and mouse usage patterns, how frequently I'm reaching out to friends, how I'm walking, but these aren't tracked and/or surfaced in a way that could easily be fed into a clinician's dashboard, or machine learning algorithm.
There are likely multiple, perhaps dozens, of data sources needed to predict mental health, and each of these require dedicated software to measure and expose. And you'd need to match these with outcome records (ideally for both diagnoses and disease progression). Then, even if you had all of this data, you would need many samples (perhaps 10^5 or more) before you unlock any value. So we have a pretty severe bootstrapping problem. There is no magic bullet, but here are some potential angles of attack:
Target a specific data source, create the model in an offline study, and then sell the point solution, as Mindstrong has done with mobile keyboards.
Try and build a product that provides data tracking for immediate user value (e.g. goal tracking, and providing input for a coach), and then connect that with outcome data, and hope you can identify patterns over time.
Raise a ton of money (VC and/or grants), build the multi-modal tracking tools and do a large-scale study to gather data train the model, and ship it whenever it's ready.
It's possible to go directly to the brain to get "ground truth" data. The problem is that brain sensors are either too expensive (MRI), too invasive (see BrainGate, Neuralink), or too low fidelity (EEG). Google X tried to extract a depression biomarker via EEG, but were unsuccessful. Perhaps with larger datasets, better sensors, and more sophisticated signal processing this will someday be possible. If not, perhaps Neuralink will arrive in the next decade or two, and solve this problem for good.
We're discussing collection and analysis of some of our most sensitive personal data, and it's important to call out potential risks. Security is paramount, and responsible usage is critical. Usage and tracking should be transparent, the individual should be the owner of the data, and they should have the right to delete the stored data if desired. These data shouldn't be used for targeted advertising without our consent, even though doing so would likely be worth billions. And the firms approaching this problem need to put consumer trust as a first priority.
Unfortunately, all attempts to automatically measure and infer mental health status are sub-scale and have big caveats and limitations. But there is a tipping point; whoever provides the first real solution in this area will start a powerful flywheel, obtaining more data and continuing to improve the quality of their algorithms over time.
There are headwinds against us “solving” mental health. The pandemic has hit young people particularly hard, which will have compounding impacts on confidence, careers, and financial security in the years to come. Inequality continues to widen as more jobs are automated and more wealth is concentrated in the tech and financial elite. Technology continues to capture more attention, at the cost of investment in the communities and relationships around us. At a certain level, "solving" mental health requires solving our structural societal issues.
However, we're at the edge of a radical transformation of how mental health is diagnosed and managed. With a massive (and increasing) societal need, some exciting early demonstrations of how technology can start to make care more effective and more accessible, and tailwinds from new developments in sensors, machine learning, better understanding of biology, and increased awareness about the importance of mental health, there is reason to be optimistic about the future.
Someday we'll be proactively alerted by algorithms reading invisible sensors when things are heading in a darker direction, and we'll be given engaging, effective, and inexpensive tools to bring us back to a better mental state. If we do, I'm sure we'll have the confidence and resolve to meet our future challenges head on, and our society will be a much happier place.