How Big Data Plays A Growing Role In Mental Health Services

The data generated by smartphone users is already being analyzed by marketers. Now, psychologists and other social scientists are using similar, if not identical data to provide guidance and treatment.

As the line between online selves and "reality" continues to shrink (as it were), the behaviors revealed by search, social media commentary and related forms of digital expression become ever more useful in assessing and predicting individuals' behavioral patterns. This could enhance the delivery of care which traditional methods are unable to do because of the data available from technology. The privacy concerns revealed by marketers' and tech companies' abuses of that information may be even more glaring in the mental health context. But just as they provide useful, even crucial, insights to business, so they do for medical professionals - and in this case help heal some of the damage done by those selfsame technologies. JL 

Daniel Barron reports in the Wall Street Journal:

Our search and social media history represents a trail of thoughts and emotions. Studies have shown the words we use to express ourselves on Facebook and Twitter can predict the conditions like postpartum depression and psychosis. A person’s recent Google search history is a better predictor of suicide than clinician’s most recent notes. Elements of speech such as coherence and the frequency of possessive pronouns (his, her, my, mine) can predict, with an accuracy of 83%, whether someone at risk for psychosis will become psychotic. The way molecular sensors reveal genetic abnormalities in oncology, digital sensors reveal behavioral patterns.

As a physician, I need to figure out three things when a new patient walks into my office: what their life is typically like, what has changed that made them seek treatment and what I can do to help them. It’s a complex problem, and most fields of medicine approach it by taking measurements. If I were a cardiologist evaluating a patient’s chest pain, for instance, I would speak with the patient, but then I would listen to their heart and measure their pulse and blood pressure. I might order an electrocardiogram or a cardiac stress test, tools that weren’t available a century ago.

Because I’m a psychiatrist, however, I evaluate patients in precisely the same way that my predecessors did in 1920: I ask them to tell me what’s wrong, and while they’re talking I carefully observe their speech and behavior. But psychiatry has remained largely immune to measurement. At no point in the examination do I gather numerical data about the patient’s life or behavior, even though tools for taking such measurements already exist. In fact, you likely are carrying one around in your pocket right now.

Our search and social media history represents a permanent breadcrumb trail of our thoughts and emotions.

In the last decade, an entire industry has been built to predict a person’s behavior based on their smartphone use and online activity. Because our search and social media history is digitized and time stamped, it represents a permanent breadcrumb trail of our thoughts and emotions. Tech companies and governments already use these data to monitor and commodify our likes and dislikes; soon psychiatrists might be able to use them to measure and evaluate our mental state.

Our smartphones measure our movements with accelerometers, our location with GPS and our social engagement with the number of calls and texts we send. These data have extraordinary potential for psychiatric diagnosis and treatment. Studies have shown that the words we use to express ourselves on Facebook and Twitter can predict the emergence of conditions like postpartum depression and psychosis. A person’s recent Google search history, it turns out, is a better predictor of suicide than their clinician’s most recent notes.

Digital tools could also help psychiatrists measure a patient’s behavior during a session. Each visit to a therapist creates a wealth of clinical data that is currently wasted because it’s not recorded or analyzed. Speech and facial recognition technologies could be used to precisely measure a patient’s expression, the words they use and the intonation of their voice. Such tools could be used to recognize the subtle changes that occur when a patient is about to become floridly manic, or analyze how they respond to treatment. A recent study by Cheryl Corcoran published in the journal World Psychiatry showed that elements of speech such as coherence and the frequency of possessive pronouns (words like his, her, my or mine) can predict, with an accuracy of 83%, whether someone at risk for psychosis will actually become psychotic. Such data is created at every clinical encounter, but it is far too subtle for a doctor to detect.

The ability to interact digitally with a patient could also smooth the transition from hospital to outpatient care. Studies have shown that up to 80% of patients do not remain in mental health treatment after they leave the hospital. This underscores a fundamental problem with the traditional model of care: I only know how a patient is doing if they show up for their next appointment, or call or email me. App-based therapies and passive behavioral measurement would allow doctors to better connect with patients after they leave, helping to reduce attrition.

Quantifying behavior by collecting large amounts of data could also help psychiatrists discover insights that aren’t intuitive. A century ago, it wasn’t at all obvious that treating hypertension could prevent heart attacks. Today, data might show that we should treat a symptom such as decreased facial expressivity to prevent suicide. In the same way that molecular sensors revealed genetic abnormalities and treatment targets in oncology, digital sensors might reveal behavioral patterns and lead to new interventions. Patients in traditional talk therapy may benefit from a big data approach to psychiatry, but it will likely prove most helpful in the diagnosis and treatment of severely ill patients.

Companies like Google, Facebook, Instagram and Twitter already use many of these tools to gain an enormous amount of information about our behavior, but the interests of the tech giants aren’t always aligned with those of the individual user or society at large. Using big data for behavioral healthcare offers a way to return ownership and benefit to the individual.

Two models have already begun to emerge. In one, doctors offer carefully calibrated laboratory tests, sitting down with patients to decide which data might be helpful to collect and why. Patients then decide whether and how much of their data to share, for how long and with which treatment providers. Some academic research groups are already piloting a new kind of medical role, a “digital navigator” who specializes in exploring data collection and sharing with patients.

The other potential approach to big data is a consumer-facing product similar to the genetic testing offered by 23andMe, in which people pay for access to their own data and then decide who to share it with. Private tech companies have already developed online dashboards for processing and viewing behavioral data in real time.

These two strategies will probably need to converge in order to succeed. Private tech companies have an abundance of technology and processing power, but they lack access to the patient data required to train their tools. Academic centers have access to patients but are forced to create in-house dashboards and data collection systems that already exist in the private sector. Bridging the divide between academic and private sectors will require a rethinking of fundamental questions about data ownership, data security and intellectual property.

Before new data technologies are employed and marketed, they must be evaluated as rigorously as other clinical tools, of course. But using big data to make psychiatry more precise and effective has the potential to help all patients.