CEO John Haskell was inspired to use machine learning to replicate human intuition.
John Haskell remembers that he took “more years than most” to get his undergraduate degree from Stanford University. He struggled to finish his major in urban studies while managing his manic depression. Ultimately he found the right treatment.
But a friend with a substance dependency came close to committing suicide following several stays in rehab and subsequent relapses. A phone call from her mother, who picked up on signs that her daughter was in distress, saved the friend’s life.
The incident stuck with Haskell, and it inspired Triggr Health, an app that uses machine learning to analyze a participant’s smartphone data to predict—and hopefully prevent—the potentially tragic consequences of substance use and addiction.
The friend’s mother listened to her intuition when she noticed that her daughter wasn’t playing Words with Friends, was texting at odd hours, and wasn’t returning phone calls. Such instincts are a human method of data analysis, Haskell says.
“She was processing data, understanding patterns, and recognizing a deviation.” Haskell wanted to replicate that process with an app that uses passive phone activity, such as screen engagement, texting patterns, phone logs, sleep history, and location and that uses machine learning to look for patterns that suggest troubling changes in behavior.
Disrupting the Traditional Approaches
What bothered Haskell most about his friend’s battle with addiction was the shortage of proven treatment options. “It’s basically rehab, abstinence, and 12-step programs,” he observes, and the long-term outcomes are poor. According to the U.S. Surgeon General, of approximately 23 million Americans who are dependent on drugs or alcohol, only about 10% seek treatment. A fraction of those (one-third, according to a study published by Evaluation Review) successfully overcome their addiction. “There’s a ton of potential to reach people earlier and to offer them support and guidance that may not require as much motivation or action” as entering a treatment program, Haskell says.
A complicating factor is that 45% of people with substance abuse disorders also have co-occurring mental health disorders, according to the National Survey on Drug Use and Health. Although mental health apps accounted for nearly a third of the healthcare apps available worldwide in 2015, according to the IMS Institute for Healthcare Informatics, Haskell found none that were specifically aimed at using data to help patients fight substance abuse and addiction.
Worse, most mental health apps do not apply advanced analytics to clients’ data to predict negative outcomes. So Haskell hired a team of engineers and data scientists, most of them ex-Google employees, to build the Triggr Health app.
Rather than take what Haskell terms a “clinical trial approach” to development—working with a small set of individuals at first to prove the app’s efficacy—Triggr Health worked with community programs and other partners to build a robust data set.
The system can predict with 92% accuracy when a participant is likely to disengage from positive recovery behavior (and possibly relapse) within three days.
Haskell says the system can predict with 92% accuracy when a participant is likely to disengage from positive recovery behavior (and possibly relapse) within three days. If the data shows that a participant is at risk, a Triggr Health guide steps in to check in on the individual or alert the client’s designated outside care team.
Triggr Health’s marketing avoids the use of words like addiction, billing itself instead as a support app for reducing drinking and drug use. That’s because it can be difficult to separate alcohol and drug use from underlying mental illnesses or other issues, such as trauma.
For people who are self-medicating with alcohol or drugs, Haskell intends Triggr Health to be an avenue for them to make positive changes without feeling stigmatized. “It’s really difficult to reduce drug use or drinking—or achieve sustainable change—without addressing the mental health side,” Haskell says. Such treatment may be beyond the scope of a 12-step program.
For such individuals, in-person mental health therapy is by far the best treatment, says Haskell, but not everyone is ready for it or has access to it. Haskell thinks that if the app can recognize through behavioral data what a participant might be experiencing without the individual having to call, text, or otherwise reach out for help, Triggr Health can intervene when those struggling with substance use might otherwise isolate themselves.
Built for Trust
When participants sign up for Triggr Health, they are paired with a counselor (called a behavioral change guide) who can help them manage cravings, practice mindfulness, and plan for situations (such as happy hour with colleagues or a family visit) that are likely to make them want to drink or use. The company has trained a team of more than 20 such guides around the United States utilizing a clinical curriculum developed in part by the company’s chief medical officer (a professor of clinical psychiatry at the University of California, San Francisco) in conjunction with advisors from the University of Pennsylvania, Northwestern University, and Dartmouth College.
These guides have the data they need to make effective, timely interventions. “We’re collecting participants’ phone data on the back end so we understand how they’re doing over time, but we combine that with 24×7 nonjudgmental support,” says Haskell. “It’s critical that the guide builds trust with the participant on the good days and the bad days.” That way, when the system alerts the guide that something may be going sideways, the participant is more likely to respond to the resulting intervention.
The guides may communicate with some participants daily, if that’s what the individual’s goals or needs require. (Other participants may spend more time with self-guided tools to manage cravings and plan for challenging situations.) Each guide today can be paired with as many as 500 participants, but Haskell says the predictive data, which alerts guides to interact with participants when needed, makes the model scalable. Haskell estimates that one guide could handle thousands of participants.
The Cost of Prevention
Triggr Health is not cheap by app standards; it costs about US$2 a day to use. Sometimes individuals pay that directly. In other cases, a health plan or recovery center pays for the app or reimburses the customer. The majority of Triggr Health customers have not yet sought formal treatment, Haskell says.
Over time, Triggr Health may help to reduce treatment costs. According to the National Institute on Drug Abuse, a 30-day substance abuse treatment program can cost between $10,000 and $19,000; some residential programs charge more than $35,000. “When we are successful at keeping people healthy, it ends up saving insurers money,” Haskell says. Sprout Health Group, one of Triggr Health’s partners, has said that since the company started using the app with its patients, the overall cost per patient has declined.
Encouraging More Help
Triggr Health’s participants tend to be highly engaged and open to sharing their phone data, says Haskell. Protecting that data by ensuring the app is HIPAA compliant has been a top priority from the start. “We make it clear through our interaction with customers that we are using their data to inform how we reach out to them and help them achieve their personal goals. What we look at is an indicator of how someone is doing, not what they are actually doing.”
Using machine learning to help treat an inherently vulnerable population carries a significant risk, particularly with individuals for whom a relapse can be fatal. When dealing with people who are addicted to heroin, which is life-threatening, Triggr Health’s behavioral guides work closely with individuals to motivate them to seek treatment beyond the app. “We have yet to find someone who loves using heroin and thinks everything is going great. They started using because something else was going wrong in their lives,” Haskell says. “Our goal is to open them up to care that is more intensive.”
At whatever stage Triggr Health’s participants enroll in the app, the approach is the same, says Haskell: “If somebody is open to higher levels of care, we encourage looking into options such as individual therapy or medication-based therapies. Our goal is always to treat the whole individual—on their terms.” D!