With the recent announcement and anticipated rollout of the Apple Watch, a lot of attention has been paid to what new technology can do for our physical well-being. New research at Dartmouth College shows how our devices might be used to monitor mental health, as well.
A team at Dartmouth has developed an app called StudentLife to help predict students’ mental health and academic performance based on objective sensor data from smartphones. The study is the first of its kind to use automatic sensing in this way.
Over the course of a ten-week study involving 48 Dartmouth students, automatic smartphone sensors collected 24/7 data about participants’ location, conversations, mobility, and sleep patterns, without any user interaction. It monitored things like the length of conversations and how much the subjects moved around inside at night. On top of this, the app would prompt the students several times per day with a short series of questions about their mood and stress levels.
As a third data set, the research team administered mental health and behavioral surveys at the start and end of the term, using well-known measurements of well-being. These surveys evaluated the participants on depression, loneliness, and stress. The researchers also gathered the students’ academic records from the administration, as well as their GPAs.
The study’s lead author, computer science professor Andrew Campbell, explained that after years of teaching on a slower-moving semester schedule at Columbia, Dartmouth’s quarter system seemed like an quick and exhausting “marathon” to his students. He often watched his students burn out over the course of the term, and wanted to examine how objective factors like sleep and social interaction influenced their mental health, and in turn, their academic performance.
At the end of the study, the researchers compared the self-reported data with the automatic sensing data and found some strong correlations. They determined that, based solely on the automatic data, the app could effectively predict certain mental health issues and academic performance levels in the students. For example, monitoring consistently low levels of physical activity and conversation often correlated with depression, and low levels of physical activity often predicted loneliness. To their surprise, the team found no correlation between academic performance and class attendance, and students who engaged in more conversations tended to earn better grades. Also, lonely students didn’t necessarily spend less time talking to others.
While this version of the study offered no feedback to students and didn’t share the data outside of the research team, Campbell says this could eventually be used as an intervention tool in cases of mental health risk or low academic performance. In this particular case, since Campbell organized the study and had access to his students’ data, he intervened in a situation where two of his students would have otherwise received failing grades for missed assignments and lectures. He instead opted to give the students “incomplete” grades for the term, because of the stress and health concerns he perceived in their data.
In other cases, due to privacy concerns, this would not have been so simple. Campbell couldn’t share this data with other professors, so the students still received failing grades in some of their other courses. Campbell acknowledges that since some of this data is very private, many individuals would likely opt not to share it. But, he explains, simply providing feedback to the user about their lifestyle patterns might serve as enough of an intervention for students to change their own habits.
Otherwise, he says, “it boils down to giving people ownership of their data.” If students were given the controls about who to share their data with—whether parents, professors, deans, or clinicians—then this information could prove to be extremely valuable, without becoming a frightening invasion of privacy. (Like some concerns that have been voiced about the Apple Watch.)
While this particular study examined a very specific population, the correlations were strong enough that they’re looking to adjust the technology for different populations. Campbell says he also hopes to test similar apps on office productivity and among high school students, though both those groups pose their own unique challenges.
The key to all this, Campbell reminds me, is to “put the control in the user’s hands about who they would like to share their data with, to put some safeguards in place.” So Campbell and his team won’t be popping their heads into any students’ dorm rooms, reminding them to get their eight hours of sleep or show up to class in the morning.