Scientists are using cellphone data to track everything from depression and mood disorders to crowd behavior.
On a difficult day, the patient’s data stream includes very few pings. Her GPS sensor shows her going slowly from the bedroom to the bathroom and back again. She does not check email, but does watch four hours of Netflix. No phone calls out, except for Chinese food. Seven phone calls in, from her sister. She answers none of them.
On a high day, her data stream shows jerky movements all over the house, plus 12 trips outside, including two to the liquor store. Twenty-five calls to people very low on her contact list—some of whom she hasn’t spoken to since high school. Three hours of sleep. And $3,000 spent via PayPal on Etsy.
Back at her doctor’s office, where the data are sent, a quick number crunch triggers a suspected diagnosis: bipolar disorder, in manic mode, with a crash coming any day.
A lifesaving diagnosis made possible by a practically invisible technology: a smartphone.
At least, that’s the brave new hope for mental health care, as envisioned by Jukka-Pekka “JP” Onnela, assistant professor in the Department of Biostatistics at Harvard T.H. Chan School of Public Health. Onnela won a National Institutes of Health 2013 Director’s New Innovator Award for his proposal to combine mental health research with big data—one of today’s cutting-edge approaches to psychiatric research.
“I’ve always been interested in the big picture,” Onnela says. “The structure of connections between individuals is the essence of public health. People are connected—and therefore our health is connected.”
Walking Data Points
In the past decade, the mobile phone has turned people into walking data points that can be measured on a vast scale. Onnela has been widely lauded for recognizing the research potential of such a tool. He is an emerging figure in the field of network science—which studies the connections among thousands of individuals, or among biological substances such as proteins and metabolites—whose findings are informing a wide range of public health questions, from psychiatric diagnosis to HIV/AIDS treatment to disaster relief.
For psychiatry in particular, the goal is to make diagnostic judgments based not on what the person says about his or her mood or state of mind—some of which may be misrepresented or misremembered—but rather on what can be objectively observed and measured. This aspect of mental illness has dogged experts for years: its deep subjectivity. Clinical depression to one practitioner may be just a rough spell to another.
Onnela is now collaborating on a project sparked by the NIH award and its $1.5 million in funding over five years. His team has hired software developers to create a smartphone app that can collect both passive data (for example, how many steps someone takes a day) and active data (such as voice samples) about mental health. The results could produce the most fine-grained scientific picture yet of the ever-changing dynamics of emotion. And if Onnela is right, using digital devices to quantify human behavior in everyday settings—an approach he calls “digital phenotyping”—could go well beyond mental health.
Physics and Public Health
The short-haired, boyish-looking Onnela, 37, was born and raised in Finland, where communications technology is more advanced than in the United States. (They’ve been banking by text there for more than a decade.) His father was a high school physics teacher, his mother ahealth sciences professional who ran a college for nurse training—twin specialties that have played out in Onnela’s career as well.
Trained in the physics of networks, he first deployed mathematical algorithms to understand complex theoretical networks. Slowly, he started looking outside the theoretical and more at how human beings function in social webs, using statistical analysis to glean how those endless interconnections play a role in health.
While on a fellowship in Oxford, England, Onnela spent a semester as a Fulbright scholar at Harvard, where he teamed up with Harvard Kennedy School political scientist David Lazer, one of the pioneers of computational social science. Then, at an international conference, he recognized Nicholas Christakis—at the time a Harvard University professor of medical sociology—in an elevator, and shyly introduced himself. Christakis was well known for his own brand of network social science, including research on obesity that linked one’s weight to that of one’s friends.
Onnela became a postdoc in Christakis’ lab, studying the ways people form social connections and how that affects their health. In 2011, he moved to Harvard T.H. Chan School of Public Health. “It’s primarily because of Nicholas’ influence that I ended up choosing a career in health,” Onnela says.
Christakis, now at Yale University, returns the compliment, noting that Onnela’s work has emerged at an important juncture in social science—just as big data meets a resurgent interest in social connectivity. “His innovative work in understanding the structure of human interaction has put him on the scientific frontier,” says Christakis.
And though it may sound counterintuitive, Christakis adds, using advanced technology helps scientists understand human beings in their natural state—the Holy Grail for social science research. “It’s very challenging to observe the natural world—it’s messy, it’s difficult, and it’s been that way since scientists first began practicing their craft,” he says. “JP is able to quantify but simultaneously simplify complex patterns in the natural world.”
Clues to Mood
When Onnela’s name started percolating through the scientific community, his discoveries attracted researchers from outside network science. One of those was Harvard psychiatric resident John Torous, who found new technology intriguing. He was struck by how Onnela—with no training in psychiatry—could immediately intuit key questions dogging the mental health world.
“He has a whiteboard in his office and began to sketch out diagrams of what we could do,” Torous says. “He understood what kind of data would be useful to mental health care—applying voice data such as pauses and pitch, location data, social data—as markers to pick up on different diseases.”
“For a long time, people have wanted to explore thesocial side of mental illness, but you can’t follow your patient all day,” explains Torous, who now collaborates with Onnela on mental health data collection. “A lot of mental health diagnosis is done retrospectively, which is unlike anything else in health care.”
Skeptical About Self-Reports
Onnela considers cellphone technology a vast improvement over many traditional methods of data collection, especially the self-reported questionnaire filled out after the fact. Most people just can’t remember what they ate or how much they slept even a day or two later, he says. “Our memories, our recollections of what we’ve done, and what others have done tend to be very biased,” he says.
Currently, Onnela, Torous, and others are piloting a daily on-screen questionnaire, in which people are prompted to record the quality of their sleep shortly after they get up. Two or three times a day, patients are pinged to answer three simple questions from a list of about a dozen traditionally used by mental health professionals as part of regular check-in office appointments. These questions can serve as bellwethers of mental well-being or illness, such as “How well can you concentrate?” and “Do you feel active and energetic?” More gravely, the pop quizzes that surface on the cellphone may ask if the patient feels he or she would be “better off dead or hurting myself.”
The team discovered that people were more likely to report suicidal feelings on the smartphone than on a printed questionnaire at a doctor’s office, though they can’t say for sure which answer is more accurate.
The smartphone questionnaire has another advantage, Onnela says. It can mimic a human interviewer by adapting to previous answers as it goes along. So if the answer to “How well did you sleep?” is always “Fine,” it will stop asking that question so often, but if the answer to “How well can you concentrate?” varies widely, it will home in on that behavior, asking more and more detailed follow-up questions.
Eventually, Onnela says, the tool could evaluate how well treatments are working. In the past, it could take months to assess the efficacy of a psychiatric drug. Today, he says, smartphone sensors can pick up more quickly whether a person reacts to a new drug, for example, with jitteriness or sleeplessness or other unpleasant side effects. “Now that we can more accurately quantify behavior, it might take us just weeks instead of months to fine-tune medications and dosages.”
Privacy and the Personal Touch
Onnela’s brand of network science is not without controversy. For one thing, he will need to navigate sticky ethical issues of privacy. In studies using data from operators, all cellphone data are anonymized—meaning no one’s specific movements can be traced back to his or her identity, so what Onnela is learning is attributed to the community at large. But for scientists to create an app that would work on the individual level, patients will need to agree to having their activities tracked by researchers.
“If we look at the past five or 10 years, our notion of privacy has evolved and continues to evolve. The boundaries between what’s public and private are getting blurred,” Onnela says. “And today, what’s private for somebody in their 30s is probably different from somebody in their 60s.”
Critics also suggest that warm human interaction can’t be replaced by cool high-tech probing devices—however accurate they may be. Onnela agrees that cellphones could never supplant caring conversation or a pat on the shoulder or, more broadly, the rich therapeutic relationship. Rather, the new technology would complement it. “Nobody would argue that because you have a thermometer or can take your blood pressure at home that it negates the need for doctors,” he says.
Torous agrees. “Just as cardiologists use echocardiograms, psychiatrists want to bring in more objective data,” he notes. “That means they will spend less time asking about symptoms and spend more time on treatment.”
Mapping Social Networks
As technology improves and sensors get smaller, collecting behavioral data will only get easier and less intrusive. Given this, Onnela would like to see classic long-term public health studies—such as the Framingham Heart Study or the Nurses’ Health Study, which follow individual habits and health outcomes over long periods—updated in the future to include social connectivity.
“People have of course considered social networks and the role of social support in the past, but because it’s been traditionally very difficult to collect data—social network data at the large scale—these studies have mostly been limited to much smaller groups of people,” he says. “We’d like to understand the structure of the social network at the societal level. That’s what’s going to determine how information and misinformation flow, or how pathogens spread, among a large group of people.”
To that end, Onnela is mapping broad social networks in markedly different settings. In 2013, he studied the Kumbh Mela festival in India, a three-month Hindu religious event that draws millions of people. Most of these festivalgoers carry cellphones. Drawing on their electronic exchanges, Onnela’s team was able to follow the ebb and flow of attendance. Over time, Onnela says, similar research techniques could be deployed to predict the dynamics of refugees during natural disasters, or to help governments predict crowd and traffic patterns to avoid major jams or even stampedes.
Onnela has also applied social mapping techniques to infectious-disease control. His ideas helped guide scientists in the Botswana Combination Prevention Project, part of the Harvard T.H. Chan School of Public Health AIDS Initiative, as they conducted trials for HIV treatments. Typically in such studies, researchers choose one village to receive treatment and another nearby village as a control group, receiving standard care but no additional treatment. If lots of people have relationships across the villages, however, the effect of the interventions is diluted and harder to measure.
With Botswana’s unusually high rate of cellphone coverage, Onnela aims to assess data to estimate how much social mixing occurs between villages. “If we can understand the extent of mixing, we can learn how effective different treatments really are,” he says.
The Next Big Thing?
Onnela is sensitive to the potential pitfalls of any “next big thing” in health care or mental health treatment. “Data alone won’t do it,” he says. The flood of information must be collected efficiently, analyzed thoughtfully, and validated clinically.
Moving forward, Onnela wants to use his math, physics, and statistical skills to manipulate bigger and bigger data sets, and to refine methods to analyze them, to improve population health. “I love being able to combine my passion for quantitative sciences with the desire to change the world, the lives of people, even in a very small way.”
And while he is devoting his career to complex theories and ever-more-advanced technology, his inspiration tends to come in notably low-tech moments. “I get my best ideas when I’m not actively trying to think, such as when taking my dog for a walk.”
So what would a researcher learn from studying Onnela’s cellphone records? “They would probably learn that I get to the office early and I leave the office late,” he says with a laugh. “But everyone who knows me already knows that.”
Karen Brown is a freelance writer and public radio reporter in Western Massachusetts who specializes in health and mental health issues.
Photo: Kent Dayton/ Harvard Chan School