When signals say more than speech
Software to analyze chats may help tell if a patient’s troubled
In any conversation, the dialogue is only half the story. Embedded in pregnant pauses, pacing, and fluctuations in a speaker’s tone of voice are signals that hint “I’m not really paying attention’’ or “I’m feeling down’’ or “I’m interested!’’
Now, Cogito Health, a company in Charlestown, is mining the unspoken parts of a conversation for insights that could improve people’s health. With technology that analyzes the tempo and syncopation of speech, the company is trying to identify indicators that someone may be at risk of depression or is unlikely to adhere to a plan for managing a chronic disease.
“What we’re working toward is enabling organizations to be more intuitive and empathic with regard to customers,’’ said Joshua Feast, chief executive of Cogito Health. “We pull out what we call social signals underlying interactions, hidden below the actual words spoken.’’
The company was founded in 2007, spun out of the laboratory of Alex Pentland, a professor at the MIT Media Lab who studies such signals. Using various technologies, Pentland has studied factors ranging from heart rate to patterns of speech among a wide variety of people, from depressed patients undergoing electroconvulsive therapy to poker players. What has emerged, he said at a recent symposium, is that signals that underlie people’s words are often more honest than what they say.
Those findings build on a century of observations by mental health specialists, who have noted that depressed patients talk differently.
One of Feast’s favorite descriptions of the different way depressed people speak comes from German psychiatrist Emil Kraepelin, who described patients in the early 20th century who “in a low voice, slowly, hesitatingly, monotonously, sometimes stuttering, whispering, try several times before they bring out a word, become mute in the middle of a sentence.’’
To turn that research into useful data for health insurers and companies that provide wellness services, Cogito Health has developed software that analyzes phone interactions.
For example, a health coach who guides a patient through a program to help her quit smoking might get feedback on how engaged the patient is in what is being said.
A nurse calling people enrolled in a program that helps them manage diabetes or heart disease might learn that the pauses and hesitations in their speech indicate they may be at risk of depression.
David Veroff, senior vice president of analytic development at Health Dialog, a Boston company that specializes in health care analytics and support, said his company became interested in finding ways to get more out of the phone interactions its employees have with patients.
Typically, health coaches interact with patients to help them navigate through the health care system and access preventive care that could reduce emergency hospitalizations or surgery.
“Their notion that you could detect key behavioral or health indicators by assessing voice signals is really interesting,’’ Veroff said. “If you can predict whether [patients] are likely to have depression, and we’re doing that process on every single call we have,’’ it becomes a powerful screening tool.
“We’re assessing the speech patterns of our employees interacting with people, to ensure that they’re showing up on the call in the best way possible,’’ said Nick Balog, a company vice president. “Think if you’re driving in a car and you’re looking at your speed, your odometer reading. We’re going to be bringing a capability that will allow our coaches to essentially monitor themselves right when they’re talking with these folks on the telephone.’’
Dr. Anthony Massey, a medical director at the insurer
In a study presented at a meeting of the American Telemedicine Association, CIGNA and Cogito offered data that suggested it is possible to tell if yes actually means no when a patient promises to attend an appointment.
Ultimately, the company plans to move beyond health care applications.
That could include programs for monitoring the quality of interactions between employees and clients, assessing customer satisfaction, and being able to tell when someone is bluffing.
Carolyn Y. Johnson can be reached at firstname.lastname@example.org.