advertisement
advertisement

An App To Diagnose Bipolar Mood Swings By How People Talk On The Phone

A system that detects early signs of a bipolar patient slipping into a manic episode could save untold heartbreak. But is phone-tapping the mentally ill the answer?

An App To Diagnose Bipolar Mood Swings By How People Talk On The Phone
[Image: Woman on cellphone via Shutterstock]

Psychiatrist Melvin McInnis was sitting in his University of Michigan office late one afternoon when the phone rang. It was a bipolar patient’s wife. “He’s talking faster and louder,” she told him, worry edging her tone. “He sounds different now. There’s something different about him.”

advertisement
advertisement

Three days later, McInnis saw his patient. Immediately, he knew the wife had predicted something potentially life-threatening. “He was revved up, charged up, and he was in this manic state,” McInnis says. “We had to petition him to go to the hospital.”

Full-on manic episodes don’t happen too regularly, but when they do, they leave a trail of destruction behind them. Livelihoods and relationships are often leveled in their wake. For clinicians, bipolar patients, and their families, finding an early predictor or biomarker of a manic episode would be something of a holy grail.

McInnis believes that vocal clues could be the answer. That’s why, over the past year, he and his team have been collecting over a thousand hours of phone call audio from bipolar patients to develop an app that can alert them to when they’re losing touch with reality.

Bipolar disorder manifests a little bit differently in everyone, and so do the severity and frequency of mood swings. But the problem for bipolar patients and their loved ones is that those episodes are not always so recognizable. Not at first. Before a bipolar person slips into a full-blown manic episode, he often experiences something called hypomania–a lesser sensation, but also a very enjoyable high, McInnis explains. And while other people might notice subtle changes in a bipolar person, like a faster speaking voice and quick jumps between disparate thoughts, the change often remains stubbornly invisible to the person experiencing it.

“The speech patterns of someone who’s becoming ill are rather prominent,” McInnis says. “We asked the question: If a human being can detect a change in the properties of speech, can we train a computer using specified algorithms? Can a computer identify small shifts and changes in the acoustics that would predict changes?”

McInnis and his team, supported by the Prechter Bipolar Research Fund, are in the early experimental stages of testing an app that records audio from bipolar patients’ phones. The app then sends that data to an engineering lab at the University of Michigan, where a machine-learning program analyzes physical characteristics of the data, like pitch and energy. Ideally, the system would then send relevant information back to the patient, or maybe even a clinician. That way, a bipolar person might recognize the need to get help before it’s too late.

advertisement

The researchers have gotten promising analyses from a dozen patients so far, but the program hasn’t yet cracked the code to depressive swings. Manic episodes are much easier to pick up on, McInnis says.

His lab isn’t the only one developing algorithmic apps to aid the mentally ill. Microsoft researchers have built tools to detect postpartum depression from social media feeds, and others are analyzing how shifts in depression can be read from Facebook. But while these experiments could help untold numbers of people, they also represent a more sinister possibility: What’s keeping this data secure? And who would have access to the tapped phones of the mentally ill?

The University of Michigan team has the explicit consent of their subjects to record audio from separate phones purchased for the experiment. But in the fast-growing, volatile world of big data and mobile health apps, that’s not always a given. Much of the time, explicit consent is sidestepped or couched in gibberish privacy policies. And health data happens to be one of the most valuable items on the data market. When health data company IMS Health raised a whopping $1.3 billion the morning of its initial public offering, it also came to market touting 400 million longitudinal “anonymous” profiles of patient data.

It appears that there’s still some time left to grapple with those questions. The University of Michigan app isn’t going to arrive on the market for quite a while, if it could ever be commercialized at all. But McInnis has high hopes.

“This could be applied to other illnesses, other entities in which speech is a major component,” he says. “Parkinson’s, Lou Gehrig’s, even an illness in oncology, if someone is a smoker. In medicine, we’re searching high and low for biological mechanisms these days.”

advertisement
advertisement

About the author

Sydney Brownstone is a Seattle-based former staff writer at Co.Exist. She lives in a Brooklyn apartment with windows that don’t quite open, and covers environment, health, and data.

More