Written by Melvin McInnis, Thomas B. and Nancy Upjohn Woodworth Professor of Bipolar Disorder and Depression, and Principal Investigator of the Heinz C. Prechter Bipolar Research Fund
Do acoustic patterns in speech predict impending manic or depressive episodes?
The Prechter Bipolar Research Program at the University of Michigan Frances and Kenneth Eisenberg and Family Depression Center has recently begun a highly innovative research project with the Department of Computer Science and Engineering at the University of Michigan. The project – "PRIORI" (Prioritizing Care) – aims to test the hypothesis that physical properties and patterns of speech can be used to predict whether a person with bipolar disorder will develop a manic or depressive episode. Dr. McInnis, the Principal Investigator of the Prechter Bipolar Research Fund, was awarded a grant of $450,000 from the National Institute of Mental Health to study acoustic properties of speech as predictors of episodes in bipolar disorder. The co-investigators are Emily Mower Provost, Ph.D. and Zahi Karam, Ph.D. from the College of Engineering.
It’s not what you say but how you say it!
Bipolar disorder is a dynamic illness - the activities, expressions (facial and gestures) and speech characteristics change in accordance with the mood state of the individual. Family and friends note this in the appearance and the sound of the voice of their loved ones. In talking to family members, one frequently hears: "I could here it in his/her voice... I knew something was wrong." Health care providers are trained to observe and note changes in speech and language; we pay attention to the content of what the individual says as well as the form. "I feel bad" has meaning, but drawn out over several seconds "I......feel......bad" in a low tone without intonation provides further information on the diminished level of activity in the brain that is generating the speech. The Prechter Bipolar investigative team is asking: Can we use speech from an individual to predict if they are heading for an episode or significant mood change? Can we predict it in sufficient time to intervene and prevent the mood episode?
The ideas behind this project sprung from listening to family members over the years and hearing them tell us of their suspicions that something was amiss based on the speech patterns of their loved ones. We thought: if the brain of someone who knows the person can detect a subtle change that raises suspicions about an impending mood change – can we train a computer to analyze the acoustic patterns of recorded speech to predict changes in mood? Would the participants in the Prechter Longitudinal Study of Bipolar Disorder be willing to participate in such a project?
The Prechter Longitudinal Study of Bipolar Disorder is a unique and detailed longitudinal outcomes study that has engaged over 900 individuals in a clinical study aimed at identifying biological, genetic, and environmental causes of bipolar disorder as well as finding cures for this potentially devastating illness.
The Michigan Institute for Clinical Health Research (MICHR) provided initial support for the PRIORI project and provided resources for preparation of the ethical, computer security, and programming parameters of the project. Recruitment has begun and the participants have found the project to be "user friendly." The essence of the research is the gathering of acoustic speech samples from the individual when they use a study-provided smart phone. The acoustics are gathered only from the person talking into the phone; the incoming component of the conversation is not gathered. The acoustic data are stored and analyzed on a central University of Michigan server securely managed by University of Michigan computer experts.
With the funding from the NIMH, the project will run for the next 3 years. It is the aim of the project to study acoustic patterns of speech from 50 individuals with rapid cycling (4 or more mood episodes per year) bipolar disorder and 10 unaffected healthy people, as well as gathering regular weekly mood ratings to determine if episodes can be predicted using speech patterns.
Why is this a big deal?
- Predicting when a person is headed for a manic or depressive episode is VERY important. Manic episodes have an enormous cost in personal, vocational, and financial terms. Two years after a full-blown manic episode, less than half of the individuals are back to their desired level of functioning. Further, once a manic episode has begun and is escalating, the individual often has little or no insight and does not listen to family or seek medical care. If we can predict an impending episode while we and the family can reason with the individual, they are more likely to accept modification of the treatment plan and head off an episode.
- Prioritization of care is an increasing concern in management of health. We believe that the acoustic patterns of speech will assist the medical profession in management of bipolar disorder as well as other diseases. IF we can predict those individuals that are unstable and in need of care – we can prioritize those people for care and maintain those that are doing well with minimal intervention. In general, most people have better things to do than go to their health care provider and the PRIORI project is testing the hypothesis that acoustic patterns of speech may be used to monitor individuals AND prioritize them for medical care. This maximizes the use of everyone’s time!
Prioritizing Health Care – A Rising Global Challenge
The challenges of access to medical and psychiatric care are enormous and the standard approach to delivering care will rapidly overwhelm the health care system in the U.S.A. There are insufficient resources to train and equip health care providers world-wide to care for the infirm globally. It is simple arithmetic. The Prechter investigators are collaborating with clinicians and computer scientists at the Tjao Tong University and Shanghai Mental Health Center to test the adaptation of this project to be used in the Mandarin language. The use of cell phones has sky-rocketed to around 8 billion users worldwide and offers an unprecedented opportunity for health care monitoring using speech acoustic patterns. The ability to prioritize and personalize health care may be soon upon us.
We remain grateful to our research participants (who are true collaborators!) and the donors to the Heinz C. Prechter Bipolar Research Fund and Programs who make this work possible. You make the world a better place.