Dr. Michael Aratow, Co-founder & CMO at Ellipsis Health , which is a leader in AI-generated vocal biomarker technology pioneering a new standard in mental health by harnessing the unique power of voice for earlier and better detection, assessment and monitoring of the severity of clinical anxiety and depression.
COVID-19 brought to the fore a mental health crisis in America. Recent studies have found that four in ten adults in the United States reported symptoms of depression or anxiety, up from one in ten in 2019. Yet up to two-thirds of all cases of depression in the United States go undiagnosed, much less treated. The crisis is equally dire for young people: more than 10 percent of America’s youth suffer from major depression each year, and 60 percent of them receive no treatment.
Today, more than three years into the pandemic, tens of millions of people in the U.S. are on their own, without professional help, suffering needlessly. It is more than time to create a new paradigm for mental health care. Just as the introduction of MRIs some 40 years ago transformed medical diagnostics, surgery and cancer care, emerging technologies can revolutionize the current inadequate standards of screening, diagnosing and treating mental health disorders.
The most common screening tools currently used are imprecise patient-reported surveys, in which people may minimize or exaggerate symptoms. In-person clinical assessments are more effective, but a nationwide shortage of mental health care providers means there simply are not enough professionals trained to identify depression, anxiety, and other conditions, especially at the early and most treatable stages. Studies have found that primary care providers, the first medical professionals that most people with mental health disorders visit, correctly identify depression only 50 percent of the time.
Artificial intelligence and the unique power of the human voice can bridge this gap, offering the opportunity to reimagine new standards in mental health care. Researchers have long recognized that a person’s voice can convey state of mind, not only through words and ideas but also through tone, rhythm and emotion. When a person is depressed, speech patterns typically become more monotone, softer and lower in pitch with frequent pauses. Anxiety creates a different pattern, causing people to speak faster and breathe harder. It isn’t easy for a physician to pick up on these often-minute vocal features, even with extensive training. Computers can.
With diverse data collected from the vocal samples of thousands of individuals, artificial intelligence algorithms can analyze voice patterns and identify characteristics that may indicate a change in mental health. These objective and scalable algorithms offer a way forward that addresses both the lack of accessibility and the imprecise nature of current screening methods.
Such a machine-learning solution is being developed by Ellipsis Health, a company I co-founded to help create a new standard of mental health care. The company’s machine learning scientists started by taking thousands of voice samples of depressed and nondepressed individuals and coding their speech patterns, including pitch, cadence, and enunciation. They then added data collected at the same time from the standard PHQ-9 and GAD-7 mental health questionnaires, and professional mental health assessments. The software was trained to identify vocal features indicative of depression and anxiety after listening to short samples of patients’ speech. The feasibility of the technology was demonstrated in a recent study published in the peer-reviewed journal Frontiers in Psychology.
The AI system isn’t meant to make a final diagnosis. It is instead a valuable support tool, alerting a clinician that further assessment may be needed to determine if a patient needs mental health interventions and to follow that patient’s condition over time. Notably, such an AI-based tool doesn’t require a clinic visit—an important advantage for patients who aren’t able to physically travel for an initial evaluation due to distance, cost, or even a lack of awareness that treatment is available or needed. The technology can be integrated into mobile apps and telehealth consultations, making screening accessible and economical for patients, providers, and payers. Clinicians can also use it to regularly check patients’ voice samples remotely in order to monitor them over time for improvement or the need for more intensive interventions.
Artificial intelligence can never fully replace a clinician’s interactions with a patient, nor is it meant to. But it does offer an easy-to-use and affordable method for reaching and caring for people who would otherwise be adrift. This new technology starts us on a path toward an improved standard of mental health care that can help end the suffering of millions.