NIMH Research Seeks to Harness Power of Technology, says Gordon


When Joshua Gordon, M.D., Ph.D., became director of the National Institute of Mental Health in 2016, he wanted to pay special attention to three areas among the many in the institute’s portfolio: suicide, neurocircuits, and computation. Suicide takes an unacceptable toll every year. Neurocircuits were revolutionizing knowledge of anxiety-like patterns in the mouse brain and needed to be applied to human illness. Computational psychology was an all-encompassing technology, embracing biophysics, computational phenotyping, and machine learning.

“We wanted to look at research that can influence clinical treatment in the next few years, but also how to put into action what we know now and who will respond to treatment,” said Gordon at the 175th History Track session titled “Shaping the Future of Psychiatry Through Research and the Delivery of Care” on Tuesday. “We also want to support investigation into how the brain works and how it integrates with the social and physical environment.”

One approach combines the focus on suicide with the power of technology and computation, said fellow panelist Matthew Nock, Ph.D., a professor of psychology at Harvard University. Research to date usually has focused on the same risk factors, most of which have small effect sizes. Nock and his colleagues with the Army Study to Assess Risk and Resilience in Servicemembers (STARRS) employed machine-learning techniques to analyze combined risk factor data from medical and administrative records on 53,769 psychiatric hospitalizations. They found that the people in the top 5% of the risk scores accounted for 53% of suicides.

Machine learning may also help identify new risk factors, too. Nock is currently replicating this study in five civilian health care systems.

He also is exploring how to identify imminent risk for suicide. Most research has looked at risk as far back as five to 10 years prior to suicide and is confounded by recall bias. About 7% covers one to six months before the event, and just 0.1% has looked at the 30 days before suicide—presumably the most useful period for clinicians.

The dearth of research on the weeks immediately prior to suicide “is the biggest limiting factor to help us understand suicide,” he said.

Nock is looking at the multiple technologies incorporated in cell phones for help in “digital phenotyping” of people at risk for suicide. “How do people’s thoughts and feelings change as they move into suicidal episodes?” he asked.

Researchers have installed apps on the phones of high-risk individuals and “ping” them four to six times a day to ask about suicidality. Certain patterns in their responses have correlated so far with increased risk for suicide attempts. A more passive approach that combines the phone’s global positioning system (or GPS), accelerometer, calling, texting, and Bluetooth system is also under investigation. First results: More texting is linked to fewer suicide attempts, possibly because the individuals are engaging with friends and family.

These technologies produce volumes of data that can be analyzed in ways unavailable in smaller, conventional studies and provide new insights for diagnosis, intervention, and treatment.

(Image: David Hathcox)