<< Back

Research team earns $2.25M grant to explore ways to tailor mental health treatment

August 14, 2024

A research team headed by Godfrey Pearlson, MD, of the Institute of Living (IOL), part of the Hartford HealthCare Behavioral Health Network, will employ a form of artificial intelligence (AI) called machine learning (ML) to identify predictive markers in patients that will allow specialists to personalize treatment for a wide variety of mental health conditions.

Backed by a $2.25-million, five-year grant from the National Institute of Mental Health, the team will analyze large quantities of data from 2,400 psychiatric patients to see if similarities can predict outcomes, says Dr. Pearlson, head of the Olin Neuropsychiatry Research Center and one of three principal investigators with Christopher Pittenger, MD, PhD, and Sarah Yip, PhD, of Yale School of Medicine.

“Globally, mental illness causes tremendous suffering, disability, premature death and cost to society, yet our ability to successfully treat or prevent symptoms has barely budged,” Dr. Pearlson says. “We hope ML can help us understand the underlying cognitive, emotional and behavioral characteristics of many mental health disorders, visualize their longitudinal course and effectively treat them.”

Digging into data

The research team will gather information elicited from adults at IOL and Yale outpatient clinics from many sources including:

  • Electronic medical records
  • Patient ratings of symptom severity at each visit
  • Biological markers identified through bloodwork, genetic tests and electroencephalogram (EEG) testing
  • Details captured on cell phones, such as mood assessments
  • Measures of physical activity and sleep gathered by smart watches
  • Emotional responsiveness and mood through video clips on patient phones
  • Measures of social media use
  • Analysis of speech patterns from patient-recorded samples, directed by Manu Sharma, MD, a natural language processing expert with the IOL

The information will be added to a central repository where data scientists – coordinated by Michael Stevens, PhD, at the IOL – will employ a series of sophisticated ML algorithms to analyze it.

“We want to find out if information within diagnoses and across multiple diagnoses can predict relapse and better guide treatment,” Dr. Pearlson explains, adding the team will examine psychiatric diagnoses such as depression, anxiety, PTSD, bipolar disorder, obsessive-compulsive disorder, substance abuse and psychosis.

Information will be gathered in three waves of 800 patients. Details from the first wave will help guide interactions with the second wave and then waves one and two will guide wave three, he says.

What this means

The new research builds upon the group’s ground-breaking work using patient subgroups based on biological similarities to guide treatment. Matching this with advanced predictive analytics will enable them to help mental health clinicians choose the treatment that will be most effective for individual patients.

One patient, Dr. Pearlson offers, might self-report their mood as normal while other data – social media use increase, slower speech, sadder emotional expression, disrupted sleep, and worse performance on cognitive tests – tells a different story.

“If all these things occur together and are consistent over time, it may indicate the person is at increased risk for an episode,” he says. “These are subtle signs the patient might not be able to report but alert clinicians to act.”

In the big picture, he says the work should help:

  • Predict individual patient outcomes
  • Characterize long-term trajectories of traditional diagnoses
  • Identify biotypes and assess their clinical utility in predicting outcomes

“Through recent ML studies, we’ve found ways to move beyond traditional research comparing behavior in subjects against a comparison group and, instead, compare individual variation on cognition and behavior,” Dr. Pearlson says. “Group-level findings likely have limited use predicting individual clinical trajectories.”

In place of group differences, this project centers on “computational fingerprints,” or each participant’s unique qualities. Evolution of the fingerprint over time, he says, may signal periods of high illness burden needing treatment.

Machine learning augments traditional research

One of the benefits of ML is its ability to process vast amounts of data, work that would take humans years, says Rocco Orlando, MD, Hartford HealthCare’s chief academic and research officer. The data being examined here will help scientists find trends that inform faster diagnosis and more effective treatment.

“In using machine learning, the research team can much more quickly pull data from patient reports and medical records to identify observable traits that signal mental illnesses. That knowledge can help specialists better care for these patients,” Dr. Orlando notes.

Find out more about the Olin Neuropsychiatry Research Center.

 

> Want more health news? Text StartHere to 85209 to sign up for text alerts