Researchers Develop Digital Twins of Psychiatric Patients for Clinical Training

Mental health clinician students need a safe, consistent way to practice these complex conversations before they sit in a room with a real person. Often times a human patient’s symptoms can overlap, change over time, and look completely different from person to person.

Researchers from the University of Pennsylvania, New York University, and Penn’s Linguistic Data Consortium are building an AI platform called STELLAR. It creates virtual patients, or “digital twins,” so trainees can practice interviews.

Building Digital Twin Patients

These AI patients are composites built from a massive set of clinical data from Penn Medicine and the Children’s Hospital of Philadelphia. The team is also pulling data from social media.

“Many mental health symptoms do not appear only in formal clinical settings,” Sharath Chandra Guntuku, a project lead from Penn, explained. “They also come through in the way people talk day to day, including online.”

This setup lets trainers adjust exactly what the student sees. They can make the virtual patient show mild anxiety, or dial it up so the anxiety overlaps with depression.

“STELLAR brings together behavioral data, clinical expertise and AI to ask a very practical question,” said Guntuku. “Can we build training tools that better prepare clinicians for how varied and complex patients are?”

Keeping It Real

An interesting and helpful feature is how these digital twin bots actually talk.

“The promise of this approach is that we can move beyond stylized and potentially biased simulations,” João Sedoc, a project lead from NYU, added. “If we can create digital patients that simulate controllable plausible symptom expression and responsibly evaluate, we can augment current clinician training practices with the kinds of conversations that are essential to better mental health care.”

To make sure the AI does not rely on stereotypes, the team is bringing in people with actual mental health conditions and their families. They will test the avatars and give feedback on how they act.

“By involving individuals with lived experience throughout the project, STELLAR can help us ask not only whether a digital patient is clinically accurate, but whether the interaction feels respectful, realistic and attentive to experiences that are too often missed,” says Raquel Gur from Penn Medicine.