According to the American Journal of Psychiatry, about 10 million Americans undergo psychotherapy treatment each year.
For USC Viterbi School of Engineering Professor Shrikanth Narayanan, this prompts the question, “How do we know how well psychotherapy works? What works and what doesn’t?”
Narayanan is part of an interdisciplinary team developing a new way to evaluate psychotherapy and therapists, relying on the science of signal processing to measure empathic processes. This includes automatically detecting “empathetic opportunities,” as well as characterizing “expressions of empathy.” Eventually, this system could lead to better matches — and better outcomes — between patients and therapists.
Signal processing measures and analyzes data, including human-generated data such as speech, language, text and non-verbal behavior.
Narayanan’s current research on empathy only evaluates vocal behavior —speech, language and non-verbal cues such as laughter. Computers analyze audio recordings of therapist-patient interactions — the system may eventually analyze visual behavior, such as gestures and facial expressions.
“In our experiment, we have accomplished a prototype system that is able to assess therapist empathy from audio recordings and give quantitative scores of empathy in a human, interpretable way,” said Bo Xiao, a Ph.D. candidate in electrical engineering who has worked on the project since 2011.
Although the experiment is in the early stage, a recent test demonstrated that the system is over 80 percent accurate in its rating ability, Xiao said. The system analyzes audio recordings from therapist sessions, identifying empathetic opportunities hidden within conversations.
Laughter is a non-verbal cue that tells you something important about the interaction.
“We’re looking at audio, speech and language data. We look at both verbal and non-verbal cues. For instance, laughter is a non-verbal cue that tells you something important about the interaction,” Narayanan said.
An empathetic opportunity is a situation when a patient expresses an emotion, and a doctor has the opportunity to respond empathetically by demonstrating that he or she understands the patient’s feelings.
Phrases that demonstrate reflective listening, such as “what I hear is,” are considered empathetic. On the other hand, purely instructional phrases like “you need to” are not deemed empathetic. Additionally, frequent use of high volume and high-pitch voice by the therapist is associated with lower empathy.
The research blends psychology with electrical engineering and computer science. Narayanan works with Professor Dave Atkins from the University of Washington’s Department of Psychiatry and Behavioral Science. Other key collaborators on the effort include Professor Panos Georgiou, Ph.D. student Dogan Can from USC and Professor Zac Imel from the University of Utah.
The key difference between Narayanan’s method of psychiatric evaluations and prior ones is objectivity. Typically, therapists are evaluated by someone simply observing them during a session, taking notes on their behavior. This approach has limitations because it’s difficult for the human brain to quantitatively measure mechanisms behind something as abstract as empathy.
Humans are good at certain things, but our perceptual and cognitive abilities aren’t as good.
“[This current method] is limited by human capabilities,” Narayanan said. “Humans are good at certain things, but our perceptual and cognitive abilities aren’t as good and do not scale well.”
For consistency’s sake
By automating the evaluation system through signal processing, Narayanan and his colleagues are creating an objective, consistent evaluation process. But Narayanan aims to support the current system — not replace it.
The new system assigns therapists an empathy level, which evaluators can use when measuring a therapist’s overall competence and offer appropriate training. The empathy level ranks therapists based on their responses to empathetic opportunities; the more they recognize such opportunities, and respond empathically, the higher their empathy level.
“This is very human-centered engineering,” Narayanan said. “The role of technology is to support, not to supplant human expertise and intelligence.”
“Empathetic opportunities” is not a new term within the medical field. Physicians often struggle to identify and respond to empathetic opportunities. A 2008 study suggested that, on average, general physicians miss 90 percent of empathetic opportunities, according to The New York Times.
Narayanan’s automated evaluation system could improve those results by providing therapists with real-time feedback on their responses to empathetic opportunities. In this way, Narayanan hopes the technology will help psychiatrists adjust to their patient’s individual needs.
“Different physicians have different expertise, and they apply this expertise differently,” he said. “The therapist-patient fit is important, and recognizing empathetic opportunities and responding appropriately is an important part of this.”
Narayanan compares therapist evaluations to drug trials: We test the efficacy of drugs to understand what works, when and for whom. So it makes sense to objectively evaluate the efficacy of therapists. Such evaluations can improve psychotherapy, as well as the therapist-patient fit.
“If you want truly personalized health care, it’s not just in pharmaceutical care but also in psychological care,” Narayanan said.