Using artificial intelligence, pediatric screening to predict psychosis

Often, persons for whom a first episode of psychosis is impending will begin to have muddled speech, and often what is called in the field, “flat affect”. That's just a schmancy way of saying a person isn’t modulating their speech to reflect and mimic the tone and emotion of whomever they are talking to.

Now, the Atlantic Monthly is reporting that researchers are developing artificial intelligence (AI) to help primary care physicians screen patients who will go on to experience serious mental illness (SMI). The AI tool would algorithmically analyze a person’s speech pattern to detect signs of an impending psychosis. If it works, the technology could help make the difference between whether a person with SMI -- a potentially debilitating disease -- can live a productive life, contributing to society while managing symptoms, or will be largely alienated from society, and statistically more likely to be unemployed, homeless, or incarcerated.

But what happens once a person is screened and is determined to need care? As I reported earlier, we already have a successful-so-far clinical trial of delivering immediate intervention in schizophrenia by way of a smart phone app. But that would that be enough to meet the needs of one percent of the world’s population who live with schizophrenia?

Evidence from an astonishingly large population cohort study (9,500—a very large number in clinical studies, and virtually unheard of in psychiatry) of young people, ages 8 to 21 years, who were seen in a pediatric clinic for a variety of reasons other than psychiatric concerns, has identified risk factors that can predict later episodes of psychosis. One key is if a child starts to develop some cognitive difficulties, however slight; for example, those who begin to suffer difficulty in school, regardless of intellect, could be in danger of psychosis later.

“The expectation that somebody who is on the psychotic spectrum at age 16 would perform cognitively like a typically developing teen will not be met. This is something that can be tracked both individually and as a group,” the study’s principal investigator, Dr. Raquel Gur, told me during an interview as the study’s data were just being announced back in 2014. Earlier intervention strategies “to improve stress resilience, optimize brain maturation, and prevent or alleviate adverse environmental circumstances,” are what Dr. Gur and her colleagues called for in a review article earlier this year. “Our current ability to recognize several risk groups at an early age not only provides an opportunity, but also implies a clinical imperative to act.”

Meanwhile, Massachusetts General Hospital psychiatrist Arshya Vahabzadeh, who researches AI in SMI, thinks screening will eventually lead to more effective treatment, more often, and independent of whether we ever overcome our country's shortage of mental health specialists. “If humans aren't going to be the cost-effective solution, we have to leverage tech in some way to extend and augment physicians' reach,” he told The Atlantic.

Where will the resources to treat people come from once they've screened positive? I don’t think we know. But it would be a damn site cheaper –and more humane—to treat people before their lives become a shambles and then leave them to a system doesn’t meet their needs.