Share This Page
Multi-Modal Model of Psychosis
October 4, 2016
Merriam-Webster defines psychosis as a “fundamental derangement of the mind characterized by defective or lost contact with reality especially as evidenced by delusions, hallucinations, and disorganized speech and behavior.” Approximately 30 percent of people who have been diagnosed with the neuropsychiatric disorder schizophrenia will eventually also succumb to psychosis. But, despite decades of work trying to understand what heralds the onset of this state, why those particular 30 percent will do so remains unknown. Now, a new theoretical simulation published in the Schizophrenia Bulletin suggests that clinicians might be able to predict psychosis with 97 percent accuracy by sequentially measuring a variety of biological measures in a computational model.
The cost of psychosis
Psychosis is often described as a break with reality, issues with perception and comprehension so significant that people who are experiencing a psychotic episode cannot successfully navigate, or even cope with, the world around them. In A Beautiful Mind, author Sylvia Nasar describes famed Princeton mathematician John Nash’s initial descent from schizophrenia into psychosis as a period of disorganized thinking and hypervigilance. She writes, “[Nash] walked into the common room one winter morning in 1959 carrying the New York Times and remarked, to no one in particular, that the story in the upper left-hand corner of the front page contained an encrypted message from inhabitants of another galaxy that only he could decipher.” Cheryl Corcoran, a research psychiatrist at Columbia University, says that preventing psychosis – both its onset and recurrence—is critical to maintaining the overall health of people with schizophrenia as well as young people at risk for schizophrenia.
“Psychosis is similar to a heart attack, in that it is a late event following a long period of risk and morbidity,” she says. “Psychosis causes all kinds of further problems for patients. It’s a neurotoxic event, associated with a reduction in gray matter in the brain as well as poor functional outcomes. That’s why being able to prevent it is so important.”
Researchers involved in large-scale studies like the North American Prodrome Longitudinal Study (NAPLS), and similar projects in Australia and Europe, have been working towards trying understand what may cause psychosis. Barbara Cornblatt, a neuroscientist and the director of the Recognition and Prevention Program (RAP), a research program dedicated to understanding the factors underlying psychosis, says that such studies look at a person’s “clinical high risk” for a psychotic episode.
“These patients are at high risk, not because of genetic factors or family background, but because of their clinical characteristics,” she says. “These adolescents and young adults often display unusual thoughts, subtle signs of thought disorder, suspiciousness, and odd perceptions. Any of these symptoms can closely resemble psychotic symptoms—but the person still retains some insight. For example, a patient may tell you that everyone at school is making fun of him. You may be able to convince him that it is not necessarily true by saying, for example, ‘Well, maybe they were talking about someone else and you just happened to walk by.’ If that person is at clinical high risk, he would likely hear what you are saying and reconsider a bit. Someone who is already psychotic would be closed to any alternate explanation and insist they know that people at school are out to get them. So in this prodromal state, which we refer to as high clinical risk, you have attenuated or less severe symptoms that closely resemble psychosis, but are much less intense, so rather than being already ill, the person is considered to be at risk for developing a psychotic disorder in the future.”
Yet, despite this categorization of clinical high risk patients, to date, clinicians have still been unable to successfully predict who will go on to develop psychosis. That’s led some working groups to try to create simulated models, using genetic, biological, and symptom data, to offer clinicians better prognostic tools.
Creating a working model
In 10 years of research, labs like Corcoran’s and Cornblatt’s have tried a variety of different predictive models. Some models used genetic information; others have tried to use performance on perceptual or auditory processing tasks. But despite decades of important research into the neurobiological underpinnings of schizophrenia and psychotic states, most of those models have only resulted in predictive values (level of success in predicting who would go on to develop psychosis) of approximately 60 percent. André Schmidt, a clinical neuroscientist at King’s College London, says that the wide range of differences, in terms of both biology and symptoms, that clinicians see in their patients, as well as studies with small sample sizes, have made it difficult to create a strong model. He and his colleagues wanted to find a way to improve upon what had already been done.
“Previous studies have analyzed biological, cognitive, and environmental data. But that data can be quite noisy because of the heterogeneity across patients,” he says. “We knew, from other fields of medicine, that sequential multi-stage testing approaches—so not putting all of the data in one model together at once, but sequentially adding one domain after another—can help deal with the noise and make more accurate models. So we decided to try that.”
Using this approach, Schmidt and colleagues found that a model that sequentially added clinical and electroencephalogram (EEG) data, then structural magnetic resonance imaging (MRI) data, and then blood markers measuring inflammation one at a time, resulted in a prognostic predictive value of about 97 percent. While Schmidt cautions that the results are based on theoretical and not real data—and still need to be replicated—he says this modeling method has great potential.
“The nice thing in this approach is that it allows us to stratify people into different risk classes and then, perhaps, tailor some treatments,” he says. “This multi-stage sequential testing seems quite promising.”
Moving towards a practical model
Cornblatt says that while the results are striking, there is a huge difference between theoretical and practical models of prediction—especially when it comes to treating patients in the real world. She’s worried that EEG and MRI data may be too expensive, as well as difficult to obtain, for a practical prediction tool that could be used with the majority of patients at risk for psychotic disorders.
“Past models of risk calculation gave us predictive values around 60 or 70 percent. It’s not 97 percent but it is comparable to risk calculators that you see in cancer, heart disease, and other physical illnesses,” she says. “But mental illness is much messier. There’s not just a simple blood test or clinical marker in the near future. We still have a lot left to learn. And, in my view, as someone who has been studying this for decades, being able to predict psychosis with high accuracy is going to require laborious, step-by-step science. We’ve already made a lot of advances. They may seem like baby steps, but things are evolving and those baby steps, taken all together, are going to eventually get us where we need to go.”
Both she and Corcoran hope that work toward preventing the illness, both in terms of traditional research studies and computational simulations, will offer insights about not only prevention but also treatment for people who do develop psychosis.
“There’s a lot of interesting research going on that is helping us better understand what’s happening in the brain. It would be wonderful if we had something better, in terms of a treatment or intervention, to offer patients,” says Corcoran. “Today, we can offer support and pharmacotherapy. But with a better understanding of how people get to psychosis—and it’s possible that there are a variety of different ways to get there—we can hopefully get to more targeted treatments that will work, and work well, for different individuals. But there’s a lot of very important work to do before we can get there.”