For decades, scientists have searched for a reliable biological signal that could tell clinicians which young people showing early warning signs of psychosis will go on to develop the condition and which will not. A comprehensive new review of the evidence suggests that, despite significant scientific progress, no such test is ready for routine clinical use.
The review, published in Neuroscience and Biobehavioral Reviews, analysed 46 studies involving around 15,500 individuals identified as being at clinical high risk of psychosis. These are people experiencing subtle changes in thinking or perception, often referred to as an at-risk mental state, who may or may not go on to develop a full psychotic disorder such as schizophrenia. Between 15% and 30% of people in this category develop psychosis within two years.
Researchers examined evidence across several categories of potential biomarkers, including brain imaging, genetic risk scores, blood-based markers and neurophysiological measures recorded using EEG. Of all the approaches tested, electrical brain signals known as mismatch negativity and the P300 response showed the most consistent results across independent studies. These signals, which reflect how the brain processes unexpected information, were repeatedly found to be weaker in people who later developed psychosis, with effect sizes described as small to moderate.
Brain imaging studies added further detail, with multiple large-scale research consortia documenting progressive thinning of the prefrontal cortex and grey matter loss in the insular cortex and superior temporal gyrus in those who went on to develop psychosis. One study found that grey matter loss in the insular cortex occurred approximately ten times faster in those who converted to psychosis compared to those who did not.
Genetic risk scores, which calculate an individual’s inherited likelihood of developing schizophrenia, performed more modestly. In people of European ancestry they explained roughly 10% of the variation in who developed psychosis, but performed significantly worse in non-European populations, raising serious concerns about equity and generalisability in diverse clinical settings.
Blood-based and proteomic biomarkers largely disappointed. A large multi-cohort study testing a plasma protein prediction model found it performed no better than chance, with a predictive accuracy score of 0.51 out of a possible 1.
Perhaps most strikingly, clinical prediction models relying solely on symptoms and functioning often matched or outperformed the more expensive biological approaches. Simple models using just two or three clinical variables, such as speech disorganisation and unusual thought content, achieved competitive accuracy without any biological testing.
The review’s authors concluded that, while biomarker research has advanced scientific understanding of how psychosis develops, none of the measures examined currently meets the standard required for clinical implementation. They called for larger, more diverse multi-site studies with rigorous external validation, and stressed the need for research from countries beyond North America and Europe.
For now, careful clinical assessment remains the most reliable tool available to those working in early intervention and psychosis prevention services.