New Insights Into Psychotic Disorders

It is estimated that 1.5%–3.5% of the population suffers from some form of affective or nonaffective psychotic disorder (1). While estimates vary, a relatively recent survey revealed that the annual incidence of first affective and nonaffective psychotic episodes to be 86 per 100,000 per year for individuals 15–29 years of age and about half of that for individuals that are 30–59 years of age (2). Only a relatively small portion of these first-episode cases turn out to be schizophrenia, which has an incidence estimated to be around 15 per 100,000 individuals (1). This issue of the Journal presents new data that deepens our understanding of psychotic episodes and psychotic disorders. Specific topics that are addressed include 1) racial and ethnic disparities associated with the diagnosis of psychotic disorders; 2) shared genetics between schizophrenia and cardiovascular risk factors; 3) predicting the conversion from clinical risk to a first psychotic episode; and 4) predicting outcomes in individuals treated for their first psychotic episode.

The issue begins with a very relevant overview on first psychotic episodes in individuals with schizophrenia, authored by Dr. Jennifer Forsyth from the University of Washington and Dr. Carrie Bearden from the University of California, Los Angeles (3). Forsyth and Bearden reframe the concept of first-episode psychosis by emphasizing the neurodevelopmental nature of schizophrenia as it relates to alterations in developmental neuroplasticity and the early onset of cognitive and brain alterations. The new research in this issue includes data from the Kaiser Permanente registry that characterizes racial and ethnic disparities in the diagnosis of affective and nonaffective psychotic disorders. Another study uses GWAS data from the Psychiatric Genetics Consortium to explore the genetic underpinnings of the increased incidence of cardiovascular disease in individuals with schizophrenia. Two additional studies use different types of data with machine learning methods attempting to improve the prediction of the onset of a first psychotic episode and as well as their treatment outcomes. The first, uses machine learning with functional MRI data obtained prior to treatment to predict treatment responses in individuals experiencing their first psychotic episode. In the other, a priority data letter, demographic and clinical data and cortisol measures from individuals in the North American Prodromal Longitudinal Study 3 (NAPLS-3) cohort are analyzed with machine learning techniques with the goal of improving the ability to predict a first psychotic episode.

Racial and Ethnic Disparities in the Incidence of Psychotic Disorders

Prior work has demonstrated racial and ethnic disparities related to various psychiatric diagnoses. To further understand such disparities, Chung et al. (4) use the Northern Kaiser Permanente patient database from 2009 to 2019 to characterize demographic features, emphasizing race and ethnicity, in relation to the incidence of affective and nonaffective psychotic disorders and other comorbid illnesses. Data from 5,994,758 individuals was used in the analyses with the primary motivation of assessing potential differences among individuals that self-identified as Black, American Indian or Alaskan Native, Native Hawaiian or other Pacific Islander, Latino or Hispanic, non-Hispanic White, Asian American, or other. Over the 10-year period, 0.22% of individuals had diagnoses of nonaffective psychotic disorders and 0.098% had a diagnosis of an affective psychotic disorder. When compared to White individuals, Black individuals had a 2.13-fold greater risk in the incidence of nonaffective psychoses, and American Indian or Alaskan Native individuals had a 1.85-times greater risk. However, Hispanic/Latino and Asian American individuals had lower risks when compared with White individuals (0.91 and 0.72, respectively). When assessing the incidence of affective psychoses, Black individuals had a 1.76-times increased risk compared with White participants. In relation to comorbidity, all individuals with psychotic disorders had at least a 4.83 greater likelihood of having other psychiatric comorbidities, with the greatest likelihoods seen for receiving a diagnosis of bipolar disorder or substance use disorder. Additionally, individuals with psychotic disorders had increased odds of suicide and premature death, as well as various metabolic-related and cardiovascular illnesses. While this study clearly demonstrates racial disparities in the diagnosis of psychotic disorders, it is important to note that this study cannot address issues related to the cause of the racial disparities. Nonetheless, it is highly likely that the multiple negative influences of structural racism are major contributors. In the Discussion section of the paper, the authors point out that the identified disparities could be due to actual differences in risk or other factors related to racial biases such as racial-related differences in the likelihood of providers’ making the diagnosis of a psychotic disorder. In their editorial (5), Dr. Els van der Ven from Vrije Universiteit Amsterdam and Dr. Ezra Susser from Columbia University discuss the consistency of these findings in relation to earlier work and strongly suggest that the disparities found in this study are directly related to the negative impacts of structural racism.

Shared Genetics Linking Schizophrenia and Cardiovascular Risk Factors

Rødevand and colleagues (6) build on previous studies examining factors that may contribute to the well-established higher incidence of cardiovascular disease that occurs in individuals with schizophrenia. At the genetic level, prior studies have demonstrated genetic correlations between schizophrenia and cardiovascular disease as well as with various cardiovascular risk factors. The current study uses statistical methods that move beyond traditional genetic correlations allowing for not only an understanding of the genetic alterations that are shared across schizophrenia and cardiovascular risk factors, but also allows insights into whether the shared single nucleotide polymorphisms (SNPs) have effects in the same or opposite direction in relation to the phenotypes. In this study, GWAS data for schizophrenia was obtained from a large cohort of individuals of European descent that participated in the Psychiatric Genetics Consortium, and the GWAS data for the cardiovascular risk factors was obtained from a separate set of participants. The cardiovascular phenotypes that were used included: BMI, type 2 diabetes, lipids, systolic and diastolic blood pressure, regular smoking, and cigarettes/day. Not surprisingly, the results demonstrated that schizophrenia and all the cardiovascular phenotypes examined were polygenic. With the method used, 9,600 SNPs were found to be associated with schizophrenia. The number of schizophrenia-related SNPs shared across the cardiovascular risk phenotypes ranged widely: 89.6% with regular smoking and 83.5% with BMI to 5.2% with coronary artery disease and 3.1% with low-density lipoproteins. Using a different method with a more conservative statistical approach, 800 SNPs were found to be associated with both schizophrenia and cardiovascular risk factors. It is of interest that those SNPs that were shared between schizophrenia and regular cigarette smoking, and the amount of smoking, had effects in the same direction whereas those shared with BMI tended to have effects in the opposite direction of that for schizophrenia. These findings suggest that genes underlying schizophrenia increase the risk for smoking behavior whereas they may decrease the risk for having an increased BMI. In their editorial (7), Drs. Rona Strawbridge and Nicholas Graham from the University of Glasgow further discuss these results and their implications in relation to understanding cardiovascular comorbidities in individuals with schizophrenia.

Using fMRI to Predict Treatment Response in First-Episode Psychosis

The study by Cao and colleagues (8) is aimed at developing functional brain-based predictors of antipsychotic treatment in individuals treated for their first psychotic episode. Unique features of the study include obtaining within-subject data from multiple different fMRI paradigms prior to treatment and using an analytic strategy that combines the data from these different paradigms, “cross-paradigm connectivity.” The authors argue that this method provides a more reliable view of trait-like, not state-like, brain connectivity. In the study, machine learning was used to develop a predictive connectomic model at the individual level in a discovery sample of 49 patients and then, for further validation, an independent sample of 24 patients was used. All patients in the study underwent four different fMRI paradigms prior to treatment: at rest with eyes closed, reward gambling task, oddball task, and a multisource interference task aimed at achieving a high level of engagement of the anterior cingulate cortex. The medication treatment phase was double blind with participants randomized to receive 12 weeks of either aripiprazole or risperidone. Results from the discovery sample demonstrated that 14 connections could be used in the predictive connectome model. Lower levels of connectivity between the cerebellum and cortical nodes and higher levels of connectivity between some of the cortical areas were found to predict treatment response. It is important to note that this connectomic model was validated in the second sample as a significant predictor of treatment response. Also, the authors point out that the current findings are generally consistent with their earlier work linking connectivity between the cerebellum and cerebral cortical regions to psychotic disorders. Drs. Donald Goff and Joshua Roffman from New York University along with Dr. Daphne Holt from Harvard provide an informative editorial (9) that elaborates on the findings.

Improving Models With Machine Learning to Predict the Transition From Risk to Psychosis

Smucny et al. (10) provide a priority data letter that uses machine learning to analyze known risk factors that were derived from earlier studies to see if this approach improves upon current methods to predict conversion to psychosis at the individual level. In the prior North American Prodromal Longitudinal Study 2 (NAPLS-2) cohort, clinical and demographic risk factors were assembled into a psychosis conversion calculator that was validated and found to be quite accurate (11). Building on this study the current researchers used new data from the NAPLS-3 study, which comprised 710 high-risk individuals and 96 comparison subjects that were followed for 2 years. The approach used was to examine the extent to which different statistical models and different types of machine learning could most accurately predict psychosis conversion. In addition to using the previously identified risk factors, the researchers also included salivary cortisol levels in their predictive models. Cortisol was included as it has been shown to be elevated in at-risk individuals, and some data support that it may add to the predictive power for psychosis conversion. Results demonstrated numerous cognitive and psychopathology-related differences between the at-risk and healthy groups as well as increases in cortisol concentrations in the clinical risk group. Over the 2 years of follow-up, 62 of 598 at-risk individuals converted to psychosis. When comparing the different statistical modeling approaches, the supervised machine learning approach, random forest model, performed the best with an overall 90% accuracy with greater than 75% sensitivity (i.e., true positives: correctly identifying individuals that convert) and greater than 85% specificity (true negatives: correctly identifying individuals that do not convert). This level of predictive capability is higher than that previously reported, and the authors speculate as to why the random forest model, a classifier method that uses multiple decision trees and can use continuous values, may have performed so well in this sample to predict conversion to psychosis. In his editorial (12), Dr. Tyrone Cannon from Yale University puts the interpretation of these findings in perspective in relation to issues relevant to the machine learning methods used, as well as commenting on the potential clinical utility of this and other algorithms for predicting conversion to psychosis.

Conclusions

This issue of the Journal provides insights into the altered neurodevelopment and early symptom presentation associated with the risk of developing schizophrenia, racial and ethnic disparities in the diagnosis of psychotic episodes, predictors of first psychotic episodes and its treatment, and the genetics associated with the increased risk of cardiovascular disease in individuals with schizophrenia. Major findings from these studies include 1) Black individuals have a significantly increased risk of being diagnosed with an affective or nonaffective psychotic disorder than White individuals; 2) schizophrenia and cardiovascular risk factors have some shared underlying genetics with a finding suggesting that genes underlying schizophrenia increase the risk for smoking behavior; 3) machine learning methods have the potential to be quite accurate in predicting the conversion from clinical risk to first psychotic episode; and 4) evidence supports that levels of functional connectivity between the cerebellum and cortex and increased interconnectivity between some cortical regions can be used to predict treatment response in individuals being treated for their first psychotic episode.