Welcome to the 22nd WCP Congress Program Scheduling

Abstract Body

Schizophrenia is a severe mental disorder that can affect the whole lifespan. While the trajectory of the disease presents with high inter-individual variability, evidence suggests the role of structural white matter abnormalities in the pathophysiology. To characterize these abnormalities, previous studies focused on specific ages. Here, we aimed to model trajectories across a wide age range by combining multiple samples from individuals at different disease stages into one analysis.

To combine data from 13 international sites and over 1000 individuals, we developed novel standardization and harmonization methods for demographic, clinical, cognitive, and diffusion imaging data. We modeled white matter trajectories across ages 16 to 65 and demonstrated abnormal white matter in individuals with schizophrenia across all ages. Observed abnormalities pointed to pathologies related to abnormal development, abnormal maturation, and accelerated aging. More white matter abnormalities were associated with more severe symptoms and a longer duration of illness. Notably, general intelligence mediated the relationship between white matter abnormalities and symptom severity, specifically for females. This finding suggests a role of the cognitive reserve for females.

In addition to white matter abnormalities, we observed pronounced cognitive deficits for all ages related to medication and symptom severity. White matter abnormalities partially explained the association between schizophrenia and cognitive deficits, further highlighting the critical role of white matter in schizophrenia.

Our studies demonstrate the feasibility of retrospectively combining data from different countries and sites. They highlight that schizophrenia is characterized by clinical and cognitive deficits directly related to lifelong, dynamic structural white matter pathologies.

Abstract Body

This talk focuses on applications of normative modelling in schizophrenia to examine i) heterogeneity in the loci of cortical abnormalities and ii) the biological basis of this heterogeneity. While gray and white matter abnormalities are indeed a core neuropathological feature of schizophrenia, our findings suggest that the anatomical loci affected by these abnormalities differ vastly across individuals to the extent that group-average maps of schizophrenia pathology do not accurately resemble individual patients. For this reason, we cannot neatly pin down schizophrenia-related brain changes to circumscribed cortical regions using conventional case-control paradigms. What underlies such vast heterogeneity? Our follow-up studies indicate that marked heterogeneity results from interindividual variation in cellular pathologies across individuals with schizophrenia, as inferred from systematically combining deviations from normative ranges of brain structure with spatially-resolved gene expression data and person-specific genomic variation. Collectively, these studies call to question the viability of ongoing endeavors to derive group-average cortical maps of schizophrenia pathology. Decomposing cortical heterogeneity through combining normative models of brain phenotypes with multi-scale omics enables prioritization of schizophrenia subsets and cell types of interest for future disease modeling efforts.

Abstract Body

Psychiatric studies are striving to gain a better understanding of common and distinct pathophysiological mechanisms underlying psychiatric disorders. With sufficiently large data, robust and advanced machine learning approaches can identify biomarkers that effectively predict clinical course and functional outcomes in psychosis at the level of individuals.

This presentation outlines recent consortium efforts, such as the Accelerating Medicines Partnership® – Schizophrenia study that are designed to support the collection of large, multivariate, harmonized data suited for the application of machine learning tools.

We will discuss the application of multivariate prediction models for clinical and functional outcomes that are: 1) Robust – pass rigid cross-validation procedures. 2) Flexible – applicable for a large number of features, including missing and unmatching entries. 3) Clinically informed – guided by informative features that can be mapped back to individual features. Additional utility is the identification of homogenous latent subtypes, based on clinical and functional outcome trajectories, for understanding of pathophysiology and to facilitate future clinical trials. Finally, machine learning approaches such as crowdsourcing can produce superior individualized risk calculators, such as those widely used in nonpsychiatric medical disorders. Predictive models can assess a combination of risk and protective factors to estimate probability of outcomes that go beyond psychosis in a given period.

The approaches proposed above can be applied on any set of input parameters. However, and importantly, including input parameters that are most likely to contribute to outcomes may substantially increase prediction power, speed up the development of models, and improve interpretability of future studies.