Search

Machine learning (ML) holds promise for precision psychiatry, but its predictive performance is unclear. We assessed whether ML provided added value over logistic regression for prediction of schizophrenia, and compared models built using polygenic risk scores (PRS) or clinical/demographic factors.\ud\udLASSO and ridge-penalised logistic regression, support vector machines (SVM), random forests, boosting, neural networks and stacked models were trained to predict schizophrenia, using PRS for schizophrenia (PRSSZ), sex, parental depression, educational attainment, winter birth, handedness and number of siblings as predictors. Models were evaluated for discrimination using area under the receiver operator characteristic curve (AUROC) and relative importance of predictors using permutation feature importance (PFI). In a secondary analysis, fitted models were tested for association with schizophrenia-related traits which had not been used in model development.\ud\udFollowing learning curve analysis, 738 cases and 3690 randomly sampled controls were selected from the UK Biobank. ML models combining all predictors showed the highest discrimination (linear SVM, AUROC = 0.71), but did not significantly outperform logistic regression. AUROC was robust over 100 random resamples of controls. PFI identified PRSSZ as the most important predictor. Highest variance in fitted models was explained by schizophrenia-related traits including fluid intelligence (most associated: linear SVM), digit symbol substitution (RBF SVM), BMI (XGBoost), smoking status (XGBoost) and deprivation (linear SVM).\ud\udIn conclusion, ML approaches did not provide substantial added value for prediction of schizophrenia over logistic regression, as indexed by AUROC; however, risk scores derived with different ML approaches differ with respect to association with schizophrenia-related traits.

We investigated whether associations between area deprivation, urbanicity and elevated risk of severe mental illnesses (SMIs, including schizophrenia and bipolar disorder) is accounted for by social drift or social causation.\ud \ud We extracted primary and secondary care electronic health records from 2004 to 2015 from a population of 3.9 million. We identified prevalent and incident individuals with SMIs and their level of deprivation and urbanicity using the Welsh Index of Multiple Deprivation (WIMD) and urban/rural indicator. The presence of social drift was determined by whether odds ratios (ORs) from logistic regression is greater than the incidence rate ratios (IRRs) from Poisson regression. Additionally, we performed longitudinal analysis to measure the proportion of change in deprivation level and rural/urban residence 10 years after an incident diagnosis of SMI and compared it to the general population using standardised rate ratios (SRRs).\ud \ud Prevalence and incidence of SMIs were significantly associated with deprivation and urbanicity (all ORs and IRRs significantly >1). ORs and IRRs were similar across all conditions and cohorts (ranging from 1.1 to 1.4). Results from the longitudinal analysis showed individuals with SMIs are more likely to move compared to the general population. However, they did not preferentially move to more deprived or urban areas.\ud \ud There was little evidence of downward social drift over a 10-year period. These findings have implications for the allocation of resources, service configuration and access to services in deprived communities, as well as, for broader public health interventions addressing poverty, and social and environmental contexts.

22q11.2 deletion syndrome (22q11.2DS) is associated with high rates of psychotic disorder, particularly schizophrenia. The deletion is considered to be a biological model for understanding this debilitating psychiatric disorder. It is unclear whether the psychotic manifestations in 22q11.2DS are similar to those in schizophrenia patients without the deletion. Catechol-O-methyltransferase (COMT), a positional candidate gene for schizophrenia, resides within the 22q11.2 region. It remains unknown whether hemizygosity for this gene is associated with risk of psychotic disorder. This study includes 83 adults with 22q11.2DS, 90 non-deleted individuals with schizophrenia, and 316 normal controls. Psychopathology was assessed using the Schedules for Clinical Assessment in Neuropsychiatry, the Schedules for the Assessment of Positive and Negative Symptoms and the Global Assessment Scale. Schizotypy was assessed with the Kings Schizotypy Questionnaire and Oxford Liverpool Inventory of Feelings and Emotions. IQ estimates were also obtained. Adults with 22q11.2DS were genotyped for a number of COMT polymorphisms as well as the Ashkenazi risk haplotype. This study confirms high rates of psychotic disorder (29%) in individuals with 22q11.2DS of which the majority had schizophrenia (22%). There does not appear to be a differential expression of schizophrenic symptom clusters in 22q11.2DS in relation to sporadic schizophrenia, though schizophrenia in 22q11.2DS seems to be less severe in terms of global assessment scores. Psychosis proneness seems to be of genetic origin in 22q11.2DS as individuals with 22q11.2DS without schizophrenia had higher schizotypy scores than normal controls. Finally, COMT was not associated with schizophrenia status or schizotypy.

PsyCymru was initially established as a proof of concept to investigate the feasibility of linking a prospectively ascertained, well-characterised (linked clinical cohort) of people with psychosis in Wales, UK with large amounts of anonymised routinely collected health record data. We are now additionally linking genetic data. PsyCymru aims to create a research platform and infrastructure for psychosis research in Wales by the establishment of two cohorts. The first is a well characterised clinically-assessed cohort of 490 individuals aged 16 and over, including genetic data. Consented individuals underwent a structured interview using a series of well-validated questionnaires and gave blood for the purpose of DNA extraction for sequencing and candidate gene identification. This data was linked to routinely collected health and social datasets with identity encryption used to protect privacy. The second is a much larger (12,097 individuals) but less well characterised population-based e-cohort of prevalent psychosis cases created using a previously validated algorithm applied to anonymised routine data. Both cohorts can be tracked prospectively and retrospectively using anonymised routinely collected electronic health and administrative data in the Secure Anonymised Information Linkage (SAIL) databank. This unique platform pools data together from multiple sources; linking clinical, psychological, biological, genetic and health care factors to address a wide variety of research questions. This resource will continue to expand over the coming years in size, breadth and depth of data, with continued recruitment and additional measures planned.

Eske M. Derks, Thomas Gladwin, GROUP XX, Marcella Rietschel, Manuel Mattheisen, Rene Breuer, Thomas G. Schulze, Markus M. Nothen, Douglas Levinson, Jianxin Shi, Sven Cichon, Nelson B. Freimer, Rita Cantor, Roel A. Ophoff Rudolf Magnus Institute of Neuroscience, Department of Psychiatry, University Medical Center Utrecht, Utrecht, Netherlands; Division of Genetic Epidemiology in Psychiatry, Central Institute of Mental Health, Mannheim, Germany; Institute of Human Genetics, University of Bonn, Bonn, Germany; Institute for Medical Biometry, Informatics and epidemiology, University of Bonn, Bonn, Germany; Department of Genomics, Life & Brain Center, University of Bonn, Bonn, Germany; Unit on the Genetic Basis of Mood and Anxiety, Bethesda, MD, USA; Department of Psychiatry, Stanford University, Palo Alto, CA, USA; National Cancer Institute, Bethesda, MD, USA; UCLA Centerfor Neurobehavioral Genetics, Semel Institute for Neuroscience and Human Behavior, Los Angeles, CA, USA