Identifying inflammation-related subgroups of psychopathology using machine learning

Evidence suggests that there is a relationship between inflammation and multiple psychiatric conditions including depression [1], schizophrenia [2], and psychosis [3]. Inflammation is shown to be specifically associated with somatic symptom profiles within depression, and negative symptomatology in schizophrenia and psychosis. This suggests that inflammation may be a potential transdiagnostic risk factor for psychiatric outcomes, understanding of which could lead to enhanced insight into shared mechanisms between heterogenous and overlapping psychiatric conditions. This study proposes a data-driven clustering approach to identifying and characterising inflammation related subgroups across the spectrum of psychopathology. We hypothesise that there are distinct subgroups of psychopathology that have a significant immunological component, characterisation of which will lead to enhanced classification and treatment of psychiatric illness, as well as helping to disentangle heterogeneity in specific clinical populations. The aims of this work are to identify psychiatric subgroups with shared immunological pathologies and characterise these subgroups with regards to relevant demographic and clinical features. Similar approaches to disease stratification have been explored within psychosis research [4,5] leading to improved clinical precision and understanding of disease taxonomy, but there is currently a deficit in research on whether psychopathological populations can be stratified according to inflammatory markers and other parameters of immunological function.

Using data from the Avon Longitudinal Study of Parents and Children (ALSPAC) birth cohort study, this research will use non-negative matrix factorisation (NMF), a data-driven, unsupervised clustering technique to identify clinically separable and interpretable subgroups of psychopathology. NMF will be applied to a wide range of immunological, psychiatric, and lifestyle data to explore potential inflammation related subgroups, including blood biomarker data, cognitive data, family mental health history, indices of substance use, body mass index (BMI), and data on individual psychiatric symptoms. Nested cross-validation (CV) will be used to evaluate and validate the stability of each cluster assignment, with additional consensus clustering methodologies used to further validate cluster solutions. Additional statistical methods will be deployed to characterise these subgroups, exploring whether inflammatory responses are related to specific diagnostic categories, persistence of symptoms over time and trajectories of mental health, responses to treatments such as antidepressants, and other relevant characteristics. Associations with measures of physical health and comorbidities will also be explored, such as glucose homeostasis, insulin resistance, autoimmune disease, cardiometabolic and physiological parameters, and additional immunological risk factors.

Application of data-driven methods to highly dimensional psychiatric and immunological health data provides the opportunity to interrogate shared mechanisms of disease between comorbid and overlapping disorders. This approach facilitates exploring drivers of heterogeneity in the clinical population within specific psychiatric conditions such as depression, as well as investigating shared immunological mechanisms between overlapping yet clinically distinct conditions. Understanding these potentially shared mechanisms paves the way for biologically based classifications of psychopathology, enhancing treatment development and furthering clinical understanding of psychiatric conditions.