NIMH seeks to ?identify biomarkers and behavioral indicators with high predictive value, as early in the course of illness development as possible?, in order to reduce the overall burden of mental illness. However, the number of potentially important psychological, environmental, and biological factors of mental health disorders is vast, and a key challenge is to narrow down to the most important predictors of disorder. This challenge is made especially difficult as 1) recent advances in neuroscience begin to reveal neural substrates of psychopathology, and 2) many predictors are themselves correlated, making it difficult to disentangle which factors are reliably related to disease, after controlling for other factors. Currently used statistical methods are inadequate to overcome this challenge. Powerful Bayesian variable selection methods, called stochastic search variable selection (SSVS), can be used to identify predictors with the most robust relationships for a given criterion, however these methods have not been developed for use in psychology and are currently only available to specialized statisticians. The goal of this project is to develop guidelines to enable mental health researchers to use SSVS to overcome current methodological barriers. I will also develop user-friendly online applications to make SSVS easily available. For the first Aim of this study I will use computer simulation studies to evaluate how SSVS works across a range of conditions and develop guidelines and software for researchers to use. In the second Aim of this study I will apply SSVS to predict obsessive compulsive disorder (OCD) symptoms in the Nathan Kline Institute Rockland sample, which is a large, publicly available database. OCD is a common, chronic, and debilitating disorder. Much regarding risk for OCD remains unknown, which limits efforts aimed at treatment and prevention. Previous research to identify potential risk factors and triggers for illness onset has relied heavily on evaluation of individuals long after symptoms began. The predictors in this sample include a wide range of theoretically-derived risk factors, including measures of potential psychological vulnerabilities, brain connectivity, stressful life events, and key comorbidities. This proposed research is embedded in a training and mentoring plan that will provide training in 1) the etiology and assessment of psychopathology, 2) neuroscience approaches to determine neural substrates of psychopathology, and 3) Bayesian variable selection methods. This K01 mentored research award will provide the training, time and resources for me to make substantial advances towards addressing this important problem and establish myself as an independent, R01-funded investigator.
There is an urgent need for powerful and effective statistical tools to determine the most important risk factors of mental illness, especially as the list of potential risk factors grows. The proposed research will develop guidelines and freely-available online applications to allow researchers to use novel Bayesian variable selection methods to overcome current challenges to reliably predict mental illness. These methods are then applied to understand important predictors of Obsessive Compulsive Disorder symptoms in a large sample of adults across the lifespan. The findings of this research can inform efforts to identify and intervene to help at risk individuals to ultimately reduce the burden of mental illness.
