Statistical Methods for Analyzing Complex, Multi-dimensional Data from Cross-sectional and Longitudinal Mental Health Studies

点击翻译按钮获取中文摘要

Project Summary To address the burden of mental illness, National institute of Mental Health encourages development of computational approaches that provide novel ways to understand relationships among complex, large datasets to further the understanding of the underlying pathophysiology of mental diseases. These datasets are multi- dimensional, including clinical assessments, behavioral symptoms, biological measurements such as neu- roimaging and psychophysiological data. The overall objective of this grant is to advance methodology for analyzing such data to more effectively extract relevant information that are predictive of disease, to improve the understanding of individual variability in clinical and neurobiological phenotypes, and to provide the capac- ity to handle both cross-sectional and longitudinal data. Our proposal will leverage two civilian trauma cohorts recruited through the Grady Trauma Project and the Grady Emergency Department Study, and an external validation cohort from the Hill Center study with a similar distribution of trauma exposure. We propose to develop statistically principled, computationally effi- cient statistical learning methods for addressing key challenges in analyzing these large datasets. Challenges include multi-type outcomes, high dimensional data with sparse signals and high noise levels, spatial and tem- poral dependence of neuroimaging data, and heterogeneous effects across patient population. The scientific premise of this computational psychiatry research is that analytical methods integrating information from brain, behavior, and symptoms will provide much-needed data driven platforms for improving diagnosis and prediction of PTSD and other mental disorders. In this application, we propose: (1) to develop partial generalized tensor regression methods and partial tensor quantile regression methods that can simultaneously achieve accurate prediction of clinical outcomes and efficient feature extraction from high dimensional neuroimaging biomarkers; (2) to develop tensor response quantile regression methods and global inference that can achieve comprehensive and robust understanding of the heterogeneity in high-dimensional neuroimaging phenotypes in terms of environmental factors such as trauma exposure; and (3) to develop and extend methods in Aims 1 and 2 for longitudinal multi-dimensional data that will enable prediction of future post-trauma symptom severity trajectories in terms of neuroimaging biomarkers and robustify the evaluation of the impact of psychophysiological factors on neuroimaging phe- notypes. The proposed methods will be applied to the two Grady studies to address scientific hypotheses relevant to PTSD research. We will use the Hill Center study as an independent validation cohort to evaluate the reproducibility and generalizability of the findings. User-friendly software will be developed. The proposed methodology is generally applicable to many other mental health studies with complex multi-dimensional data.