Frontiers in Neuro-Epidemiology: Advancing Methods Across Diverse Modalities of Data

Abstract

Defined as the study of the distribution and risk factors of neurological disorders, Neuro-epidemiology has been a widely evolving field for the last few decades. The earliest epidemiological studies of neurological disorders attempted to explore often poorly understood neurological manifestation of various conditions. Due to the rather blunt nature of scientific tools in earlier times, most preliminary studies limited themselves to analyzing cross-sectional counts and identifying crude risk factors such as age, sex, and location. Later, there was value seen in the utilizing longitudinal cohorts to better appreciate the temporal occurrences of many diseases including those of neurological origin. While limited in certain aspects, such data provided an additional means of appreciating the development of neurological disease in populations overtime with respect to measured variables. While the epidemiological study of neurological disease evolved, the scientific community oversaw an exponential growth in medical technology and practice including radiological imaging, genotyping, human biomarker assays, and robust diagnostic criteria. Such technological advancements enhanced the ability to study the etiological pathways of many neurological disorders from granular biochemical pathways to neuroimaging aberrations. These tools also provided epidemiologists greater ability to explore their once blunt risk factor associations in finer detail within large populations. The recent explosion of genetic, proteomic, imaging data in large cohorts in conjunction with developments in predictive modeling, such as machine learning, and causal inference, particularly in the field of statistical genetics, have presented a unique opportunity to interrogate new questions. Considering such developments, this thesis explores the cutting edge of the various subfields within modern neuro-epidemiology to answer some of the most pertinent questions in the field using most robust and innovative methods. In Chapter One, we observe the continued relevance of population level data and analyses by determining the trends in acute ischemic stroke presentations and treatments during the 2020 Coronavirus pandemic. In Chapter Two, we leverage geometric machine learning methods in a neuroimaging dataset to explore how we can improve the generalizability of diagnostic computer vision models across various neurological disease. In Chapter Three, we utilize population level genetic and proteomic data in various cohorts to understand possible interaction between protein levels and neuroimaging phenotypes with an eye towards understanding implicated biological pathways and disease phenotypes.