The theme of this dissertation focuses on developing statistical models to learn progression dynamics and mechanisms of neuropsychiatric disorders using data from various domains. Due to limited knowledge about the underlying pathological processes in neurological disorders, it remains a challenge to establish reliable diagnostic criteria and predict disease prognosis in the presence of substantial phenotypic heterogeneity. As a result, current diagnosis and treatment of neurological disorders often rely on late-stage clinical symptoms, which poses barriers for developing effective interventions at the premanifest stage. It is crucial to characterize the temporal disease progression course and study the underlying mechanisms using clinical assessments, blood biomarkers, and neuroimaging biomarkers to evaluate disease stages, identify markers that are useful for early clinical diagnosis, compare or monitor treatment effects and accelerate drug discovery. We propose three projects to tackle challenges in leveraging multi-domain biomarkers and clinical symptoms to learn disease dynamics and progression of neurological disorders: (1) A nonlinear mixture model with subject-specific random inflection points to jointly fit multiple longitudinal markers and estimate marker progression trajectories in a single modality; (2) A multi-layer exponential family factor model integrating multi-domain data to learn lower-dimensional latent space of disease impairment and fully map disease risk and progression; (3) A latent state space model that jointly analyzes multi-channel EEG signals and learns dynamics of different sources corresponding to brain cortical activities. In addition, motivated by the ongoing COVID-19 pandemic, we propose a parsimonious survival-convolution model to predict daily new cases and estimate the time-varying reproduction numbers to evaluate effects of mitigation strategies. In the first project, we propose a nonlinear mixture model with random time shifts to jointly estimate long-term progression trajectories using multivariate discrete longitudinal outcomes. The model can identify early disease markers, their orders of occurrence, and the rates of impairment. Specifically, a latent binary variable representing disease susceptibility status incorporates subject covariates (e.g., biological measures) in the mixture model to capture between-subject heterogeneity. Measures of disease impairment for susceptible patients are modeled jointly under the exponential family framework. Our model allows for subject-specific and marker-specific inflection points associated with patients' characteristics (e.g., genetic mutation) to indicate a critical time when the fastest degeneration occurs. Furthermore, it uses subject-specific latent scores shared among markers to improve efficiency. The model is estimated using an EM algorithm. Extensive simulation studies are conducted to demonstrate validity of the proposed method and algorithm. Lastly, we apply our method to the Parkinson's Progression Markers Initiative (PPMI), and show utility to identify early disease signs and compare clinical symptomatology for the genetic form of Parkinson's Disease (PD) and idiopathic PD. In the second project, we tackle challenges to leverage multi-domain markers to learn early disease progression of neurological disorders. We propose to integrate heterogeneous types of measures from multiple domains (e.g., discrete clinical symptoms, ordinal cognitive markers, continuous neuroimaging and blood biomarkers) using a hierarchical Multi-layer Exponential Family Factor (MEFF) model, where the observations follow exponential family distributions with lower-dimensional latent factors. The latent factors are decomposed into shared factors across multiple domains and domain-specific factors, where the shared factors provide robust information to perform behavioral phenotyping and partition patients into clinically meaningful and biologically homogeneous subgroups. Domain-specific factors capture the remaining unique variations for each domain. The MEFF model also captures the nonlinear trajectory of disease progression and order critical events of neurodegeneration measured by each marker. To overcome computational challenges, we fit our model by approximate inference techniques for large-scale data. We apply the developed method to Parkinson's Progression Markers Initiative (PPMI) data to integrate biological, clinical and cognitive markers arising from heterogeneous distributions. The model learns lower-dimensional representations of Parkinson's disease and the temporal ordering of the neurodegeneration of PD. In the third project, we propose methods that can be used to analyze multi-channel electroencephalogram (EEG) signals intensively measured at a high temporal resolution. Modern neuroimaging technologies have substantially advanced the measurement of brain activities. EEG as a non-invasive neuroimaging technique measures changes in electrical voltage on the scalp induced by cortical activities. With its high temporal resolution, EEG has emerged as an increasingly useful tool to study brain connectivity. Challenges with modeling EEG signals of complex brain activities include interactions among unknown sources, low signal-to-noise ratio and substantial between-subject heterogeneity. In this work, we propose a state space model that jointly analyzes multi-channel EEG signals and learns dynamics of different sources corresponding to brain cortical activities. Our model borrows strength from spatially correlated measurements and uses low-dimensional latent sources to explain all observed channels. The model can account for patient heterogeneity and quantify the effect of a subject's covariates on the latent space. The EM algorithm, Kalman filtering, and bootstrap resampling are used to fit the state space model and provide comparisons between patient diagnostic groups. We apply the developed approach to a case-control study of alcoholism and reveal significant attenuation of brain activities in response to visual stimuli in alcoholic subjects compared to healthy controls. Lastly, motivated by the ongoing COVID-19 pandemic, we propose a robust and parsimonious survival-convolution model aiming to predict COVID-19 disease course and compare effectiveness of mitigation measures across countries to inform policy decision making. We account for transmission during a pre-symptomatic incubation period and use a time-varying effective reproduction number to reflect the temporal trend of transmission and change in response to a public health intervention. We estimate the intervention effect on reducing the infection rate using a natural experiment design and quantify uncertainty by permutation. In China and South Korea, we predicted the entire disease epidemic using only early phase data (two to three weeks after the outbreak). A fast rate of decline in reproduction number was observed and adopting mitigation strategies early in the epidemic was effective in reducing the infection rate in these two countries. The nationwide lockdown in Italy did not accelerate the speed at which the infection rate decreases. In the United States, the reproduction number significantly decreased during a 2-week period after the declaration of national emergency, but declines at a much slower rate afterwards. If the trend continues after May 1, COVID-19 may be controlled by late July. However, a loss of temporal effect (e.g., due to relaxing mitigation measures after May 1) could lead to a long delay in controlling the epidemic.

Preterm birth is a key indicator of maternal and child health, affecting 1 in 10 deliveries in the United States (US) and contributing to long-term morbidity and healthcare costs. The COVID-19 pandemic and policies to mitigate the spread of infection may have indirectly impacted preterm birth, but the results of early epidemiological studies were mixed and declines were largely concentrated in high-income countries and populations. Moreover, while most studies focused on stress-related pathways associated with lockdown policies, healthcare disruptions may have also played a role. The goal of this dissertation was to investigate changes in preterm birth and healthcare disruptions related to the COVID-19 lockdown in a low-income population in the US. In the first aim, I conducted a systematic review of the literature on the pandemic and preterm birth, with a focus on studies that examine heterogeneity by income. In the second aim, New York State (NYS) Medicaid claims were used to examine changes in preterm birth rates during the state’s lockdown policy (NYS on PAUSE) using difference-in-difference methods. In the third aim, changes in preterm were further stratified into those that were spontaneous or medically induced, which may reflect a healthcare pathway. Weekly rates of healthcare utilization, antenatal surveillance, and maternal complications were also assessed using interrupted time series models to characterize healthcare disruptions over the course of the lockdown and across the state. Results from the systematic review documented the rapid growth in research on this topic since the beginning of pandemic. Among the 67 articles included, most reported some decline in preterm birth rates; however, there was large variation by country, methods of exposure assessment, and onset of delivery. Only seven studies focused on differences by individual income (or income proxies) and those that did were inconsistent. Results from Aim 2 suggested that NYS on PAUSE was associated with nearly a percentage point decline in preterm birth rates in the Medicaid-insured population, without a concomitant increase in stillbirth. Aim 3 demonstrated that the change in preterm was largely driven by declines in medically induced preterm. Interrupted time series models showed substantial, but time-limited, declines in pregnancy-related healthcare utilization at the beginning of NYS on PAUSE. Overall, the findings in this dissertation suggest there were modest declines in preterm birth during the COVID-19 lockdown among low-income women in NYS, particularly in medically induced preterm. Healthcare disruptions were common for Medicaid-insured women and may partially explain the reduction in preterm birth in this population. Future research is needed to determine whether this change was positive for some and negative for others, and what that might mean for efforts to improve pregnancy outcomes in the future.

The SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) pandemic has led to widespread socioeconomic and clinical damage. The coalescent response from the global scientific community has been unparalleled, both in speed and furor. Numerous efficacious interventions have been developed and deployed, including several vaccines, antibody therapies, and drugs. Yet, SARS-CoV-2 embodies the quintessential virological issue which threaten these achievements; rapid evolution in the face of selective pressure. This dissertation investigates such adaptations by SARS-CoV-2, and accordingly, modalities to combat this virus despite such evasive measures. To this end, we first studied the antigenic properties of several members of the B.1.1.529 or Omicron lineage of SARS-CoV-2. We observed that B.1.1.529.1 (BA.1), B.1.1.529.1.1 (BA.1.1), and B.1.1.529.2 (BA.2) are the most antibody resistant SARS-CoV-2 variants to-date, while being antigenically unique between each other. Consequently, we turned to explore modalities which may withstand such formidable resistance. We undertook some of the first explorations of a heterologous booster vaccination regimen, finding expanded breadth and potency against SARS-CoV-2, suggesting it may be one simple measure that could be utilized. We also sought to identify broadly neutralizing SARS-CoV-2 antibodies, isolating several with breadth against coronaviruses beyond that of SARS-CoV-2. One of these antibodies, 10-40, was determined to be the broadest receptor-binding domain-directed antibody reported to-date. Finally, we examined an alternative viral target, the 3CL protease. We discovered several SARS-CoV 3CL protease inhibitors that could be repurposed for inhibition of SARS-CoV-2 and determined their crystal structures, which could allow for their use as lead compounds. We further developed and conducted a deep mutational scan of the 3CL protease to examine the activity of all possible single point mutants, revealing that the enzyme had unexpected malleability, as well as several conserved sites that may be targeted by future inhibitors. The SARS-CoV-2 pandemic has been a remarkable trial, but has also served to demonstrate the good that science can do. We hope that this work has been a small contribution among such difficult times.

This study aimed to i) investigate the prevalence of unhealthy attributes, common diseases, and inequities in social determinants of health across a large and representative sample of adults in New York City; and ii) identify common key predictors of COVID-19 seropositivity by comparing various regression models using a hierarchical regression method among a sample of New York City adults. The study will use the New York City Community Health Survey (NYC CHS) 2020 dataset for this analysis. An exploratory approach is used to data to understand the social, environmental, and individual determinants of health in the New York City population at the peak of the pandemic and their effects on COVID-19 seropositivity. The study also emphasizes on using a predictive modeling approach to develop and select an optimal ML model that accurately predicts COVID-19 seropositivity from various ML algorithms. Hierarchical logistic regression was carried out on a sample of 928 participants. It was found that age group 65-75, Black and Hispanic race and being born in the US were statistically significant factors in model 1 of the hierarchical regression where only socioeconomic factors were considered. With the inclusion of health behaviors, tobacco smoking behaviors, and physical activity were statistically significant. In the full model, BMI, asthma prevalence, and suicidal thoughts were statistically significantly correlated with COVID-19 seropositivity. The findings are consistent with public health literature highlighting the importance of healthy behaviors and public health efforts in maintaining overall health and immunity.

The COVID-19 pandemic elucidated geographical disparities in COVID-19 burden on a globalscale. Geographical disparities in adverse COVID-19 outcomes may suggest population-level drivers of disease, such as environmental exposures. Epidemiological literature provides strong evidence that greater exposure to ambient air pollution, an environmental exposure, is associated with a greater risk of COVID-19 hospitalization and fatality. The pathways by which ambient air pollution exposure influences adverse COVID-19 outcomes are currently unknown. I propose that cardiovascular morbidity is relevant in this pathway, given that cardiovascular morbidity is a predominant risk factor of adverse COVID-19 outcomes, and there are strong and consistent associations between air pollution and cardiovascular morbidity. I suggest that the role of cardiovascular morbidity will be different for historical air pollution (period > 30 days) and short-term air pollution (period < 30 days). By proposing clear causal structures for the relationship between air pollution and adverse COVID-19 outcomes, we can explicate how air pollution leads to greater COVID-19 burden and address the larger goal of reducing geographic disparities in adverse COVID-19 outcomes. This dissertation is comprised of three specific aims. For the first aim, I performed a systematic review of the literature that examined the relationship between ambient air pollution and individual-level adverse COVID-19 outcomes. I identified if and how researchers conceptualized the causal role of comorbidities, specifically cardiovascular morbidities, in the relationship between air pollution and adverse COVID-19 outcomes. For the second aim, I examined if cardiovascular morbidity mediates the relationship between historical air pollution and adverse COVID-19 outcomes. For the third aim, I examined if there was evidence of synergistic interaction between short-term air pollution and cardiovascular morbidity in influencing the risk of adverse COVID-19 outcomes, suggesting that the effect of both short-term air pollution and cardiovascular morbidity on adverse COVID-19 is greater than the sum of the individual effects. In conducting the first aim, I used Covidence, a software used to manage systematic reviewstudies, to identify studies that examined the relationship between ambient air pollution exposure and individual-level adverse COVID-19, using the Embase, MEDLINE, and Web of Science databases. In conducting the empirical aims, I used a retrospective cohort study design using INSIGHT-Clinical Research Network (CRN) data, a harmonized repository of inpatient electronic health records in New York City (NYC) across metropolitan healthcare systems (3/1/2020-2/28/2021). INSIGHT-CRN included data pertaining to sociodemographics, diagnoses, outcomes, and residential ZIP Code to link air pollution exposure. For the second aim, I used the New York City Community Air Survey (NYCCAS) to estimate historical air pollution exposure to particulate matter (PM2.5), black carbon (BC), nitrogen dioxide (NO₂), and ozone (O₃) on a ZIP Code level (2009-2019). For the third aim, I used the 2020 Environmental Protection Agency (EPA) Community Multiscale Air Quality (CMAQ) downscaler modeled data, which estimated 2020 daily exposure to PM2.5 and O3 on a census tract level. I aggregated the census tract data to ZIP Code using a spatial weighting approach and estimated short-term air pollution as a 7-day average of daily PM2.5 and O3 exposure prior to patient hospitalization. For the first aim, the systematic review included 42 studies that examined the relationship between ambient air pollution, such as exposures to PM2.5, NO₂, and O₃, and individual-level adverse COVID-19, such as hospitalization, intensive care unit (ICU) admission, intensive respiratory support (IRS), and fatality. The studies were primarily retrospective cohort study designs, and were conducted in the United States and Europe (2020 to 2021). The majority of studies adjusted for cardiovascular morbidity without causal role specification, whereas some studies identified cardiovascular morbidity as a mediator or an effect modifier. For the second aim, I found evidence of cardiovascular morbidity mediating the relationship between historical air pollution and risk of acute respiratory distress syndrome (ARDS), dialysis use, ventilation use, and COVID-19 fatality, but not risk of pneumonia from March to June 2020, within areas of greater hospital catchment. Indirect effects suggest that historical air pollution increases the risk of atrial fibrillation and myocardial infarction, which increases risk of adverse COVID-19. For the third aim, I found evidence of synergistic interaction between short-term PM2.5 and presence of cardiovascular morbidities for only risk of COVID-19 pneumonia, in the latter half of 2020. Overall, there was evidence that cardiovascular morbidity mediates the relationship betweenhistorical air pollution and more severe COVID-19 outcomes, while cardiovascular morbidity synergistically interacts with short-term air pollution for risk of acute respiratory infections, such as pneumonia. This dissertation assesses the pathways by which air pollution may influence risk of adverse COVID-19, in better examining the causal role of cardiovascular morbidity. Knowledge gained could be used to mitigate population-level vulnerabilities to air pollution, and encourage population-level pandemic preparedness in the future.

COVID-19 continues to have a major impact on US health and society. Robust research on the epidemiology of acute and post-acute COVID-19 remains fundamentally important to informing policy makers, scientists, as well as the public. This dissertation reports on the development of a large, diverse, United States general population-based meta-cohort with standardized, prospective ascertainment of SARS-CoV-2 and COVID-19, integrated with comprehensive pre-pandemic phenotyping from 14 extant cohort studies. Meta-cohort data were used to investigate risk and resilience factors for incident severe (hospitalized or fatal) and non-severe COVID-19 and correlates of time-to-recovery from SARS-CoV-2 infection. Results support the major acute and post-acute public health impact of COVID-19 and the vital role of modifiable (e.g., obesity, diabetes, cardiovascular disease) and non-modifiable (e.g., age, sex) risk factors for adverse COVID-19 outcomes. Findings suggest that standard primary care interventions—including obesity and cardiometabolic disease prevention and treatment, depression care, and vaccination—remain fundamental to COVID-19 risk mitigation among US adults. Given its longitudinal design and comprehensive pre-pandemic and pandemic-era measurements, the meta-cohort is well suited to support ongoing work regarding the public health impact of SARS-CoV-2 infection, COVID-19, post-acute sequelae, and pandemic-related social and behavioral changes across multiple health domains.

Background: The COVID-19 pandemic posed an unparalleled challenge to worldwide public health systems, characterized by its high transmissibility and the initial absence of accessible testing, treatments, and vaccines. The deficiency in public awareness and the scarcity of readily available public health information regarding this century's disaster further intensified the critical need for innovative solutions to bridge these gaps. In response, Shaman Labs1,2, leveraging its deep expertise in forecasting for influenza3, Ebola, and various SARS viruses, initiated the development of country-wide COVID-19 projections within weeks following the WHO's declaration of the pandemic4–6. Almost immediately thereafter, it became necessary to create a sophisticated online platform—a system capable of displaying county-specific COVID-19 forecasts, including daily estimated infections, cases, and deaths. This platform was designed to allow users to select any county, state, or national geography and compare it with another, under various scenarios of social distancing measures. Additionally, the architecture of this system was required to facilitate the regular integration of updated data, ensuring the tool's ongoing relevance and utility. Columbia University's data visualization system aimed to communicate epidemiological forecasts to various stakeholders. At the onset of the COVID-19 pandemic, amid escalating uncertainty and the pressing need for reliable data, Dr. Rundle played a pivotal role in briefing key stakeholders on the unfolding crisis. His efforts were directed towards providing Congressman Ron Johnson, Chairman of the U.S. Senate Committee on Homeland Security & Governmental Affairs, and Congresswoman Anna Eshoo, as well as their staff, with up-to-date projections and analyses derived from the Classic Data Visualization tools. Dr. Rundle’s consultative role extended to a diverse array of institutions including the U.S. Army Corps of Engineers, the U.S. Air Force, and the Federal Reserve Board, as well as advising private entities such as Pfizer, MetLife, and Unilever. His expertise facilitated informed planning and response efforts across various levels of government and sectors, underscoring the critical role of sophisticated data visualization from the earliest stages of the pandemic. This Integrated Learning Experience (ILE) examines the development and implementation of the Time Machine platform, focusing on its application in visualizing and analyzing COVID-19 epidemiological forecasts. The study explores methods for improving forecast data presentation, analysis, and accuracy assessment. Methods: The body of this work unfolds through a series of critical chapters that collectively address the multifaceted functionality and impact of the Time Machine platform. Initially, the work focuses on the construction of the Time Machine platform, a web-based R interactive user interface coupled with cloud-based database system, specifically tailored for the intuitive visualization of epidemiological forecasts, detailing the technical and design considerations essential for enabling users to interpret complex data more effectively. Following this, the implementation of a rigorous data-discovery framework is presented, examining case reporting inconsistencies across different regions, using low-level GitHub and Windows scripting technologies, thereby highlighting the significance of accurate data collection and the impact of discrepancies on public health decisions. The narrative then transitions to the implementation of advanced statistical models, such as strictly proper scoring and weighted interval scoring, to assess the accuracy of the forecasts provided by the Time Machine platform, using a dedicated R library and testing with the help of MS Excel sandbox, underscoring the importance of reliable predictions in the management of public health crises. Lastly, a detailed analysis is conducted, encompassing countrywide data (3142 counties) over an extended period (147 weeks), utilizing Generalized Estimating Equations (GEE) to identify key predictors that influence forecast accuracy, offering valuable insights into the factors that either enhance or detract from the reliability of epidemiological predictions. Results: The deployment of the Classic Data Visualization and the subsequent evolution of the Time Machine platform have significantly advanced epidemiological forecast visualization capabilities. The Time Machine platform was designed with an automated data refresh system, allowing for regular updates of epidemiological forecast data and reported actuals. The project developed tools for monitoring and evaluating the quality of public health reporting, aiming to improve the accuracy and timeliness of data used in public health decisions. Additionally, the research implemented methods for standardizing forecast accuracy assessments, including the normalization of scores to enable comparisons across different geographical scales. These approaches were designed to support both local and national-level pandemic response efforts. The accuracy analyses throughout different phases of the pandemic revealed a 42% improvement in forecast accuracy from Phase 1 to Phase 7. Larger populations (27% increase per unit increase on a base-10 logarithmic scale) and higher county-level activity (45% increase from the lowest to the highest quartile) resulted in better estimations. Additionally, the analysis highlighted the significant impact of reporting quality on forecast accuracy. On the other hand, the study identified the challenges in predicting case surges, showing a 27% decline in accuracy during periods of rising infections compared to declining periods. The regression results highlight the potential benefits of improving data collection and providing timely feedback to forecasting teams. Conclusion: This study demonstrates the potential of advanced data visualization and accuracy measurement techniques in improving epidemiological forecasting. The findings suggest that factors such as urbanicity, case reporting quality, and pandemic phase significantly influence forecast accuracy. Further research is needed to refine these models and enhance their applicability across various public health scenarios.