Pilot Project Program 2021Recipients

PI: Chen Liang, Ph.D.

Because of the large base of population known to have been infected by SARS-CoV-2 in the US, it is estimated that millions of people will experience long-lasting health consequences, denoted as Post-Acute Sequela of SARS-CoV-2 infection (PASC). This projected disease burden will have a profound public health impact during post-COVID-19 care. An emergent national initiative has been established to rapidly advance the understanding of PASC. However, preliminary findings are inconclusive in part because of the heterogeneous findings from existing small and single-center cohorts, the complex pathophysiological nature of SARS-CoV-2 infection, and the time constraints for collecting data from longitudinal clinical trials. To address these challenges, we propose a high-throughput Deep Phenotyping approach to identify and characterize phenotypes (e.g., morphological, physiological, biomedical, socio-behavioral data and nuanced phenotypic traits) of individuals with PASC using multi-modal, longitudinal, extended Electronic Health Records (EHR) data from the South Carolina COVID-19 Cohort Collaborative (S3C). Leveraging clinical Natural Language Processing and Semantic Web Ontology methods, we will design and construct a Graph representational model on top of S3C. We will develop unsupervised Graph-based clustering algorithms, including Deep Learning, and incorporate multi-level and iterative clinical evaluation to detect, identify, and validate Graph-based phenotypes for individuals with PASC. We will exploratory analyze validated phenotypes to answer a list of high-profile research questions (e.g., the prevalence of symptoms, associated multi-system/organ dysfunction, risk factors, health disparity, and treatment/intervention strategies).

PI: Jihong Liu, Ph.D.

Annually in the U.S., nearly 60,000 women experience severe maternal morbidity and mortality (SMMM) with substantial health disparities by race/ethnicity, even prior to the COVID-19 pandemic. The unprecedented COVID-19 pandemic has hit communities of color the hardest. Non-Hispanic Black and Hispanic women who are pregnant appear to have disproportionate SARS-CoV-2 infection and death rates. Questions regarding the impact of the COVID-19 pandemic on racial disparities in SMMM and the dynamics and interactions of multilevel determinants such as broader social contexts of SMMM remain unanswered. The overarching goal of this study is to investigate racial/ethnic disparities in maternal morbidity and mortality (MMM), the contributing roles of social contexts (e.g., residential segregation, racial discrimination in poverty, education, unemployment, and homeownership). We will achieve our goal by studying the distributions of COVID-19 cases and multilevel determinants of maternal health in South Carolina (SC), a state with persistent racial disparities in SMMM within historical systemic Southern contexts. We will build upon our existing statewide SC COVID-19 Cohort (S3C) by creating a pregnancy cohort that will link COVID-19 testing data, electronic health records (EHR), and birth certificate data for all births in SC in 2019-2020 (N~114,000). We will use the socio-ecological framework and employ a retrospective cohort study design to achieve three specific aims: 1) to expand our existing S3C by creating a pregnancy cohort (S3C-P) for SC; 2) to examine the impacts of the COVID-19 pandemic on racial/ethnic disparities in MMM; and 3) to initially examine the association between social contexts (including structural racism and racial discrimination) and the potential widening of racial/ethnic disparities in MMM during the pandemic.

PI: Anita Nag, Ph.D.

The current state of the coronavirus-induced disease (COVID-19) as a global pandemic caused by SARS coronavirus 2 (SARS-CoV-2) demands immediate action to contain and manage the spread of the virus. However, the continuous mutation in the Spike protein present on the viral envelope makes it difficult to localize the spread by using vaccination only. Nonstructural viral proteins have shown minimum mutations and may serve as better targets for repurposing drugs to control the spread of the mutated strains. Nonstructural protein 1 of SARS-CoV-2 is a host shutoff protein that is produced in infected cells once the viral genome is translated and blocks host gene expression by inhibiting host translation and degrading host mRNAs. Here, we propose to use a combination of bioinformatics and computational data mining tools on mass spectrometry-based existing data and the publicly available protein database to identify interacting partners of nsp1 in host cells. The blending of experimentally available data with computational data mining will provide suitable candidates in protein-protein interaction that could be disrupted by drug therapy in the future.

PI: Shan Qiao, Ph.D.

Increasing literature reports that a considerable proportion of COVID-19 positive people show symptoms that last
for weeks or even months after recovery from acute illness (i.e., past-acute-sequela of SARS-CoV-2 [PASC]). The persistent and multi-organ, multi-system manifestations of PASC delay COVID-19 patients’ return to usual health and normal life. Building large prospective longitudinal cohorts of COVID-19 patients is a critical step to advance our understanding of PASC. In this study, we will test the feasibility of identifying COVID-19 patient cohorts using US-based Twitter data retrieved from 2/01/2020 to 12/31/2021 (Aim 1) and further ascertain, in a preliminary fashion, if we can use longitudinal information collected from this cohort for PASC symptoms analysis (Aim 2). For Aim 1, we will design and validate a set of queries that can be used to retrieve posts that are highly indicative of COVID-19 positive Twitter users; develop and evaluate a supervised machine learning approach that can accurately distinguish between real COVID-19 positive tweets and false positives; and design an end-to-end pipeline for collecting longitudinal data from the identified COVID-19 positive Twitter users. For Aim 2, we will extract all PASC symptom terms from the longitudinal Twitter data of the COVID-19 patient cohort; validate these symptoms with the assistance of domain experts (infectious disease clinicians); and demonstrate the frequency distribution of symptoms, symptom clusters, and temporal pattern of symptoms since the onset of COVID-19 diagnosis and compare the results with the evidence-based on EHR data (i.e., NIH N3C dataset and South Carolina COVID-19 cohort data). Social media has a rich description of related symptoms, a longitudinal nature to allow flexible observational/analysis windows for PASC, historical information on health-related information (e.g., lifestyles and behaviors), and relatively low costs. Using Twitter data to identify a COVID-19 patient cohort could serve as a complementary and innovative strategy in data collection and analysis for investigating PASC symptoms. Our study will lay important groundwork for several initiatives at the intersection of social media data mining, COVID-19 recovery, and PASC-related healthcare services.

PI: Elizabeth Regan, Ph.D.

Planning for End-of-Life (EOL) care includes a patient’s verbal and written instructions about the care he/she wishes to receive if no longer able to make his/her own decisions. The goal is to optimize care for dying patients to prevent and relieve pain and suffering. In 2016, the Centers for Medicare and Medicaid Services
(CMS) started paying providers for EOL care discussions with patients. However, studies have shown a mixed association between EOL care planning and EOL care intensity. Despite spending about 30% of Medicare’s $800 billion annual budget on EOL care, patients continue to receive high-intensity or aggressive, low-value, and costly EOL care discordant with the wishes of patients and family members. The proposed study will assess the relationship between oral and written EOL care instructions and EOL care intensity after controlling for several supply-side and demand-side factors. Several studies have identified a need for such research to address recognized gaps in the literature. Our primary aim is to identify the determinants of quality EOL care to design interventions that will lead to more consistent delivery of care that is concordant with the wishes of patients and their family caregivers and avoids emergency and intensive care that may provide little benefit with potential adverse side effects and significantly increased costs.

This interdisciplinary, convergent research between scientists and clinicians from the University of South Carolina (UofSC) and Prisma Health Upstate addresses the complex problem of end-of-life care from a systemic care delivery perspective. The study has four interlinked objectives are to:
1. determine the frequency of preference-discordant EOL care and frequency and predictors of aggressive EOL care.
2. analyze the intersection of care delivery with EOL care planning and predictors of high-intensity care.
3. study implications for new models of delivery for end-of-life care to better optimize the patient experience.
4. provide a foundation for a more comprehensive research proposal (to National Institute of Health (NIH) or Agency for Healthcare Research and Quality (AHRQ)) to conduct prospective research with patients, family caregivers, and providers to design and test interventions for new care delivery models that can
improve the quality of end-of-life care at Prisma Health.

This novel research will determine the predictors of high-intensity or aggressive EOL care using a comprehensive framework proposed by Prigerson and Maciejewski. The framework uses a hierarchical nested approach to explain the relationship between the several demand-side and supply-side EOL care intensity determinants. The nested model suggests that the regional determinants of EOL care intensity influence the healthcare institutions’ practices, which affect providers’ practices, which affect patients’, caregivers’, and family members’ attitudes and behaviors towards EOL care. The study will review the existing Prisma Health Electronic Medical Records (EMR) and chart data from patients who died between January 2016-July 2021. We will also link this date with the Dartmouth Atlas of Healthcare, Area Health Resource File, and hospital data from the SC Revenue and Fiscal Affairs databases to include regional demand-side and supply-side determinants of EOL care. Outcomes will inform a more holistic view of the factors impacting EOL care delivery

PI: Gregory Trevors, Ph.D.

Vaccine hesitancy (VH) threatens to undermine the effectiveness of SARS-CoV-2 vaccines and may result in prolonging the current pandemic, which every day is estimated to cause hundreds of deaths in the US and will cost $16 trillion over the next decade under optimistic assumptions for containment. Effective and scalable interventions are urgently needed to promote vaccine acceptance. The long-term goal of this project is to inform the design of subsequent digital health promotion interventions that will dynamically personalize to participants based on empirical insights gained from machine learning analyses on our prior intervention data. Equipped with these insights, we will be able to capitalize on the growth potential of intelligent gamified health promotion
interventions. To accomplish these long-term goals, the immediate objectives in this proposal are (1) to identify distinct vaccine hesitancy subgroups of our prior intervention participants; and (2) identify optimal content that is associated with an increased likelihood of healthy behavioral intentions and attitudes, including vaccine intent and confidence, for each VH subgroup. We will draw on and integrate data from our three digital game interventions that totals over 220,000 participants and numerous features across digital platforms. The overall results from this project will lay the foundation for intelligent and personalized digital health promotion interventions that are optimized for positive engagement and healthy behavioral intentions and attitudes.

Pi: Dezhi Wu, Ph.D.

Substance use among young pregnant women is a significant public health concern. During the COVID-19 pandemic, recent studies have shown an alarming increase in substance use due to economic hardships, persistent social isolation, and prolonged fear and uncertainty attributed to stay at home orders with no definitive timeline for a return to normalcy for social gatherings. How the pandemic has impacted vulnerable pregnant women during the pandemic is unknown but critical, since pregnant women who use substances, such as alcohol and opioids, etc., can negatively impact fetal birth outcomes including low birth weight, poor brain development, inadequate nervous system, poor behavioral and memory issues. This proposed project aims to use natural language processing, text mining, and machine learning techniques to extract just-in-time substance use (SU) data among pregnant women on Twitter, and then explore their communication patterns, risky health perceptions, sentiment, and maternal and fetal health outcomes. These results will serve to inform clinicians, public health officials, and policymakers on intentional messaging needed for substance use prevention among this vulnerable group.