Advancing Real-time Pandemic Forecasting Using Large Language Models: A COVID-19 Case Study (preprint)

Forecasting the short-term spread of an ongoing disease outbreak is a formidable challenge due to the complexity of contributing factors, some of which can be characterized through interlinked, multi-modality variables such as epidemiological time series data, viral biology, population demographics, and the intersection of public policy and human behavior. Existing forecasting model frameworks struggle with the multifaceted nature of relevant data and robust results translation, which hinders their performances and the provision of actionable insights for public health decision-makers. Our work introduces PandemicLLM, a novel framework with multi-modal Large Language Models (LLMs) that reformulates real-time forecasting of disease spread as a text reasoning problem, with the ability to incorporate real-time, complex, non-numerical information that previously unattainable in traditional forecasting models. This approach, through a unique AI-human cooperative prompt design and time series representation learning, encodes multi-modal data for LLMs. The model is applied to the COVID-19 pandemic, and trained to utilize textual public health policies, genomic surveillance, spatial, and epidemiological time series data, and is subsequently tested across all 50 states of the U.S. Empirically, PandemicLLM is shown to be a high-performing pandemic forecasting framework that effectively captures the impact of emerging variants and can provide timely and accurate predictions. The proposed PandemicLLM opens avenues for incorporating various pandemic-related data in heterogeneous formats and exhibits performance benefits over existing models. This study illuminates the potential of adapting LLMs and representation learning to enhance pandemic forecasting, illustrating how AI innovations can strengthen pandemic responses and crisis management in the future.

FABP4 as a Therapeutic Host Target Controlling SARS-CoV2 Infection (preprint)

Host metabolic fitness is a critical determinant of infectious disease outcomes. In COVID-19, obesity and aging are major high-risk disease modifiers, although the underlying mechanism remains unknown. Here, we demonstrate that fatty acid binding protein 4 (FABP4), a critical regulator of metabolic dysfunction in these conditions, regulates SARS-CoV2 pathogenesis. Our study revealed that elevated FABP4 levels in COVID-19 patients strongly correlate with disease severity. In adipocytes and airway epithelial cells we found that loss of FABP4 function by genetic or pharmacological means impaired SARS-CoV2 replication and disrupted the formation of viral replication organelles. Furthermore, treatment of infected hamsters with FABP4 inhibitors alleviated lung damage and fibrosis and reduced lung viral titers. These results highlight a novel host factor critical for SARS-CoV2 infection and the therapeutic potential of FABP4-targeting agents in treating COVID-19 patients.

Unraveling Attributes of COVID-19 Vaccine Hesitancy in the U.S.: A Large Nationwide Study (preprint)

SARS-CoV-2 vaccines are powerful tools to combat the COVID-19 pandemic, but vaccine hesitancy threatens these vaccines’ effectiveness. To address COVID-19 vaccine hesitancy and ensure equitable distribution, understanding the extent of and factors associated with vaccine hesitancy is critical. We report the results of a large nationwide study conducted December 2020-January 2021 of 34,470 users from COVID-19-focused smartphone-based app How We Feel on their willingness to receive a COVID-19 vaccine. Nineteen percent of respondents expressed vaccine hesitancy, the majority being undecided. Vaccine hesitancy was significant among females, younger people, minority and low-income communities, healthcare and essential workers, rural residents, geographical regions with higher COVID-19 burden, those who did not use protective measures, and those who did not receive COVID-19 tests. Our findings support the need for targeted efforts to develop education and outreach programs to overcome vaccine hesitancy and improve equitable access, diversity, and inclusion in the national response to COVID-19.

Multi-resolution characterization of the COVID-19 pandemic: A unified framework and open-source tool (preprint)

Amidst the continuing spread of COVID-19, real-time data analysis and visualization remain critical to track the pandemic's impact and inform policy making. Multiple metrics have been considered to evaluate the spread, infection, and mortality of infectious diseases. For example, numbers of new cases and deaths provide measures of absolute impact within a given population and time frame, while the effective reproduction rate provides a measure of the rate of spread. It is critical to evaluate multiple metrics concurrently, as they provide complementary insights into the impact and current state of the pandemic. We describe a unified framework for estimating and quantifying the uncertainty in the smoothed daily effective reproduction number, case rate, and death rate in a region using log-linear models. We apply this framework to characterize COVID-19 impact at multiple geographic resolutions, including by US county and state as well as by country, demonstrating the variation across resolutions and the need for harmonized efforts to control the pandemic. We provide an open-source online dashboard for real-time analysis and visualization of multiple key metrics, which are critical to evaluate the impact of COVID-19 and make informed policy decisions.

HYPER: Group testing via hypergraph factorization applied to COVID-19 (preprint)

Large scale screening is a critical tool in the life sciences, but is often limited by reagents, samples, or cost. An important challenge in screening has recently manifested in the ongoing effort to achieve widespread testing for individuals with SARS-CoV-2 infection in the face of substantial resource constraints. Group testing methods utilize constrained testing resources more efficiently by pooling specimens together, potentially allowing larger populations to be screened with fewer tests. A key challenge in group testing is to design an effective pooling strategy. The global nature of the ongoing pandemic calls for something simple (to aid implementation) and flexible (to tailor for settings with differing needs) that remains efficient. Here we propose HYPER, a new group testing method based on hypergraph factorizations. We provide theoretical characterizations under a general statistical model, and exhaustively evaluate HYPER and proposed alternatives for SARS-CoV-2 screening under realistic simulations of epidemic spread and within-host viral kinetics. We demonstrate that HYPER performs at least as well as other methods in scenarios that are well-suited to each method, while outperforming those methods across a broad range of resource-constrained environments, being more flexible and simple in design, and taking no expertise to implement. An online tool to implement these designs in the lab is available at http://hyper.covid19-analysis.org .

Federated Learning used for predicting outcomes in SARS-COV-2 patients (preprint)

‘Federated Learning’ (FL) is a method to train Artificial Intelligence (AI) models with data from multiple sources while maintaining anonymity of the data thus removing many barriers to data sharing. During the SARS-COV-2 pandemic, 20 institutes collaborated on a healthcare FL study to predict future oxygen requirements of infected patients using inputs of vital signs, laboratory data, and chest x-rays, constituting the “EXAM” (EMR CXR AI Model) model. EXAM achieved an average Area Under the Curve (AUC) of over 0.92, an average improvement of 16%, and a 38% increase in generalisability over local models. The FL paradigm was successfully applied to facilitate a rapid data science collaboration without data exchange, resulting in a model that generalised across heterogeneous, unharmonized datasets. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as set the stage for broader use of FL in healthcare.

Unraveling US National COVID-19 Racial/Ethnic Disparities using County Level Data Among 328 Million Americans (preprint)

Identifying areas with high COVID-19 burden and their characteristics can help improve vaccine distribution and uptake, reduce burdens on health care systems, and allow for better allocation of public health intervention resources. Synthesizing data from various government and nonprofit institutions of 3,142 United States (US) counties as of 12/21/2020, we studied county-level characteristics that are associated with cumulative case and death rates using regression analyses. Our results showed counties that are more rural, counties with more White/non-White segregation, and counties with higher percentages of people of color, in poverty, with no high school diploma, and with medical comorbidities such as diabetes and hypertension are associated with higher cumulative COVID-19 case and death rates. We identify the hardest hit counties in US using model-estimated case and death rates, which provide more reliable estimates of cumulative COVID-19 burdens than those using raw observed county-specific rates. Identification of counties with high disease burdens and understanding the characteristics of these counties can help inform policies to improve vaccine distribution, deployment and uptake, prevent overwhelming health care systems, and enhance testing access, personal protection equipment access, and other resource allocation efforts, all of which can help save more lives for vulnerable communities. Significance statementWe found counties that are more rural, counties with more White/non-White segregation, and counties with higher percentages of people of color, in poverty, with no high school diploma, and with medical comorbidities such as diabetes and hypertension are associated with higher cumulative COVID-19 case and death rates. We also identified individual counties with high cumulative COVID-19 burden. Identification of counties with high disease burdens and understanding the characteristics of these counties can help inform policies to improve vaccine distribution, deployment and uptake, prevent overwhelming health care systems, and enhance testing access, personal protection equipment access, and other resource allocation efforts, all of which can help save more lives for vulnerable communities.

Unraveling US National COVID-19 Racial/Ethnic Disparities using County Level Data Among 328 Million Americans (preprint)

Racial and ethnic disparities in COVID-19 outcomes reflect the unequal burden experienced by vulnerable communities in the United States (US). Proposed explanations include socioeconomic factors that influence how people live, work, and play, and pre-existing comorbidities. It is important to assess the extent to which observed US COVID-19 racial and ethnic disparities can be explained by these factors. We study 9.8 million confirmed cases and 234,000 confirmed deaths from 2,990 US counties (3,142 total) that make up 99.8% of the total US population (327.6 out of 328.2 million people) through 11/8/20. We found national COVID-19 racial health disparities in US are partially explained by various social determinants of health and pre-existing comorbidities that have been previously proposed. However, significant unexplained racial and ethnic health disparities still persist at the US county level after adjusting for these variables. There is a pressing need to develop strategies to address not only the social determinants but also other factors, such as testing access, personal protection equipment access and exposures, as well as tailored intervention and resource allocation for vulnerable groups, in order to combat COVID-19 and reduce racial health disparities.

Population-scale Longitudinal Mapping of COVID-19 Symptoms, Behavior, and Testing Identifies Contributors to Continued Disease Spread in the United States (preprint)

Despite social distancing and shelter-in-place policies, COVID-19 continues to spread in the United States. A lack of timely information about factors influencing COVID-19 spread and testing has hampered agile responses to the pandemic. We developed How We Feel, an extensible web and mobile application that aggregates self-reported survey responses, to fill gaps in the collection of COVID-19-related data. How We Feel collects longitudinal and geographically localized information on users' health, behavior, and demographics. Here we report results from over 500,000 users in the United States from April 2, 2020 to May 12, 2020. We show that self- reported surveys can be used to build predictive models of COVID-19 test results, which may aid in identification of likely COVID-19 positive individuals. We find evidence among our users for asymptomatic or presymptomatic presentation, as well as for household and community exposure, occupation, and demographics being strong risk factors for COVID-19. We further reveal factors for which users have been SARS-CoV-2 PCR tested, as well as the temporal dynamics of self- reported symptoms and self-isolation behavior in positive and negative users. These results highlight the utility of collecting a diverse set of symptomatic, demographic, and behavioral self- reported data to fight the COVID-19 pandemic.

Full-spectrum dynamics of the coronavirus disease outbreak in Wuhan, China: a modeling study of 32,583 laboratory-confirmed cases (preprint)

Vigorous non-pharmaceutical interventions have largely suppressed the COVID-19 outbreak in Wuhan, China. We extended the susceptible-exposed-infectious-recovered model to study the transmission dynamics and evaluate the impact of interventions using 32,583 laboratory-confirmed cases from December 8, 2019 till March 8, 2020, accounting for presymptomatic infectiousness, and time-varying ascertainment rates, transmission rates, and population movements. The effective reproduction number R0 dropped from 3.54 (95% credible interval: 3.41-3.66) in the early outbreak to 0.27 (0.23-0.32) after full-scale multi-pronged interventions. By projection, the interventions reduced the total infections in Wuhan by 96.1% till March 8. Furthermore, we estimated that 87% infections (lower bound: 53%) were unascertained, potentially including asymptomatic and mild-symptomatic cases. The probability of resurgence was 0.33 and 0.06 based on models with 87% and 53% infections unascertained, respectively, assuming all interventions were lifted after 14 days of no ascertained infections. These results provide important implications for continuing surveillance and interventions to eventually contain the outbreak.

Building an International Consortium for Tracking Coronavirus Health Status (preprint)

Information is the most potent protective weapon we have to combat a pandemic, at both the individual and global level. For individuals, information can help us make personal decisions and provide a sense of security. For the global community, information can inform policy decisions and offer critical insights into the epidemic of COVID-19 disease. Fully leveraging the power of information, however, requires large amounts of data and access to it. To achieve this, we are making steps to form an international consortium, Coronavirus Census Collective (CCC, coronaviruscensuscollective.org), that will serve as a hub for integrating information from multiple data sources that can be utilized to understand, monitor, predict, and combat global pandemics. These sources may include self-reported health status through surveys (including mobile apps), results of diagnostic laboratory tests, and other static and real-time geospatial data. This collective effort to track and share information will be invaluable in predicting hotspots of disease outbreak, identifying which factors control the rate of spreading, informing immediate policy decisions, evaluating the effectiveness of measures taken by health organizations on pandemic control, and providing critical insight on the etiology of COVID-19. It will also help individuals stay informed on this rapidly evolving situation and contribute to other global efforts to slow the spread of disease. In the past few weeks, several initiatives across the globe have surfaced to use daily self-reported symptoms as a means to track disease spread, predict outbreak locations, guide population measures and help in the allocation of healthcare resources. The aim of this paper is to put out a call to standardize these efforts and spark a collaborative effort to maximize the global gain while protecting participant privacy.

Evolving Epidemiology and Impact of Non-pharmaceutical Interventions on the Outbreak of Coronavirus Disease 2019 in Wuhan, China (preprint)

BACKGROUND We described the epidemiological features of the coronavirus disease 2019 (Covid-19) outbreak, and evaluated the impact of non-pharmaceutical interventions on the epidemic in Wuhan, China. METHODS Individual-level data on 25,961 laboratory-confirmed Covid-19 cases reported through February 18, 2020 were extracted from the municipal Notifiable Disease Report System. Based on key events and interventions, we divided the epidemic into four periods: before January 11, January 11-22, January 23 - February 1, and February 2-18. We compared epidemiological characteristics across periods and different demographic groups. We developed a susceptible-exposed-infectious-recovered model to study the epidemic and evaluate the impact of interventions. RESULTS The median age of the cases was 57 years and 50.3% were women. The attack rate peaked in the third period and substantially declined afterwards across geographic regions, sex and age groups, except for children (age <20) whose attack rate continued to increase. Healthcare workers and elderly people had higher attack rates and severity risk increased with age. The effective reproductive number dropped from 3.86 (95% credible interval 3.74 to 3.97) before interventions to 0.32 (0.28 to 0.37) post interventions. The interventions were estimated to prevent 94.5% (93.7 to 95.2%) infections till February 18. We found that at least 59% of infected cases were unascertained in Wuhan, potentially including asymptomatic and mild-symptomatic cases. CONCLUSIONS Considerable countermeasures have effectively controlled the Covid-19 outbreak in Wuhan. Special efforts are needed to protect vulnerable populations, including healthcare workers, elderly and children. Estimation of unascertained cases has important implications on continuing surveillance and interventions.