Contact surveys reveal heterogeneities in age-group contributions to SARS-CoV-2 dynamics in the United States (preprint)

SARS-CoV-2 is spread primarily through person-to-person contacts. Quantifying population contact rates is important for understanding the impact of physical distancing policies and for modeling COVID-19, but contact patterns have changed substantially over time due to shifting policies and behaviors. There are surprisingly few empirical estimates of age-structured contact rates in the United States both before and throughout the COVID-19 pandemic that capture these changes. Here, we use data from six waves of the Berkeley Interpersonal Contact Survey (BICS), which collected detailed contact data between March 22, 2020 and February 15, 2021 across six metropolitan designated market areas (DMA) in the United States. Contact rates were low across all six DMAs at the start of the pandemic. We find steady increases in the mean and median number of contacts across these localities over time, as well as a greater proportion of respondents reporting a high number of contacts. We also find that young adults between ages 18 and 34 reported more contacts on average compared to other age groups. The 65 and older age group consistently reported low levels of contact throughout the study period. To understand the impact of these changing contact patterns, we simulate COVID-19 dynamics in each DMA using an age-structured mechanistic model. We compare results from models that use BICS contact rate estimates versus commonly used alternative contact rate sources. We find that simulations parameterized with BICS estimates give insight into time-varying changes in relative incidence by age group that are not captured in the absence of these frequently updated estimates. We also find that simulation results based on BICS estimates closely match observed data on the age distribution of cases, and changes in these distributions over time. Together these findings highlight the role of different age groups in driving and sustaining SARS-CoV-2 transmission in the U.S. We also show the utility of repeated contact surveys in revealing heterogeneities in the epidemiology of COVID-19 across localities in the United States.

The impact of mobility network properties on predicted epidemic dynamics in Dhaka and Bangkok (preprint)

1Properties of city-level commuting networks are expected to influence epidemic potential of cities and modify the speed and spatial trajectory of epidemics when they occur. In this study, we use aggregated mobile phone user data to reconstruct commuter mobility networks for Bangkok (Thailand) and Dhaka (Bangladesh), two megacities in Asia with populations of 16 and 21 million people, respectively. We model the dynamics of directly-transmitted infections (such as SARS-CoV2) propagating on these commuting networks, and find that differences in network structure between the two cities drive divergent predicted epidemic trajectories: the commuting network in Bangkok is composed of geographically-contiguous modular communities and epidemic dispersal is correlated with geographic distance between locations, whereas the network in Dhaka has less distinct geographic structure and epidemic dispersal is less constrained by geographic distance. We also find that the predicted dynamics of epidemics vary depending on the local topology of the network around the origin of the outbreak. Measuring commuter mobility, and understanding how commuting networks shape epidemic dynamics at the city level, can support surveillance and preparedness efforts in large cities at risk for emerging or imported epidemics.

GPS-estimated foot traffic data and venue selection for COVID-19 serosurveillance studies (preprint)

Tracking the dynamics and spread of COVID-19 is critical to mounting an effective response to the pandemic. In the absence of randomized representative serological surveys, many SARS-CoV-2 serosurveillance studies have relied on convenience sampling to estimate cumulative incidence. One common approach is to recruit at frequently visited community locations ("venue-based" sampling), but the sources of bias and uncertainty associated with this strategy are still poorly understood. Here, we used data from a venue-based community serosurveillance study, GPS-estimated foot traffic data, and data on confirmed COVID-19 cases to report an estimate of cumulative incidence in Somerville, Massachusetts, and a methodological strategy to quantify and reduce uncertainty in serology-based cumulative incidence estimates obtained via convenience sampling. The mismatch between the geographic distribution of participants' home locations (the "participant catchment distribution") and the geographic distribution of infections is an important determinant of uncertainty in venue-based and other convenience sampling strategies. We found that uncertainty in cumulative incidence estimates can vary by a factor of two depending how well the participant catchment distribution matches the known or expected geographic distribution of prior infections. GPS-estimated business foot traffic data provides an important proxy measure for the participant catchment area and can be used to select venue locations that minimize uncertainty in cumulative incidence.

Frequent testing and immunity-based staffing will help mitigate outbreaks in nursing home settings (preprint)

Background: Nursing homes and other long term care facilities have been disproportionately impacted by the COVID-19 pandemic. Strategies are urgently needed to reduce transmission in these vulnerable populations. We aimed to evaluate the reduction in transmission in nursing homes achieved through contact-targeted interventions and testing and to evaluate the effectiveness of two types of screening tests conducted with varying frequency: 1) rapid antigen testing and 2) PCR testing. Methods: We developed an agent-based Susceptible-Exposed-Infectious(Asymptomatic/Symptomatic)-Recovered (SEIR) model to examine SARS-CoV-2 transmission in nursing homes. Residents and staff are modelled individually; residents are split into two cohorts based on COVID-19 diagnosis. In the resident cohorting intervention, recovered residents are moved back from the COVID (infected) cohort to the non-COVID (susceptible/uninfected) cohort. In the immunity-based staffing intervention, recovered staff, who we assume have protective immunity, are assigned to work in the non-COVID cohort, while susceptible staff work in the COVID cohort and are assumed to have high levels of protection from personal protective equipment. These interventions aim to reduce the fraction of people's contacts that are presumed susceptible (and therefore potentially infected) and replace them with recovered (immune) contacts. Results: The frequency and type of testing has a larger impact on the size of outbreaks than the cohorting and staffing interventions. The most effective testing strategies modeled are daily antigen testing of everyone and daily antigen testing of staff with weekly PCR testing for residents. Under all screening testing strategies, the immunity-based staffing intervention reduces the final size of the outbreak. The resident cohorting intervention reduces the final outbreak size under some, but not all, testing scenarios. Conclusions: Increasing the frequency of screening testing of all residents and staff, or even staff alone, in nursing homes has the potential to greatly reduce outbreaks in this vulnerable setting. Immunity-based staffing can further reduce spread at little or no additional cost and becomes particularly important when daily testing is not feasible.

Participatory syndromic surveillance as a tool for tracking COVID-19 in Bangladesh (preprint)

Limitations in laboratory diagnostic capacity and reporting delays have hampered efforts to mitigate and control the ongoing coronavirus disease 2019 (COVID-19) pandemic globally. To augment traditional lab and hospital-based surveillance, Bangladesh established a participatory surveillance system for the public to self-report symptoms consistent with COVID-19 through multiple channels. Here, we report on the use of this system, which received over 3 million responses within two months, for tracking the COVID-19 outbreak in Bangladesh. Although we observe considerable noise in the data and initial volatility in the use of the different reporting mechanisms, the self-reported syndromic data exhibits a strong association with lab-confirmed cases at a local scale. Moreover, the syndromic data also suggests an earlier spread of the outbreak across Bangladesh than is evident from the confirmed case counts, consistent with predicted spread of the outbreak based on population mobility data. Our results highlight the usefulness of participatory syndromic surveillance for mapping disease burden generally, and particularly during the initial phases of an emerging outbreak.

Modeling the impact of human mobility and travel restrictions on the potential spread of SARS-CoV-2 in Taiwan (preprint)

Background: As COVID-19 continues to spread around the world, understanding how patterns of human mobility and connectivity affect outbreak dynamics, especially before outbreaks establish locally, is critical for informing response efforts. Methods: Here, in collaboration with Facebook Data for Good, we built metapopulation models that incorporate human movement data with the goals of identifying the high risk areas of disease spread and assessing the potential effects of local travel restrictions in Taiwan. We compared the impact of intracity vs. intercity travel restrictions on both the total number of infections and the speed of outbreak spread and developed an interactive application that allows users to vary inputs and assumptions. Findings: We found that intracity travel reductions have a higher impact on overall infection numbers than intercity travel reductions, while intercity travel reductions can narrow the scope of the outbreak and help target resources. We also identified the most highly connected areas that may serve as sources of importation during an outbreak. The timing, duration, and level of travel reduction together determine the impact of travel reductions on the number of infections, and multiple combinations of these can result in similar impact. Interpretation: In Taiwan, most cases to date were imported or linked to imported cases. To prepare for the potential spread within Taiwan, we utilized Facebook's aggregated and anonymized movement and colocation data to identify cities with higher risk of infection and regional importation. Both intracity and intercity movement affect outbreak dynamics, with the former having more of an impact on the total numbers of cases and the latter impacting geographic scope. These findings have important implications for guiding future policies for travel restrictions during outbreaks in Taiwan.

U.S. county-level characteristics to inform equitable COVID-19 response (preprint)

The spread of Coronavirus Disease 2019 (COVID-19) across the United States confirms that not all Americans are equally at risk of infection, severe disease, or mortality. A range of intersecting biological, demographic, and socioeconomic factors are likely to determine an individual's susceptibility to COVID-19. These factors vary significantly across counties in the United States, and often reflect the structural inequities in our society. Recognizing this vast inter-county variation in risks will be critical to mounting an adequate response strategy. Using publicly available county-specific data we identified key biological, demographic, and socioeconomic factors influencing susceptibility to COVID-19, guided by international experiences and consideration of epidemiological parameters of importance. We created bivariate county-level maps to summarize examples of key relationships across these categories, grouping age and poverty; comorbidities and lack of health insurance; proximity, density and bed capacity; and race and ethnicity, and premature death. We have also made available an interactive online tool that allows public health officials to query risk factors most relevant to their local context. Our data demonstrate significant inter-county variation in key epidemiological risk factors, with a clustering of counties in certain states, which will result in an increased demand on their public health system. While the East and West coast cities are particularly vulnerable owing to their densities (and travel routes), a large number of counties in the Southeastern states have a high proportion of at-risk populations, with high levels of poverty, comorbidities, and premature death at baseline, and low levels of health insurance coverage. The list of variables we have examined is by no means comprehensive, and several of them are interrelated and magnify underlying vulnerabilities. The online tool allows readers to explore additional combinations of risk factors, set categorical thresholds for each covariate, and filter counties above different population thresholds. COVID-19 responses and decision making in the United States remain decentralized. Both the federal and state governments will benefit from recognizing high intra-state, inter-county variation in population risks and response capacity. Many of the factors that are likely to exacerbate the burden of COVID-19 and the demand on healthcare systems are the compounded result of long-standing structural inequalities in US society. Strategies to protect those in the most vulnerable counties will require urgent measures to better support communities' attempts at social distancing and to accelerate cooperation across jurisdictions to supply personnel and equipment to counties that will experience high demand.

Modeling the Comparative Impact of Individual Quarantine vs. Active Monitoring of Contacts for the Mitigation of COVID-19 (preprint)

Background: Voluntary individual quarantine and voluntary active monitoring of contacts are core disease control strategies for emerging infectious diseases, such as COVID-19. Given the impact of quarantine on resources and individual liberty, it is vital to assess under what conditions individual quarantine can more effectively control COVID-19 than active monitoring. As an epidemic grows, it is also important to consider when these interventions are no longer feasible, and broader mitigation measures must be implemented. Methods: To estimate the comparative efficacy of these case-based interventions to control COVID-19, we fit a stochastic branching model to reported parameters for the dynamics of the disease. Specifically, we fit to the incubation period distribution and each of two sets of the serial interval distribution: a shorter one with a mean serial interval of 4.8 days and a longer one with a mean of 7.5 days. To assess variable resource settings, we consider two feasibility settings: a high feasibility setting with 90% of contacts traced, a half-day average delay in tracing and symptom recognition, and 90% effective isolation; and low feasibility setting with 50% of contacts traced, a two-day average delay, and 50% effective isolation. Findings: Our results suggest that individual quarantine in high feasibility settings where at least three-quarters of infected contacts are individually quarantined contains an outbreak of COVID-19 with a short serial interval (4.8 days) 84% of the time. However, in settings where this performance is unrealistically high and the outbreak continues to grow, so too will the burden of the number of contacts traced for active monitoring or quarantine. When resources are prioritized for scalable interventions such as social distancing, we show active monitoring or individual quarantine of high-risk contacts can contribute synergistically to mitigation efforts. Interpretation: Our model highlights the urgent need for more data on the serial interval and the extent of presymptomatic transmission in order to make data-driven policy decisions regarding the cost-benefit comparisons of individual quarantine vs. active monitoring of contacts. To the extent these interventions can be implemented they can help mitigate the spread of COVID-19.

Using predicted imports of 2019-nCoV cases to determine locations that may not be identifying all imported cases (preprint)

Cases from the ongoing outbreak of atypical pneumonia caused by the 2019 novel coronavirus (2019-nCoV) exported from mainland China can lead to self-sustained outbreaks in other populations. Internationally imported cases are currently being reported in several different locations. Early detection of imported cases is critical for containment of the virus. Based on air travel volume estimates from Wuhan to international destinations and using a generalized linear regression model we identify locations which may potentially have undetected internationally imported cases.