key: cord-0857449-jx4kspen
authors: Tatapudi, Hanisha; Das, Rachita; Das, Tapas K.
title: Impact assessment of full and partial stay-at-home orders, face mask usage, and contact tracing: An agent-based simulation study of COVID-19 for an urban region
date: 2020-10-19
journal: Glob Epidemiol
DOI: 10.1016/j.gloepi.2020.100036
sha: ac4d7f7decf5a0e65208b3342c606fe34ed04d7d
doc_id: 857449
cord_uid: jx4kspen

PURPOSE: Social intervention strategies to mitigate COVID-19 are examined using an agent-based simulation model. Outbreak in a large urban region, Miami-Dade County, Florida, USA is used as a case study. Results are intended to serve as a planning guide for decision makers. METHODS: The simulation model mimics daily social mixing behavior of the susceptible and infected generating the spread. Data representing demographics of the region, virus epidemiology, and social interventions shapes model behavior. Results include daily values of infected, reported, hospitalized, and dead. RESULTS: Results show that early implementation of complete stay-at-home order is effective in flattening and reversing the infection growth curve in a short period of time. Whereas, using Florida's Phase II plan alone could result in 75% infected and end of pandemic via herd immunity. Universal use of face masks reduced infected by 20%. A further reduction of 66% was achieved by adding contact tracing with a target of identifying 50% of the asymptomatic and pre-symptomatic. CONCLUSIONS: In the absence of a vaccine, the strict stay-at-home order, though effective in curbing a pandemic outbreak, leaves a large proportion of the population susceptible. Hence, there should be a strong follow up plan of social distancing, use of face mask, contact tracing, testing, and isolation of infected to minimize the chances of large-scale resurgence of the disease. However, as the economic cost of the complete stay-at-home-order is very high, it can perhaps be used only as an emergency first response, and the authorities should be prepared to activate a strong follow up plan as soon as possible. The target level for contact tracing was shown to have a nonlinear impact on the reduction of the percentage of population infected. Increase in contact tracing target from 20% to 30% appeared to provide the largest incremental benefit.

Emergence of the severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) was first reported on December 31, 2019 in Wuhan, China and subsequently declared a global pandemic on March 11 by the World Health Organization (WHO) [3, 52] . As of Sept. 9, 2020, the number of reported cases worldwide has reached over 27.5 million causing 897,789 deaths. The number of infected cases continues to rise quite significantly [2] . The U.S. has been among the hardest hit by the coronavirus pandemic with 6.3+ million reported infections and 189,538

reported deaths (>21% of the total reported deaths worldwide) so far. However, as the new cases, hospital admissions, and deaths began to decline in mid-May, most States in the U.S. began phased lifting of their social intervention measures. For example, Florida adopted a three phased approach: Phase I (which began in May 18, 2020) allowed most business and workplaces to reopen with up to 50% of their building capacities and with large events constrained to 25%; Phase II began in June 5, 2020 and allowed all businesses to reopen for up to 50-75% of their capacities and also permitting events in large venues with no more than 50% of their capacities; Phase III will be akin to a complete reopening for which neither a date nor the criteria have been declared. A summary of Florida's phased intervention plan can be seen in Figure A2 (in the Appendix). As the reopening entered Phase II, Florida, along with many other states, began to see sharp increases in daily new infections (e.g., Florida reported over 15,000 new cases on July 11, 2020 along with a test positivity rate reaching over 15%).

In this paper, we investigate a few 'what-if' scenarios for social intervention policies including if the stay-at-home order were not lifted, if the Phase II order continues unaltered, what impact will the universal face mask usage have on the infections and deaths, and finally, how do the benefits of contact tracing vary with various target levels for identifying asymptomatic and pre-symptomatic. We conduct our investigation by first developing a comprehensive agent-based simulation model for COVID-19, and then using a major urban outbreak region (Miami-Dade County hospitalization (if infected with acute illness); and 10) recovery or death (if infected). The AB model reports daily and cumulative values of actual infected, doctor visits, tested, reported cases, hospitalized, recovered, and deaths, for each age category. A schematic diagram depicting the algorithmic sequence and parameter inputs for the AB simulation model is presented in Figure 1 .

Our AB simulation model works as follows. It begins by generating the individual people according to the U.S. census data that gives population attributes including age (see Table A1 ) and occupational distribution (see Table   A4 ). Thereafter, it generates the households based on their composition characterized by the number of adults and children (see Table A2 ). The model also generates, per census data, schools (see Table A3 ) and the workplaces and other community locations (see Table A4 ). Each individual is assigned a household, while maintaining the average household composition, and, depending on the age, either a school or a workplace (considering employment levels).

A daily (hour by hour) schedule is assigned to every individual, chosen from a set of alternative schedules, based on their attributes. The schedules vary between weekdays and weekends and also depend on the prevailing social intervention orders (see Table A5 ).

Simulation begins on the day when one or more infected people are introduced to the region (referred to as simulation day 1). Simulation model tracks hourly movements of each individual (susceptible and infected) every day, and records for each susceptible the number of infected contacts and their identification at each location. Based on the level of infectiousness of each infected contact (which depends on the day of his/her infectiousness period), the model calculates the daily force of infection received by each susceptible from all infected contacts at all hours of the day [18] . Daily force of infection is considered to accumulate. However, it is assumed that if a susceptible does not gather any additional force of infection (i.e., does not come in contact with any infected) for two consecutive days, the cumulative force of infection for the susceptible reduces to zero. At the end of each day, the Table A6 for average lengths of the periods) Epidemiological models and other parameters that guide the AB model are described next. Figure 3 presents a schematic of the disease natural history of COVID-19, parameters of which are given in Table A6 . Once infected, an individual simultaneously begins the latency and the incubation periods. The individual becomes infectious after the latent period is complete but displays symptoms (unless asymptomatic) at the end of the incubation period. The period between end of latency and end of incubation is referred to as pre-symptomatic, a time when the infectiousness grows rapidly and almost reaches its peak. Symptomatic cases either follow a non-acute progression (majority of cases, not requiring hospitalization) or acute progression (requiring hospitalization). Cases for whom disease does not become acute, enter a recovery period after infectiousness ends. Those with acute disease progression (generally toward the end of the infectious period) are hospitalized. After the hospital stay period, cases either recover or die. For average lengths of recovery and hospitalized periods that are used in the AB model, see Table A6 . There is some evidence based on animal experimentation that recovered individual may become immune to reinfection [31, 40] . But other studies remain inconclusive [32] . Hence, due to lack of established data on this matter, our model considers the recovered cases to be immune to further COVID-19 infections.

The duration and intensity of infectiousness is considered to be guided by a lognormal density function (see Figure   4 ). The function is truncated at the average length of the infectiousness period (which is considered to be 9.5 days).

Asymptomatic cases are assumed to follow a similar infectiousness intensity profile but scaled by a factor ( in the force of infection calculation (1), see Table A7 ). ` The AB model estimates the probability of infection for a susceptible using the accumulated value of daily force of infection ( ), which is calculated as follows.

The first component in (1) accounts for the force experienced by susceptible individual at home from another infected household members . The second component captures the force experienced at schools/workplaces/community places for work and also at community places visited for daily errands; this happens when a susceptible is in the same location type where infected is at hour . The definition and values of the parameters of (1) are given in Table A7 . Equation (1) is a modified version of the force of infection equation given in [18] , which has three components that separately calculate force of infection received at home, at indoor workplaces, and at the outdoor community. For the sake of simplicity, we have considered only the first two components, home and indoor workplaces, where most of the COVID-19 transmission is assumed to be taking place. We have assumed that the mode of virus transmission at indoor community places that are routinely visited by people as part of their daily errands (like grocery stores, home goods stores, dine-in/take-out restaurants, etc.) is similar to that of indoor workplace transmission.

The force of infection is gathered by a susceptible individual each day from all infected contacts in his/her mixing groups (home, school/workplace, and community places). The cumulative value of is used at the end of each day to calculate the probability of infection as .

J o u r n a l P r e -p r o o f

The AB model incorporates all applicable intervention orders like stay-at-home, school and workplace closure, isolation of symptomatic cases at home, and quarantine of household members of those who are infected. The model also considers: varying levels of compliances for isolation and quarantine, lower on-site staffing levels of essential work and community places during stay-at-home order, restricted daily schedule of people during various social intervention periods, phased lifting of interventions, use of face masks in workplaces, schools and community places with varying compliance levels, and contact tracing with different target levels to identify asymptomatic and presymptomatic cases. The timeline for social interventions implemented in the model are summarized in Table A8 .

Other salient considerations in the implementation of our AB model are as follows. Across all age groups, 35% of the infected cases were considered asymptomatic [7] . Approximately, twenty percent (20%) of Florida residents are reported as uninsured and do not have access to a primary care physician [49] . Uninsured people thus considered not to have the doctor referral required for most of the testing facilities in Miami-Dade County, and hence not tested. All symptomatic cases with health insurance were assumed to visit/consult with a doctor. Depending on their symptoms, travel history, and contact history, some of them were given referrals for testing. We considered that only a small percentage of cases visiting/consulting a doctor were given referrals in the early months of the pandemic (until the middle of April 2020), due to the shortage of testing and restrictive CDC guidelines for who could be tested [6] . However, as CDC relaxed its test eligibility guidelines [39] and the capacity to test increased in Florida, we gradually increased the probability of getting a test referral from a doctor closer to 100% by early June 2020 for symptomatic cases (see Table A9 ). We also considered in our model that a small fraction (reaching only up to 10% over time) of the asymptomatic cases are randomly tested through various community testing protocols, e.g., at elderly care facilities, healthcare facilities, workplaces, etc. Note that we did not consider co-infection, and therefore all cases that were tested in our simulation model had COVID-19. Hence, each test yielded a positive outcome with a probability equal to the test sensitivity (see Table A9 ). Based on the data reported on Florida COVID-19

dashboard, a test result reporting delay of up to 10 days on average was considered at the start of the pandemic, which was progressively reduced (see Table A9 ). All symptomatic cases with or without testing were considered to isolate at home with a given probability of compliance. The probability of compliance was considered to vary during the length of the symptomatic period of infection. For this purpose, we divided the symptomatic period into three parts: I, II, and III, and assumed a lower isolation compliance in parts I and III and higher in part II, when the illness J o u r n a l P r e -p r o o f is more apparent. See Table A10 for the isolation compliance probabilities. Susceptible members of the households with one or more infected cases are considered to quarantine themselves. We also assumed a level of compliance for the quarantine (see Table A10 ). We used hospitalization and death data reported for Miami-Dade County [20] for each age group to obtain probabilities of hospitalization of the reported cases, and probabilities of death for those who are hospitalized (see Table A11 ).

Though we have implemented our AB simulation model for a specific region, it is quite general in its usability for other urban regions with similar demography, societal characteristics, and intervention measures. In our model, Tables A1-A4 summarize the demographic inputs (age and household distribution, number of schools for various age groups, and number of workplaces of various types and sizes). These data will need to be curated from both national and local census records. Social interventions vary from region to region and hence the data in Table A8 will need to be updated. Similarly, testing availability, test sensitivity, and test outcome reporting delay may also vary significantly from region to region, and thus Table A9 will also need to be updated. The rest of the data (in Tables A5, A6 , A7, A10, and A11) are related to epidemiology of COVID-19. These are unlikely to be significantly different, though some adjustments of these based on population demographics may be needed.

The AB model utilizes a large number of parameters, which are demographic parameters, epidemiological parameters, and social intervention parameters. We kept almost all of the above parameters fixed at their respective Once the model is calibrated and validated with available reported data on infected and dead, we extended the model into the future to predict outcomes. The only parameters that were altered after the calibration period are to reflect the expected changes in social interventions, e.g., order mandating use of face mask, re-closing some community places, expected increase in contact tracing, and changes in community response via daily schedule restrictions.

Hence, the parameters that were changed after the calibration period included those for daily schedules, transmission coefficients, testing and contact tracing rates, and compliance to isolation and quarantine. Most of the parameter values used in the AB model were obtained from government archives and research literature, for which references are provided (see Tables A1-A11 ). For some of the parameters for which we could not find an archived data source, we used expert opinion and current media reports. We first present a summary of the key results of our study (see Table 1 ), from which a number of key insights can be derived that may apply to other similar urban regions experiencing respiratory/influenza type virus outbreaks.

Early imposition of stay-at-home order appears to have been quite effective in first flattening and then reversing the growth curve. Per our model, if the stay-at-home order was allowed to remain enforced, the pandemic would have subsided with a relatively low percentage of the population (5.8%) infected and approximately 0.04% dead within six months of inception; 50 or below daily new infections was used as the criterion to consider that pandemic has subsided in Miami-Dade County. If the extent of social mixing akin to Phase II reopening of Florida is in place for an urban region (without the use of face mask and contact tracing), the pandemic would likely have raged for 8-9

months and subside only after reaching herd immunity with over 75% of the population infected and 1.3% of the population dead. Universal use of face masks of surgical variety was shown by the model to reduce average total infected, hospitalized, and dead by 20%, 19%, and 15%, respectively. Aggressive contact tracing with a goal to identify 50% of the asymptomatic and pre-symptomatic was also projected to have a very significant positive impact with an average reduction of 66% of total infected. The average reductions in total infected with 40%, 30%, and 20% contact tracing targets were found to be 58%, 41%, and 14%, respectively.

In what follows, we expound the results from our study. Figures 6 and 7 fig. (a) ) and hospitalizations and deaths ( fig. (b) ) if stay-athome order were not lifted Figure 6 shows a strong influence of continuing with the stay-at-home order in curbing the COVID-19 growth J o u r n a l P r e -p r o o f within approximately 6 months from its inception with on average less than 5.8% of the population infected, 0.15% hospitalized, and 0.037% dead; 50 or below daily new infections was used as the criterion to consider that pandemic has subsided in Miami-Dade County. Such a quick suppression of a virus outbreak always leaves the possibility of resurgence, for which an effective plan of contact tracing, testing, isolation, and support for those isolated (when needed) should be in place. . (a) and fig.  (b) ) and Phase II reopening without face mask and contact tracing ( fig. (c) and fig. (d) ) Figure 7 shows the expected outcomes of continuing with the Phase I order and the Phase II order. Figure 7 outbreaks, it is shown that adjusted odds ratio (aOR) of getting an infection after wearing surgical variety face masks versus without wearing face mask is 0.33 on average [12] . This can be interpreted as the likelihood of getting J o u r n a l P r e -p r o o f infected if wearing a surgical variety face mask is one third of what it would be for not wearing a mask. Hence, we considered a 67% reduction in the transmission coefficient ( ) used in calculating the force of infection (see equation (1)), assuming a 100% compliance in the use of surgical variety masks at workplaces, schools and community places. We also tested the impact of 30% and 45% reductions in the transmission coefficient value ( ), which translate to approximately 50% and 70% compliance for face mask usage, respectively. The anticipated impact of face mask usage together with Phase II order on the average cumulative numbers of infected are shown in Figure 8 (a). It also depicts the risk difference between the average values of cumulative infected without and with the universal use of face mask. It may be noted that since the infections grow slower with the use of face mask, the cumulative risk difference rises to almost 875K in the middle of August and then settles down close to 430K when pandemic is predicted to subside by the end of November 2020. Figure 8 (b) depicts the daily values of the average infected for Phase II without and with universal face mask policy. As expected, the peak of daily infection with face mask usage is shifted to a slightly later date and the downward trend begins after a smaller percentage (31%) of the total population are infected compared to 36% without the use of face mask. Our agent-based model has several limitations. First and foremost, the simulation model is an abstraction of how a pandemic impacts a large and complex society. Though our model deliberately introduces some variabilities, somewhat pre-defined daily schedules are used to approximate a highly dynamic contact process of an urban region.

Also, the contact process does not account for significant variabilities in the types and lengths of interactions even within each mixing groups. We did not assign geographic locations (latitude and longitude) for households, businesses, schools, and community places, and assumed them to be uniformly distributed over the region. It is common for urban population centers and associated establishments to grow in clusters, for which the contact patterns are expected to be different from those in uniformly dispersed regions. We did not consider special events like parties, games, and street protests, some of which are known to have caused superspreading of the virus and case increases. Finally, and perhaps most importantly, the model uses a large number of parameters (listed in Tables Each scenario of our case study with 10 replicates (with different seeds) takes approximately 8-12 hours to run in a standard desktop computer with Intel Core i7 with 16GB memory. In the interest of presenting our observations quickly to the public health decision makers, while COVID-19 is still rampant in the region, we chose to use a limited number (10) of replicates. As the main purpose of this paper is to conduct a broad what-if analysis, we do not believe that use of a small number of replicates has negatively influenced our observations. The trends and observations derived from our results are only intended to be used for planning and guidance of public health decision makers.

As part of our continuing (future) work, we plan to use our model to examine the impact of reopening of K-12 schools and colleges/universities for the new academic year, which began at the end of August and early September.

We also plan to use our AB model to assess efficacy of various prioritization strategies (based on age, risk, and work groups) for the vaccines that are anticipated to be available in limited quantities by the beginning of 2021.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

County can be found in [A11] . c) Use of face mask: Boston implemented a facemask policy in early May in an executive order by the State [A9] . d) Implementation of contact tracing: Massachusetts state government provides a dashboard on the community health outcomes for COVID-19. Details on the success of contact tracing in the communities along with the outcome measures varying over time can be found in [A10] . e) Policy for returning to school: School reopening policies also widely vary from state to state and also among counties within a state. Information on Boston's public school reopening policy can be found in [A12] . It is important to frequently check sources on school policy as they are transient. For example, Boston planned to reopen on October 15, but shifted to October 22 after seeing an increase in the number of cases. 10) Time varying testing of symptomatic and asymptomatic: Limited testing availability has been a serious concern in many U.S. regions that suffered from a high level of disease spread. Time varying data on test availability and test outcome reporting delay are difficult to find in indexed literature during a pandemic. Hence, these can be assumed from regional news reports, test reporting data, and/or other grey literature. 11) Number of reported, hospitalized and deaths: Daily data and archived data on the number of people reported positive, hospitalized and dead can be found from dashboard in [A10] . 12) Probability of hospitalization for reported cases and probability of death for hospitalized: This can be calculated from [A10] based on age specific reporting.

The information contained in the following tables for Miami Dade County are likely to be same for other regions like Boston city/Suffolk County. 13) Daily schedules for people can be assumed to be same as in Table A5 . 14) Disease natural history parameters for COVID-19 can be assumed to be same across regions within a country, see Table A6 . 15) Some of the parameters in Table A7 for calculating the force of infection need to be calibrated (see Step 3) .

However, the remaining parameters in Table A7 can be assumed to be same. 16) Self-isolation compliance for symptomatic cases, and quarantine compliance for household members can be assumed to be same for different urban regions within a country, as in Table A10 .

Step 2: Updating of the simulation model Once the input data collection is complete, the next step is to update the model parameters as follows. 1) Update the simulation model with all gathered input data from step 1: After gathering data, it needs to be curated and transformed into .txt files to be read by the simulation model. Some of the data are directly coded in the model, where applicable. 2) Decide simulation begin date: Simulation begin date depends on the outbreak region and is based on the date of the first reported case. Up to 14 days before the first reported case can be used as a potential date for simulation model to begin. 3) Decide simulation end date: Simulation end date is chosen as desired by the modeler. 4) Number of initial infected cases: Most Departments of Health (DOH) provide a count and characterization of the number of initial infected cases with travel histories. One can identify these initial infected cases during the first month or so of the outbreak and use those cases to initiate social mixing and community spread.

Step 3: Calibrate and validate simulation model Once the simulation model is updated with the input data for the region, the model is calibrated using a small applicable subset of input data and the model output is validated with actual surveillance data from the region, as follows. 1) Generate multiple seeds for the uniform random variables that are used to calculate the probabilities of infection, hospitalization, death, testing, symptomatic, disease severity, test sensitivity, compliance for isolation and quarantine, among others. Simulation output from each seed is considered a replicate. Using output data from all replicates, an average value and a corresponding confidence interval for each output measure are calculated. 2) A set of initial values of the transmission coefficients for home, school, work, and community places are assumed (based on current literature and published models for outbreaks of similar diseases). These transmission coefficients (along with other parameters, see Table A7 ) are used for calculating force of J o u r n a l P r e -p r o o f infection, which is then use to calculate the probability of infection. Different sets of transmission coefficient values are selected for different reference points in time in the simulation, depending on changes in social intervention status and significant current events. For example, the transmission coefficients are appropriately calibrated (reduced) on the day universal use of facemask is announced. Also, percentage testing of asymptomatic and pre-symptomatic are increased when contact tracing begins. Street protests combined with Independence Day holiday in early July, 2020 are examples of current events that may require adjustment of transmission coefficient values. 3) Other parameters that are considered suitable for the AB model calibration are probability of running errand that guides daily schedule and probability of employees reporting to work for essential and nonessential businesses. These values can also be assumed to change over time during a pandemic depending on the phased intervention policies implemented by the government in the outbreak region. 4) The simulation is calibrated for a chosen period. In this study, we chose to calibrate the model up to July 15th, as we had reported data available until that date for validation purposes at the time of calibrating the model. 5) Results for reported cases, hospitalized, dead for all age groups are gathered from the simulation model for each seed. 6) Average values (with confidence interval) are computed for the numbers of reported, hospitalized and dead. 7) For model validation, the simulated average values for the reported, hospitalized, and dead and compared with actual surveillance data. 8) Alter calibration parameters as needed to obtain desired level of validation accuracy. Measure validation accuracy is calculated as the difference in the seven-day moving average between simulated and surveillance data.

Step 4: Implement calibrated model for prediction 1) Run calibrated simulation model for all seeds for a desired prediction period beyond the calibration/validation time. 2) Extract age specific data for total infected, reported cases, hospitalized, and dead from simulation for each seed. 3) Report mean and confidence interval. Figure A1 : Florida's phased social intervention plan for COVID-19 pandemic [44] J o u r n a l P r e -p r o o f Table A10 : Self-isolation compliance for symptomatic cases and quarantine compliance for household members 

Modeling the impact of social distancing, testing, contact tracing and household quarantine on second-wave scenarios of the COVID-19 epidemic

The proximal origin of SARS-CoV-2

Estimating the infection horizon of COVID-19 in eight countries with a datadriven approach

Evolution and epidemic spread of SARS-CoV-2 in Brazil

Centers for Disease Control and Prevention'. CDC releases consolidated COVID-19 testing recommendations

Centers for Disease Control and Prevention. COVID-19 Pandemic Planning Scenarios

Centers for Disease Control and Prevention. Interim Clinical Guidance for Management of Patients with Confirmed Coronavirus Disease

Modeliling tranmission and control of the COVID-19 pandemic in Australia

FluTE, a publicly available stochastic influenza epidemic simulation model

COVID-19 virus outbreak forecasting of registered and recovered cases after sixty day lockdown in Italy: A data driven model approach

Physical distancing, face masks, and eye protection to prevent person-to-person transmission of SARS-CoV-2 and COVID-19: a systematic review and metaanalysis

Modeling the Worldwide Spread of Pandemic Influenza: Baseline Case and Containment Interventions

A large-scale simulation model of pandemic influenza outbreaks for development of dynamic mitigation strategies

How to restart? An agent-based simulation model towards the definition of strategies of COVID-19 "second phase

An influenza simulation model for immunization studies

Transmission dynamics of the COVID-19 outbreak and effectiveness of government interventions: A data-driven analysis

Strategies for mitigating an influenza pandemic

Strategies for containing an emerging influenza pandemic in Southeast Asia

Coronavirus: summary of persons being monitored, persons under investigation, and cases

COVID-19: summary of persons being monitored, persons tested, and cases

Division of Disease Control and Health Protection. Florida's COVID-19 Data and Surveillance Dashboard

Hospitalization Rates and Characteristics of Patients Hospitalized with Laboratory-Confirmed Coronavirus Disease

Mitigation strategies for pandemic influenza in the United States

Containing bioterrorist smallpox

Evidence summary for COVID-19 viral load over course of infection

The Effectiveness of Quarantine of Wuhan City Against the Corona Virus Disease 2019 (COVID-19): A Well-Mixed SEIR Model Analysis

Infectious Disease Modeling: Creating a Community to Respond to Biological Threats. Statistical Communications in Infectious Diseases

A contribution to the mathematical theory of epidemics

COVID-19 and Postinfection Immunity

Lack of Reinfection in Rhesus Macaques Infected with SARS-CoV-2. bioRxiv

Containing Pandemic Influenza with

FACS: A geospatial agent-based simulator for analyzing COVID-19 spread and public health measures on local regions

Design of non-pharmaceutical intervention strategies for pandemic influenza outbreaks

Miami Matters. Households/Income Data for County: Miami-Dade

COVID-19/parameter_estimates

Coronavirus Screening Test Developed at Johns Hopkins

Will we see protection or reinfection in COVID-19?

Epidemic analysis of COVID-19 in China by dynamical modeling. medrXiv

COVID-ABS: An agent-based model of COVID-19 epidmeic to simulate health and economic effects of social distancing interventions

Estimating disease burden of a potential A(H7N9) pandemic influenza outbreak in the United States

The City of Miami -Florida. COVID-19 Updates

All Sectors: County Business Patterns by Legal Form of Organization and Employment Size Class for U.S., States, and Selected Geographies

All Sectors: County Business Patterns by Legal Form of Organization and Employment Size Class for U.S., States, and Selected Geographies

Annual Business Survey: Statistics for Employer Firms by Industry, Sex, Ethnicity, Race, and Veteran Status for the

A predictive decision-aid methodology for dynamic mitigation of influenza pandemic

A new coronavirus associated with human respiratory disease in China

Early estimation of the case fatality rate of COVID-19 in mainland China: a data-driven analysis

Modified SEIR and AI prediction of the epidemics trend of COVID-19 in China under public health interventions

The impact of social distancing and epicenter lockdown on the COVID-19 epidemic in mainland China: A data-driven SEIQR model study. medRxiv

The Boston Planning and Development Agency

Annual Business Survey: Statistics for Employer Firms by Industry, Sex, Ethnicity, Race, and Veteran Status for the

State-by-State Guide to Face Mask Requirements

COVID-19 Response Reporting

COVID-19 State of Emergency

J o u r n a l P r e -p r o o f Journal Pre-proof Table A4 : Distribution of different types of workplaces in Miami-Dade County. All industries and community places are classified into essential or non-essential. Essential industries remain functional with a percentage of their workforce reporting during interventions like stay-at-home or phased reopening. Non-essential industries are considered to operate remotely. Essential industries include wholesale trade, waste management and remediation services, agriculture, forestry, fishing and hunting, mining, quarrying, oil and gas extraction, utilities, construction, manufacturing, transportation and warehousing. Non-essential industry includes finance and insurance, real estate and rental and leasing, professional, scientific and technical services, management of companies and enterprises, administrative and support for waste management and remediation services, educational services, other services except public administration. Essential community includes grocery stores, convenience stores, pharmacies and drug stores, home centers, health care and social assistance. Non-essential community includes retail, arts, entertainment and recreation and accommodation and food services. For details on education institutions, see J o u r n a l P r e -p r o o f