key: cord-0892840-zuvk7k84
authors: Politis, M. D.; Hua, X.; Ogwara, C. A.; Davies, M. R.; Adebile, T. M.; Sherman, M. P.; Zhou, X.; Chowell, G.; Spaulding, A. C.; Fung, I. C.-H.
title: Spatially refined time-varying reproduction numbers of SARS-CoV-2 in Arkansas and Kentucky and their relationship to population size and public health policy, March - November, 2020
date: 2021-05-29
journal: nan
DOI: 10.1101/2021.05.26.21257862
sha: 08b1e58ca85efa9ec25960f49d81294fa9a3af96
doc_id: 892840
cord_uid: zuvk7k84

Purpose: To examine the time-varying reproduction number, Rt, for COVID-19 in Arkansas and Kentucky and investigate the impact of policies and preventative measures on the variability in Rt. Methods: Arkansas and Kentucky county-level COVID-19 cumulative case count data (March 6-November 7, 2020) were obtained. Rt was estimated using the R package 'EpiEstim', by county, region (Delta, non-Delta, Appalachian, non-Appalachian), and policy measures. Results: The Rt was initially high, falling below 1 in May or June depending on the region, before stabilizing around 1 in the later months. The median Rt for Arkansas and Kentucky at the end of the study were 1.15 (95% credible interval [CrI], 1.13, 1.18) and 1.10 (95% CrI, 1.08, 1.12), respectively, and remained above 1 for the non-Appalachian region. Rt decreased when facial coverings were mandated, changing by -10.64% (95% CrI, -10.60%, -10.70%) in Arkansas and -5.93% (95% CrI, -4.31%, -7.65%) in Kentucky. The trends in Rt estimates were mostly associated with the implementation and relaxation of social distancing measures. Conclusions: Arkansas and Kentucky maintained a median Rt above 1 during the entire study period. Changes in Rt estimates allows quantitative estimates of potential impact of policies such as facemask mandate.

Coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-COV-2), was first reported in humans in Wuhan in December 2019. From the early stages of the pandemic to November 2020, there has been a rise in both cases and deaths among states that contain large rural areas in the United States (US). 1 Arkansas, one of eight states that did not implement a stay-at-home order, and Kentucky, a state that has been more proactive from the beginning of the pandemic, are two southern states that have very similar COVID-19 morbidity and mortality rates, yet differed in their approach in addressing this pandemic. Both states have regions that are classified as rural (the Delta in Arkansas and Appalachia in Kentucky), which face higher percentages of health disparities and socioeconomic stress compared to their respective state counterparts.

In both states, disparities in rurality, poverty, health conditions, and healthcare access have a significant role. In Arkansas, 41% of Arkansans live in rural counties, 2 compared to only 14% of the US population who live in nonmetropolitan counties. In Kentucky, 25.3% of individuals in Appalachia live in poverty compared to 15.3% in non-Appalachia. 3 These rural communities face challenges with the pandemic and may be unsuited to handle large surges within their healthcare systems. 4, 5 Fifty percent of rural residents are at a higher risk of hospitalization and serious illness if they became infected with COVID-19 compared to 40% of metropolitan residents because of pre-existing health conditions. 6 Rural residents are more likely to be older, poorer, and have more comorbidities including obesity, diabetes, hypertension, heart disease, and chronic lower respiratory disease than urbanites. [6] [7] [8] [9] [10] The time-varying reproduction number, Rt, represents a pathogen's changing transmission potential over time. As the average number of secondary cases per case at a certain time t, Rt>1 indicates sustained transmission and <1 epidemic decline. [11] [12] [13] Examining the Rt among these two states will provide a better indication of COVID-19 transmission, especially among vulnerable rural areas. Our study aimed to estimate the Rt for COVID-19 within Arkansas and Kentucky and to compare the Rt among the two states, as well to determine if it differs among the urban and rural areas of each state, and to investigate the impact of policies, and preventative and relaxation measures on the Rt.

Using data from the New York Times GitHub data repository, 14 we downloaded the cumulative confirmed case count from March 6 -November 7, 2020, for Arkansas and Kentucky, including the counties located in each state. We used the Delta Regional Authority 15 and the Appalachian Regional Commission 16 to classify the counties in Arkansas as Delta and non-Delta, and Appalachian and non-Appalachian in Kentucky. A detailed list of all 75 and 120 counties of Arkansas and Kentucky are provided in Supplementary Tables 1 and 2. The first case in Arkansas was reported on March 11, 2020 , and the first case in Kentucky was reported on March 6, 2020. The study cutoff point was November 7, 2020. The management of negative incident case counts is described in Appendix A. We merged the county-level data to obtain the regional-level data (Delta, non-Delta, Appalachian, and non-Appalachian). To generate Rt, from the reported cumulative case count numbers, we utilized the daily number of new confirmed COVID-19 cases. We accessed 2019 county-level population data for Arkansas and Kentucky from the U.S. Census Bureau. 17

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint

We downloaded the executive orders from the governors' offices of each state and identified the date of the implementation and relaxation of public health interventions in each state respectively ( Table 1) .

Rt was estimated using the instantaneous reproduction number method as implemented in the R package 'EpiEstim' version 2.2-3. This measure was defined by Cori et al. 11 as the ratio between It, the number of incident cases at the time t, and the total infectiousness of all infected individuals at the time t. This method has been implemented worldwide in multiple studies to estimate the Rt of SARS-CoV-2 and is briefly described in Appendix B. [18] [19] [20] [21] [22] [23] [24] [25] We shifted the time series by nine days backward (assuming a mean incubation period of 6 days and a median delay to testing of 3 days) 26 for generating Rt by the assumed date of infection, 13 and we specified the serial interval (mean = 4.60 days; standard deviation = 5.55 days). 27 Besides using the 7-day sliding window, we also analyze Rt by the different non-overlapping time periods when different combinations of non-pharmaceutical interventions have been implemented, known as policy change Rt (PCRt) thereafter. We estimated the 1-week sliding window Rt and PCRt for both states at the state and regional levels. We calculated the median Rt difference percentage changes and the 95% credible interval (CrI), comparing with the previous policy interval, by bootstrapping (1000 random samples for each Rt distribution) for each state-level PCRt, each respective state region, and the hot-spot analyses for each state (Supplementary Tables 3-6 ).

We also performed the similar analysis at the county-level in which we identified as hot spots based on the reported data and local news (Appendix C). For Arkansas, we analyzed data from Washington, Benton, Lincoln, and Yell Counties, respectively, and combined data from Washington County and adjacent Benton County for analysis as they are one metropolitan area . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint (Supplementary Figure 1) . For Kentucky, we analyzed Jefferson, Shelby, Elliott, and Warren Counties, respectively, and combined data from Jefferson County and adjacent Shelby County for analysis as they are one metropolitan area (Supplementary Figure 2) .

We conducted linear regression between the log10-transformed per capita cumulative case count and the log10-transformed population size, 28,29 at four different dates: May 7 th , July 7 th , September 7 th , and November 7 th . See Appendix D for details and results.

Statistical analysis was performed using R 4.0.3 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria). Maps were created using ArcGIS Pro Version 2.4.0 (Esri, Redlands, CA, USA), with color codes arranged according to quintiles of the values.

The Georgia Southern University Institutional Review Board made a non-human subjects determination for this project (H20364) under the G8 exemption category.

As of November 7, 2020, there were 119,057 cumulative confirmed COVID-19 cases in Arkansas (57,836 for Delta and 61,221 for non-Delta) and 122,024 cases in Kentucky (27,480 for Appalachian and 94,544 for non-Appalachian). Figures 1 and 2 present the spatial variation of cumulative case count and cumulative incidence per 100,000 population by county in Arkansas and Kentucky at four different dates: May 7 th , July 7 th , September 7 th , and November 7 th , 2020, respectively.

From March 11 to November 7, 2020, Arkansas revealed two major surges of new cases in July and October (Figure 3 ). The 7-day sliding window Rt estimates in Arkansas was high at the beginning, nearing an Rt estimate of 3, dropping below 1 in mid-April, and having peaks . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint above 1 for a few months before steadily staying around 1. At the end of the study, the median 7day sliding window Rt estimate was 1.15 (95% CrI, 1.13, 1.18). In the Delta region, the 7-day sliding window Rt estimates had more pronounced decreased peaks in mid-May and mid-June, whereas the non-Delta region had two peaks below 1 in the early stages, an increased peak in mid-May that was above 1, and then stabilized around 1. At the end of the study, the Delta and non-Delta median 7-day sliding window Rt estimates were 1.14 (95% CrI, 1.10, 1.17) and 1.17 (95% CrI, 1.13, 1.20), respectively, with both regions demonstrating extensive community transmission of SARS-CoV-2, with a median Rt >1.

At the beginning, the PCRt estimates were high in Arkansas and both the Delta and non-Delta regions. The PCRt estimates declined statewide (median Rt difference percentage: -53.56%, 95% CrI, -53.1%, -54.1%) and both Delta (-44.56%, 95% CI, -43.4%, -45.8%) and non-Delta regions (-62.67%, 95%, -62.4%, -63.0%) after schools closed on March 17 th . The PCRt estimate remained stable statewide and in the Delta region when gatherings were restricted to 10 individuals or fewer on March 23 rd , but declined by -10.81% (95% CrI, -26.9%, +8.35%) to below 1 in the non-Delta region. The PCRt estimates increased statewide (+6.68%; 95% CrI, +5.58%, +7.75%) and the non-Delta region (+14.29%; 95% CrI, -5.14%, +23.68%) after May 11 th , when restaurant dine-in operations could resume. Both regions (Delta region: -12.08%; 95% CrI, -11.9%, -12.3%; Non-Delta region: -10.97%; 95% CrI, -10.6%, -11.3%), as well as Arkansas as a whole (-10.64%; 95% CrI, -10.60%, -10.70%), saw a decrease in the PCRt estimate when face masks were required in public beginning on July 20 th . There was an increase in the PCRt estimates statewide (+11.56%; 95% CrI, +9.88%, +13.27%) and both regions (Delta region: +9.07%; 95% CrI, +6.85%, +11.18%; Non-Delta region: +14.51%; 95% CrI, +12.3%, +16.7%) after August 24 th , when schools reopened with in-person instruction. 7 From March 6 to November 7, 2020, Kentucky's daily incidence data showed a steady increase (Figure 4) . In Kentucky, the 7-day sliding window Rt estimate was high in March and decreased in April. The Rt estimate had peaks that stayed around 1 and by the end of the study its median was 1.10 (95% CrI, 1.08, 1.12). Both regions (Appalachian and non-Appalachian) demonstrated an extensive community transmission of SARS-CoV-2, with a median 7-day sliding window Rt larger than 1. The Appalachian and non-Appalachian regions' median 7-day sliding window Rt estimates were 1.07 (95% CrI, 1.04, 1.11) and 1.11 (95% CrI, 1.09, 1.14), respectively, at the end of the study.

The PCRt estimates were high among Kentucky and both regions, as the pandemic began spreading through the states. Out-of-state travel restrictions were issued on March 30 th , decreasing the PCRt estimate statewide and in the non-Appalachian region, yet PCRt increased in the Appalachian region (+32.85%; 95% CrI, +30.3%, +35.8%). The PCRt estimate decreased to below 1 in the Appalachian region (-53.51%; 95% CrI, -45.16%, -61.2%) and remained stable in the entire state, after April 4 th , when the state adopted on a voluntary basis guidance from the Centers for Disease Control and Prevention (CDC) recommending that individuals wear cloth masks in some situations. The PCRt estimates statewide (+5.19%; 95% CrI, +4.47%, +5.91%) and both regions (Appalachian region: +33.46%; 95% CrI, +20.7%, +46.8%; Non-Appalachian region: +1.93%; 95% CrI, +1.3%, +2.51%) increased after gatherings of 10 or less were allowed on May 14 th . The PCRt estimates decreased to near 1 statewide (-5.93%; 95% CrI, -4.31%, -7.65%) and both regions (Appalachian region: -13.34%; 95% CrI, -11.5%, -15.2%; Non-Appalachian region: -4.39%; 95% CrI, -2.56%, -6.33%) beginning on July 9 th , with the executive order requiring face coverings in public. There was an increase in the PCRt estimates statewide (+8.97%; 95% CrI, +8.86%, +9.08%) and both regions after September 28 th , when schools . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint reopened with in-person instruction (Appalachian region: +7.49%; 95% CrI, +7.48%, +7.51%; Non-Appalachian region: +9.39%; 95% CrI, +9.23%, +9.56%).

The purpose of this paper was to estimate and compare state and county-level Rt Rt estimates as predictive models and quantitative measures of epidemic growth or decline. [30] [31] [32] Here, the Rt trajectories of Arkansas and Kentucky differed among rural and urban areas, increasing or decreasing, depending on the implementation of preventative and relaxation measures. The Rt will be useful as the pandemic progresses to inform policymakers and public health professions of the direction of potential outbreaks, assisting in preventing health care surges and implementing more preventative measures and policies. For example, both Kentucky and Arkansas implemented mandated facial coverings or masks in July, 2020, which was reflected by a decrease in COVID-19 transmission.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Our study sought to further examine if differences in COVID-19 transmission occurred among location, specifically urban versus rural, since we observed that the role of population size in counties has had a less significant effect on the spread of COVID-19. One study examined trends in the distribution of COVID-19 hotspot counties and found that more hotspot counties were occurring in the southern states of the US during summer months. 33 This follows the trend and wave progression that occurred in the US, hitting the large metropolitan areas first, followed by spread in the Southern region and then in the Mid-West region. Another study found that that many of the less vulnerable counties that had a low Social Vulnerability Index had slightly higher average incidence and death rates early in the pandemic, and as the pandemic progressed, the trends crossed, with many of the most vulnerable counties facing higher rates. 34 Many of the urban metropolitan areas and cities were impacted first, before spreading to the rural areas. This may be due to the linkage of metropolitan areas, through social, economic, and commuting relationships.

Arkansas, one of eight states in the US that did not implement a stay at home or lockdown order, lacked the immediate response, as seen by other states, could explain the higher Rt estimate, as it was at 2 or higher at the beginning of the pandemic. 35 Arkansas had 22 cases before the first preventative measure, the closing of schools on March 17 th , was implemented.

Additionally, the only time the PCRt estimate was below 1 was when face coverings were implemented in July, demonstrating a decrease in COVID-19 transmission. One of the biggest drivers in COVID-19 transmission in Arkansas was the poultry plant outbreaks that occurred among employees and spread through community transmission. 36 In Lincoln County, Arkansas, many COVID-19 cases were attributable to the correctional facility outbreak, rather than community transmission. 37 Additionally, there was an increase in mass testing at the correctional . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2021. ; facility in Lincoln County, which could explain the large peaks in Rt estimates that we observed. 38 One study conducted among a correctional facility in Arkansas observed that if testing for COVID-19 was only among symptomatic individuals, then fewer cases would have been detected, allowing for a greater transmission of disease to occur. 39 At the beginning of the pandemic, many states in the South and Midwest of the US observed increased COVID-19 infection rates, yet Kentucky's rate was notably low. 40 Kentucky took a very conservative method in their approach, as was observed by the policies and measures implemented, to slow the transmission of COVID-19. A decrease in COVID-19 transmission was seen in the Appalachian region, when the state adopted the guidance from the CDC recommending that people wear cloth masks in some situations and when Kentucky passed an executive order requiring face coverings in public. The Kentucky Appalachian region has high rates of comorbidities, especially respiratory diseases due to the coal industry, but saw an increase in mask wearing when required. 41 In Jefferson, Shelby, and Warren Counties in Kentucky, a decrease in PCRt was observed in transmission towards the beginning of the pandemic, when an order was issued to restrict out-of-state travel. This decrease in transmission may have been due to less travel that occurred across state lines, as Warren County is near the Tennessee border and Nashville, the Tennessee capital, and Jefferson and Shelby Counties border Indiana, and is near Cincinnati in Ohio. 42 While the Rt differed among rural and urban areas at the beginning of the pandemic, as the pandemic progressed, the Rt was similar across the urban and rural counties in both states.

Although population size has been found to have a less significant effect on COVID-19 spread than hypothesized at the early pandemic, it is still important to discuss the disparities that occur between rural and urban locations and the implications the pandemic has on rural locations.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021.

Rural areas have had lower testing rates, as well as poorer health care infrastructures to handle cases. 43 Rural health care and public health systems are more vulnerable and have struggled to respond to the COVID-19 crisis. 44 Additionally, most healthcare systems do not have the capacity to handle surges in cases, and only one percent of the nation's intensive care unit beds are located in rural areas. 45 There were several limitations in this study. One limitation was the lack of data on superspreading events that occurred in each state (for example, within prisons 46 and nursing homes, 47 as well as in religious settings, schools and sport camps, and social events 48 ). Many of the counties located in both Arkansas and Kentucky contained large prison populations. The counties of Lincoln, Arkansas and Elliot, Kentucky, both contain county correctional facilities and prisons. 37,49 The reason for the unstable Rt in these counties may stem from disease amplification in prison outbreaks rather than community spread. However, it is difficult to pinpoint certain related outbreaks, and there is limited county-level data specific to correctional facilities. Additionally, there were 1,755 unknown county-level cumulative cases in Arkansas.

These cases were included in our state-level data analysis, but they were excluded from the Delta, non-Delta, and county-level hot spots analyses. Kentucky had all county-level data and all reported cases were used in all analyses.

This study observed that both Arkansas and Kentucky, as well as the respective regions, had an extensive spread of COVID-19, since both states maintained a median Rt above 1. The direction of the trend of the Rt estimates were reflected by the implementation of preventative measures and their subsequent relaxation as the pandemic progressed. This study was able to examine the changing transmission potential of COVID-19 over time in rural and urban areas in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; two socio-demographically similar Southern states. Further research is needed to examine the rural and urban differences in the spread of the COVID-19 pandemic in the US.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ;

Lundeen . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ;

15.

Delta . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ;

Wang Y, Siesel C, Chen Y, et al. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ;

Aspinwall C, Neff J. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ;

Gale . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; Governor Hutchinson announced that the state will be moving into Phase 2 of reopening beginning on June 15, 2020. Under Phase 2, social distancing and facial coverings are still recommended, and restaurants and businesses will be allowed to operate at two-thirds capacity, as opposed to the one-third capacity allowed during Phase 1.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint recommending that people wear cloth masks in some situations. April 9

Ordered Natural Bridge and Cumberland Falls state resort parks to close. April 20

Governor Beshear dvised the commonwealth's education leaders to keep facilities closed to in-person instruction for the rest of the school year. May 6

Governor Beshear issued new executive order that continued to ban anyone with a positive or presumptively positive case of COVID-19 from entering Kentucky, except as ordered for medical treatment. It also kept in place requirements of social distancing on public transportation. Those traveling from out of state into Kentucky and staying were being asked to selfquarantine for 14 days.

Everybody working for an essential business that was reopening should be wearing a mask. D May 14 Groups of 10 people or fewer could gather. E July 9 Required use of face coverings/masks in public. July 20

Cabinet for Health and Family Services has issued new order that pulls back on guidance covering social, non-commercial mass gatherings. The Kentucky Department of Public Health issued a new travel advisory that recommended a 14-day self-quarantine for travelers who went to any of eight states -Alabama, Arizona, Florida, Georgia, Idaho, Nevada, South Carolina and Texasthat were reporting a positive coronavirus testing rate equal to or greater than 15%. The advisory also included Mississippi, which was quickly approaching a positive testing rate of 15%, and the U.S. Commonwealth of Puerto Rico. July 27

Announced the closing of bars for two weeks, effective, Tuesday, July 28. Announced that restaurants will be limited to 25% of prepandemic capacity indoors; outdoor accommodations remain limited only by the ability to provide proper social distancing.

Recommended that public and private schools avoid offering in-person instruction until the third week of August. G August 6 Extended the state's mandate requiring face coverings in some situations for another 30 days. August 10 Governor Beshear recommended that schools waited to begin in-person classes until Sept. 28.

Issued an executive order allowing bars and restaurants to operate at 50% of capacity, as long as people could remain six feet from anyone who was not in their household or group. Bars and restaurants would be required to halt food and beverage service by 10 p.m. and close at 11 p.m. local time.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ;

Governor Beshear offered an update on his administration's travel advisory, which recommended a 14-day self-quarantine for Kentuckians who traveled to states and territories that were reporting a positive coronavirus testing rate equal to or greater than 15%. Extended the state's mandate requiring face coverings in some situations for another 30 days.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; 

If there were any negative daily case counts in the data (i.e., when public health agencies made corrections to their cumulative case counts at a specific date or dates), they were identified and adjusted by changing the negative cases counts to zero and correcting the daily case counts on previous days such that the cumulative case counts of the previous days would not exceed the cumulative case counts of the day that reported negative daily case counts. We adjusted the negative daily case counts at both state and county levels. These data management steps were done in R 4.0.3.

Time-varying reproduction number, denoted as Rt, was estimated using the R package is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; method of instantaneous reproduction number as implemented in EpiEstim assumes that the Rt is constant over a given time frame of size τ ending at time t. This will enable a precise estimate to be estimated with limited variability. Given that transmissibility is assumed to be constant over a period of time, from (t-τ+1) to t, and is denoted by a reproduction number, Rt,τ, the possibility of the number of daily new cases during the time period, from I(t-τ+1) to It, is contingent on daily number of new cases prior to the time period, i.e., from I0, to I(t-τ). 1 Cori et al. 1 derived an analytical expression of the posterior distribution of Rt and thus estimated its median, the variance, and the 95% credible interval by using a Bayesian framework with a Gammadistributed prior to Rt,τ,

In this paper, we estimated Rt using both a 7-day sliding window and non-overlapping time periods specified by the dates of policy changes. The data was analyzed using EpiEstim version 2.2-3. 1

We chose four Arkansas counties for further analysis: Washington County, Benton County, Lincoln County, and Yell County. We combined Washington County (where Fayetteville is) and Benton County (second most populous county in Arkansas) together for analysis because these two counties are geographically next to each other and large metropolitan areas, and therefore the frequency of daily communication is high.

Washington County and Benton County's daily number of new cases revealed two major surges in June and September (Supplementary Figure 1) . The 7-day sliding window Rt estimates in Washington County and Benton County was low at the beginning of the pandemic, either near 1 or below 1, reaching a peak of an Rt estimate of 2 in mid-May, before steadily staying around 1 and increasing to above 1 in a peak occurring in mid-August. At the end of the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint study, the median Rt estimate was 1.30 (95% CrI, 1.23, 1.36). For Lincoln County, a major outbreak occurred in April. Lincoln County started with peaks with the 7-day sliding window Rt estimates nearing 3, and then decreasing to below 1 and then having steady increasing and decreasing peaks as the pandemic progressed. At the end of the study, the median Rt estimates for Lincoln County was 0.66 (95% CrI, 0.42, 0.99). For Yell County, the major surge of new cases occurred between June and July. Yell County had peaks in the beginning with the 7-day sliding window Rt estimates above 3, before having smaller increasing and decreasing peaks that remained around 1. At the end of the study, the median Rt was 1.24 (95% CrI, 0.90, 1.65).

Except Lincoln County, the rest of the three counties demonstrated an extensive community transmission of SARS-CoV-2 transmission with the median 7-day sliding window Rt more than 1.

Among the policy change Rt plots, Washington and Benton counties followed a similar pattern that was seen for the overall state of Arkansas. There was a decrease in the policy change Rt estimate when gatherings were restricted to 10 or fewer on March 23 rd (median Rt difference percentage: -30.66%; 95% CrI, -28.0%, -34.0%), but increased (+68.04%; 95% CrI, +15.7%, +147.6%) again to above 1 when businesses, manufacturers, construction companies and places of worship were required to implement social distancing protocols on April 4 th . There was another decrease (-9.77%; 95% CrI, -8.24%, -11.25%) in the policy change Rt estimate to below 1 when face coverings and masks were required in public on July 20 th , but it increased (+22.08%; 95% CrI, +19.5%, +24.7%) once schools reopened for in-person instruction on August 24 th . Lincoln County did not have a clear pattern in terms of policy change Rt estimates.

For example, the largest decline in the policy change Rt estimate was when the state allowed for the reopening of restaurants, with the continuation of dine-in operations on May 11 th (-46.87%;

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint 95% CrI, -25.5%, -62.4%). There was also a slight, but not significant increase in the policy change Rt estimate (+6.15%; 95% CrI, -10.4%, +23.9%) when face coverings and masks were required, and also a decline below 1 in the policy change Rt estimate (-22.91%; 95% CrI, -15.7%, -29.9%) when schools reopened again with in-person instruction. However, Lincoln County contains a large correctional facility that may be driving the transmission in a congregate setting instead of community transmission. 2 Yell County did not have a significant decrease when face coverings and masks were required (-2.31%; 95% CrI, -14.7%, +13.6%), nor a significant increase when schools reopened with in-person instruction (+4.18%; 95% CrI, -12.4%, +23.8%).

We chose four Kentucky counties for further analysis: Jefferson County, Shelby County, Elliot County, and Warren County. Jefferson County is where Louisville is and Shelby County is its suburb. We combined Jefferson County and Shelby County together for analysis because these two counties are geographically next to each other therefore the frequency of daily communication is high. The Jefferson County and Shelby County's daily number of new cases reveal steady increase (Supplementary Figure 2) , and the median 7-day sliding window Rt estimates was 1.05 (95% CrI, 1.01, 1.09) for the assumed date of infection. For Elliott County, the major outbreak happened on October. As of November 7, 2020, the median 7-day sliding window Rt estimate Elliott County was 0.94 (95% CrI, 0.70, 1.24). For Warren County, it had a steady increase with the major surges of new cases in May. As of November 7, 2020, the median 7-day sliding window Rt was 1.23 (95% CrI, 1.11, 1.35). All four counties demonstrated an extensive community transmission of SARS-CoV-2 transmission with the median 7-day sliding window Rt very close to and more than 1.

Jefferson County and Shelby County in Kentucky followed a similar pattern that was seen for the overall state of Kentucky for the policy change Rt plots. There was a decrease in the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; policy change Rt estimate when out of state travel was restricted on March 30 th (median Rt difference percentage: -57.09%; 95% CrI, -48.1%, -64.5%), but increased again to above 1 on April 4 th (+59.92%; 95% CrI, +32.0%, +95.6%), when the state adopted on a voluntary basis the new guidance from the CDC recommending that individuals wear cloth masks in some situations. The policy change Rt estimate decreased when groups of 10 could gather (-30.66%; 95% CrI, -28.0%, -34.0%), remained the same when face coverings and masks were required in public (+1.79%; 95% CrI, -0.08%, +3.64%), and increased once schools reopened for in-person instruction (+6.21%; 95% CrI, +5.8%, +6.63%). Sustained low transmission was found in Elliot County even after facemasks was mandated in July. However, the real change was the reopening of schools with in-person instructions. An outbreak in October ensued with a policy change Rt>1.

Warren County had a policy change Rt estimate of 1.52 (95% CrI, 0.59, 3.12) at the beginning of the pandemic, and 1.97 (95% CrI, 1.31, 2.81) after schools were ordered to close (+29.33%; 95% CrI, -42.9%, +253.7%). The policy change Rt estimate decreased to below 1, when the order was issued to restrict out-of-state travel (-54.23%; 95% CrI, -14.90%, -76.2%). However, there was no significant increase when facial coverings were mandated (+2.37%; 95% CrI, -5.07%, +10.0%) and increased to above 1 when schools reopened with in-person instruction (+11.59%; 95% CrI, +7.45%, +16.83%).

As in Fung et al., 3 we explored the power-law relationship between the cumulative case count of COVID-19 (C) and their Census-estimated population size (N), C~N g (where g is the exponent), as follows:

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2021. ; https://doi.org/10.1101/2021.05.26.21257862 doi: medRxiv preprint log 10 (cumulative case count) log 10 (population size) = log 10 (per capita cumulative case count × population size) log 10 (population size) = log 10 (per capita cumulative case count) log 10 (population size) + log 10 (population size) log 10 (population size) = log 10 (per capita cumulative case count) log 10 (population size)

Where m is the slope of the regression line between the log-transformed per capita cumulative case count and the log-transformed population size.

Per capita cumulative incidence would be exactly proportional to population size, and there was no heterogeneity of per capita cumulative incidence across geographic units of different population sizes if m=0 (i.e., g=1). Geographical units with lower population sizes would have a higher per capita cumulative incidence if m<0 (i.e., g<1) and lower per capita cumulative incidence if m>0 (i.e., g>1).

We only present the 1-week sliding window Rt and policy change Rt with y-axis ranging between zero to four because anything larger than 4 with small data size is not reliable. Given that the R0 of COVID-19 is accepted to be around 3 with some viral variants with a slightly higher R0 4, 5 and that Rt would be < R0, it is reasonable to assume 0 < Rt < 4. The same method applied to the starting date of Rt estimation: the R package EpiEstim has a default Rt estimation (mean value of the prior) of 5 and the 95% CrI is wide with limited daily case count. Therefore, in the county-level hot spot analyses, we only recorded the Rt estimation from the month that those locations experienced first major local outbreaks (e.g., for Elliot County since it had a small daily case count in the beginning of the pandemic until August, 2020).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. indicating that counties with a higher population size would have a higher per capita cumulative incidence. However, for other regions, there was no evidence to reject the hypothesis of homogeneity of per capita cumulative incidence across counties of different population sizes.

At the beginning of the pandemic, it was hypothesized that population density and size was the main factor for the rapid spread of COVID-19, especially in cities such as Wuhan, New York, and Milan. 6 However, many high-population, high-density cities, including Hong Kong 7 and Seoul, 8 were able to successfully limit both COVID-19 cases and deaths with populationlevel interventions, such as mask wearing, social distancing, and contact tracing. One study found that denser places were not linked to higher infection rates and were associated with lower COVID-19 death rates. 9 Another study found that larger metropolitan size was linked to higher infection and mortality rates over time, however, during the same time period, higher population density was linked to lower infection and mortality rates. 10 Although our study examined the differences in Rt among urban and rural locations to determine if rurality had a significant role in the spread of COVID-19, we did not examine population density; however, we did find that population size was not a factor, with the exception of the non-Appalachian region. This supports the hypothesis that population size has not played a major role in the spread of COVID-19.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Another study found that population outflow and migration in Wuhan contributed to the spatial-geographical spread of COVID-19 in that region. 11 We observed higher levels of COVID-19 transmission in the hot spot analyses of the metropolitan counties of Washington and Benton in Arkansas, as well as the metropolitan counties of Jefferson and Shelby in Kentucky. This may be due to a larger number of commuters from surrounding counties that travel for work or school; however, further research is needed to investigate if the COVID-19 transmission in these metropolitan counties was due to commuters or the population outflow and inflow since this study did not examine the spatial-geographical spread among communities and counties. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

The Worst Virus Outbreaks in the U.S. Are Now in Rural Areas. The New York Times

Rural Profile of Arkansas 2019: Social & Economic Trends Affecting Rural Arkansas

The Appalachian Region: A Data Overview From The

Pre-Existing Health Disparities Could Affect COVID-19's Impact In Rural Communities. Health News Florida

Far from immune, rural areas face unique COVID-19 challenges

Half of Rural Residents at High Risk of Serious Illness Due to COVID-19, Creating Stress on Rural Hospitals

A New Framework and Software to Estimate Time-Varying Reproduction Numbers During Epidemics

More prison deaths … COVID-19 related

Assessing Early Heterogeneity in Doubling Times of the COVID-19 Epidemic across Prefectures in Mainland China

Epidemiology of SARS-CoV-2

Estimated transmissibility and impact of SARS-CoV-2 lineage B.1.1.7 in England

High population densities catalyse the spread of COVID-19

What can countries learn from Hong Kong's response to the COVID-19 pandemic?

Response to COVID-19 in South Korea and implications for lifting stringent interventions

Does Density Aggravate the COVID-19 Pandemic

Longitudinal analyses of the relationship between development density and the COVID-19 morbidity and mortality rates: Early evidence from 1,165 metropolitan counties in the United States

Population flow drives spatio-temporal distribution of COVID-19 in China