Time-varying Reproduction Numbers of COVID-19 in Georgia, USA, March 2, 2020 to November 20, 2020

In 2020, the COVID-19 pandemic caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) impacted the state of Georgia as well as other jurisdictions within the US. Within Georgia, Metro Atlanta counties have been the hardest hit by the virus, with thousands of confirmed cases cumulatively: 37,238 in Fulton, 36,407 in Gwinnett, 25,853 in Dekalb, 26,255 in Cobb, and 12,314 in Hall County as of November 20, 2020.¹ Dougherty County, with Albany as the county seat, was an early COVID-19 hotspot in southeastern Georgia and reported a large number of cases (as of November 20, 2020: cumulative number, 3431; incidence rate, 3816 per 100,000 individuals).¹ In Georgia, every county government had the power to impose preventive measures to reduce viral transmission as they see fit, before the state imposed a state-wide emergency that overrode the autonomy of county governments (Table 1). On March 23, 2020, the Georgia state government issued an executive order requesting citizens with underlying conditions and those with a COVID-19 diagnosis to shelter in place.² Certain businesses were to remain closed and no more than 10 individuals could gather in a location without maintaining a distance of at least 6 feet. The order also called for restaurants to offer curbside pickup or delivery only.² On April 2, 2020, a state-wide shelter-in-place ordinance was enacted by the governor, allowing only essential services to operate (implemented on April 3).³ The Georgia state government announced on April 27, 2020, during a press conference, that services such as beauty salons, barber shops, stores, and restaurants can reopen if they follow pertinent social distancing measures specified by the state.⁴ On May 12, 2020, the state government recommended residents and visitors to the state wear face coverings, practice social distance, and limit gatherings. On July 28, 2020, and with a renewal on November 1, 2020, all individuals in the state of Georgia with a positive or suspected COVID-19 diagnosis should isolate until their infectious period is over, and those exposed to the virus should comply with a 14-day quarantine.⁵ As the COVID-19 epidemic in Georgia continues, it is important to quantify the epidemiologic characteristics of COVID-19 so that we may formulate policies and implement interventions to minimize transmission and mortality.

To characterize the transmission potential of an epidemic, it is necessary to calculate the reproduction number based on the trajectory of the incidence curve.⁶ The basic reproduction number, R₀, is the average number of secondary cases that 1 primary case can generate in a completely susceptible population in the absence of behavioral changes or public health interventions.⁶ The estimated values of R₀ for SARS-CoV-2 vary across geographic locations. An early study of the epidemic in Wuhan reported an R₀ value of 2.2, assuming a serial interval of 7.5.⁷ A more recent study of the epidemic in China, adjusted for the changing case definition, estimated an R₀ value of 1.8 to 2.0 (assuming a serial interval of 7.5) or 1.4 to 1.5 (assuming a serial interval of 4.7).⁸ Assuming a serial interval of 4.4, our analysis of confirmed COVID-19 cases in Iran estimated the mean R₀ value as 3.5 or 4.4, depending on the statistical method chosen.⁹

In contrast, the time-varying reproduction number, R_t, is a time-dependent estimate of the secondary cases that arise from 1 case at time t, when depletion of the susceptible population, behavioral changes, and measures to control transmission of disease have taken place.^10,11 As with R₀, if R_t > 1, it indicates there is sustainable transmission in the population. When R_t < 1, disease transmission cannot be sustained, and it is used as an indication of the effectiveness of infection control measures.^6,10

Various statistical methods have been proposed to estimate R_t. Their strengths and weaknesses have been recently assessed by researchers who compared the performance of different methods using synthetic epidemic data^12,13 and observed COVID-19 incidence data.¹⁴ We chose to use an oft-used method, known as the instantaneous reproduction number method, as implemented in the R package EpiEstim version 2.2-3 (R Version 1.2.5033 Windows NT 10.0; R Core Team, Vienna, Austria).^10,11 This Bayesian method provides an estimate of the average R_t over a short time window specified by the user (in our study, a 7-day window that ends at time t). It treats the fluctuation in incidence data as signals of an increasing or decreasing reproduction number. This method has been used to estimate COVID-19 R_t values in jurisdictions such as mainland China,¹⁵ Hong Kong,¹⁶ Iran,¹⁷ South Korea,¹⁸ Italy,¹⁹ Nigeria,²⁰ and Switzerland.²¹

Our study aimed to estimate R_t for COVID-19 in Georgia, urban Metro Atlanta counties, and rural Dougherty County and its surrounding counties, analyzing data from March 2, 2020, through November 20, 2020, as the state incrementally implemented and then relaxed social distancing interventions (Table 1).

This study uses data from the COVID-19 pandemic, March 2, 2020 to November 20, 2020, in the state of Georgia, all Metro Atlanta counties (Supplemental Tables 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, and 12^a), and Dougherty County and its surrounding counties. Metro Atlanta is defined by the US Office of Management and Budget as the “Atlanta-Sandy Springs-Alpharetta, Georgia Metropolitan Statistical Area.”²² The list of Metro Atlanta counties is provided in Table 2.

Community transmission of SARS-CoV-2 remained ongoing in Georgia based on incidence data by assumed date of infection from February 22, 2020, to November 11, 2020 (date of report: March 2, 2020–November 20, 2020). As of November 11, 2020, the median EpiEstim R_t estimate was 1.03 (95% CrI: 1.03, 1.03). The same results were observed for Metro Atlanta, with an R_t estimate equal to 1.03 (95% CrI: 1.03, 1.03). The transmission may have been under control for Dougherty County, with the R_t estimate being 1.01 (95% CrI: 1.00, 1.02) (Table 2).

As social distancing measures unfolded and then relaxed in Georgia during our study period, the median EpiEstim R_t estimate in Georgia dropped from 1.14 (95% CrI: 1.11, 1.17) on June 14, 2020 to 1.03 (95% CrI: 1.03, 1.03) until November 11, 2020, as the assumed date of infection. The median R_t estimate fluctuated around 1 from mid-March to November 11, 2020 (Figure 1).

Regarding Metro Atlanta (Figure 1), the EpiEstim R_t estimate fluctuated above 1.5 before the end of March and gradually decreased to around 1 by May through November 11. The R_t estimates for each of the Metro Atlanta counties fluctuated around 1 during our study period (Table 2, Supplemental Figures 2, 3, 4, 5, and 6^a).

For Dougherty County (Figure 1), we observed a speedy decline in EpiEstim R_t estimates from around 2 in mid-March to around 1 in mid-April; these values were maintained around 1 up to November 11, 2020, when the mean R_t estimate was observed at 1.01 (95% CrI: 1.00, 1.02). This finding was driven primarily by the early epidemic observed in Dougherty County, where large clusters of cases were infected via 2 funerals that happened to be superspreading events.²⁶ On March 13, 2020, the mean R_t estimate for Dougherty County was 2.63 (CrI: 2.27, 3.02). Counties surrounding Dougherty (Baker, Calhoun, Lee, Mitchell, Terrell, and Worth counties) presented mean EpiEstim R_t estimates around 2.00, as its outbreak developed (Figure 2, Supplemental Figure 1^a). It was observed that the surrounding counties around Dougherty also obtained mean R_t estimates reaching 2 in all of them, except for Baker County, which had the greatest R_t median estimate of 1.78 (CrI: 1.14, 2.56) for March 26, 2020 (Figure 2). A week after the first case was reported in Dougherty County, Calhoun, Mitchell, Terrell, and Worth counties presented estimates greater than 2 (Table 3). The R_t median estimates decreased to near 1.00 up to November 11, with the exception of Calhoun County, which presented a median point estimate of 0.99 (95% CrI: 0.95, 1.04) (Table 2).

Community transmission of SARS-CoV-2 remained ongoing in Georgia as of November 11, 2020 (ie, approximately 3 weeks after the presidential election). On April 27, after implementing strict social distancing measures, Georgia reopened some sectors of the economy, with specific guidelines pertinent to social distancing.⁴ As the economy slowly reopened and unprotected social mixing increased, and events such as the presidential election occurred, an increase in the daily number of new confirmed cases was observed starting in June and continuing until November (study period) as SARS-CoV-2 transmission continued unabated.¹ Our study documents the decrease in R_t following social distancing interventions in Georgia and provides further evidence that social distancing measures remained important to keep COVID-19 under control. Our findings are supported by the analysis of Lau et al (2020), in which they also registered a decreased in the effective reproductive number for Dougherty County after the shelter-in-place order was mandated, with estimates decreasing from 5.19 (95% CrI: 5.01, 5.31) to less than 1, and then fluctuating around 1 weeks later.²⁷

Furthermore, many residents in both rural and urban Georgia are medically vulnerable. A recent analysis by The Surgo Foundation²⁸ estimated the COVID-19 community vulnerability index for Dougherty County, by combining epidemiologic risk factors for infection and sociodemographic factors, at very high levels (COVID-19 community vulnerability index = 0.87) when compared with counties in Metro Atlanta (Fulton county’s COVID-19 community vulnerability index = 0.42). The EpiEstim R_t estimates for Dougherty County fluctuated near 1 since April, with a slight increase near August and then continuing to fluctuate near 1 up to November 11, 2020.

The relaxation of social distancing measures should be implemented with an abundance of caution because of the population’s vulnerability. With the mean R_t estimates in Georgia and almost all counties included in our study remaining near 1 for more than 6 months, we believe that mandating nonpharmaceutical interventions, such as wearing facemasks when outdoors,²⁹ could help decrease the mean R_t estimates even more. Another important factor for consideration is access to health care and surge capacity in hospitals, especially in rural Georgia. The health-care system in Dougherty County was impacted heavily by the surge of COVID-19 patients driven by superspreading events.²⁶

Our study evidences the negative effects the superspreading event in Dougherty County caused in surrounding counties. One week after the increase in cases in Dougherty County, neighboring areas showed an increase in their mean R_t estimates. These results reflect local transmission of SARS-CoV-2 in rural areas in Georgia as the epidemic spread from Dougherty County to neighboring counties.

The resumption of economics activities, mobility of young adults, and reopening of educational institutions led to the resurgence of COVID-19 cases in Georgia, as observed in July and August.³⁰ Further research into the spatiotemporal variation of SARS-CoV-2 transmission potential, and its association with economic and medical vulnerability will shed light on the disease and economic burden of COVID-19 in Georgia.

Our study estimated R_t values using the instantaneous reproduction number method implemented in the R package EpiEstim.^10,11 The EpiEstim estimate is sensitive to fluctuation in daily incident case counts as the instantaneous reproduction number method treats such changes as meaningful signals reflecting genuine increases or decreases in transmission potential. The instantaneous reproduction number method in the EpiEstim package can be used if the purpose is to identify time-dependent changes in the R_t estimate that reflect the implementation or relaxation of social distancing measures over time. However, cautious interpretation is needed, especially at the beginning of the outbreak, as the case count was small and the R_t estimate was unstable.

Regarding the time window chosen for EpiEstim, we used a window of 7 days in our analysis. We did not use a window of < 7 days, because a weekend effect was observed in the data (ie, the daily number of cases reported during the weekend was consistently less than those reported during the weekdays before or after the weekend).

The R_t estimate of SARS-CoV-2 has been fluctuating around 1 for Georgia, Metro Atlanta, and Dougherty County and its neighboring counties since the Georgia economy reopened in late-April 2020. Social distancing and other personal protective behavior (such as face coverings) appear to keep the SARS-CoV-2 transmission potential at a reduced level. Government agencies should weigh carefully the next steps of their COVID-19 response plans for their communities, considering ongoing transmission across Georgia and the potential surge after the holiday season.

We acknowledge Bryan O. Sepulveda-Bahamundi, MS, for his contribution to data collection for this project.