key: cord-0765004-dkb95zvi authors: Wang, Y.; Siesel, C.; Chen, Y.; Lopman, B.; Edison, L.; Thomas, M.; Adams, C.; Lau, M. S.; Teunis, P. F. M. title: Transmission of COVID-19 in the state of Georgia, United States: Spatiotemporal variation and impact of social distancing date: 2020-10-26 journal: nan DOI: 10.1101/2020.10.22.20217661 sha: d557224790d2b3d66ebccf42b9c7dd3953729292 doc_id: 765004 cord_uid: dkb95zvi Background Beginning in early February 2020, COVID-19 spread across the state of Georgia leading to 258,354 cumulative cases as of August 25, 2020. The time scale of spreading (i.e., serial interval) and magnitude of spreading (i.e., Rt or reproduction number) for COVID-19, were observed to be heterogenous by demographic characteristics, region and time period. In this study, we examined the COVID-19 transmission in the state of Georgia, United States. Methods During February 1-July 13, 2020, we identified 4080 transmission pairs using contact information from reports of COVID-19 cases from the Georgia Department of Public Health. We examined how various transmission characteristics were affected by disease symptoms, demographics (age, gender, and race), and time period (during shelter-in-place and after reopening). In addition, we estimated the time course of reproduction numbers during early February-mid-June for all 159 counties in the state of Georgia, using a total of 118,491 reported COVID-19 cases. Findings Over this period, the serial interval appeared to decrease from 5.97 days in February-April to 4.40 days in June-July. With regard to age, transmission was assortative and patterns of transmission changed over time. COVID-19 mainly spread from adults to all age groups; transmission among and between children and the elderly was found less frequently. Younger adults (20-50 years old) were involved in the majority of transmissions occurring during or after reopening subsequent to the shelter-in-place period. By mid-July, two waves of COVID-19 transmission were apparent, separated by the shelter-in-place period in the state of Georgia. Counties around major cities and along interstate highways had more intense transmission. Interpretation The transmission of COVID-19 in the state of Georgia had been heterogeneous by area and changed over time. The shelter-in-place was not long enough to sufficiently suppress COVID-19 transmission in densely populated urban areas connected by major transportation links. Studying local transmission patterns may help in predicting and guiding states in prevention and control of COVID-19 according to population and region. Funding Emory COVID-19 Response Collaborative. Introduction time-varying effective reproduction number (R t ) using the methods developed by Wallinga and Teunis [14, 15] . With R t s over time, we can study the spatial distribution of transmission across all 159 Georgia 104 counties, as well as the effects of shelter-in-place and subsequent gradual reopening. Based on reported contacts with confirmed cases, pairs of primary case-patients (infectors) and their 118 secondary case-patients (infectees) were identified, assuming the symptom onset of any infector occurred 119 before the symptom onset of his/her infectees. The serial interval for symptom onset was defined as the 120 number of days between symptom onset for a primary case and an associated secondary case. Thus, 121 serial intervals were assumed to be always positive. Transmission pairs with serial intervals longer than 122 15 days were dropped as such long intervals are unlikely, as shown in previous studies [16, 17] . We 123 modeled the serial interval as a gamma distribution and maximum likelihood estimators of shape and 124 scale parameters were obtained. Furthermore, we explored whether the duration of the serial interval 125 varied by demographic characteristics, various disease symptoms, and time periods. For example, do 126 infectious cases with specific symptoms (e.g., coughing) lead to a shorter serial interval when infecting a 127 descendant case compared to cases without those specific symptoms? The large numbers of linked cases 128 enabled us to examine the variation in transmission within and between different groups by age, sex, and 129 race. these two case-patients can be calculated using the serial interval distribution as a kernel density [15] . 136 Additional information at an individual level (e.g., evidence of social contact between case-patients i and 137 j) is accounted for by a n ⇥ n weighting matrix [15] . The transmission probability matrix V can be 138 estimated in a Markov chain Monte Carlo procedure [15, 4] . 139 When the transmission probability matrix is known, it can be used to calculate reproduction numbers. 140 Elements of row i show the probabilities of case-patient i having received their infection from any other 141 case-patient in the observed population. Rows of V must therefore add to 1. Likewise, elements of 142 column j show the probabilities that case-patient j has transmitted their infection to any other case-143 patient in the observed population. Columns of V therefore add to an estimate of the number of cases 144 infected by case-patient j: its reproduction number. Using GDPH data on cases confirmed with COVID-19 during February 1-July 13, 2020, we estimated 146 effective reproduction numbers (R t ) by date, using dates of symptoms onset and social contact informa-147 tion (wherever available) in each county independently by estimating the transmission probability matrix. Among 118,491 confirmed cases, 48,887 (41.3%) had a missing date of symptom onset. These missing 149 symptom onset dates were imputed based on dates of first specimen collection when available, or dates 150 of laboratory report otherwise. The number of days between symptom onset and date of first specimen 151 collected (or date of laboratory report) was modeled using negative binomial regression with the date of 152 first specimen collected (or date of laboratory report) as the predictor. Estimation of R t approaching present date (i.e., nowcasting) was not feasible because not all contribut-154 ing cases had become symptomatic and been reported. To avoid this right censoring issue, the most 155 recent four weeks (June 16-July 13) were removed from the analysis. It is likely to have multiple 156 waves of COVID-19 transmission in the state of Georgia with considerable variation among Georgia Role of the funding source 170 The funder of this study had no role in study design, data collection, data analysis, data interpretation, or 171 writing the report. All authors had full access to all the data in the study, and the corresponding author 172 had final responsibility for the decision to submit for publication. Tracked Pairs: Serial Interval 175 Based on 4080 tracked pairs of primary case-patients (infectors) and their secondary case-patients (in-176 fectees) in the state of Georgia (Table S2) , the serial interval distribution in Georgia was estimated as a 177 gamma distribution with shape parameter 2.00 and scale parameter 2.49. Table S3 shows the variation 178 in serial interval distribution by subgroup, for clinical and demographic categories. Generally, the serial 179 interval was longer when primary case-patients had severe clinical outcomes, such as hospitalization, un-180 dergoing ventilation, having an abnormal chest X-ray result, or death. Specific symptoms in the primary 181 cases including fever, cough, shortness of breath, or diarrhea did not shorten serial intervals. Serial inter-182 vals were not differentiated across demographic categories including age, sex, race, or location. Figure 183 1 shows estimated serial interval distributions for the three different time periods May, 184 and June-July). The average serial interval became shorter over time: from 5.97 days in February-April, 185 to 5.03 days in May, and then to 4.40 days in June-July. Tracked Pairs: Characteristics of Transmission To study the variation in transmission by age, the frequencies of tracked links can be shown in an age 188 matrix. Figure 2 (a)-2(b) show the disease transmission within and between age, sex, and race groups. Males were twice as likely to infect a female as to infect a male, while females were equally as likely to 190 infect a male or a female. Transmission between races was strongly assortative. Compared with people 191 of other races, white people were 4.4 times as likely to infect white people and black people were 5.6 192 times as likely to infect black people. Based on these data, COVID-19 seemed to mainly spread from adults aged 20-60 years during July; transmission among and between children (<20 years old) and the elderly (>60 years old) was ob-195 served less often, suggesting that transmission occurred more frequently between people of similar ages. 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 October 26, 2020. ; https://doi.org/10. 1101 /2020 Temporal and Spatial Patterns of Transmission 204 During February and March, reproduction numbers started higher than 1 and then decreased until late 205 April and early May, which can be considered as the "first wave" in Georgia. R t usually decreased to 206 a (mathematical) local minimum during the shelter-in-place period but started to increase again as the 207 "second wave" took off. As was observed during the first wave, the R t s peaked and then started to 208 decrease ( Figure 5 ). Although the number of reported cases was lower in first wave, the R t (around 3) 209 was much higher compared with the second wave. Although the general pattern of COVID-19 transmission was similar across all counties, the dates of 211 local maxima/minima (i.e., first peak, local minimum, and second peak) and the magnitude of R t at 212 these extremes varied among counties. Figure 6 (a) shows the peak dates for the first wave in counties 213 with cumulative case numbers higher than 200 cases by July 13, 2020. At that time, counties with high 214 numbers of COVID-19 cases were located around cities and along highways. Starting in early February, 215 COVID-19 spread radially and along the interstate highway from Atlanta and Albany, the two initial 216 outbreak sources. Other cities, including Augusta and Savannah had outbreaks later. Figure 7 shows that 217 74.7% (65/87) of counties with more than 200 cumulative cases by July 13th reached a local minimum in 218 R t during the shelter-in-place period (April 3-April 30). After reopening, many counties experienced a 219 strong second wave with increased numbers of COVID-19 cases reported. Based on the magnitude of R t at first peak, local minimum, and second peak, we categorized case data 221 into the following patterns ( Figure 5 ): • Consistent spreading: sustained transmission of COVID-19 (R t > 1) during the shelter-in-place 223 period. Consequently, numbers of cases were high and increased rapidly upon reopening. • Two strong waves: a first wave of early transmission followed by a slowdown (R t < 1) during 225 the shelter-in-place period. After reopening, a new surge in cases (1  R t < 2) appeared. • Strong first wave: a considerable number of cases during the initial period of the outbreak. During 227 the shelter-in-place period spreading was controlled and after reopening no new surge in cases 228 occurred (R t < 1). • Strong second wave: there were few cases during the early transmission period, but a surge 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 October 26, 2020. to unwillingness or inability of some contact persons to respond to contact tracers. Therefore, during 261 February-April, older cases with more severe symptoms may have been over-represented in the notified 262 case data. More recent data collected when availability of testing has improved are less likely to have this 263 bias. As Figure 1 shows, contraction of serial intervals continued into May-early July, so that the changes 264 may still be explained at least partly by increased prevalence and increased contact rates. tracked case pairs in this study were more likely to be observed when case-patients knew each other (e.g., family members, friends, or colleagues), while transmissions in public spaces (e.g., stores or restaurants) 272 usually could not be traced. Transmission occurs frequently among subjects of the same age group, and 273 less likely among different age groups. Figure 3 shows that some off-diagonal (between different age 274 groups) transmission occurred in Georgia, although this may have been across generations, for example, 275 9 . 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 October 26, 2020. ; https://doi.org/10. 1101 /2020 between parents and children, or grandparents and grandchildren [21] . 276 As to differences in transmission by sex, a primary case-patient who was male was more likely to pass 277 their infection to a female contact than to a male contact. Figure 3 (b) also shows that female case-patients 278 were infected by male case-patients across a wide range of ages, while Figure 3 (a) shows that male case-279 patients were mainly infected by young male case-patients. A possible explanation may be that females 280 tend to be caregivers, taking care of sick people in the household, while young males may be more likely 281 to acquire infection from outside the household. Like the serial interval, transmission patterns also changed as the pandemic continued. The major con- During the shelter-in-place period (April 3-April 30), COVID-19 transmission slowed and the reproduc-300 tion numbers reached a local minimum in most counties. However, before reopening, the reproduction 301 numbers were still above 1 in many counties even at the local minimum, indicating continued disease 302 spreading ( Figure 6(b) ). After reopening, transmission increased again across the state of Georgia. These 303 data suggest that the three or four weeks of shelter-in-place orders were not long enough to sufficiently 304 suppress COVID-19 transmission (local and out-of-state imported) in densely populated urban areas con-305 nected by major transportation links (i.e., airports and interstate highways). Thus far, the second wave has been heterogenous in time and magnitude in different counties. Local 307 prevalence was different at the time of reopening, and counties with high prevalence (i.e., counties bor-308 dering cities and along interstate highways) experienced a stronger second wave. Counties (e.g., those 309 counties around Albany) not connected by major transportation links often saw a second wave of COVID-310 19 cases as well, but on a relatively small scale. Finally, some counties (e.g., Lee, Sumter, Terrell, and 311 Mitchell), that saw an early and intense first wave, did not experience an obvious second wave. Possibly, 312 inhabitants of these counties were more compliant with the disease prevention and control measures. . 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 October 26, 2020. ; https://doi.org/10. 1101 /2020 Limitations Despite having information for more than one hundred thousand cases, a majority of the records had 315 missing reported contacts with a confirmed case. The tracked pairs were missed not at random, since 316 contact tracing is voluntary and its capacity was limited at the early stage of the pandemic. Also, tracked 317 pairs were more likely to be recorded when they involved known contacts; and identifying transmission 318 links in public spaces or in a cluster of cases is challenging. In addition, testing capacity of COVID-19 319 showed a spatiotemporal variation with only cases with severe symptoms detected during February-April. The raw data used in this study are available from the corresponding author upon reasonable request. . 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 October 26, 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. The copyright holder for this preprint this version posted October 26, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 Tables 404 Table 1 : Descriptive statistics of demographic and clinical information for people with confirmed COVID-19 cases in the state of Georgia in three time periods (February-April, May, and June-July 2020). May June-July Total . k and ✓ are the scale and shape parameters for the gamma distribution. The y-axis represents the probability density of having certain serial interval. The unit of probability density is one per unit of 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 October 26, 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. . 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 October 26, 2020. ; https://doi.org/10.1101 https://doi.org/10. /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. The copyright holder for this preprint this version posted October 26, 2020. ; https://doi.org/10.1101 https://doi.org/10. /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. The copyright holder for this preprint this version posted October 26, 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. The copyright holder for this preprint this version posted October 26, 2020. Figure 7 : Distributions of estimated dates of first maximum, minimum, and second maximum in R t for 87 counties with cumulative 200 cases by July 13, 2020, and key events possibly driving 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. The copyright holder for this preprint this version posted October 26, 2020. ; https://doi.org/10. 1101 /2020 March 14, 2020 Georgia governor declared a public health emergency. March 23, 2020 Large gatherings were banned and shelter-in-place order was issued for "medically fragile" population. March 24, 2020 Bars and clubs were ordered to close. April 1, 2020 All K-12 schools were closed. April 3, 2020 Statewide shelter-in-place order was issued. April 24, 2020 Some businesses (gyms, fitness centers, bowling alleys, body art studios, barbers, cosmetologists, hair designers, nail care artists, estheticians, their respective schools, and massage therapists) were allowed to reopen w/minimum basic operations. April 27, 2020 More businesses (theaters, private social clubs, and restaurant dine-in services) were allowed to reopen with social distancing and sanitation mandates. April 30, 2020 Statewide shelter-in-place order was lifted. June 1, 2020 Limits on the size of public gathering were relaxed: bars and nightclubs were allowed to reopen, sports events could resume, and summer schools and camps were allowed to begin sessions. . 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 October 26, 2020. ; Female ( S2 . 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 October 26, 2020. ; https://doi.org/10. 1101 /2020 For clinical characteristics, "+" represents yes and "-" represents no. The shape and scale parameters of gamma distributions were estimated using maximum likelihood estimator. . 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 October 26, 2020. ; https://doi.org/10. 1101 /2020 3%) 54,282 (45.8%) Missing 0%) 36,478 (30.8%) Missing 18,822 (59.6%) 10,695 (55.5%) 48,658 (71.9%) 78,175 (66.0%) Yes 2,127 (6.7%) 558 (2.9%) 320 (0.5%) 3,005 (2.5%) No 1%) 8,960 (46.5%) 23,542 (34.8%) 42,309 (35.7%) Missing