key: cord-0897478-zueprmdz authors: Clouston, S.; Morozova, O.; Meliker, J. title: Outdoor transmission of COVID-19: Analysis of windspeed date: 2021-02-08 journal: nan DOI: 10.1101/2021.02.05.21251179 sha: 302ab3403f5eb0a61e3527796edef7c02baafa73 doc_id: 897478 cord_uid: zueprmdz Background To examine whether outdoor exposures may contribute to the COVID-19 epidemic, we hypothesized that slower outdoor windspeed is associated with increased risk of transmission when individuals socialize outside. Methods Daily COVID-19 incidence reported between 3/16/2020-12/31/2020 was the outcome. Average windspeed and maximal daily temperature were derived from the National Oceanic and Atmospheric Administration. Negative binomial regression was used to model incidence, adjusting for susceptible population size. Results Cases were very high in the initial wave but diminished quickly once lockdown procedures were enacted. Unadjusted and multivariable-adjusted analyses revealed that warmer days with windspeed <5.5 MPH had increased COVID-19 incidence (aIRR=1.50, 95% C.I.=[1.25-1.81], P<0.001) as compared to days with average windspeed [≥]5.5 MPH. Conclusion This study suggests that outdoor transmission of COVID-19 may occur by noting that the risk of transmission of COVID-19 in the summer was highest on days when wind was reduced. To examine whether outdoor exposures may contribute to the COVID-19 epidemic, we hypothesized that slower outdoor windspeed is associated with increased risk of transmission when individuals socialize outside. Daily COVID-19 incidence reported between 3/16/2020-12/31/2020 was the outcome. Average windspeed and maximal daily temperature were derived from the National Oceanic and Atmospheric Administration. Negative binomial regression was used to model incidence, adjusting for susceptible population size. Cases were very high in the initial wave but diminished quickly once lockdown procedures were enacted. Unadjusted and multivariable-adjusted analyses revealed that warmer days with windspeed <5.5 MPH had increased COVID-19 incidence (aIRR=1.50, 95% C.I.=[1.25-1.81], P<0.001) as compared to days with average windspeed ≥5.5 MPH. This study suggests that outdoor transmission of COVID-19 may occur by noting that the risk of transmission of COVID-19 in the summer was highest on days when wind was reduced. . CC-BY 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 February 8, 2021. ; https://doi.org/10.1101/2021.02.05.21251179 doi: medRxiv preprint 3 The novel coronavirus SARS-CoV-2, which causes a potentially deadly disease called COVID-19 began spreading in China (1) , and Italy (2) COVID-19 transmits via aerosolized viral particles that begin shedding before symptoms are evident (4), making it difficult to trace patterns or locations where exposures are occurring. As a result, approximately half of those diagnosed with COVID-19 report not knowing where they may have become infected (5) . The most likely explanation for this lack of known exposures is that COVID-19 transmits in spaces that are thought to be safe. A handful of studies have made some headway in identifying such situations. For example, one study found that COVID-19 could transmit through the air over relatively long distances (6) and another highlighted the impact of air conditioning vents (7) . A third study found that a cluster of 17 cases could be traced to indirect transmission in shared spaces at a shopping mall in Wenzhou, China (8) . Still other studies have revealed that individuals in a constricted space could spread COVID-19 via inhaled transmission over potentially large distances by following air flow within a restaurant (7) and within the Diamond Princess cruise ship (6) . A recent review concluded that transmission within constricted indoor spaces is critically important, but outdoor exposures may be possible yet little is known about their dynamics or specific pathways (9) . There are reports of sporadic outbreaks in outdoor environments, including at a construction site in Singapore (10, 11) , jogging (10), or during conversation (12) . . CC-BY 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) Because of much lower risk outdoors, close outdoor contacts are often discussed as though they are risk-free and exposure-mitigating strategies have focused on promoting the use of exterior spaces when conducting social activities in efforts to mitigate risk of exposure. If exposure occurs outside, it is likely to be hampered by the same factors as are commonly seen in studies of indoor transmission including the air turnover rate. In the current study, we hypothesized that lower exterior wind speed would be associated with increased risk of transmission during warmer days, when individuals were most likely to be socializing outside. To examine the potential for exterior exposure risk, we modeled incidence of cases reported to the [COUNTY] Department of Health from March 16 th , when data first began being recorded reliably using an electronic interface, until December 31 st , 2020, at which time the COUNTY was enduring a second wave. Data were shared with [INSTITUTION BLINDED] for the purposes of supporting the COVID-19 modeling efforts at the local level. The analysis of publicly available deidentified case counts retrieved from the internet are considered to be not human subjects research and are exempt from ethics review. The main outcome was the number of daily confirmed incident cases as reported by the [COUNTY] Department of Health. We limited analysis to dates following March 16 th , 2020 with the opening of multiple drive-through testing sites throughout the area and when case-reporting . CC-BY 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 February 8, 2021. ; https://doi.org/10.1101/2021.02.05.21251179 doi: medRxiv preprint routines were established. Population at risk estimates come from population estimates for the county derived from the U.S. census; susceptible population counts were updated for daily death counts, and for the reported number of COVID-19-related disease counts. Since daily case counts exhibit temporal dependence that is primarily determined by the underlying community force of infection, which cannot be measured directly, in a secondary analysis, we use an alternative outcome measure of a relative change in daily case counts compared to an 8-day forward/backward moving average defined as: The 8-day forward/backward moving average serves as a proxy measure of underlying force of infection allowing to partially capture the variability in absolute case counts that is due to "natural" transmission patterns rather than external shocks such as wind speed. It is important to note that, on average, this measure would be zero when case counts remain relatively constant over time, however during the periods of exponential rise and decay of an epidemic, this measure would on average be negative, and it would be positive around the peak of an epidemic curve. It is therefore important to take these distinct behaviors into account. Maximal daily temperature as well as average windspeed were derived from the National Oceanic and Atmospheric Administration (NOAA) data portal (w2.weather.gov); data were recorded at a central location at the [LOCATION BLINDED]. Total snowfall and rainfall were also recorded in inches. While warmer temperatures are likely to be protective, as warm days allow individuals to socialize outside, where exposure appears to be markedly lower, increased windspeed may have diverging effects depending on temperature. In the summer, higher . CC-BY 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 February 8, 2021. ; https://doi.org/10.1101/2021.02.05.21251179 doi: medRxiv preprint windspeed increases air flow and may reduce risk versus in the winter when it may work to push social contacts that were occurring outside to shelter in indoor spaces. When exterior temperatures are warm enough to allow for outdoor social contacts to occur comfortably, we anticipated that increased windspeed would reduce overall outdoor risk. In contrast, on days where exterior temperatures were cooler, increased windspeed might cause individuals to retreat indoors for social occasions. We adjusted for number of days since lockdown (March 16 th , 2020) and days since reopening began in [COUNTY] . To account for differences in daily reporting patterns, we incorporated a categorial variable indicating the day of the week that cases were reported. Noting that there has been significant spread in [COUNTY] following holidays, we incorporated an indicator of holidays that also incorporated the most significant weekend nearby. We also included covariates measuring rainfall and snowfall as these weather conditions are hypothesized to correlate with windspeed as well as social activities outdoors. In the primary analysis, we also adjust for the 8day forward/backward moving average daily case count. Descriptive characteristics include time-related trends in maximal temperature, daily windspeed, and daily case counts. Trends in maximal temperature and in average windspeed were provided alongside smoothed polynomial best-fitting trend lines. In the main analysis, incidence of COVID-19 positive caseload is reported as counts per day and, therefore, multivariable-adjusted modeling relied on negative binomial regression (13) . Negative . CC-BY 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 February 8, 2021. ; https://doi.org/10.1101/2021.02.05.21251179 doi: medRxiv preprint binomial regression was chosen over alternatives including Poisson because we were concerned about the potential for over-dispersion in the outcome (14) since the infectious disease case load is highly variable, and because COVID-19 appears to spread commonly through superspreading clusters (15) . A nine-day lag between exposure and case registration was assumed, consistent with epidemiological estimates of the incubation period for COVID-19 (16, 17) coupled with a two-day testing and one-day reporting lag period for available estimates online. Unadjusted and multivariable-adjusted incidence rate ratios (IRR) and 95% confidence intervals (95% C.I.) were reported. The interval between infection and disease ascertainment is not well known and varies geographically because it depends substantially on local testing availability and reporting systems -it can be reduced in places where testing is easy to find and lengthened in places where testing is difficult or requires hospitalization. As such, we conduct a sensitivity analysis considering the values of time intervals between exposure and case reporting between 5-15 days. For our lagging period, we allowed five days was chosen because our experience suggests that it takes two days to report results of testing to the DOH and an additional day to report those results online. Fifteen days was selected as a ceiling for index case analysis to reduce the risk of sequential effects of prior case/exposure cycles; however, sensitivity analyses reported results from 5-21 days to clarify the impact of those choices. The log-likelihood estimate was reported to compare model fit for different lags. We analyzed the secondary outcome -a relative measure of daily case counts -using linear regression with the same set of covariates as the primary outcome measure and exploring the results for a range of reporting lags. . CC-BY 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) Since we hypothesized the effect heterogeneity of windspeed on transmission depending on the temperature, cutoffs for "warm" days and for days when windspeed was sufficiently fast were determined by comparing Akaike's information criterion (AIC) across multiple models using different details as modeled parameters. We compared AIC between models to determine that >60°F [15°C] was an optimal cutoff for temperature. Because cutoffs may be useful when adjudicating risk at the local level, we similarly used AIC and compared model fit using multivariable adjusted models to identify optimal cutoffs for windspeed and ultimately identified low windspeed as days where windspeed was <5.5 MPH as the optimal cutoff for these models. Since the relative measure of daily case counts only partially adjusts for the community force of infection and underlying "natural" epidemic dynamics, we also conducted additional stratified sensitivity analyses cut into time periods when case counts were relatively flat (06/07/2020-11/03/2020) and when epidemic was exponentially increasing (03/16/2020-04/10/2020 and 11/04/2020-12/31/2020) or decaying (04/11/2020-06/06/2020). We use two criteria: daily temperature (warm/cool) and epidemic dynamics pattern (flat versus rising or decaying) to determine subsets for stratified analyses. Analyses were completed using Stata 16/MP [StataCorp] . Data used in this study are secondary analyses of de-identified case counts reported on a publicly available website and therefore this was not human subject's research. . CC-BY 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) We begin by showing the number of daily cases over the entire observational window (Figure 1 ). Cases were very high in the initial wave but diminished quickly once lockdown procedures were enacted. The average temperature was 67.6 ± 14.4 °F, the average daily windspeed was 8.7 ± 3.6 MPH. Trends in daily temperature and windspeed are depicted throughout the analytic period ( Figure 2 ). Most days between May 1 st , 2020 and October 24 th , 2020 were characterized by temperatures exceeding 60°F (blue dashed lines; solid trend line). Windspeed diminished slowly over time, and then began to increase again in December 2020. Further interrogating the functional shape of the relationship between the windspeed and incidence (Figure 3 ), we found that during "warm" time period higher windspeed was associated with diminishing degree of protection. Using the logarithmic transformation to capture this tapering in a multivariable-adjusted model, we found that while an increase in windspeed from 1 to 2 MPH is associated with a 4.2% reduction in caseload as compared to days when the air was still, a similar increase from 10 to 11 MPH was only associated with a 0.08% decrease in caseload on the "warm" days. Table 2 . Note that the Incidence was lagged from windspeeds by nine days. Unadjusted analyses revealed statistically significant associations between higher COVID-19 incidence and lower windspeed in warmer weather ( Table 1) . Multivariable-adjusted analyses similarly revealed that results remained statistically significant upon adjusting for confounders. As noted in the Methods section, cutoffs were determined to be >60°F [21°C] in temperature, and <5.5 MPH in windspeed. Using these cutoffs, in Table 2 we examined the risk associated with lower windspeed (<5.5 MPH) on warmer days (>60°F). Analyses revealed that on warmer days, having windspeed <5.5 MPH was associated with a 50% increase in incidence in multivariable adjusted models. Note: *Warm days were defined as >60°F while slow windspeed was defined as <5.5 MPH. MPH: Miles Per Hour; °F: degrees Fahrenheit; IRR: incidence rate ratio; 95% C.I.: 95% confidence interval. All models adjust for day of the week in which cases were reported and for the size of the county population adjusted for reductions due to individuals who had died or become immune due to COVID-19 during the period of observation. Alpha is a measure of dispersion. P-values derived from Student's T test. We examined the sensitivity of the results to analytic choices by first examining whether reliance on different outcomes made differences to the results. For the relative change in daily case counts compared to an 8-day forward/backward moving average, the results were substantively similar (B = -16.12 [-27 .78, -4.45], P=0.007) on warmer days; in other words, as windspeed decreased by one MPH, incidence increased by 16.12% (Table S1 ). We also examined whether choices in the lag between exposure and case reporting changed our results. While the results shown theoretically represent the appropriate timing, we also examined variation in periods between exposure and case recording from 5-21 days. We found that while the nine-day reporting average was the best performing within our hypothesized observational window ( Figure S1 ), that the 16-day reporting lag was the best performing lag structure. Across all lags, a consistently association was identified linking slower windspeed days with lower follow-up case counts (Table S2) To obtain a measure of windspeed for this analysis, we relied on data from a central airport. While this provided highly consistent measures of windspeed for the island, it also provides measures that may not be generalizable to microclimates that can occur in the lea of hills, fenced-in backyards, or forests. Notably, this choice may mean that cutoffs used here may not apply in other situations and more analysis is necessary if weather data are going to be relied upon to help understand caseload in other areas. We reported results from a 9-day exposure-test positive reporting lag structure; however, sensitivity analyses suggested that a 16-day lag structure may work better. The 16-day lag is outside of the expected lag period for cases in our area, but we felt that it might indicate that case dynamics could proceed from asymptomatic younger individuals to cause secondary cases in older individuals reported 16 days later. As such, future work should anticipate that different cutoffs will be necessary when windspeeds are measured in different places and in locations where wind is highly sensitive to local geography. Throughout the U.S. epidemic, the role of outdoor shared spaces such as parks and beaches has been considered and ultimately beaches and parks remained open. This analysis does little to suggest that either should be closed, since the level of risk due to outdoor exposures should be weighed in relation to the much higher risk of exposure in shared interior spaces such as houses, restaurants, or public transport. Instead, this study suggests that individuals socializing outdoors are not entirely safe by virtue of being outdoors and should remain vigilant. In this case, outdoor use of increased physical distance between individuals, improved air circulation, and use of masks in outdoor environments may be useful. The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19) in China. Zhonghua liu xing bing xue za zhi= Zhonghua liuxingbingxue zazhi The early phase of the COVID-19 outbreak in Socioeconomic inequalities in the spread of coronavirus-19 in the United States: A examination of the emergence of social inequalities It Is Time to Address Airborne Transmission of Coronavirus Disease 2019 (COVID-19) Characteristics of adult outpatients and inpatients with COVID-19-11 academic medical centers Air recirculation role in the infection with COVID-19, lessons learned from Diamond Princess cruise ship. medRxiv COVID-19 outbreak associated with air conditioning in restaurant Indirect virus transmission in cluster of COVID-19 cases Outdoor Transmission of SARS CoV-2 and Other Respiratory Viruses, a Systematic Review. The Journal of Infectious Diseases Group CC-W. What settings have been linked to SARS-CoV-2 transmission clusters? Work-related COVID-19 transmission in six Asian countries/areas: A follow-up study Indoor transmission of SARS-CoV-2. medRxiv Regression models for categorical dependent variables using Stata: Stata press Regression analyses of counts and rates: Poisson, overdispersed Poisson, and negative binomial models Evidence that coronavirus superspreading is fat-tailed Early insights from statistical and mathematical modeling of key epidemiologic parameters of COVID-19. Emerging infectious diseases Incubation period of COVID-19: a rapid systematic review and meta-analysis of observational research Covid-19: Cases still rising in at least 23 US states as health officials warn against gatherings