key: cord-0913624-gz2gd1ey authors: Guo, Cui; Bo, Yacong; Lin, Changqing; Li, Hao Bi; Zeng, Yiqian; Zhang, Yumiao; Hossain, Md Shakhaoat; Chan, Jimmy W.M.; Yeung, David W.; Kwok, Kin on; Wong, Samuel Y.S.; Lau, Alexis K.H.; Lao, Xiang Qian title: Meteorological factors and COVID-19 incidence in 190 countries: an observational study date: 2020-11-23 journal: Sci Total Environ DOI: 10.1016/j.scitotenv.2020.143783 sha: 521828a4601971dec3f167bcdca032efe87a078f doc_id: 913624 cord_uid: gz2gd1ey Novel corona virus disease 2019 (COVID-19), which first emerged in December 2019, has become a pandemic. This study aimed to investigate the associations between meteorological factors and COVID-19 incidence and mortality worldwide. This study included 1,908,197 confirmed cases of and 119,257 deaths from COVID-19 from 190 countries between 23 January and 13 April, 2020. We used a distributed lag non-linear model with city-/country-level random intercept to investigate the associations between COVID19 incidence and daily temperature, relative humidity, and wind speed. A series of confounders were considered in the analysis including demographics, socioeconomics, geographic locations, and political strategies. Sensitivity analyses were performed to examine the robustness of the associations. The COVID-19 incidence showed a stronger association with temperature than with relative humidity or wind speed. An inverse association was identified between the COVID-19 incidence and temperature. The corresponding 14-day cumulative relative risk was 1.28 [95% confidence interval (CI), 1.20–1.36] at 5°C, and 0.75 (95% CI, 0.65–0.86) at 22°C with reference to the risk at 11°C. An inverse J-shaped association was observed between relative humidity and the COVID-19 incidence, with the highest risk at 72%. A higher wind speed was associated with a generally lower incidence of COVID-19, although the associations were weak. Sensitivity analyses generally yielded similar results. The COVID-19 incidence decreased with the increase of temperature. Our study suggests that the spread of COVID-19 may slow during summer but may increase during winter. The spread of COVID-19 may slow during summer but may increase during winter. • There was an inverse J-shaped association between relative humidity and the incidence of COVID-19. • The association between wind speed and the incidence of COVID-19 was nonlinear. Keywords: COVID-19 incidence, COVID-19 mortality, Meteorology, Temperature, Relative humidity, Wind speed The coronavirus disease 2019 , caused by the severe acute respiratory countries/areas/territories (Organization 2020). The COVID-19 outbreak has aroused unprecedented public health concern worldwide. stable in cold environments but sensitive to increased temperature (Chin et al. 2020) . Evidence regarding the effects of ambient meteorological factors on COVID-19 incidence/mortality are controversial. Some epidemiological/ecological studies have reported a negative association between temperature and the number of confirmed COVID-19 cases in one specific country, including China (J Liu et al. 2020; Qi et al. 2020 ; P. , Brazil (Prata et al. 2020) , US (Runkle et al. 2020) , Turkey (Sahin 2020 ) and other countries (Abdollahi and Rahbaralam 2020) . In contrast, other studies have observed positive associations (Ahmadi et al. 2020; Bashir et al. 2020; Tosepu et al. 2020; Zhu and Xie 2020) or no association . The inconsistency may be ascribed to the relatively small sample size and short study period, in addition to the fact that most previous studies have considered a narrow range of meteorological factors. Literature on this topic from large-scale studies based on various climate zones and countries were limited and inconsistent (Awasthi et al. 2020; Baker et al. 2020; Chiyomaru and Takemoto 2020; Islam et al. 2020; Le et al. 2020; Pan et al. 2020; Sajadi et al. 2020; Sobral et al. 2020) . Furthermore, few studies have examined the potential nonlinear associations between meteorological factors and COVID-19 and the delayed effects of the meteorological factors on the incidence of COVID-19. Such information is important to allow us to understand the activity and viability of SARS-CoV-2 in the natural environment and may thus be useful in the development of appropriate strategies to control the spread of COVID-19. of 190 countries. Hourly meteorological data, including temperature, relative humidity, and wind speed were obtained from the ground-based monitoring network of the World Meteorological Organization global telecommunications system. The hourly data were then aggregated as daily average meteorological data. If a study site had more than one monitoring stations, the average values of the meteorological factors were weighted by population density using the following formula: Population weighted average = 1 × 1 + 2 × 2 + ⋯ + × 1 + 2 + ⋯ + In which, is the meteorological value at the i th monitoring station and is the corresponding weight. = 1, 2, 3, 4 and 5 for an area with a population density of <10, ≥10 ~ <100, ≥100 ~ <1,000, ≥1,000 ~ <10,000, and ≥10,000 persons/km 2 , respectively. Information on demographic and socio-economic status, geographic locations and intervention policies for each city/country was also collected, including population size (number of persons) and density (persons per square kilometre), median age (years), Global Health Security Index (GHSI) (Bannister-Tyrrell et al. 2020; Chang and McAleer 2020; Kandel et al. 2020) , and public health interventions (i.e., intervention policy on transportation restriction, large-scale testing, wearing of masks, Index is a common indicator for population morbidity between countries. The GHSI data were collected from the GHSI report (Johns Hopkins 2019). GHSI is mainly used to assess health security capabilities of a country in the following six categories: emergence prevention or pathogens release, early detection and reporting, rapid response and mitigation, patient treatment and health worker's protection, compliance with global norms, as well as risk environment. The Education Index data were obtained from the United Nations Development Programme (Programme 2020). It is measured by the average of school years of adults and the expected school years of children. Because the Henley Index, GHSI and Education Index data are only available at the country level, we applied them to all cities within the same country. We also searched the official webpages of the 415 cities/countries and obtained information on the policy of public health interventions. We used the distributed lag non-linear model (Gasparrini et al. 2015 ), a standard time-series regression model, to investigate the associations between the meteorological factors and the COVID-19 incidence. This model can present non-linear exposure-response associations between meteorological factors and health outcomes. The lag-response associations can also be examined to show the delayed effects on health. We also included the first date of reported cases to control for the variations in the beginning of COVID-19 epidemic in different countries. In this study, negative binomial distribution was used to control potential over-dispersion. We applied the natural cubic spline function with one internal knot Lauer et al. 2020). The cumulative associations (7 consecutive days and 10 consecutive days) were presented. Various numbers of knots were also examined in the sensitivity analyses. Akaike Information Criteria was used to assess the goodness of fit and the simplicity of models. The following covariates were included in the model: logarithm of population to offset population effects across regions, a natural cubic spline of time with 2 degree of freedom (Liu et al. 2019) to control seasonal and long-term trends over the study period, an indicator of the day of the week and public holidays to account for weekly or periodically variations in the number of people who accepted testing, and the number of confirmed cases reported on the previous day to account for temporal autocorrelation. The following city-/country-level confounders were also included: the date of the first reported cases to account for the detection time, the population density to account for the higher transmission rate in regions with a high population density, the national median age to account for the higher incidence among older adults, the GHSI to account for countries' capacity to prevent and mitigate epidemics and pandemics, the latitude and longitude to account for potential spatial autocorrelation, and the intervention policies implemented (i.e., yes vs no for whether implementing the policies of transportation restriction, large-scale nucleic acid examinations, mask wearing, propaganda strategies, declaration of public health emergency and social distancing) to account for the effectiveness of the implemented intervention. Both cumulative and single-day relative risks (RRs) with a 95% confidence interval (CI) at the 25 th and 75 th percentiles of the meteorological factors were used to present the strengths of the associations. Because the associations were generally J o u r n a l P r e -p r o o f Journal Pre-proof non-linear, we used overall median value of meteorological factors as the reference levels. We also plotted the exposure-response curves to show the overall associations. A series of sensitivity analyses were performed to examine the robustness of the estimated associations: 1) we applied the same method to investigate the association between COVID-19 mortality and the meteorological factors; 2) we extended the lag period from 14 days to 23 days to examine the possible long incubation period reported in previous literature (Lauer et al. 2020) ; 3) we compared various numbers of knots of meteorological factors (i.e., from 2 to 4 placed at equally spaced values) to examine whether the number of knots may affect the associations; 4) we analysed the associations by including only a single meteorological factor in the model to avoid potential correlation between these factors; 5) we investigated the associations by excluding Hubei province in China, which was a significant outlier in the model ; 6) we excluded sites with the highest or lowest 5% of the values for temperature, relative humidity and wind speed to avoid potential outlying effects; 7) we excluded the covariates "mask wearing" and "propaganda" separately in the model to avoid collinear effects (the Spearman correlations coefficient for "mask wearing" and "propaganda" was 0.7); 8) we excluded 50 study sites with one or no monitoring station because population weight averages of meteorological factors were unavailable; 9) we excluded the 10 countries with the largest area (including Russian All data analyses were performed using R 3.6.1 (R Core Team, Vienna, Austria) with 'dlnm' and 'lme4' packages. A two-tailed P value of < 0.05 was considered statistically significant. In this study, 415 sites were included in the analysis ( Table S2 in the Appendix shows the characteristics of each country included. Figure 1 shows the spatial distributions of cumulative confirmed cases and the ground-level meteorology monitoring stations. The 415 study sites covered four climate zones, including cold (e.g., Canada), temperate (e.g., the United Kingdom), subtropical (e.g., South Africa) and tropical (e.g., Singapore). Figure S2 presents the time-series plots of daily confirmed cases, deaths, temperature, relative humidity, and wind speed between 23 Jan 2020 and 13 Apr 2020. The spatial distributions of meteorological factors, GHSI, median age and population density are shown in Figures S3-S4 in Appendix. Figure 2 shows the associations between the 14-day cumulative risk of COVID-19 incidence and the three meteorological factors. An inverse association was observed between temperature and the COVID-19 incidence (Figure 2A ). The cumulative RR decreased sharply with increased temperature as the temperature lower than 11℃. The curve flattened when the temperature was greater than 11℃. The corresponding cumulative RRs were 1.28 (95% CI, 1.20-1.36) at 5℃, and 0.75 (95% CI, 0.65-0.86) at 22℃, respectively, with reference to 11℃ (Table 1 ). An inverse J-shaped association was observed between relative humidity and the COVID-19 incidence, and the cumulative RR peaked at 72% ( Figure 2B ). The 14-day cumulative RRs were 0.82 (95% CI, 0.79-0.86) at a relative humidity of 59% and 0.99 (95% CI, 0.94-1.04) at a relative humidity of 79% with reference to 72% (Table 1 ). The overall associations between wind speed and COVID-19 incidence were weak. The 14-day cumulative RR of the COVID-19 incidence increased a little as the wind speed increased up to 6 m/s and then started to decrease as the wind speed continue to increase ( Figure 2C ). No significant association was observed when wind speeds was above 6 m/s. The 14-day cumulative RRs were 0.9 (95% CI, 0.83-0.98) at a wind speed of 2 m/s, and 1.08 (95% CI, 1.02-1.14) at 4 m/s with reference to 3 m/s (Table 1 ). Figure 3 shows the associations of the daily risk of COVID-19 incidence with the 25 th , 50 th and 75 th percentiles of the meteorological factors over a period of 14 consecutive days. The patterns were generally consistent across percentiles except for the relative humidity at the 75 th percentile. The risk of COVID-19 incidence was generally higher between the 6 th and 10 th days due to the harvest effects and the J o u r n a l P r e -p r o o f Journal Pre-proof effects generally disappeared after 12 days. Figure S5 shows the associations between the COVID-19 incidence and the meteorological factors on the current day (Lag 0), the 7 th day (Lag 7) and the 14 th day (Lag 14). Sensitivity analyses generally yielded similar results. The association of meteorological factors with the COVID-19 mortality was similar to their association with COVID-19 incidence (Table S3) . We did not observe substantial changes in the results by extending the lag period from 14 days to 23 days (Table S4) , including various numbers of internal knots in the model (Table S5) , including only a single meteorological factor in the model (Table S6) , excluding outlier province (Hubei, China) and extreme meteorological values (Tables S7-S8 ), excluding the covariates of "mask wearing" or "propaganda" (Table S9) , excluding the countries with one or no monitoring station (Table S10) , excluding the 10 countries with largest area but without state-/province-level data (Table S11) , and using linear meteorological variables and shorter lag days (Table S12) . This study investigated the associations between the COVID-19 incidence and meteorological factors in 415 sites from 190 countries. We observed significant non-linear associations for temperature, relative humidity, and wind speed. The association of COVID-19 incidence with temperature was much stronger than those with relative humidity and wind speed. An increase in temperature from 7℃ to 22℃ was associated with a decrease of 25% in the cumulative risk of COVID-19 incidence over a 14-day period. Sensitivity analyses generally yielded similar results, thus suggesting the robustness of the associations. In line with our study, most previous studies have reported a negative association J o u r n a l P r e -p r o o f Journal Pre-proof between temperature and COVID-19 incidence or mortality in China (Gupta 2020; J Liu et al. 2020; Qi et al. 2020; Peng Shi et al. 2020) , Brazil (Prata et al. 2020) , US (Runkle et al. 2020) , Turkey (Sahin 2020) , other countries (Runkle et al. 2020; Triplett 2020) , and multi-city regions (Baker et al. 2020; Chiyomaru and Takemoto 2020; Islam et al. 2020; Le et al. 2020; Sobral et al. 2020) . We observed an overall inverse association between temperature and the COVID-19 incidence at a global scale. The COVID-19 incidence decreased sharply with temperature as the temperature was lower than 11℃. The temperature-related effects on the COVID-19 incidence flattened when the temperature was higher than 11℃. We did not find an optimum temperature for the COVID-19 incidence. In contrast to our study, some previous studies have reported no significant association (Awasthi et al. 2020; or a positive association in some specific countries (Bashir et al. 2020; Bu et al. 2020; Tosepu et al. 2020; Zhu and Xie 2020) . This inconsistency is likely due to the limited study sites in previous studies, and the fact that most studies used a linear or log-linear model to examine the association. The non-linear association and the delayed effects of temperature on COVID-19 should be further explored and confirmed in future studies. We observed a reverse J-shaped association between relative humidity and the COVID-19 incidence. The cumulative risk of COVID-19 incidence peaked at a relative humidity of 72%. There is relatively sparse information on the association between COVID-19 incidence and relative humidity at a global scale. A few studies have reported the associations between COVID-19 and humidity in China Wang et al. 2020 ) and in the world (Baker et al. 2020; Islam et al. 2020; Sajadi et al. 2020) , whereas others have found positive association (Sobral et al. 2020) , no significant association (Awasthi et al. 2020 ; Chiyomaru and Takemoto 2020; Pan et al. Journal Pre-proof study areas may contribute to these inconsistent findings. We observed an inverse U-shaped association between wind speed and COVID-19 incidence. The higher wind speed was generally associated with a lower risk of COVID-19 incidence as wind speed was higher than 6 m/s. Two previous studies reported similar findings (Islam et al. 2020; Qiu et al. 2020) . The negative associations were also observed in Iran (Ahmadi et al. 2020) , Turkey (Sahin 2020) and China (Zhu and Xie 2020) . In contrast, three multi-city studies (Chiyomaru and Takemoto 2020; Pan et al. 2020; Sajadi et al. 2020 ) did not observe significant linear association between wind speed and the cases/basic reproduction number of COVID-19. Further studies on this topic are warranted. Although the measurements of temperature (℃), relative humidity (%) and wind speed (m/s) differ, the association of COVID-19 incidence with temperature seems more apparent and much stronger than those with relative humidity and wind speed in this study. The 14-day cumulative effect of temperature on COVID-19 incidence was the greatest when the three factors changed by the same percentile (e.g., an increase from the 25 th to 50 th percentile in temperature, relative humidity and wind speed was associated with decreases of 28%, -18% and -10% in the risk of COVID-19 incidence, respectively). The potential mechanism underlying the associations between meteorological factors and COVID-19 incidence remains unclear. The negative association of COVID-19 incidence with temperature found in our study is supported by a few laboratory studies, in which SARS-CoV-2 was shown to be highly stable in a cold environment J o u r n a l P r e -p r o o f but sensitive to increased temperature (Chin et al. 2020) . SARS-CoV-2 is a member of the coronavirus family. A few studies have reported that other coronaviruses such as SARS-CoV-1 or MERS-CoV may lose viability at a high temperature (Chan et al. 2011 ). A laboratory experiment shows that coronaviruses were inactivated more rapidly at 20 °C than at 4°C. In addition to SARS-CoV-1's sensitivity to temperature, we speculated that poor indoor ventilation during winter is another possible reason for the strong association between low ambient temperature and a high risk of COVID-19. People generally stay indoors more during winter with the doors and windows closed Little is known about the relationship between humidity and SARS-CoV-2, but a previous study showed that both titre and genome copies of MERS-CoV decreased more in an environment with a relative humidity of 70% than in an environment with a relative humidity of 40% at the same temperature of 20 °C (Van Doremalen et al. ). COVID-19 is transmitted mainly via respiratory droplets generated by coughing and sneezing. Aerosol transmission may also contribute to the spread of COVID-19. Higher wind speeds may blow away the droplets and reduce the concentration of infected aerosol particles. First, the large sample size enabled us to have more stable results and to conduct a series of sensitivity analyses to confirm the robustness of the associations. Second, our study included 190 countries and covered four major climate zones with a wide temperature range (from -30 °C to 40 °C), which allowed us to investigate the effects Figure 2 Overall association of the cumulative risk of COVID-19 incidence with temperature, relative humidity, and wind speed Red lines represent the estimated cumulative relative risks of COVID-19 incidence with shaded bands as 95% confidence intervals. A, B, and C represent the associations of the 14-day cumulative risk of COVID-19 incidence with daily temperature, relative humidity, and wind speed, respectively. The corresponding reference levels were 11℃, 71% and 3 m/s, respectively. Effect of temperature on the transmission of covid-19: A machine learning case study in spain Investigation of effective climatology parameters on covid-19 outbreak in iran Temperature and humidity do not influence global covid-19 incidence as inferred from causal models Susceptible supply limits the role of climate in the early sars-cov-2 pandemic Preliminary evidence that higher temperatures are associated with lower incidence of covid-19, for cases reported globally up to 29th Correlation between climate indicators and covid-19 pandemic in new york Analysis of meteorological conditions and prediction of epidemic trend of 2019-ncov infection The effects of temperature and relative humidity on the viability of the sars coronavirus Alternative global health security indexes for risk analysis of covid-19 Aerodynamic analysis of sars-cov-2 in two wuhan hospitals Effects of temperature variation and humidity on the death of covid-19 in Wuhan Global burden of respiratory infections due to seasonal influenza in young children: A systematic review and meta-analysis Warmer weather unlikely to reduce the covid-19 transmission: An ecological study in 202 locations in 8 countries Temperature significantly changes covid-19 transmission in (sub)tropical cities of Brazil Covid-19 transmission in mainland china is associated with temperature and humidity: A time-series analysis Impacts of social and economic factors on the transmission of coronavirus disease Short-term effects of weather parameters on covid-19 morbidity in select us cities Impact of weather on covid-19 pandemic in turkey Temperature, humidity, and latitude analysis to estimate potential spread and seasonality of coronavirus disease 2019 (covid-19) The impact of temperature and absolute humidity on the coronavirus disease 2019 (covid-19) outbreakevidence from China Impact of temperature on the dynamics of the covid-19 outbreak in China Association between climate variables and global transmission of sars-cov-2 Correlation between weather and covid-19 pandemic in jakarta, indonesia Evidence that higher temperatures are associated with lower incidence of covid-19 in pandemic state, cumulative cases reported up to United Nations Development Programme Stability of middle east respiratory syndrome coronavirus (mers-cov) under different environmental conditions Aerosol and surface stability of sars-cov-2 as compared with sars-cov-1 High temperature and high humidity reduce the transmission of covid-19 World Health Organization. 2020. Coronavirus disease (covid-2019) situation reports. Wind. 2020. Wind financial database No association of covid-19 transmission with temperature or uv radiation in chinese cities Association between ambient temperature and covid-19 infection in 122 cities from China