key: cord-0826277-wdlvo4n2 authors: Xiao, Shuang; Qi, Hongchao; Ward, Michael P.; Wang, Wenge; Zhang, Jun; Chen, Yue; Bergquist, Robert; Tu, Wei; Shi, Runye; Hong, Jie; Su, Qing; Zhao, Zheng; Ba, Jianbo; Zhang, Zhijie title: Meteorological conditions are heterogeneous factors for COVID-19 risk in China date: 2021-04-16 journal: Environ Res DOI: 10.1016/j.envres.2021.111182 sha: c138b50f5fb31887030e67c41a4851eadb771e82 doc_id: 826277 cord_uid: wdlvo4n2 Whether meteorological factors influence COVID-19 transmission is an issue of major public health concern, but available evidence remains unclear and limited for several reasons, including the use of report date which can lag date of symptom onset by a considerable period. We aimed to generate reliable and robust evidence of this relationship based on date of onset of symptoms. We evaluated important meteorological factors associated with daily COVID-19 counts and effective reproduction number (R(t)) in China using a two-stage approach with overdispersed generalized additive models and random-effects meta-analysis. Spatial heterogeneity and stratified analyses by sex and age groups were quantified and potential effect modification was analyzed. Nationwide, there was no evidence that temperature and relative humidity affected COVID-19 incidence and R(t). However, there were heterogeneous impacts on COVID-19 risk across different regions. Importantly, there was a negative association between relative humidity and COVID-19 incidence in Central China: a 1% increase in relative humidity was associated with a 3.92% (95% CI, 1.98% to 5.82%) decrease in daily counts. Older population appeared to be more sensitive to meteorological conditions, but there was no obvious difference between sexes. Linear relationships were found between meteorological variables and COVID-19 incidence. Sensitivity analysis confirmed the robustness of the association and the results based on report date were biased. Meteorological factors play heterogenous roles on COVID-19 transmission, increasing the possibility of seasonality and suggesting the epidemic is far from over. Considering potential climatic associations, we should maintain, not ease, current control measures and surveillance. The novel coronavirus disease , caused by severe acute respiratory 56 syndrome coronavirus 2 (SARS-CoV-2), has been rapidly spreading across the world. 57 on provincial or national scales; 4) fidelity of meteorological data: mostly based the 85 crude provincial or national average values; and 5) inconsistent modelling approaches. 86 In addition, previous studies used publicly released aggregated COVID-19 datasets 87 based on date of case reports, which includes around a 10-day delay from the date of 88 symptom onset (WHO, 2020) . This can potentially cause an important bias, which 89 needs to be carefully addressed. Therefore, the effect of meteorological factors on the 90 spread of COVID-19 remains unclear and the possible effect modification of 91 demographic characteristics (including sex and age) on COVID-19 is unknown. In this study, we collected information on date of symptom onset and demographic 94 characteristics of COVID-19 cases, and aggregated confirmed cases based on the date 95 of symptom onset instead of the date of case reporting. We used a widely applied 96 standard two-stage time-series modelling approach (Liu et al., 2019) to examine the 97 association between short-term exposure to meteorological factors and COVID-19 98 risk, and also determined for the first time the modifying effects of sex and age on the 99 association. for suspected, clinically diagnosed and laboratory confirmed cases follows those 106 guidelines; detailed descriptions have been reported in previous studies (Li et al., 2020; 107 Pan et al., 2020; Zhang et al., 2020) . (Deng et al., 2020; Dong et al., 2020; Pan et al., 2020; Zhang et al., 2020) and news 118 flashes issued by local governments. Daily aggregated counts for each city were 119 calculated  beginning from the onset date of the first case to the epidemic peak  for 120 all the patients and for different subgroups of sex and age (0-64 years and ≥ 65 years). 121 Furthermore, the effective reproduction number (R t ) (Mubayi et al., 2009 ) defined as 122 the mean number of secondary cases infected by a typical primary case at any specific 123 time in a population where there is some immunity or intervention measures in place, 124 was calculated for each city to measure the transmission rate of SARS-COV-2. Calculations were based on standard SIR disease transmission models analogous to 126 J o u r n a l P r e -p r o o f the time-series SIR (TSIR) approach using an algorithm of real time Bayesian 127 estimation (Bettencourt and Ribeiro, 2008) . The serial interval (SI) was estimated 128 with a gamma distribution (mean 7.5, SD 3.4 days) (Li et al., 2020) . i.e., the difference between maximum and minimum temperatures, was then 140 calculated to represent the temperature variation. Population data for each city was 8 The associations between meteorological factors and COVID-19 in each city were 148 first assessed with a standard time-series approach and then the region-specific (7 149 geographical regions based on the comprehensive physical regionalization  North, 150 Northeast, East, South, Central, Southwest, and Northwest China) and 151 national-average associations were calculated via meta analysis. In the first stage, a quasi-Poisson generalized additive model (GAM) was applied. Considering the validity and stability of the results, only cities with >100 cases were 155 included in the analyses. We calculated the Spearman correlation coefficient matrix 156 among meteorological factors for each city, and the correlation matrices were then 157 pooled by averaging the correlation coefficients. A correlation coefficient threshold of were also considered to control for the time trend of the COVID-19 incidence, and its optimum degrees of freedom was determined based on QAIC. The GAM is as 169 follows: Where is the daily count of comfirmed cases at day in city , μ is the In the second stage, a random effects meta regression model was used to obtain the 177 national-average associations between meteorological factors and COVID-19 by 178 pooling association estimates across cities, thus accounting for between-city 179 heterogeneity that was tested and quantified with Cochran Q test and I 2 statistic. We 180 applied the above two-stage approach to each of 7 geographical regions to obtain 181 region-specific effect estimates. The formula in the second stage was as followed: 182 183 β ij = α + + , Where β ij denotes the effect of temperature or relative humidity of city j in regions 184 in GAM, α denotes the overall mean of the effect, is the city-specific effect for city j, and is the random error term. We also fitted meta regression models with population size, longitude, latitude, road 188 density, and distance from Wuhan as meta predictors to explain the between-city 189 heterogeneity of the effects of meteorological variables on 190 we used the same two-stage approach to estimate the associations between 191 meteorological factors and the R t of To estimate the overall shape of the association between the meteorological variables There was a total of 86,576 confirmed cases from December 1 to March 31 2020 with 217 most cases occurring in the cities of Central and East China. The study periods from 218 December 1 to the epidemic peak (February 15, 2020) were selected in the analysis. Cases from February 15th, 2020 to March 31, 2020 were not included in the analysis 220 due to sparse cases per city per day in this period. Central China ( Fig 1A) . Three quarters of all cases (74.74%) were aged ≤65 years and 224 the male/female sex ratio was 0.97:1. Sunshine hour, temperature range, rainfall and wind speed were excluded due to multilinearity or the Wald test (Table A1) . A 14-day moving average (lag 0 to 13) for 228 daily average temperature, a 15-day moving average (lag 0 to 14) for relative 229 humidity (Table A2) , and the natural splines of time with 3 degrees of freedom were 230 determined to be optimal according to QAIC (Table A3 ). The spatial distributions of 231 temperature and relative humidity are shown in Fig 1B & (Table A5) . Region-specific analysis also confirmed their modification effects (Table 1, Table A6 ). There were similar region-specific effects of meteorological factors between the 257 genders, but individuals aged ≥65 years tended to be more vulnerable to the impacts 258 of meteorological factors compared to those aged 0-64 years, although most of the 259 differences in effect did not reach statistical significance (Table 2) . Overall, temperature and relative humidity were associated with daily case counts in 262 an approximately linear fashion without discernible thresholds ( Fig A1) . Sensitivity analysis indicated that the results in general were robust. However, the use 264 of aggregated counts based on the date of reporting instead of the date of symptom 265 onset could generate biased results and a higher uncertainty (Table 3) . We examined the associations between short-term exposure to meteorological factors and COVID-19 incidence and transmission rate (R t ) based on the date of symptom 270 onset, instead of the date of reporting, which is commonly available. In general, 271 temperature and relative humidity neither showed any effect on the R t of COVID-19 272 at the national nor at regional scales, but it was associated with COVID-19 incidence 273 in some areas. In this study, spatial heterogeneity in effects of temperature and relative humidity on 276 COVID-19 incidence was found. Our data showed a negative correlation between 277 relative humidity and COVID-19 incidence in Central China, which is consistent with intra-provincial movement were associated with a higher provincial infection rate. The attenuation of the effects of meteorological factors on the incidence of Table A4 ). The association between meteorological factors and COVID-19 incidence was Through sensitivity analysis, significant changes in the results were found for the data 380 We declare that we have no conflicts of interest. Tables Table 1. Percentage change of daily count of COVID-19 with one centigrade increase of temperature and one percent increase of relative humidity and the effects on R t in country-based and region-based scales. Tables Table A. 1. Correlation coefficient matrix among meteorological factors. Fig A. 1. The exposure-response curves for temperature and relative humidity at country-based scale. J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f 29.69) -23.66 (-57.68, 37.71) North China -9 Xinyu 8 Shangrao 8 Zhengzhou 2.23±2.54 68 Wuhan 6 Huangshi 6 Shiyan 5 Xiangfan 5 Ezhou 6 Jinmen 6 Xiaogan 5 Huanggang 6 Xianning 6 Suizhou 5 Mueang Hubei 6 Chengdu 3