key: cord-0263643-z8kcgoa2 authors: Liang, J.; Yuan, H.-Y. title: Assessing the impact of temperature and humidity exposures during early infection stages on case-fatality of COVID-19: a modelling study in Europe date: 2021-09-26 journal: nan DOI: 10.1101/2021.09.23.21264017 sha: 768ebd3dfa70f55b9a175d93b596005319632ba0 doc_id: 263643 cord_uid: z8kcgoa2 Background Although associations between key weather indicators (i.e. temperature and humidity) and COVID-19 mortality has been reported, the relationship between these exposures among different timing in early infection stages (from virus exposure up to a few days after symptom onset) and the probability of death after infection (also called case fatality rate, CFR) has yet to be determined. Methods We estimated the instantaneous CFR of eight European countries using Bayesian inference in conjunction with stochastic transmission models, taking account of delays in reporting the number of newly confirmed cases and deaths. The exposure lag response associations between fatality rate and weather conditions to which patients were exposed at different timing were obtained using distributed lag nonlinear models coupled with mixed effect models. Results Our results showed that the Odds Ratio (OR) of death is negatively associated with the temperature, with two maxima (OR=1.29 (95% CI, 1.23, 1.35) at -0.1 degrees Centigrade, OR=1.12 (95% CI, 1.08, 1.16) at 0.1 degrees Centigrade) occurred at the time of virus exposure and after symptom onset. Two minima (OR=0.81 (95% CI: 0.71, 0.92) at 23.2 degrees Centigrade, OR=0.71 (95% CI, 0.63, 0.80) at 21.7 degrees Centigrade) also occurred at these two distinct periods correspondingly. Low humidity (below 50%) during the early stages and high humidity (around 89%) after symptom onset were related to the lower fatality. Conclusion Environmental conditions may affect not only the initial viral load when exposure to viruses but also individuals immunity response around symptom onset. Warmer temperatures and higher humidity after symptom onset were related to the lower fatality. Introduction temperature and humidity) during the non-severe symptomatic period can affect patients' innate 23 immune responses 4, 7, 8 and hence their risk of death, preventive measures can be designed to reduce 24 COVID-19 severity for these cases. However, until now, no such preventive measures are proposed 25 due to that the impact of those conditions during the symptomatic period is unknown. Although many studies have reported that low temperatures may increase the COVID-19 deaths or 28 mortality 9-14 , these studies did not have a direct measurement on the risk of death. Case fatality rate 29 (CFR) 15, 16 is an important index to measure the disease severity, but one limitation is that this rate 30 only represents the average proportion of deaths among all confirmed cases over a duration of time, 31 without the ability to reflect the instantaneous probability of death. This time-varying instantaneous 32 probability, also called instantaneous CFR (iCFR) 17 , can be influenced by many factors, such as The study aimed at assessing the impact of weather conditions COVID-19 patients were exposed 39 to at different timing of the early infection course on the death risk. To resolve the above issues in 40 delays and to estimate the iCFR, stochastic modelling 22 was used, taking into account of the delays 41 in reporting the number of newly confirmed cases and deaths in each of the European countries. After 42 adjusting delays in reporting cases and deaths, the correlation between the iCFR and daily weather 43 conditions at different timing since the infection was obtained using distributed lag non-linear 44 models (DLNMs) coupled with generalized linear mixed models. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. European countries from 16th February to 31st June 2020 were collected from 'Our World in Data' 1 . 51 We defined the community outbreak began from the first two consecutive days those average case 52 number exceeded the country's baseline (the 5% quantile of the maximal daily number of new cases) 53 and ended at the first two consecutive days those average case number was less than that baseline. Methods). In this study, the daily mean temperatures and relative humidity values for each country 61 were calculated using the average records from all monitoring stations. Estimating COVID-19 transmission patterns by SEIR model 64 We constructed a stochastic model that extended Susceptible -Exposed -Infectious -Recovered effective reproduction number (Re) for each of countries during their outbreaks. The extended SEIR calculate the iCFR of the date of case confirmation, newly confirmed cases were divided into HR 70 and HD compartments following the probabilities of 1-iCFR and iCFR upon the date of case confir- To explore the non-linear association between weather conditions and the iCFR with taking account 85 of other unknown local factors for each of countries, a combination of distributed lag nonlinear 86 models (DLNMs) 27 and generalized linear mixed models (GLMM) was adopted to estimate the 87 differential effects of weather conditions exposed during the early infection stages on the iCFR. For All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. To assess convergence of parameters in the SEIR model, we constructed three independent chains of 92 algorithm sets with 100,000 iterations and calculated the Gelman-Rubin convergence diagnostic 93 statistics 28 across the three chains. For DLNM models, we used different combinations of tempera- Table 5 ). The prediction performance of the best-fitting model was presented by comparing its All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. responses at about a few days after symptom onset were affected by environmental temperatures. 156 Furthermore, results showed the impacts of temperatures on the risk of death were greater at virus 157 exposure time and few days after symptom onset than other periods during early infection stages. For example, a decrease from 5 • C to 0 • C at one day after symptom onset increased the risk of 159 deaths significantly (OR increased from 1.03 to 1.07; see Figure 5B ). This increase was significantly 160 greater than during the presymptomatic transmission period (e.g. 6-folds greater than that) at three 161 days before symptom onset (OR only increased from 1.006 to 1.012). Figure 5D showed the associations between the humidity and risk of death, with a relative humidity 173 of 62% as the reference. The highest OR (1.08, 95% CI: 1.07, 1.10) was observed at 79.6% relative 174 humidity at symptom onset time. Figure 5E showed the cumulative OR increased when the humidity 175 raised from 30% to 80%. However, the cumulative OR clearly reduced when the humidity increased 176 from 80% to 89%. This reduction was mainly resulted from the effect of humidity after symptom 177 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. In summary, the variations of iCFRs within the eight European countries were associated with 181 changes in weather conditions. Furthermore, the OR of fatality was clearly associated with the 182 temperature and humidity that patients were exposed to at two distinct infection stages: virus 183 exposure and after symptom onset. During the first epidemic wave in Europe, certain countries suffered high mortality rates. We found 198 that warm conditions reduced the risk of deaths especially when the temperature was greater than 199 All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. people to be cautious and reduce outdoor activities). In addition, because in some countries, many 256 cases were confirmed and isolated in hospitals around two days after symptom onset on average 257 (see Figure 1B) , hence their environmental exposure would mainly be determined by the hospital air 258 conditioning system. Therefore, we only assess the impact of weather exposure no longer than that 259 time. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Ethics approval is not needed as the study uses publicly available country-level (aggregated) morbid-264 ity, mortality, and weather data. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted September 26, 2021. ; Figure 1 . (A) Schematic of the extended Susceptible, Exposed, Infectious and Recovered model with case confirmation and death statuses. The total population was divided into seven compartments: S (susceptible), E (exposed), I (infectious after the incubation period), HR (hospital confirmed cases who later recovered), HD (hospital confirmed cases who later died), R (recovered), and D (death). β is the transmission rate, 1 σ is the incubation period, µ is the proportion of pre-symptomatic infectious individuals among the total number of exposed individuals 41 , ε is the recovery rate for un-reported cases (mainly asymptomatic cases), 1 ρ is the confirmation delay, 1 ϕ is the infection outcome delay, and γ is the recovery rate for hospital confirmed cases. iCFR t is the iCFR at time t. (B) Estimated confirmation delay and infection outcome delay in the eight European countries. Black dots represent the posterior mean value and horizontal lines represent the 95% posterior credible intervals. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted September 26, 2021. ; where the mortality rates were relatively low. Red and purple curves represent the estimated mean iCFR and Re, light red and dark red shaded areas represent the 95% and 50% credible intervals for iCFR. Purple shaded areas represent the 95% credible intervals for Re. The black vertical lines refer to the dates when Re reduced to below 1.5. iCFRs after these dates were used for estimating the effects of weather conditions. Daily mean iCFR, Re, and their credible intervals were obtained from the PMCMC posterior samples. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted September 26, 2021. ; Daily mean temperature and humidity in countries where the mortality rates were relatively high. (B) Daily mean temperature and humidity in countries where the mortality rates were relatively low. The blue curve represents the trend of daily temperature and humidity, which was obtained from a smoothing curve (a 18th order polynomial function). The grey shadows represent 95% credible intervals for the trends. The black vertical lines refer to the dates when Re was reduced to below 1.5. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted September 26, 2021. ; https://doi.org/10.1101/2021.09.23.21264017 doi: medRxiv preprint Figure 5 . The effect of weather conditions on the COVID-19 fatality rate. (A) The timeline of COVID-19 infection course while taking account of the effects of weather conditions (i.e. temperature and relative humidity). The duration of home-isolation was approximately equal to the confirmation delay because many cases were isolated at hospitals after being confirmed. The duration of hospital-isolation for cases who later died was indicated by the infection outcome delay. Weather conditions mainly affected infected individuals during the early infection period including exposure to virus, incubation time and home-isolation period. During hospital-isolation period, indoor temperature and humidity were controlled by hospitals. (B and D) Relationships between weather conditions and the OR of fatality at different time point between exposure to virus and two days after symptom onset. Redder colours indicate higher OR. 11 • C of the temperature and 62% of the relative humidity were used as references. (C and E) The estimated cumulative effects of the temperature and the relative humidity on the fatality. The red lines are the mean ORs, and the grey shaded areas are the 95% credible intervals. All rights reserved. No reuse allowed without permission. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted September 26, 2021. ; https://doi.org/10.1101/2021.09.23.21264017 doi: medRxiv preprint Inverse correlation between average monthly high temperatures and covid-309 19-related death rates in different geographical areas The association between covid-19 deaths and short-term ambient air 311 pollution/meteorological condition exposure: a retrospective study from wuhan, china Effects of temperature and humidity on the daily new cases and new deaths of 314 covid-19 in 166 countries Monitoring transmissibility and mortality of covid-19 in 316 europe Cross-country comparison of case fatality rates of covid-19/sars-318 cov-2 Decreased case fatality rate of covid-19 in the second wave: a study in 53 countries