key: cord-0870696-i4p1jdnz
authors: Christophi, Costas A.; Sotos-Prieto, Mercedes; Lan, Fan-Yun; Delgado-Velandia, Mario; Efthymiou, Vasilis; Gaviola, Gabriel C.; Hadjivasilis, Alexandros; Hsu, Yu-Tien; Kyprianou, Aikaterini; Lidoriki, Irene; Wei, Chih-Fu; Rodriguez-Artalejo, Fernando; Kales, Stefanos N.
title: Ambient temperature and subsequent COVID-19 mortality in the OECD countries and individual United States
date: 2021-04-22
journal: Sci Rep
DOI: 10.1038/s41598-021-87803-w
sha: 38ab462e5ed39b606ac9b2c8ba7668a391cd90c5
doc_id: 870696
cord_uid: i4p1jdnz

Epidemiological studies have yielded conflicting results regarding climate and incident SARS-CoV-2 infection, and seasonality of infection rates is debated. Moreover, few studies have focused on COVD-19 deaths. We studied the association of average ambient temperature with subsequent COVID-19 mortality in the OECD countries and the individual United States (US), while accounting for other important meteorological and non-meteorological co-variates. The exposure of interest was average temperature and other weather conditions, measured at 25 days prior and 25 days after the first reported COVID-19 death was collected in the OECD countries and US states. The outcome of interest was cumulative COVID-19 mortality, assessed for each region at 25, 30, 35, and 40 days after the first reported death. Analyses were performed with negative binomial regression and adjusted for other weather conditions, particulate matter, sociodemographic factors, smoking, obesity, ICU beds, and social distancing. A 1 °C increase in ambient temperature was associated with 6% lower COVID-19 mortality at 30 days following the first reported death (multivariate-adjusted mortality rate ratio: 0.94, 95% CI 0.90, 0.99, p = 0.016). The results were robust for COVID-19 mortality at 25, 35 and 40 days after the first death, as well as other sensitivity analyses. The results provide consistent evidence across various models of an inverse association between higher average temperatures and subsequent COVID-19 mortality rates after accounting for other meteorological variables and predictors of SARS-CoV-2 infection or death. This suggests potentially decreased viral transmission in warmer regions and during the summer season.

Data was collected on all US States and the District of Columbia (DC) as well as the other 35 OECD countries from publicly available sources (see supplementary methods). Colombia joined OECD on April 28, 2020 after the start of the pandemic, and therefore, was not included in the current analysis (Table 1) . We chose the OECD countries for our study because they provide a group of nations/regions with varied climatic conditions, all of which have been affected by the pandemic. Additionally, to achieve OECD membership, all countries have met certain developmental criteria regarding their economies, infrastructure and political systems, such that they are likely to provide reliable and comparable COVID-19 mortality and covariate data. Data included information on COVID-19 mortality as well as other characteristics, such as average temperature, humidity, and precipitation, air www.nature.com/scientificreports/ pollution, measures of social distancing, measures of population density, economic and health indices. These are described in further detail below and in Supplementary Fig. S1 . In order to adjust for the onset of the epidemic in each region, we used the date of the first local COVID-19 death to define a baseline date for each location from which to collect subsequent mortality outcomes, days of social distancing/confinement metrics, and meteorological data before and after the index date.

Exposure (average ambient temperature) and meteorological covariates. The primary exposure variable was average ambient temperature over the 25 days prior to the first reported COVID-19 death. The secondary exposure variable was average ambient temperature over the 25 days after the first reported COVID-19 death in order to account for possible differences in the duration of COVID-19 incubation and clinical progression. We also collected relative humidity, and cumulative precipitation, air pollution (average PM 2.5 levels) data for the same time periods for each region. For data that were available on the city level, we used the first and second largest cities 5,6 of the individual states and countries. These values were averaged before using in the analysis.

Non-meteorological covariates. We collected data on the population size, population density, days of social distancing prior to first reported COVID-19 death, the Gini index as a measure of socioeconomic inequality, ICU beds, prevalence of obesity, smoking prevalence, and proportion of the population older than 75 years. Further details on the sources used for the different variables are provided in the "Supplemental methods information S1", including the sources for other characteristics that were collected but were eventually not used in the current manuscript. For the exposure and covariates, we retained for analyses only one variable for scenarios where two or more variables were correlated or otherwise not independent (e.g. nursing home residents and persons older than 75 years; PM2.5 and PM10, etc.) and chose the factor with the most complete and reliable information for all regions.

Outcome variables. The primary outcome of interest was COVID-19 mortality. The data were extracted from open-access databases, namely NY Times 25 for the US States and DC and Worldometer 26 for the other OECD countries. We used the date of the first reported local COVID-19 death as the reference date. We then collected the cumulative number of deaths for the period of 25, 30, 35, and 40 days after the first death and computed the corresponding mortality rates per 100,000 population. Statistical analysis. Data are presented as mean ± Standard Deviation (SD) for characteristics with symmetric distributions or as median (Quartile 1, Quartile 3) for characteristics with skewed distributions. Deaths at 30 days after the first COVID-19 death was chosen as the main outcome as all regions had accumulated 30 days from the first local death at the time of data collection. Given the over-dispersion in the data, 30-day mortality was modeled using negative binomial regression models with the natural log of the population size used as the offset variable, which has been widely used in other COVID-19 mortality studies [27] [28] [29] . The main predictor of interest was the average temperature 25 days prior to the first death and the models were adjusted for a number of other meteorological and non-meteorological covariates, namely average humidity and average precipitation 25 days prior to 1st death, average PM2.5, days of social distancing, density of the largest city, the Gini index, the proportion of the population over 75 years old, the prevalence of obesity and smoking, and ICU beds per million population. In addition to the mortality at 30 days, the 25-day, 35-day, and 40-day mortality were also analyzed. For 35-and 40-day mortality (i.e. secondary outcomes), we used the secondary exposure variable, or the average The robustness of the estimates was also checked in other sensitivity analyses, including analyses excluding areas with less than ten deaths over the 30-day period, excluding very cold areas with average temperatures < 0 °C, or excluding areas that could be influential, such as New York, Italy, Spain, Japan, and South Korea for distinct local reasons. In addition, while the NY Times dataset is largely derived from the Johns Hopkins University (JHU) COVID-19 database 30 , which has been widely used for scientific research 31 , we conducted an additional validation analysis using the mortality data directly extracted from the JHU database. The Mortality Rate Ratio (MRR), which is derived by exponentiating the beta coefficient from the negative binomial regression model, together with the corresponding 95% Confidence Interval (CI) and p-value are presented. The Statistical Analysis System SAS version 9.4 (SAS Inc, Cary, NC, USA) was used for all analyses and statistical significance was set at p = 0.05 with all tests performed being two-tailed. The GENMOD procedure in SAS was used for building the negative binomial regression models.

There was available exposure and outcome information for 86 geographic areas selected for study: including the 50 US States and the District of Columbia as well as the other 35 OECD countries ( Table 1 ). The cumulative number of COVID-19 deaths reported over 30 days after the first local COVID-19 death in each area ranged from 6 (South Dakota) to 9378 (New York).

Summary measures of the major study characteristics considered in the analysis are presented in Table 2 . The median mortality rate 30-days after the first reported COVID-19 death was 2.7 per 100,000 population (quartiles: 1.3, 5.6 per 100 K) and the average temperature over the 25-day period preceding the first COVID-19 death had a mean value of 6.7 ± 5.7 °C. Furthermore, the mean number of days of social distancing prior to the first reported death was − 1.0 ± 8.8 days and the proportion of people in the population over the age of 75 years old was 7.4 ± 1.9%. The relationship of the 30-day mortality rates (per 100,000 population) and the average ambient temperature over the 25 days prior to the first death is shown in Fig. 1; Fig. 1A is presenting the scatterplot for all geographic areas together, Fig. 1B for the OECD countries only, and Fig. 1C for the individual USA states only.

In multivariable negative binomial regression models (Table 3) , the average temperature over 25 days prior to the first death was associated with a statistically significant lower mortality. For a 1 °C increment in average temperature there was a 6% lower mortality rate at 30 days (MRR: 0.94; 95% CI 0.90, 0.99; p = 0.016). Similarly, the number of days enforcing social distancing before the first death was also associated with a lower mortality (MRR: 0.97; 95% CI 0.95, 1.00; p = 0.044). In addition, the density of the largest city (in hundreds of persons/ km 2 ) and the percentage of the population older than 75 years were associated with increased mortality (MRR: 1.02; 95% CI 1.01, 1.03; p = 0.001 and MRR: 1.21; 95% CI 1.03, 1.41; p = 0.020, respectively).

In secondary analyses, we assessed the effect of average temperature on COVID-19 mortality in different scenarios to assess the robustness of our results. These scenarios included excluding New York, further excluding Japan and South Korea, excluding areas with less than ten deaths, and excluding areas with average temperature less than 0 °C. The results remained similar when excluding various outliers, including areas with less than ten deaths or with average temperatures less than 0 °C ( Table 4 ). The association of average ambient temperature with lower mortality showed similar effect in stratified analyses within OECD countries and the US States as well (Table 4) .

We performed other sensitivity analyses. The effect of average temperature on mortality was also statistically significant when the outcome was mortality at 25 days after the first death. Similarly, changing the exposure to the average temperature during the period 25 days after the first death was significantly associated in multivariable models with mortality at 35 days and at 40 days as the outcome (Supplementary Table S1 online). Days of social distancing, density of the largest city, and the proportion of the population over 75 years old also remained significant in these sensitivity analyses.

In the additional validation analysis using the dataset extracted directly from the JHU mortality database, the main findings remain consistent. The multivariate adjusted MRRs for the mortality rate at 30 days were 0.94 (95% CI 0.90, 0.99; p = 0.017) for a 1 °C increment in average temperature, 0.97 (95% CI 0.95, 1.00; p = 0.045) for the number of days enforcing social distancing before the first death, 1.01 (95% CI 1.00, 1.03; p = 0.020) for the www.nature.com/scientificreports/ density of the largest city (in hundreds of persons/km 2 ), and 1.20 (95% CI 1.02, 1.40; p = 0.026) for the percentage of the population older than 75 years, respectively.

In this study we found that after accounting for other meteorological variables and factors predictive of SARS-CoV-2 infection/mortality rates, including age distribution, population density, social distancing, income inequality, and availability of ICU beds, a 1 °C increase in ambient temperature is associated with about 6% lower COVID-19 mortality in the subsequent 30 days, across many countries and the individual United States and District of Columbia. The results were consistent when average temperature was measured during the same and subsequent periods before and after the first reported COVID-19 death and for subsequent mortality at 25, 35 and 40 days; also, they were robust across several sensitivity analyses.

Only a few studies have analyzed the effect of ambient temperature on COVID-19 mortality rates, and to our knowledge none following our approach, thus direct comparison of our results is difficult. Ma et al. found a positive correlation (r = 0.44) between diurnal temperature range and daily deaths, but their analysis is severely limited by the failure to use sufficient time lag periods (only 0-5 days, with non-stable results) and included only the Wuhan population 32 . Wu et al., studied the association between climate and daily new cases and deaths in 166 countries. After adjusting for median wind speed, age, Global Health Security, Human Development Index, and population density, no significant association with temperature and daily deaths was found, although after limiting the analysis to countries with over 10 days since the first reported case and over 100 cumulative cases, a 1 °C increase was associated with a 1.2% reduced number of daily deaths 8 . Again, this study was also limited by short lag periods and by not adjusting for other important prognostic factors, such as social distancing. Finally, Falcao et al. found no association between temperature and COVID-19 deaths but did find an inverse relationship between country average temperature and COVID-19 infections 33 . Our study overcomes the limitations in previous studies by (1) Several epidemiological studies have also analyzed the impact of temperature and other weather conditions on COVID-19 incidence, global transmission or reproduction rate with inconclusive results 9, [12] [13] [14] [15] [16] [17] [33] [34] [35] . While some studies showed no association between temperature in the range 5-18 °C and cumulative COVID-19 cases such as in Spain 15 , or temperature (median 12.8 °C) and epidemic growth 17 , more recent evidence conducted using worldwide data or data of particular regions including China, Indonesia, Spain, and US are in line with our findings and suggest a beneficial effect of higher temperature on new daily COVID-19 cases 9,34,36-40 . Our results are consistent given that average temperature 25 days prior to the first death was associated with mortality 30 days after, because mortality reflects earlier transmission and clinical disease progression. Specifically, one study has shown that COVID-19 growth rates peaked in temperate regions of the Northern Hemisphere with mean temperature of about 5 °C, and specific humidity of 4-6 g/m 3 , during the outbreak period, while rates were lower both in warmer/wetter and colder/dryer regions 39 , after adjustment for socio-economic variables. Our crude results (Fig. 1 ) also suggest low mortality in very cold regions, but conclusions are difficult given the small number of observations, the low number of deaths in these areas, and generally the lower population densities of very cold regions. Also in agreement with our results, a recent preprint study has reported an inverse relationship between temperatures above 25 °C and the estimated SARS-CoV-2 reproduction number (3.1% reduction) 34 . Another data-driven study looks into both short-term and long-term COVID-19 transmission patterns, and builds prediction models based on meteorological factors using worldwide data. For the long-term prediction, average temperatures are negatively associated with incident cases two weeks later with a non-linear relationship 41 .

Experimental studies have shown that SARS-CoV-2 can be stable at 4 °C but it is very sensitive to heat: with an incubation temperature of 70 °C the survival time of the virus is 5 min, it achieves over 3 log-unit reduction after 7 days of incubation at 22 °C, and no virus is detected at 14 days; also 1 day of incubation at 37 °C leads to over 3 log-unit reduction in viral level, and no detection afterward 42 . This is in line with evidence on other respiratory viruses, such as SARS-CoV, which shows less stability at 22-25 °C 43 . Moreover, seasonal viruses, like the influenza, peak at winter months with cold weather compromising the host immunity, but also increasing the conditions for person-to-person transmission indoors if ventilation is low and humidity outside is 40-60% 3 . Taking together, this information makes plausible the association of temperature and other weather conditions with SARS-CoV-2 viral infection and death.

Under the assumption that warmer temperatures may help slow down the viral transmission, the summer season may offer an opportunity for public health authorities in each country to strengthen the preparedness and response strategies against a new wave of COVID-19 in the fall. During summer, when in most countries COVID-19 metrics fall, authorities may accelerate vaccination programs to ensure high vaccine coverage before new virus variants, potentially resistant to the existing vaccines, could emerge. Policymakers must also be aware that extreme hot and humid weather may increase social gathering in air-conditioned indoor spaces which, because they are cool and sometimes poorly ventilated, may become ideal environments for virus transmission. Indeed, social distancing measures helped to contain the virus and flatten the peak around the world; in our study, days of social distancing before the first local death was negatively associated with mortality rates at 25-40 days, which is in line with recent research 17 . Finally, unexpected increases in virus transmission or disease mortality, resulting from currently unknown or poorly understood factors, cannot be ruled out. This possibility must also be seriously considered by policymakers and public health officials when enacting or updating COVID-19-related policies. www.nature.com/scientificreports/ Of note is that the observed association between ambient temperature and lower COVID-19 mortality holds after accounting for variables, which, as expected, were strong mortality predictors, such as population density, age structure, and economic inequality. There is previous evidence of a greater COVID-19 fatality rates among older people, living either in communities with crowded housing or especially in nursing homes 44 . In our model, we captured these aspects through the Gini index, population density and the proportion of people older than > 75 years 44 . Indeed, there was a significant correlation between the proportion of people over 75 years old and nursing homes (r = 0.46, p < 0.001). Adjustment for population density is also important because it is known that cities with higher density also tend to have more people in crowded housing, higher proportion of socioeconomically disadvantaged persons, and greater income and health disparities, also reflected by the Gini index, which are important drivers of COVID-19 contagion and mortality.

Strengths of our study are: (1) accounting for a good number of meteorological and socio-demographic predictors of COVID-19 infection and mortality; (2) using data on weather conditions before the first reported COVID-19 death, and therefore before most of the countries and US states declared the confinement actions, which could minimize the effect of ambient temperature; (3) analyzing COVID-19 mortality rates, rather than incidence rates, which are less subject to screening and reporting biases, (4) examining the exposure and outcomes across different time periods and in different sensitivity models. Some limitations should also be acknowledged. The most important is that a single measure of ambient temperature may not be fully representative of an entire country or State if it is rather large. Also, we cannot rule out small errors in the dates of the first reported death in each country, due to the different data sources as well as differences in epidemiological surveillance across countries. 

In conclusion, our results showed that a higher average temperature is associated with a subsequent reduction in COVID-19 mortality rates across various developed countries and the individual United States, after accounting for average humidity, precipitation, PM2.5, social distancing, population density, socioeconomic level, the proportion of the elders in the population, the prevalence of obesity and smoking, and ICU beds availability. While the study is limited by possible measurement errors in ambient factors due to using data from the two biggest cities as a proxy, and inherent heterogeneity in the data sources across regions, our main findings remained robust in several sensitivity analyses. Thus, they suggest that mortality from COVID-19 is likely to be reduced during the summer season as a result of warmer weather, possibly decreasing viral survival/infectivity/ virulence 45 and environmental conditions that favor greater time spent outdoors and less indoor gatherings, and thus, less viral transmission.

Environmental factors on the SARS epidemic: Air temperature, passage of time and multiplicative effect of hospital infection

Seasonality of respiratory viral infections

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in Wuhan, China: A modelling study

Homeland Security's Science And Technology Directorate

Influenza virus transmission is dependent on relative humidity and temperature

Effects of temperature and humidity on the daily new cases and new deaths of COVID-19 in 166 countries

Impact of meteorological factors on the COVID-19 transmission: A multi-city study in China

Temperature significantly changes COVID-19 transmission in (sub)tropical cities of Brazil

No Association of COVID-19 transmission with temperature or UV radiation in Chinese cities

The role of absolute humidity on transmission rates of the COVID-19 outbreak

Asymmetric nexus between temperature and COVID-19 in the top ten affected provinces of China: A current application of quantile-on-quantile approach

The nexus between COVID-19, temperature and exchange rate in Wuhan city: New findings from partial and multiple wavelet coherence

A spatio-temporal analysis for exploring the effect of temperature on COVID-19 early evolution in Spain

Effects of climatological parameters on the outbreak spread of COVID-19 in highly affected regions of Spain

Impact of climate and public health interventions on the COVID-19 pandemic: A prospective cohort study

Geographic differences in COVID-19 cases, deaths, and incidence-United States

Effect of non-pharmaceutical interventions to contain COVID-19 in China

The role of community-wide wearing of face mask for control of coronavirus disease 2019 (COVID-19) epidemic due to SARS-CoV-2

Evaluating the effectiveness of social distancing interventions to delay or flatten the epidemic curve of coronavirus disease

Fluctuations in environmental pollutants and air quality during the lockdown in the USA and China: Two sides of COVID-19 pandemic

Co-variance nexus between COVID-19 mortality, humidity, and air quality index in Wuhan, China: New insights from partial and multiple wavelet coherence

The Organization for Economic Co-operation and Development (OECD)

Covid-19) Data in the United States

COVID-19 CORONAVIRUS PANDEMIC

Low dispersion in the infectiousness of COVID-19 cases implies difficulty in control

Factors associated with COVID-19 infections and mortality in Africa: A crosssectional study using publicly available data

Impact of COVID-19 prevalence and mode of transmission on mortality cases over WHO regions

The New York Times. U.S. Coronavirus Data: Frequently Asked Questions

An interactive web-based dashboard to track COVID-19 in real time

Effects of temperature variation and humidity on the death of COVID-19 in Wuhan

Association between climate variables and global transmission oF SARS-CoV-2

The Modest Impact of Weather and Air Pollution on COVID-19 Transmission

The correlation between the spread of COVID-19 infections and weather variables in 30 Chinese provinces and the impact of Chinese government mitigation plans

Correlation between weather and Covid-19 pandemic in Jakarta

Effect of temperature on the transmission of COVID-19: A machine learning case study in Spain

Correlation between climate indicators and COVID-19 pandemic

Climate affects global patterns of COVID-19 early outbreak dynamics

Investigating the effects of meteorological parameters on COVID-19: Case study of New Jersey, United States

Predicting the local COVID-19 outbreak around the world with meteorological conditions: A model-based qualitative study

Stability of SARS-CoV-2 in different environmental conditions

The effects of temperature and relative humidity on the viability of the SARS coronavirus

COVID-19: Towards controlling of a pandemic

Evolving virulence? Decreasing COVID-19 complications among Massachusetts healthcare workers: A cohort study

Access to the data used can be freely accessed to the scientific community (see "Supplementary data S1").Received: 14 July 2020; Accepted: 5 April 2021 Acknowledgements MSP holds a Ramón y Cajal contract (RYC-2018-025069-I) from the Ministry of Science, Innovation and Universities and FEDER/FSE. The funding agency had no role in study design, data collection and analysis, interpretation of results, manuscript preparation or in the decision to submit this manuscript for publication.

The authors declare no competing interests.

The online version contains supplementary material available at https:// doi. org/ 10. 1038/ s41598-021-87803-w.Correspondence and requests for materials should be addressed to S.N.K.Reprints and permissions information is available at www.nature.com/reprints.Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.