Impact of environmental temperature and relative humidity on spread of COVID-19 infection in India: a cross-sectional time-series analysis

Abstract Purpose: Coronavirus disease 2019 (COVID-19) has become a serious public health problem worldwide. This study sought to examine the associations of daily average temperature (AT) and relative humidity (ARH) with the percent increase in COVID-19 cases. Methods: Daily confirmed cases and meteorological factors in 38 districts of India were collected between 1st April 2020 to 30th April 2020. Taking a 5-day time lag of average values of the variables and multiple days-samples, we ran multiple models and performed appropriate hypothesis tests to decide the single preferred model for each sample data. Suitable fixed effects (FE) and random effects (RE) models with cluster-robust standard errors were applied to quantify the district-specific associations of meteorological and other variables with COVID-19 cases. Results: All FE models revealed that every one-degree rise in AT led to a decrease in 3.909 points (on average) in percent increase in COVID-19 cases. All RE models showed that with one unit increase in the malaria annual parasite index, there was a significant increase in 10.835 points (on average) in percent increase in COVID-19 cases. In both FE and RE models, ARH was found to be negatively associated with a percent increase in COVID-19 cases, although in half of these models the association was statistically insignificant. Conclusion: Our results indicate that mean temperature, mean relative humidity, and malaria endemicity might have an essential role in the stability and transmissibility of the 2019 novel coronavirus.


Introduction
Coronavirus disease 2019  is caused by the new coronavirus called 2019-nCoV that was first identified in Wuhan city of China. 1,2 Subsequently, it has spread to other regions of China and many other countries of the world and now it has become a serious public health problem worldwide. 3,4 COVID-19 has resulted in 13.4 million cases, more than 580,000 deaths, and significant social and economic distress across the world. 5 COVID-19 is characterized by fever, cough, breathlessness, myalgia, pneumonia, and may cause progressive respiratory failure and death. On 30 th January 2020, India's first case of COVID-19 was confirmed in the state of Kerala and on 24 th March 2020, the Government of India initiated nationwide strict lockdown measures such as restriction of movement of the entire population of India, restriction of mass gatherings, closure of institutions, social distancing, and other quarantine methods to prevent the COVID-19 pandemic in India. As of 30 th April 2020, data have shown that approximately 33,000 confirmed cases were identified and over 1100 deaths in the whole of India.
Early studies suggested that COVID-19 is transmitted from person to person through direct contact or droplets. [6][7][8] Also, various studies have pointed to the role of climatic conditions like ambient temperature and humidity on the transmission and survival of 2019-nCoV. [9][10][11][12] In this study, we sought to explore the effect of environmental temperature and humidity on the spread of this novel coronavirus in India. To provide useful implications for policymakers and the public, our study aimed to examine the relationship of average air temperature (AT) and relative humidity (ARH) with the percent increase in COVID-19 cases in different districts of India. We also assessed the effect of other covariates like daily average air pressure, daily average wind speed, latitude, longitude, BCG vaccination rate, Polio vaccination rate, and malaria endemicity on the spread of the 2019 novel coronavirus.

Study area and data collection
Our study included panel data of 950 observations from 38 districts of 22 states that covered the majority of the Indian mainland (11 to 35 north latitude and 73 to 93 east longitude). The dataset including daily confirmed cases for all the districts of India was collected from the website by "GRAM India's data unlocked by How India Lives (https://howindialives. com/gram/coronadistricts/) during April 2020. The website sourced the data from the health ministry website and news articles from National and regional publications. We chose the month of April as our study period as, during this period, lockdown measures were strictly followed across all the states of India which could have minimized the potential inclusion of imported cases to a particular district from another. It is estimated that the incubation period of Covid-19 can extend to 14 days and averages around 5 days. We took a five-day incubation period between exposure and symptoms as estimated in previous studies. 8,13 All the Indian districts that had at least one confirmed case on 6 th April and ! 20 cases on 30 th April were included in our study. Meteorological information regarding daily average air temperature, daily average air pressure, daily average relative humidity, and daily average wind speed of each district during the same study period was acquired from the website (https://darksky.net/forecast), one of the accurate weather networks. The website collects data using a combination of meteorologists, a global forecast engine, and other sources. We took the average hourly values of the meteorological variables from 8:00 a.m. to 8:00 p.m. Also, the latest data about other variables for each district such as latitude and longitude (http://earth.google.com); BCG vaccination rate and polio vaccination rate (http://rchiips.org); and malaria annual parasite indices (http://nvbdcp.gov.in) were retrieved using publicly accessible websites. Other relevant accessible websites were checked to verify the data.

Statistical analysis
Before the analysis of our panel data, we checked several assumptions for ordinary least squares (OLS) about the data. There was the presence of heteroscedasticity and one extreme data point that was influential in the regression analysis. Hence, we dropped that influential data point from our analysis and finally included 949 observations in the analysis. As the incubation period of Covid-19 and temperature effect could last for several days, it was reasonable to use the moving average approach to account for the cumulative effect of temperature. 12,14,15 We took multiple days-samples for 38 districts such as observations for 2 days (6 th -7 th April), 3 days (6 th -8 th April), … … … .25 days (6 th -30 th April).To analyze such panel data, the fixed effects model and random effects model are the preferred statistical tests. First, the fixed-effects model and random-effects model were tested against pooled ordinary least squares (OLS) regression model by F test and Breusch-Pagan Lagrange multiplier test respectively. As in each case, the null hypothesis was rejected, we considered fixed effects and random effects models appropriate for the analysis. Then we ran the Hausman test to decide whether fixed effects or random effects model should be preferred for the analysis of each sample data and found the fixed effects model suitable in the majority of cases. All the samples with favorable model fit statistics were included and the effect of the independent variables on the outcome variable was examined using appropriate fixed-effects or random-effects model with cluster-robust standard errors to allow for heteroscedasticity and serial correlation. 16,17 A percentage increase variable was created using the formula: Percent increase ¼ New cases (t)/ Cumulative cases (t-1) where t is daily. This was the outcome variable for our study which was examined for its association with four time-varying independent meteorological variables such as average air temperature, average relative humidity, average air pressure, and average wind speed. In the fixed-effects model (least squares dummy variable model), dummy variables were introduced for every district barring Mumbai (exhaustive dummy variables always need a reference point). This means that observations from a particular district would have a value of 1 in that district dummy and 0 in all other district dummies. The dummy variables capture variance due to factors specific to the district that do not change over time. The fixed-effects model allows us to only look at how inter-day variation in temperature, relative humidity, pressure, and wind speed explain the spread of COVID-19. In this model, we could effectively compare the same districts over time and then compared those results with other districts in the same period.
In the random-effects model, other time-invariant covariates such as latitude, longitude, malaria annual parasite index, BCG vaccination rate, and Polio vaccination rate along with the above-mentioned timevarying meteorological factors were examined. All the analyses were performed using the RStudio version 1.3.959 software. The statistical tests were two-tailed, and p < 0.05 was considered statistically significant.
Ethics approval: This study did not require research ethics approval, as we used publicly accessible, anonymized aggregate data for all analyses.

Results
We included 38 districts with 949 observations over one month (1 st -30 th April 2020) in our analyses. Figures 1 and 2 show the presence of heterogeneity across days and districts respectively. Figure 3 shows the location of selected districts in India on a map. The study included 20,373 cases during the observation period and the average daily confirmed cases were 21.47. Average daily temperature, relative humidity, air pressure, and wind speed were 31.78 C, 41.32%, 1149.86 hPa, and 8.91 m/s respectively. Mean temperature had significant negative correlations with relative humidity (r ¼ À 0.522, p < 0.01), air pressure (r ¼ À 0.480, p < 0.01), wind speed (r ¼ À 0.234, p < 0.01). Table 1 depicts that the fixed-effects model in all the samples consistently revealed that air temperature had a significantly negative impact on COVID-19 cases i.e., on average, every one-degree rise in AT led to a decrease in 3.909 points in percent increase in COVID-19 cases. In all the samples, ARH was negatively associated with a percent increase in COVID-19 cases and the association was statistically significant in samples 9 À 14. None of the samples revealed any significant relationship between air pressure and wind speed with a percent increase in COVID-19 cases. Table 2 presents that the random-effects model in all the samples showed a negative effect of AT and ARH on the percent increase in COVID-19 cases although, in most of the samples, the association was statistically insignificant. However, we observed that with one unit increase in the malaria annual parasite index, there was a significant increase in 10.835 points (on average) in percent increase in COVID-19 cases. The effect of other variables such as air pressure, wind speed, latitude, longitude, BCG vaccination rate, and Polio vaccination rate on COVID-19 cases was found statistically insignificant.

Discussion
In this cross-sectional time-series study of 38 districts with 20,373 confirmed cases of COVID-19, the percent increase in COVID-19 during the follow-up period from 1 st to 30 th April 2020, was negatively associated with AT andARH, but positively associated with the malaria annual parasite index. This cannot be mistaken that high temperature and high humidity will take care of the COVID-19 pandemic as these results only hold under the condition of strict lockdown measures in India. Thus, the intervention   observed that high temperature and high humidity significantly reduced the transmission of the virus. 11 Another study conducted in 31 different states of China and 70 cities of 11 countries revealed that air temperature and humidity had a detrimental impact on the transmission of the 2019-new coronavirus. 18 Similar results have been reported in other studies. [19][20][21] However, contrasting our results, some studies could not find any evidence supporting the fact that COVID-19 cases could decline in warm, humid regions. 12,22 Bilal et al. conducted a meta-analysis to examine the nexus between COVID-19 and temperature and did not find any significant association between them. 23 We also observed in our study that the percent increase in COVID-19 cases was positively associated with malaria endemicity of districts which indicates that there might be a limited spread of COVID-19 in malaria-free districts as compared to malaria-endemic districts. This is in contrast to the result found in an earlier study. 24 Further studies exploring the relationship of malaria endemicity with COVID-19 transmission are needed to be conducted to establish such an association.
Our study has some limitations. First, some potential risk factors such as district-specific general health policies, socioeconomic status, exposure to air pollution, etc. which could influence the spread of COVID-19 were not included in the models. Also, we could not consider the district-specific population density in our study as the latest data were not available. Second, our data only covered districts of India and thus the findings may not be generalized to other regions of the world. Third, it is difficult to quantify the compliance of the population with social distancing and

Conclusion
Our study shows that meteorological factors like air temperature, relative humidity, and malaria endemicity are associated with the spread of COVID-19. Based on our results and other recent available evidence, it seems that high temperature and relative humidity might play an essential role in the stability and transmissibility of COVID-19. However, we suggest that more rigorous studies should be done with data from other countries using different study designs and incorporating other covariates like air pollution, population density, health infrastructure, cultural practices, etc. to validate these findings.

Funding
No funds, grants, or other support was received.

Conflicts of interest
The authors have no conflicts of interest to declare that are relevant to the content of this article.

Ethics approval
Not applicable as the present study is based on publicly accessible data.