Introduction

Coronavirus Disease 2019 (COVID-19) is an acute respiratory infectious disease caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), which first emerged in Wuhan, China. According to the World Health Organization’s (WHO) report1, as of November 2, 2023, there have been a total of 771,679,618 confirmed cases and 6,997,023 deaths reported worldwide. China is one of the countries significantly affected by COVID-19, with a reported total of 99,317,967 confirmed cases and 121,764 deaths as of October 20231. Several factors play a role in the incidence and spread of COVID-19, including interpersonal interactions, population density, and levels of air pollution. Recent extensive research has explored the connection between these risk factors and COVID-19 infection2,3. However, limited attention has been given to studying the influence of these factors on COVID-19 mortality, especially regarding the impact of air pollution. Therefore, it is crucial to investigate whether air pollution is a contributing risk factor for acute respiratory infectious diseases like COVID-19.

Epidemiological studies on a global scale have revealed that both long- and short-term exposure to air pollutants can elevate the incidence and mortality rates of COVID-19. For instance, studies in Italy and Germany indicate that long-term exposure to particulate air pollutants may be a contributing factor to the increase in COVID-19 cases3,4. Meanwhile, a Swiss study found that long-term exposure to air pollution increased the risk of COVID-19 during the initial wave of the pandemic in Switzerland5. In Italy, research found that short-term exposure to air pollutants also increases the risk of COVID-19 mortality6. In the United States (U.S.), Harvard University analyzed the association between long-term PM2.5 exposure and COVID-19 mortality in over 3000 counties using a generalized additive model (GAM) with a negative binomial distribution7. Subsequently, Emory University employed a similar approach to investigate the impact of air pollutants on COVID-19 mortality in 32 U.S. counties8. Both studies concluded that prolonged PM2.5 exposure increases COVID-19 mortality.

In China, studies have also examined the relationship between air pollution and COVID-199,10,11. For example, a systematic review and meta-analysis confirmed the association between NO2 exposure and COVID-19 mortality12. Zhu et al. used GAM to analyze the impact of six major air pollutants on COVID-19 incidence in 120 cities, finding positive associations for PM2.5, PM10, NO2, and O3, while SO2 showed a negative correlation9,13. A study in Shanghai using the same method also found significant associations between PM2.5, PM10, NO2, and SO2 with daily confirmed cases14. Previous studies in China mainly used Poisson regression within the GAM framework to examine the relationship between air pollution and respiratory infectious diseases12,15. However, Poisson regression has limitations, particularly in handling overdispersed data. In contrast, negative binomial regression is more suitable for such data, providing a better model fit. Notably, two U.S. studies successfully applied this method to analyze the link between air pollution and COVID-19, producing robust results7,8.

Building upon the aforementioned discussions, the study employed a GAM model with a negative binomial distribution, utilizing data from 45 cities in China to investigate the relationship between air pollutant concentrations (PM2.5, SO2, NO2, and O3) and the COVID-19 mortality from January 23, 2020 to May 23, 2020.

Results

Descriptive statistics of air pollutants, meteorological factors, and COVID-19 cases

Table 1 presents the descriptive analysis of air pollutants, meteorological factors, and COVID-19 data from 45 cities in China spanning from January 23rd to May 6th, 2020. Throughout this period, the average concentrations of air pollutants PM2.5, SO2, NO2, and O3 were 43.79 μg/m3, 9.32 μg/m3, 24.31 μg/m3, and 66.15 μg/m3, respectively. The daily average temperature and relative humidity across the 45 cities during the study duration were 17.6 °C and 78.33%, respectively. From January 23rd to May 23rd, 2020, there were a total of 1,568,270 cumulative COVID-19 cases reported across the 45 cities in China. The maximum daily count of cases reached 48,984, while the minimum daily count was 1. Over the study period, there were 3244 reported deaths due to COVID-19, with the highest daily count of deaths being 216 and the lowest being 1. Among the 45 cities, Wuhan recorded the highest cumulative number of cases, totaling 1,166,842, along with 2670 cumulative deaths.

Table 1 Descriptive analysis of air pollutants, meteorological factors, and COVID-19 patient counts
Full size table

Regional disparities in COVID-19 impact and air pollution levels

Table 2 provides data on 45 cities, including population, cumulative COVID-19 cases and deaths during the study period. These cities are categorized into 11 northern, 24 southeastern, and 10 western cities. Chongqing has the largest population with 34,162,900 residents in 2019, while Lhasa has the smallest at 559,000. Generally, northern and southeastern cities have larger populations, often exceeding 5 million residents. The COVID-19 pandemic was more severe in the southeast, with 11 of 24 cities reporting over 1000 cases and 10 cities with more than 10 deaths. Wuhan had the highest number of confirmed cases (1,166,842), representing 13% of its population, and the highest number of deaths (2670). Air pollution was most intense in the northern region, where average pollutant concentrations were higher than in the southeast and west. Except for Beijing, all northern cities had PM2.5 levels above the WHO interim target 2 (50 μg/m3), with Shijiazhuang recording the highest at 92.3 μg/m3. In contrast, only four cities in the southeast and west exceeded this PM2.5 threshold. SO2 and NO2 levels were similarly high in the north and west, significantly surpassing those in the southeast. O3 levels were comparable across all regions (Table S1).

Table 2 Descriptive analysis of population and COVID-19 patients counts in 45 cities
Full size table

Delayed effects of air pollution on COVID-19 mortality

Figure 1 illustrates the influence of four air pollutants on daily COVID-19 mortality across 45 cities, taking into account various lag days after adjusting for average daily temperature and relative humidity. For NO2, its effect fluctuates from lag 0 to lag 14, all statistically significant. At lag 1, the effect weakens slightly compared to the current day, then trends upward at lag 2 before fluctuating and decreasing, reaching its nadir at lag 7. It then rises again, peaking on the 11th lag day (RR = 1.283, 95% CI: 1.143–1.351). SO2 exhibits a similar trend, although it differs from NO2 in terms of significance. While NO2 shows significance across all lag days, SO2 demonstrates significance only at lag 0. From the 3rd to the 9th lag day, O3 exhibits a negative association with COVID-19 mortality, acting as a protective factor. Its effect peaks on the 7th lag day (RR = 0.889, 95% CI: 0.851–0.929). Conversely, PM2.5 demonstrates no significant effect on COVID-19 mortality throughout the observed lag time interval. The cumulative lag effect of air pollutants on COVID-19 mortality is similar to the single-day lag effect, with NO2 having the most significant effects (Fig. S1).

Fig. 1: Single-day lag effects of air pollutants on COVID-19 mortality.
Fig. 1: Single-day lag effects of air pollutants on COVID-19 mortality.The alternative text for this image may have been generated using AI.
Full size image

Single-day lag effect estimates and confidence intervals of four air pollutants on COVID-19 mortality from lag 1 to lag 14 based on all data.

Differences in pollution effects between Wuhan and other cities

Figure 2 shows the influence of four air pollutants on COVID-19 mortality at lag0, lag7, and lag14 in 45 cities and Wuhan. In Wuhan, PM2.5 concentrations had significant effects at lag0 and lag7 (p < 0.05), with every 10 μg/m3 increase in PM2.5 associated with a 21.2% (95% CI: 1.110–1.323) and 12.7% (95% CI: 1.030–1.233) rise in COVID-19 mortality, respectively. In the 45 cities, PM2.5 did not show statistical significance for the three single-day lag effects. NO2 in Wuhan was significant at lag0 and lag14, with 10 μg/m3 increases associated with 51.1% (95% CI: 1.199–1.903) and 30.6% (95% CI: 1.111–1.535) rises in mortality, respectively. These effects were stronger than those in the 45 cities at the same lags (lag0: RR = 1.249, 95% CI: 1.119–1.393; lag14: RR = 1.143, 95% CI: 1.048–1.247). SO2 in Wuhan was negatively associated with COVID-19 mortality at lag7, unlike in the 45 cities. O3 showed no statistical effect at lag0 in both Wuhan and the 45 cities but had a negative association with mortality at lag7, with a slightly weaker effect in Wuhan.

Fig. 2: Effect estimates of air pollutants on COVID-19 mortality at different lag days.
Fig. 2: Effect estimates of air pollutants on COVID-19 mortality at different lag days.The alternative text for this image may have been generated using AI.
Full size image

Effect estimates and confidence intervals of four air pollutants on COVID-19 mortality at lags of 1, 7, and 14 days in 45 cities and Wuhan, China.

Long-term impact of air pollution on COVID-19 mortality

Figure S2 shows that the cumulative lag effects of the air pollutants align with the single-day lag effects from Fig. 1. Overall, the cumulative and single-day effects are not significantly different. The cumulative effects of PM2.5 and NO2 in Wuhan remained significant, particularly for PM2.5 over a lag period of 0–14 days. SO2 and O3 exhibited protective effects during the lag periods of 0–7 and 0–14 days in Wuhan, consistent with their single-day lag effects.

Regional differences in air pollution effects on COVID-19 mortality

Figures 3 and S3 present the results of stratified analysis across different regions, with Fig. 3 showing the effect of single lag days and Fig. S3 showing the effect of cumulative lag days. The figures indicate that, whether it is single-day lag effects or cumulative effects, there is no statistical significance for the impact of PM2.5 and SO2 on COVID-19 mortality in the northern, western, and southeastern cities. For NO2, only the effect size in southeastern cities is statistically significant, while the effects in western and northern cities are not. Notably, the cumulative lag effect is greater than the single-day lag effect, and the impact of NO2 becomes stronger with increasing lag days; conversely, the effect of NO2 weakens with longer single-day lags. Regarding O3, it shows a protective effect only in the southeastern cities for single-day lag effects. In cumulative lag effects, O3 has a negative association with COVID-19 mortality in both southeastern and western cities, but its effect remains statistically insignificant in northern cities.

Fig. 3: Regional effects of air pollutants on COVID-19 mortality at different lag days.
Fig. 3: Regional effects of air pollutants on COVID-19 mortality at different lag days.The alternative text for this image may have been generated using AI.
Full size image

Effect estimates and confidence intervals of four air pollutants on COVID-19 mortality at lags of 1, 7, and 14 days in 45 cities across three regions of China.

Interaction effects of multiple air pollutants on COVID-19 mortality

In addition to single-pollutant models, we also employed two-pollutant models to assess the impact of air pollution on COVID-19 mortality. The results indicate that the effects are strongest at lag0–4, lag0–9, and lag0–12 (Table S2). In the two-pollutant models, the effect size of NO2 significantly increased after adjusting for PM2.5. Similarly, the presence of SO2 also enhanced the effect of NO2. O3 increased the effect of NO2 at lag0–4 but decreased it at lag0–9 and lag0–12, though the results remained statistically significant. The effect of SO2 remained statistically insignificant after adjusting for PM2.5, NO2, and O3. After adjusting for NO2, the effect of PM2.5 was negative, indicating a negative association with COVID-19 mortality.

Discussion

In this study, we employed a generalized additive model with a negative binomial distribution to examine the impact of short-term air pollution exposure on COVID-19 mortality in 45 Chinese cities. The results from the single-pollutant model indicated a positive association between NO2 concentrations and COVID-19 mortality across the 45 cities. Exposure to elevated concentrations of SO2 increased the risk of COVID-19 mortality, which was more apparent in lagged effects, while O3 exposure showed a protective effect. The adverse health effect of PM2.5 was statistically significant only in Wuhan, where the effect of NO2 on COVID-19 mortality was notably stronger than in the other cities. Furthermore, we found that the cumulative lagged effects in the single pollutant model were stronger than the single-day lagged effects, especially for NO2. In the two-pollutant model, NO2 estimates remained statistically significant after adjusting for SO2, PM2.5, and O3, with a stronger effect compared to the single-pollutant model. Stratified analysis reveals that NO2 has a greater impact on COVID-19 mortality in southeastern cities compared to other regions, while the protective effect of O3 is generally observed, except in the northern region.

Although the pandemic has ended, there have been relatively few studies investigating the impact of air pollution on COVID-19. Most studies have focused on the relationship between pollution and incidence rather than mortality. Our findings on the short-term positive effects of NO2 on COVID-19 mortality in both Wuhan and 45 cities in China were consistent with previous research in China and evidence from other countries. Firstly, a study of 120 cities in China, using a generalized additive model, demonstrated a significant positive association between NO2 and COVID-19 incidence. Specifically, the daily case count of COVID-19 increased by 6.94% for every 10 μg/m3 rise in NO2 concentration at lag 0–14, consistent with our findings6. Another study conducted in Wuhan and Xiaogan, China, similarly found a significant association between COVID-19 incidence and NO2 concentration, aligning with our research results for Wuhan16. Studies from other countries such as Britain and Germany also indicated a positive association between NO2 concentration and both incidence and mortality of COVID-1912,15,17. A large population-based cohort in Spain found that short-term exposure to NO2 can significantly increase the risk of COVID-19 mortality18. However, another nationwide study in the Netherlands did not observe a significant effect of NO2 on the risk of COVID-1916.

A number of studies have documented the adverse health effects of NO2 on the respiratory system due to both short-and long-term exposure18. Children, who are particularly vulnerable, exhibit heightened sensitivity to respiratory infections triggered by NO2 exposure19,20,21. Moreover, studies have identified coronavirus as one of several viruses linked to NO2 induced respiratory infections21. Animal studies have further highlighted that NO2 exposure can compromise human immune responses, potentially increasing susceptibility to viral infections. Previously infected animals are more prone to reinfection upon subsequent NO2 exposure22. Future research efforts should perhaps focus on further elucidating the specific mechanisms through which NO2 increases the risk of COVID-19.

There is inconsistent evidence regarding the association between exposure to PM2.5 and COVID-19 mortality. Specifically, a positive association was observed only in Wuhan, whereas analysis across 45 cities did not reveal a consistent association between PM2.5 and COVID-19 mortality. Studies in other Chinese cities like Shanghai and Xiaogan have identified PM2.5 as a significant risk factor for transmission and morbidity in COVID-1923,24. Similarly, another research conducted in Wuhan has linked PM2.5 to increased COVID-19 incidence24. In London, England, a study confirmed that every 1 μg/m3 increase in PM2.5 concentration was associated with a 1.1% rise in COVID-19 cases and a 2.3% increase in mortality. However, nationwide studies in the U.S. and the United Kingdom (U.K.) aligned with our multi-city findings, showing uncertainty regarding the impact of PM2.5 on COVID-1925. Previous studies indicate that fine particles can serve as carriers for viruses, including influenza, facilitating their inhalation into the human body. Research in Taiwan, China suggests two potential mechanisms through which PM2.5 may promote COVID-19 transmission: First, PM2.5 can directly carry COVID-19 particles; Second, PM2.5 exposure increases expression of the ACE2 receptor in the lungs, enhancing COVID-19 adhesion. Additionally, PM2.5 may disrupt the respiratory barrier, exposing deeper lung tissues to pathogens19.

Our study found that O3 is a protective factor against COVID-19, which contrasts with the majority of previous research findings16. Other studies in China and other parts of the world have identified O3 as a risk factor for COVID-19, contributing to increased incidence and mortality of the disease26. For instance, a study in Italy found that O3 might facilitate the spread of COVID-196. Increased oxidative stress is widely recognized as a key mechanism by which pollutants exert toxicity27. O3 reacts directly with unsaturated fatty lipids in the respiratory tract, generating reactive oxygen species such as hydrogen peroxide, as well as lipid ozonation products like lipid peroxides and reactive aldehydes28. This oxidative stress can lead to mitochondrial dysfunction, DNA damage, and subsequent inflammatory responses29, potentially increasing susceptibility to COVID-19.

However, it is not entirely implausible that O3 could be protective against COVID-19 based on certain mechanisms. Research conducted in Biosafety Level 3 Laboratory (P3 Laboratory) has shown that ozone effectively inactivates the Severe Acute Respiratory Syndrome (SARS) virus, achieving a comprehensive inactivation rate of up to 99.22% in experiments conducted on green monkey kidney cells. Both the SARS virus and the novel coronavirus causing COVID-19 belong to the coronavirus family. Researchers have also found an 80% genomic sequence similarity between the novel coronavirus and the SARS coronavirus30. Therefore, there is reason to speculate that atmospheric O3 may play a role in reducing COVID-19 mortality rates. Additionally, the negative association between O3 and COVID-19 mortality observed in our study could also be attributed to scenarios where elevated outdoor ozone levels trigger warnings and advisories from meteorological stations, prompting citizens to reduce outdoor activities and thus decreasing opportunities for COVID-19 transmission.

We also observed a negative association between SO2 levels and COVID-19 mortality. Current research consistently indicates that SO2 acts as a protective factor in relation to COVID-1931. The research of 120 cities in China found that higher SO2 levels were associated with reduced COVID-19 incidence32. Similarly, the previously mentioned study in Wuhan, employing a Poisson regression model, confirmed a negative association between SO2 and COVID-19 mortality33. Research suggests that SO2 possesses anti-pneumonia and lung protection properties34. Its antimicrobial action involves penetrating viruses and bacteria, disrupting enzyme and protein activities35. Specifically, SO2 can deactivate the protein coat of viruses and impair internal enzymatic proteins, leading to structural damage and loss of function, thereby causing viral death36. Coronaviruses, which are enveloped in a lipid membrane, are particularly susceptible to SO2 due to their affinity, making them easier targets for SO2 attack37.

In stratified analysis, the impact of air pollution on COVID-19 is notably stronger in the southeastern region compared to the western and northern regions of the country. One possible explanation is that in northern areas, outdoor activities are generally reduced, especially during colder seasons38. COVID-19 patients are isolated in hospitals, thereby limiting their exposure to outdoor air pollutants, which minimizes the influence of outdoor air pollution on COVID-19 mortality rates39. Despite severe air pollution in northern regions, the pandemic began in central China (Wuhan) and spread nationwide. Consequently, as the pandemic spread from Wuhan to other regions, proactive containment measures were already in place, partially masking the impact of air pollution on the pandemic40.

One distinctive feature of our study is the utilization of a generalized additive model assuming a negative binomial distribution for the outcome variable. In contrast, many previous studies have employed linear regression or ecological regression models41. While some studies have also used generalized additive models, they typically assume a Poisson distribution rather than a negative binomial distribution42. The Poisson regression model is a fundamental approach for count data but assumes that the mean and variance are equal, which may not hold true in many real-world datasets where over-dispersion (variance exceeding the mean) is common43. The negative binomial model, as a generalization of the Poisson distribution, addresses this issue effectively44. It is widely applied in diverse fields such as health statistics and econometrics due to its ability to handle data with varying dispersion levels more robustly45. Therefore, our study opts for the negative binomial distribution, which allows for more effective statistical inference under conditions where data exhibit significant deviations from Poisson assumptions.

In addition to data analysis, the study has several strengths. Firstly, we chose the initial outbreak year as the study period to capture the early and natural development of the COVID-19 pandemic, when the influence of preventive policy and control interventions on the evaluation of the association between air pollution exposure and COVID-19 mortality was less pronounced compared to the mid to late stages of the epidemic. Secondly, every province in mainland China was represented by at least one city in the study, which reduces the risk of regional bias and provides a more accurate picture of the nationwide impact. Thirdly, the study examined the single and cumulative lag effects of six main air pollutants exposure from 0 days to 14 days. This approach captures both the immediate and delayed effects, offering a comprehensive understanding of how short-term exposure to air pollution influences COVID-19 mortality. Furthermore, the study stratified the cities into three regions, which adjusts for the region diversity in the environmental, social, health system, and economic conditions that may confound the association between air pollution levels and COVID-19 outcomes.

This study has several limitations. First, pathogen-host interactions may have influenced the results. To address this, we applied the GAM to control for meteorological factors, indirectly accounting for this confounding, as meteorological conditions can also affect pathogen-host dynamics. Second, differences in infection rates, healthcare resources, and public health interventions across cities could introduce heterogeneity. To mitigate this, we grouped cities into three categories based on environmental, social, healthcare, and economic factors, and compared differences across these groups. Third, lockdowns and industrial shutdowns may have affected air pollution levels and viral transmission, potentially introducing bias. However, by focusing on the early phase of the pandemic, before full control measures were implemented, we were able to better minimize this bias compared to later stages, when more structured interventions were in place. Fourth, the study relied on aggregated count data, lacking individual-level information such as age, gender, and residency. This limited our ability to examine the effects across specific demographic groups. Additionally, using city-level average air pollution concentrations as a proxy for individual exposure may have introduced misclassification bias, a common limitation of ecological studies. Fifth, we observed protective effects of certain pollutants on COVID-19 outcomes. While this is plausible based on existing literature, it may also reflect the study’s small sample size, highlighting the need for further validation of these findings. Finally, the generalizability of the study is limited to 45 cities in China, which may not fully represent the diversity of conditions in other regions. Future research should expand to different geographical areas, incorporate larger datasets, and collect individual-level data to provide a more comprehensive global understanding of these issues.

This research underscores the harmful effects of air pollution, particularly NO2 and PM2.5, on COVID-19 mortality risk during the early pandemic stages and highlights regional health disparities. Our findings emphasize the need to address air pollution as a key factor in the spread and severity of COVID-19. Although the global pandemic has subsided, its lasting impact on public health remains a concern. This study calls for ongoing health surveillance, particularly in areas with persistent air pollution, and provides valuable insights for future pandemic prevention and treatment strategies.

Methods

Study area

For this study, we chose 45 cities from across all provinces and municipalities in mainland China, ensuring a comprehensive analysis while considering the limitations in data availability. The selection was based on the pandemic severity in each region. Provinces with more than 10,000 COVID-19 cases included both the capital and all major cities (e.g., Hubei with Wuhan and 10 other cities). For provinces reporting between 1000 and 10,000 cases, we selected the capital and the city with the highest number of cases outside the capital (e.g., Guangdong, Zhejiang, Henan, and Hunan). In provinces with fewer than 1000 cases, only the capital city was included. This approach was designed to ensure that the selected cities provided a diverse representation of pandemic severity across different regions of China from January 23, 2020, to May 23, 2020, and to create a robust dataset for investigating the link between air pollution and COVID-19 mortality.

Data collection

Daily cumulative confirmed cases and daily new deaths of each city from January 23, 2020 to May 23, 2020, were obtained from the official websites of each city’s Centers for Disease Control and Prevention. The study concluded on May 23, 2020, marking the time when the last selected city achieved 0 new daily confirmed COVID-19 cases during the initial outbreak phase. The air pollutants included in this study are PM2.5, SO2, NO2 and O3. Regarding particulate matter, we exclusively analyzed PM2.5 and excluded PM10 for two main reasons: (1) Since PM10 essentially contains PM2.5, albeit with larger particle sizes, simultaneous inclusion of both in the analysis could lead to collinearity issues in the multi-pollutant model13; (2) Including PM2.5 rather than PM10 is more reasonable because it is widely acknowledged that particles with smaller diameters have a greater ability to penetrate deep into the respiratory tract46. COVID-19 is a lower respiratory tract disease, so PM2.5 with a smaller diameter is more likely to carry the virus into the lower respiratory tract of the human body and cause infection. We obtained daily average concentration data of four air pollutants in 45 cities from the China Air Quality Online Monitoring and Analysis Platform (https://www.aqistudy.cn). The two meteorological variables, daily average temperature and relative humidity, were obtained from the National Meteorological Information Center of China.

Descriptive analysis

The descriptive analysis in this study is divided into two parts. The first part utilizes statistical measures to describe the characteristics of air pollutants, meteorological factors, and the distribution of COVID-19 mortality. The second part employs correlational analysis to estimate the correlation between the four pollutants and two meteorological factors to examine collinearity among predictor variables. The two-pollutant model will not simultaneously include pairs with excessively high correlation.

Generalized additive model based on negative binomial distribution

GAM is particularly beneficial in environmental health studies, where non-linear and complex relationships are common in real-world datasets. It allows for smooth and flexible functions of predictors, such as splines, to capture these relationships. Unlike linear regression, where the effect of each predictor is constant, GAM enables the effect to vary smoothly across its range, making it adaptable to real-world data. GAM also avoids the limitations of polynomial models by not requiring predefining the degree of the relationship. Additionally, GAM can handle different error distributions, such as negative binomial for overdispersed count data.

GAM is also useful in controlling for meteorological variables, which are closely related to infectious disease pathogens and vector-borne diseases. By adjusting for these confounders, GAM enhances the accuracy of models in environmental health studies47. In time-series studies of air pollution and mortality, GAM is widely used as it can nonparametrically adjust for nonlinear confounding effects of seasonality, trends, and weather variables.

The negative binomial distribution is suitable for studies with count data as the independent variable47. Therefore, this study employs a generalized additive negative binomial distribution model to investigate the impact of air pollution on COVID-19 mortality. The research model for this study is as follows:

$$\log {\rm{E}}\left ({Y}_{{ti}}\right)=\beta {X}_{{ti}}+{s}\left({{\rm{temperature}}},{df}\right)+{s}\left({{\rm{relative}}\; {\rm{humidity}}},{df}\right)$$

In this model, \({{E}}\left ({{{Y}}}_{{{ti}}}\right)\) represents the expected number of deaths in city i on day t, and \({{{X}}}_{{{ti}}}\) denotes the pollutant concentration level in city i on day t, measured in μg/m3. Since the model employs a negative binomial distribution, the logarithmic function is chosen as the link function. β represents the coefficient of the exposure-response relationship. Given the nonlinear relationship between weather and health, a spline smoothing function is utilized to manage this nonlinear mixed effect. S () represents a smooth spline function, and df indicates its degree of freedom.

In this study, we used the Akaike Information Criterion (AIC) to evaluate the efficiency of the generalized additive model (GAM) with a negative binomial distribution. AIC balances model fit and complexity, with lower values indicating a more efficient model. AIC also serves as a model validation tool, ensuring that the GAM accurately represents the relationship between air pollution and COVID-19 mortality while avoiding overfitting.

To investigate whether the confounding effect of multiple pollutants influences the magnitude of the effect, this study established both single-pollutant and two-pollutant models. Given that the impact of air pollution on health outcomes may extend over several days, and the incubation period of COVID-19 can be up to approximately 14 days, this study examined the single-day lag effect (lag0–lag14) and cumulative multi-day lag effect (lag01–lag014) of air pollutants on health outcomes. Subgroup analyses were conducted to explore potential regional variations in their relationship. Following the approach of previously published literature48, the 45 Chinese cities included in the study were categorized into three regions based on geographical location: northern, southeastern, and western regions (Fig. 4).

Fig. 4: Geographic distribution of COVID-19 cases in China during the study period.
Fig. 4: Geographic distribution of COVID-19 cases in China during the study period.The alternative text for this image may have been generated using AI.
Full size image

Map of China showing cumulative COVID-19 cases in three regions and 45 selected cities during the study period.

The statistical analyses involved in this study were all conducted using R software (version 4.0.0).