Synergistic Effects of Ambient PM_2.5 and O₃ with Natural Temperature Variability on Non-Accidental and Cardiovascular Mortality: A Historical Time Series Analysis in Urban Taiyuan, China

Abstract

Climate change and air pollution are associated with a range of health outcomes, including cardiovascular and respiratory disease. Evaluation of the synergic effects of air pollution and increasing natural temperature on mortality is important for understanding their potential joint health effects. In this study, the modification effects of air temperature on the short-term association of ambient fine particulate matter (PM_2.5) and ozone (O₃) with non-accidental death (NAD) and cardiovascular disease (CVD) mortality were evaluated by using the generalized additive model (GAM) combined with the distributed lag nonlinear model (DLNM) in urban areas of Taiyuan, a representative of energy and heavy industrial cities in Northern China. The data on the daily cause-specific death numbers, air pollutants concentrations, and meteorological factors were collected from January 2013 to December 2019, and the temperature was divided into low (<25th percentile), medium (25–75th percentile), and high (>75th percentile) categories. Significant associations of PM_2.5 and O₃ with NAD and CVD mortality were observed in single-effect analysis. A statistically significant increase in the effect estimates of PM_2.5 and O₃ on NAD and CVD mortality was also observed on high-temperature days. But the associations of those were not statistically significant on medium- and low-temperature days. At the same temperature level, the effects of PM_2.5 and O₃ on the CVD mortality were larger than those on NAD (1.74% vs. 1.21%; 1.67% vs. 0.57%), and the elderly and males appeared to be more vulnerable to both higher temperatures and air pollution. The results suggest that the acute effect of PM_2.5 and O₃ on NAD and CVD mortality in urban Taiyuan was enhanced by increasing temperatures, particularly for the elderly and males. It highlights the importance of reducing PM_2.5 and O₃ exposure in urban areas to reduce the public health burden under the situation of global warming.

1. Introduction

Extensive epidemiological studies have reported that exposure to high levels of air pollutants such as ground-level ozone (O₃) and fine particulate matter (PM_2.5) is associated with human mortality [1,2,3]. Well-established evidence indicates that the ambient high temperatures could increase the risk of death from various diseases [4,5,6]. In the context of global climate change, an increase in the magnitude of mean temperatures and the frequency of hot days results in modification of the health effects caused by air pollution. Now, there has been increasing concern about the combined effects of air pollution and temperature on mortality [7,8]. To better understand the potential synergistic effects between air pollution and temperature on mortality, we have to identify the main pollutants leading to air pollution and how temperature impacts the generation of these air pollutants. In general, most cities suffer from air pollution caused by PM_2.5 and O₃ in China, leading to deterioration of ambient air quality, and temperature can affect air quality by influencing pollutant emission, transport, and chemical transformation [9,10]. In particular, the formation pathways of both PM_2.5 and O₃ exhibit distinct sensitivities to meteorological factors, particularly temperature. Elevated temperatures accelerate O₃ production by intensifying photochemical reactions of precursors like nitrogen oxides (NO_x) and volatile organic compounds (VOCs) [11]. Conversely, temperature influences PM_2.5 in seasonally divergent ways [12]. Ambient PM_2.5 comprises primary emissions and secondary inorganic/organic aerosols formed from gaseous precursors (e.g., SO₂, NO_x, VOCs), which are strongly temperature-dependent. The warmer conditions accelerate photochemistry, enhancing oxidant production and promoting the conversion of gases to particulates [13]. While in winter, lower temperatures lead to an increase in heating, which emits a larger number of primary particles [14]. Consequently, temperatures act as a critical environmental regulator driving the synergistic production of both PM₂.₅ and O₃. It is suggested that the interactions between natural temperature and air pollutants play important roles in affecting human health, particularly during pollution episodes in urban areas.

The effects of temperature modification on air pollution-associated mortality have been confirmed by numerous studies [7,8,15]. Most of these studies identified a significant interaction between air pollution and high temperatures in the cause-specific mortality, whereas other studies reported a non-significant modification effect [16,17], implying that the temperature-modification effects on the impact of air pollution on mortality are inconsistent. This phenomenon happened in the studies that investigated the modifying effect of different temperature levels on PM_2.5- and O₃-associated mortality [7,15]. Most studies found a significant effect modification between PM_2.5 or O₃ and high temperature on the cause-specific mortality [6,18], whereas a few detected a stronger synergistic effect of PM_2.5 or O₃ concentration and low-temperature level [10,19,20]. But some studies reported a negative association between cardiovascular mortality rate and O₃ level on warm days [21], and some reported no statistically significant effect modifications between PM_2.5 or O₃ and temperature [16,22]. Regarding that (a) most studies mainly examined the linear effect by applying a single-day or moving-average lag structure for temperature, the full delayed effect may not be captured, and the nonlinear association between temperature and mortality might be simplified [10]; (b) many studies were carried out in developed countries, including Australia, North America, and Europe [17,23,24,25], but only a few were conducted in developing countries such as China; (c) as the largest developing country, China is confronted with unprecedented joint challenges caused by elevated temperature and air pollution and it is necessary to conduct studies to explore the synergistic effects of air pollution and temperature on mortality. Committing to improving the current understanding of their joint health effects, we would apply a combination of the time-series generalized additive model (GAM) with the distributed lag nonlinear model (DLNM) to examine whether increasing temperature enhances the effects of PM_2.5 or O₃ on the non-accidental and cardiovascular mortality in typical cities of China.

In the study, more attention needs to be paid to the three highlights as follows. Firstly, since temperature is generally controlled as a confounder when a model for estimating air pollution health impacts is established [26], the potential modification effects of temperature on mortality related to air pollutants are often ignored. Herein, it is highlighted to accurately assess the effect modifications between air pollution and temperature on mortality. Furthermore, potential differences in risk among the vulnerable population under different temperature conditions were also examined by subgroup analysis. Secondly, though many air pollutants, including volatile organic compounds (VOCs) and polycyclic aromatic hydrocarbons (PAHs), can make adverse impacts on human health, this study specifically examines PM_2.5 and O₃ because (a) PM_2.5 and O₃ are criteria air pollutants with globally standardized monitoring networks, enabling robust exposure assessment across large populations [27]; (b) substantial epidemiological evidence confirm they have independent and temperature-modified mortality risks [28]. In many cases, VOCs and PAHs generally exert health effects indirectly through PM_2.5/O₃ formation pathways. VOCs act as primary precursors for O₃ through photochemical oxidation [29] and contribute to secondary organic aerosol (SOA) formation, which is a major PM_2.5 component [30]. Similarly, PAHs can partition onto fine particulate matter and undergo atmospheric oxidation to form quinones and other redox-active PM_2.5 constituents that drive oxidative stress [31]. Thus, PM_2.5 and O₃ represent integrative endpoints for complex pollutant mixtures, including VOCs/PAHs. Focusing on these endpoints allows us to evaluate dominant pathways through which multi-pollutant exposures interact with temperature to influence mortality. Thirdly, the urban Taiyuan is an appropriate place for exploring whether a strong interactive effect exists between air pollution and higher temperature on mortality. As the capital of Shanxi Province, Taiyuan is a typical city of heavy industry, where electricity production and industrial activities such as coking and steel-making rely heavily on large amounts of coal consumption, so the city suffers from regional persistent heavy air pollution and frequent haze events characterized by fine particles and ozone pollution. In addition, Taiyuan belongs to a warm temperate zone with a continental monsoon climate, where the population may be more susceptible to higher temperatures.

2. Materials and Methods

2.1. Data Collection

Air pollution, meteorological parameters, and daily mortality data from 1 January 2013 to 31 December 2019 were collected from six districts of urban Taiyuan (Figure 1). The daily air pollution level data, including mass concentrations of PM_2.5 and maximum 8 h average O₃, were obtained from the fixed-site stations of state air-quality monitoring networks. In each station, the daily 24 h mean concentrations of criteria air pollutants were collected to analyze the effects of co-pollutants, including sulfur dioxide (SO₂) and nitrogen dioxide (NO₂). The daily concentrations were averaged based on all valid monitoring stations, which had at least 75% of the daily available data for each pollutant. The missing monitoring data accounted for approximately 3% of each pollutant in the present study, which was supplied with a seasonal (spring: March–May, summer: June–August, autumn: September–November, and winter: December–February) average as interpolation [32,33]. All measurements were conducted according to China’s National Ambient Air Quality Standards (PRC Ministry of Ecology and Environment, GB3095-2012) [34]. The daily average concentrations of air pollutants were calculated by averaging all valid monitoring data from each urban area [35].

The meteorological data, including the daily mean temperature and relative humidity, were collected from China’s Meteorological Data Sharing Service System (http://data.cma.cn/, accessed on 5 February 2025.) without missing values. The seasons were further divided into the warm season from April to September and the cold season from October to March.

The daily death counts for the same time period of six urban areas were extracted from four large and two medium-sized hospitals in Taiyuan. The death registration system encompasses all fatalities from six urban districts. Besides demographic details like gender, date of death, age at death, and residential address, the death certificate also records the cause of death as reported by the physician. Rigorous quality assurance (QA) and quality control (QC) procedures have been implemented to maintain the accuracy of the death registration data [10]. According to the International Classification of Diseases, revision 10 (ICD-10), the causes of death were coded and classified into two categories for this study: non-accidental death (NAD, A00-R99) and cardiovascular disease death (CVD, I00-I99). Each cause-specific death was then stratified into two subgroups: age (<65 years and ≥65 years) and gender (male and female).

2.2. Statistical Analysis

2.2.1. Analysis of the Association Between Environmental Factors and Mortality

The generalized additive model (GAM) with over-dispersion was applied to investigate the short-term associations of the daily mortality with the atmospheric pollutants [36]. Several covariates were included in the basic models. Based on the existing literature [37,38] and Akaike’s Information Criterion for quasi-Poisson (Q-AIC) [39], a natural cubic spline with seven degrees of freedom (df) per year was used for time to eliminate the immeasurable long-term trends and seasonal effects. Furthermore, three df were chosen for the relative humidity to control their potential nonlinear confounding effects, and the day of the week (DOW) and public holidays (Holiday) were included as dummy variables in the basic model to reflect the short-term variation [40]. To further evaluate the model fit, diagnostic analyses were performed, including residual plots, partial autocorrelation function (PACF) plots, and quantile–quantile (Q-Q) plots of the standardized deviance residuals (Figure S1). Subsequently, within the framework of the final cause-specific optimized models, individual air pollutants were introduced separately into a single-pollutant model to examine their associations with daily mortality counts.

A cross-basis function was conducted to control the cumulative effect of temperature by applying the distributed lag nonlinear model (DLNM) [27]. A natural cubic spline with four df and two df for the temperature space and lag space were chosen according to the Q-AIC, and a maximum period of 14 days was applied for the temperature because of the longer lag periods for cold temperatures (up to several weeks) [41,42]. The calendar time and relative humidity were also controlled in this model using a natural cubic smoothing spline function with seven df per year and three df, respectively. The basic model is defined using the following function:

$$\begin{gathered} L\mathrm{og}\left[E\left(Y_t\right)\right]=\alpha +\beta X_{t,l}+{\beta }_{1}Cb.temp+n\mathrm{s}\left(Time,df=7\times year\right)+ns\left(RH,df=3\right) \\ +DOW+ Holiday \end{gathered}$$

(1)

where E(Y_t) represents the expected daily death counts on day t; α is the intercept; β and β₁ indicate the regression coefficients of air pollution concentration and temperature, respectively; X_t,l is the daily mean concentrations of air pollutants on day t, and l is the lag days; Cb.temp represents the cross-basis matrix of temperature; ns is the natural cubic smoothing spline function; RH is the relative humidity; DOW denotes the day of the week, and Holiday represents a public holiday. Based on a previous study [35], using Formula (1), we estimated the effects of pollutants with various lag structures, including a single-day lag (current day and the previous 1, 2, and 3 days [lag0, lag1, lag2, and lag3]) and the cumulative-day lag effect (two-, three-, and four-day moving average [lag01, lag02, lag03]). The two-day moving average (lag01) was chosen for air pollutants to represent their short-term exposure.

2.2.2. Evaluation of the Death Risk of Ambient Pollution Modified by Temperature

The GAM combined with a non-parametric bivariate response surface model was applied to investigate the joint effect between air pollutants and temperature on mortality by demonstrating the three-dimensional surfaces [26]. The model was constructed as shown below:

$$\begin{gathered} L\mathrm{og}\left[E\left(Y_t\right)\right]=\alpha +te(X_{t,l},temp)+n\mathrm{s}\left(Time,df=7\times year\right)+ns\left(RH,df=3\right) \\ +DOW+Holiday \end{gathered}$$

(2)

where te indicates the thin-plate spline, and the meaning of the rest of the parameters is the same as in the basic model.

GAM combined with a temperature-stratified parametric model was used to examine the potential interaction effects between air pollutants (PM_2.5 and O₃) and categorized temperature. A few studies have demonstrated that extremely low temperatures could enhance the effects of O₃ on cause-specific death [15]. However, there were inconsistent trends for PM_2.5, showing larger estimated values at high temperatures. Therefore, we examined heterogeneity in the effects of PM_2.5 and O₃ exposure across categories defined by daily mean temperature. The temperature strata were defined as low, medium, and high levels using the 25th and 75th percentiles of daily mean temperature as cut-offs according to previous studies [43,44,45]. The interaction effects model is described as follows:

$$\begin{gathered} L\mathrm{og}\left[E\left(Y_t\right)\right]=\alpha +\beta X_{t,l}+{\beta }_{1}Cb.temp+{\beta }_2(X_{t,l}:tem{p}_s)+n\mathrm{s}\left(Time,df=7\times year\right) \\ +ns(RH,df=3)+DOW+Holiday \end{gathered}$$

(3)

where temp_s represents the three temperature levels and β₂ indicates the regression coefficient of the interaction effect between pollutants and temperature categories. The other parameters are the same as in the basic model. The effect estimates of pollutants on low-temperature days were obtained from the main effects of pollutant (β), and the effect estimates of pollutants at medium and high temperatures were obtained from β + β₂.

The differences in the effect estimates between the pollutants and categorized temperature were tested by calculating both point estimates and 95% confidence interval (CI) using the following formula:

$$\left(\widehat{{\mathrm{Q}}_1}-\widehat{{\mathrm{Q}}_2}\right)\pm 1.96\sqrt{{\left(\widehat{{\mathrm{S}\mathrm{E}}_1}\right)}^2+{\left(\widehat{{\mathrm{S}\mathrm{E}}_2}\right)}^2}$$

(4)

where $\widehat{{\mathrm{Q}}_1}$ and $\widehat{{\mathrm{Q}}_2}$ represent the estimates of two categories, and $\widehat{{\mathrm{S}\mathrm{E}}_1}$ and $\widehat{{\mathrm{S}\mathrm{E}}_2}$ are their respective standard errors [46,47].

2.2.3. Sensitivity Analyses

Several sensitivity analyses were conducted to verify the robustness of the effect estimate results. First, the df for the time trend variable from 5 to 9 was changed, and different maximum lags for temperature (21 and 28 days) were adopted. Second, different temperature cut-off points (20th/80th, 15th/85th, 10th/90th, and 5th/95th percentiles of the daily mean temperature) were applied to substitute previous temperature strata in the main models. Thirdly, two-pollutant models were performed by adjusting for other pollutants (PM₁₀, SO₂, and NO₂) one at a time to assess the potential confounding effect of multiple exposures. Based on previous studies [15,48], the elderly individuals showed higher vulnerability to air pollution and temperature variations, while inconsistent effect estimates were found for gender and season groups among different regions. Therefore, some stratified analyses were also conducted in different ages (people aged ≥ 65 and <65 years old) and gender (male and female) groups, and season periods (warm and cold seasons) to explore the variation of the estimated results.

All analyses were performed using R software (version 3.3.2, R Development Core Team, 2016) with the “mgcv” and “dlnm” packages [49]. The statistical tests were two-sided, and results with p-values < 0.05 were considered significant. The effect estimates were expressed as the percentage change in daily death count associated with a 10 μg/m³ increase in air pollutant concentrations and 95% confidence intervals (CIs).

3. Results

3.1. Descriptive Statistics

During the entire study period, there was a total of 104,433 NADs, of which 48.9% were attributed to CVD. In terms of gender and age, a higher proportion of deaths were found in males and the elderly. The mean concentrations of daily PM_2.5, PM₁₀, SO₂, and NO₂ were (62.0 ± 43.8) μg/m³, (121.2 ± 64.8) μg/m³, (48.5 ± 52.7) μg/m³, and (42.5 ± 16.6) μg/m³, respectively, and the average 8 h maximum concentration of O₃ was (53.5 ± 35.7) μg/m³. The mean daily temperature and relative humidity were (11.4 ± 10.3) °C and (55.3 ± 18.2) %, respectively (

Table 1. Summary statistics of daily death counts, air pollutant concentrations, and weather conditions in Taiyuan from January 2013 to December 2019.

Variables		Mean ± SD	Min	P25	Median	P75	Max
Death number	Non-accidental	40.6 ± 8.8	16.0	34.0	40.0	46.0	76.0
	Male	23.9 ± 6.0	7.0	20.0	23.0	27.0	44.0
	Female	16.7 ± 2.7	4.0	14.0	17.0	19.0	32.0
	<65	10.5 ± 1.4	3.0	9.0	11.0	12.0	14.0
	≥65	30.1 ± 7.4	12.0	25.0	29.0	34.0	62.0
	Cold season	44.1 ± 9.0	26.0	38.0	44.0	50.0	76.0
	Warm season	38.0 ± 7.6	16.0	33.0	37.0	43.0	65.0
	Cardiovascular disease	19.8 ± 5.6	3.0	16.0	20.0	23.0	54.0
	Male	8.4 ± 3.5	1.0	7.0	8.0	11.0	24.0
	Female	6.1 ± 2.9	0	4.0	7.0	8.0	16.0
	<65	2.5 ± 1.9	0	2.0	3.0	4.0	8.0
	≥65	13.0 ± 4.6	2.0	9.0	12.0	15.0	37.0
	Cold season	22.0 ± 5.9	12.0	18.0	22.0	25.0	54.0
	Warm season	18.3 ± 4.8	5.0	15.0	18.0	21.0	36.0
Air pollutants (μg/m3)	PM2.5	62.0 ± 43.8	4.0	32.0	51.0	78.0	442.0
	PM10	121.2 ± 64.8	12.0	75.1	110.0	150.2	538.0
	SO2	48.5 ± 52.7	2.0	15.0	30.0	59.6	428.0
	NO2	42.5 ± 16.6	5.0	31.0	40.0	52.0	117.0
	O3 8 h-max	53.5 ± 35.7	3.0	26.0	46.0	74.0	226.0
Meteorological factors	Temperature (°C)	11.4 ± 10.3	−15.5	2.0	12.0	21.0	34.0
	Relative Humidity (%)	55.3 ± 18.2	11.0	41.0	56.0	69.0	98.0

). The cause-specific mortality and the concentrations of most pollutants showed similar seasonal trends, showing higher values in the cold season than in the warm season (Figure S2). By contrast, the ozone concentrations exhibited an opposite trend and were 2.3 times larger in the warm season than in the cold season.

Figure 2 shows the Spearman correlation between the meteorological factors and air pollution concentrations. Significant positive correlations existed among PM_2.5, PM₁₀, and SO₂ (p < 0.01), whereas they correlated significantly negatively with daily mean temperature and relative humidity (p < 0.01). On the other hand, positive correlations were observed for O₃ with temperature and relative humidity and NO₂ with relative humidity (p < 0.01). In addition, O₃ was negatively correlated with PM_2.5, PM₁₀, and SO₂, whereas NO₂ was positively correlated (p < 0.01).

3.2. The Interaction Between Air Pollution and Temperature on Mortality

The effects of PM_2.5 and O₃ on NAD and CVD mortality in the single pollutant model are listed in

Table 2. Percentage increase (95% CI) in the daily non-accidental death and cardiovascular mortality associated with a 10 μg/m³ increase in PM_2.5 and O₃ at different temperature levels ^a.

Pollutants	Variables	NAD		CVD
		Percentage Increase (95% CI)	p Value b	Percentage Increase (95% CI)	p Value b
PM2.5	Independent	1.05 (0.05, 2.06) *		1.29 (0.12, 2.47) *
	Low	−1.76 (−3.67, 0.19)	0.003	0.09 (−0.55, 0.74)	<0.001
	Medium	−0.42 (−1.72, 0.89)	0.014	0.35 (−0.54, 1.25)	0.002
	High	1.22 (0.27, 2.18) *	reference	1.75 (0.11, 3.42) *	reference
O3	Independent	0.57 (0.23, 0.92) **		1.03 (0.14, 1.93) *
	Low	−0.10 (−0.89, 0.70)	0.098	−0.52 (−3.63, 2.70)	0.178
	Medium	0.21 (−1.22, 1.65)	0.611	−0.18 (−2.32, 2.00)	0.093
	High	0.58 (0.02, 1.14) *	reference	1.68 (0.11, 3.28) *	reference

^a The 25th and 75th percentiles of daily mean temperature were used as temperature cut-offs. ^b Statistically significantly different from the high temperature level. * p < 0.05, ** p < 0.01.

, which shows significant associations among them. For a 10 μg/m³ increase in PM_2.5 concentrations (lag01) and the 8 h maximum O₃ (O₃ 8 h-max) concentrations (lag01), the estimated daily increases are 1.05% (95% CI: 0.05%, 2.06%) and 0.57% (95% CI: 0.23%, 0.92%) in NAD mortality, and 1.29% (95% CI: 0.12%, 2.47%) and 1.03% (95% CI: 0.14%, 1.93%) increases in the CVD mortality, respectively. Also, Table 1 lists the pooled effects of PM_2.5 and O₃ on the daily cause-specific mortality at different temperature levels. Stronger associations were observed between the air pollutants and daily mortality on high-temperature days. The results of the response surface model (Figure 3) show the potential modification effects of pollutants and daily average temperature on mortality, particularly at high temperatures.

For the 10 μg/m³ increase in PM_2.5, the estimated percentage increases were 0.09% (95% CI: −0.55%, 0.74%), 0.35% (95% CI: −0.54%, 1.25%), and 1.75% (95% CI: 0.11%, 3.42%) in cardiovascular mortality at low, medium, and high temperatures, respectively. The corresponding effect estimates for O₃ were −0.52% (95% CI: −3.63%, 2.70%), −0.18% (95% CI: −2.32%, 2.00%), and 1.68% (95% CI: 0.11%, 3.28%) in cardiovascular mortality at each temperature level.

3.3. Subgroup Analysis

Subgroup analysis revealed significant modification effects of higher temperatures for PM_2.5 exposure on the non-accidental and cardiovascular mortality in male populations and elderly people aged 65 years and older (Figure 4 and Figure 5). A similar pattern was found for O₃ exposure on non-accidental in both female and male populations and the elderly. The effects of O₃ on cardiovascular mortality were stronger for males and the elderly on high-temperature days. Furthermore, the pattern of the interactive effect estimates in the warm season was similar to that of the main model, showing that there were stronger associations between PM_2.5/O₃ and non-accidental mortality on high-temperature days compared to low- and medium-temperature days.

3.4. Sensitivity Analysis on the Model

In the sensitivity analysis, the effect estimates were consistent with the results through the main analysis when different dfs for the time trend or different maximum lag days and cut-offs for temperature categories were applied (Tables S1–S3). Furthermore, the same patterns of the effect modification on air pollution-related mortality by temperature were also found in the two-pollutant models (Tables S4 and S5). The effects of PM_2.5 and O₃ on the non-accidental death at the high-temperature levels were robust after adjusting for other pollutants except for PM₁₀. The effect modification of PM_2.5 on the cardiovascular mortality by temperature persisted when other pollutants were controlled. On the other hand, significant interactions were found between the temperature and O₃ on the cardiovascular mortality only after adjusting for SO₂.

4. Discussion

4.1. Modification Effects Made by Temperature on Air Pollution-Induced Mortality

A significant interactive effect between air pollution and high temperatures on the daily mortality was observed in urban Taiyuan. The results showed that high temperature could enhance the effects of PM_2.5 and O₃ on non-accidental and cardiovascular mortality. Elderly people and males were more sensitive to PM_2.5 and O₃ exposure on high-temperature days. Although these findings are consistent with previous studies conducted in Chinese cities and other regions [15,21,50], there are differences in the magnitudes of the effect estimates compared to others.

In the present analysis, a distributed lag nonlinear temperature term, rather than a linear, single lagged, or moving-average temperature term, was applied to capture the complicated nonlinear and lagged dependencies in both the exposure-response and lag-response associations. Comparatively, the models may accurately assess the impact of complex interactions between air pollutants and air temperature on daily mortality. However, several studies reported opposite results from those in the present study and found stronger air pollution effects on mortality on low-temperature days [51,52,53]. For example, a study conducted in Suzhou City, China, showed a synergistic effect of ozone exposure and low temperatures on mortality [26]. The inconsistency of estimates in different cities might be related to several factors, such as demographic characteristics, air pollution levels and source, geographical location, and weather conditions [52,54]. Among these, the following two factors likely contribute the majority to the discrepancy. On the one hand, there may be different exposure patterns among cities and climate regions. In North China, air suffers from severe PM_2.5 due to coal use and geography especially in winter or suffers from severe PM₁₀ due to dust storms in spring; while in South China, it has better air overall but often faces strong summer ozone and regional haze [8,26,51], resulting in exposure to different types of air pollutants. On the other hand, there are different response mechanisms for human beings to low and high ambient temperatures [55]. Generally, the heat effect was immediate (within a few days) when assessing temperature-associated mortality using different temperature lags, whereas the cold effect would show a long time lag (up to 3 or 4 weeks) [56].

In the sensitivity analysis, the modification patterns of temperature were consistent with those in the main analysis after changing the temperature lag days, which further confirmed the modification hypothesis regarding the enhanced mortality risk of air pollution under high temperatures. Consequently, supernumerary research on the potential interactive effects between air pollution and temperature on daily mortality should be conducted in multiple cities in the future to better explain the differentiation.

In subgroup analysis, higher risks of air pollution on high-temperature days were found for elderly individuals aged ≥ 65 years and males. These findings are in line with the results of many previous studies and show that elderly people are more vulnerable to air pollution and temperature [15,57,58]. The age heterogeneity may be because of the differences in physiological function, organism immunity ability, and the activity patterns between the elderly and younger [48]. Firstly, elderly individuals usually have a poor health status or preexisting chronic diseases, as well as decreased physiological processes, particularly thermoregulatory ability, causing high risks when they are exposed to high temperatures and air pollution [42,59]. Secondly, most elderly persons would spend more time outside on warmer days when the ozone concentrations are higher, making these individuals suffer from more cumulative exposure [15]. Higher vulnerability among males may be due to the divergence in occupational exposure, physiology, and thermoregulation between genders, while such gender-related effect modification varies among different regions and populations [60]. Generally, males tend to spend more time outdoors and are exposed to more adverse factors, including high temperatures and air pollution. Regarding the modification effects in different season periods, the effect estimates of PM_2.5 exposure on the non-accidental mortality were modified by temperature in the warm season and on high-temperature days, while there were opposite findings from several studies conducted in a single city in southern China [44,51,61]. Therefore, the various exposure patterns in the different cities and climate regions may be responsible for the inconsistency between the previous findings and the present study. According to the evidence on vulnerable populations, it is suggested that the government pay more attention to the subgroups with strong susceptibility and low adaptive capacity and take more preventive measures to mitigate the risk caused by air pollution and temperature.

In addition, compared to non-accidental death, it seems that cardiovascular death was particularly sensitive to both air pollution and high temperature. Some studies are consistent with the findings. Yang et al. reported higher effect estimates of heatwaves for ischemic heart diseases and stroke deaths than non-accidental deaths [62]. Gu et al. and Ma et al. also found that the effect estimates of high temperature were markedly higher for cardiovascular mortality than non-accidental deaths [63,64]. Higher mortality in cardiovascular diseases may be related to the failure of thermoregulation and physiological changes in the circulatory system [65]. Given that cardiovascular diseases were the leading cause of death in Taiyuan residents, our finding has important significance and should be taken seriously from a public health perspective.

Previous studies have primarily examined the interaction effects between temperature and mortality from PM_2.5, PM₁₀, and O₃, with limited focus on NO₂ and SO₂, with inconsistent findings. Our study found that high temperatures modulate the impact of PM₁₀ on cardiovascular mortality, particularly in the elderly and males. No interaction was observed between temperature and NO₂ or SO₂ (Table S8). PM₁₀ exposure can directly affect the respiratory tract, including the upper respiratory tract, bronchioles, and alveoli [66], potentially regulating the autonomic nervous system and influencing the cardiovascular system [28]. Sudden temperature changes can cause physiological and psychological stress, exacerbating existing conditions, especially in vulnerable populations like the elderly. Consequently, elevated PM₁₀ levels and significant temperature fluctuations may synergistically affect cardiovascular mortality. NO₂ and SO₂ can transform into water-soluble nitrates and sulfates through atmospheric reactions, forming inorganic salts that can be incorporated into particulate matter. This may explain the lack of evidence for temperature’s modifying effect on the mortality impacts of NO₂ and SO₂ compared to PM₁₀. Additionally, studies indicate a negative correlation between temperature and NO₂ and SO₂ concentrations under global warming conditions [67], as evidenced by the significant negative correlation between daily average temperature and SO₂ in this study (Figure 2).

4.2. Possible Mechanism of High Temperature Enhancing the Interactions of PM_2.5 or O₃ with Human Health

In the present study, the modification effects of PM_2.5 and O₃ exposure on mortality by temperature were examined in urban Taiyuan. Important evidence on the potential interactive effects between air pollution and temperature on non-accidental and cardiovascular mortality was provided. Although the complex underlying biological mechanism of the modification effect between high temperature and air pollutants on mortality is not completely understood, there may have been several contributions to the interactive effect, as shown in Figure 6 [66,68]. At first, there might exist the synergistic effect of air pollution and temperature on the cardiovascular system because there are common pathophysiological pathways. Under increasing air temperature, the blood viscosity and coagulability, cholesterol levels, and corresponding inflammatory responses could be increased. And the increase in PM_2.5 or O₃ concentrations could also be associated with elevated blood pressure and platelet aggregation, as well as systemic oxidative stress and inflammation [28]. Given that air pollutants can cause macrophage injury and autonomic imbalance as well as cell cycle dysregulation, the adverse effects of extremely high temperatures on the human body will be aggravated [69]. At higher temperatures, the composition and toxicity of air pollutants will accordingly change with the increase in total inhalation. In particular, the generation rate of ozone will be aggravated under higher temperatures, further increasing its health risks to individuals and exacerbating adverse health effects such as poor sleep quality, negative emotions, and even suicide [70,71]. At high temperatures, the thermoregulatory stress of individuals may increase [72], which then makes their physiological responses to air pollutants alter, leading to an increased susceptibility to the adverse effects of PM_2.5 or O₃. In addition, on high-temperature days, the individual exposed to outdoor air pollution might increase because these people would spend more time outdoors or keep the house windows open for a longer duration. In urban Taiyuan, the emission of a large number of air pollutants from coal consumption and industrial factories, and adverse diffusion conditions favor PM_2.5 and O₃ formation. And higher temperatures may further accelerate chemical reactions, resulting in higher concentrations of ambient PM_2.5 and O₃, ultimately leading to adverse health effects. The present study confirms the synergistic effects of high temperatures and PM_2.5 or O₃ on NAD and CVD mortality.

4.3. Limitations

This study had several limitations. First, as in most time-series studies, the ambient air pollution data were substituted from fixed monitoring stations for the individual exposure level, which may cause potential exposure measurement errors and lead to an underestimation of the effects. Second, the present study only examined the modification effect between the daily average temperature and ozone rather than fully considering the different temperature indicators. A previous study found that the overall effect of ozone could be weakened or even disappear when the average temperature was replaced with the maximum temperature [73]. Third, this study was conducted in a single city with typical air pollution and meteorological levels, and the findings may be difficult to generalize to other cities. In addition, we could not collect data on other confounding factors, such as air conditioner use and physical activity patterns. Therefore, caution should be taken when generalizing the results to other regions for health risk assessments.

5. Conclusions

The modification effect of temperature on the short-term association of ambient fine particulate matter (PM_2.5) and ozone (O₃) with non-accidental death (NAD) and cardiovascular diseases (CVD) mortality in Taiyuan city, China, was evaluated using a generalized additive model (GAM) combined with a penalized distributed lag nonlinear model (DLNM). Stratified analyses were further conducted in different age and gender groups. For the individual effect analysis, there were significant effect estimates for PM_2.5 and O₃ exposure on the NAD and CVD mortality. When the modification effect between air pollution and temperature was evaluated, the effect estimates of PM_2.5 and O₃ on NAD and CVD mortality on high-temperature days (>75th percentile) were increased significantly. Furthermore, this modification effect remained consistent and was more pronounced in the elderly and males. But the effect estimates for PM_2.5 and O₃ with NAD and CVD mortality in both medium (25–75th percentile) and low-temperature days (<25th percentile) were not statistically significant.

In short, the adverse effects of air pollution on the daily mortality in Taiyuan were enhanced by the higher temperature, particularly for the elderly and males, indicating that under the conditions of high temperature and high concentration of pollutants, the subgroups with strong susceptibility should be paid more attention to mitigate the health risk.