The association between ozone exposure and blood pressure in a general Chinese middle-aged and older population: a large-scale repeated-measurement study

Abstract

Background

The relationship between ozone (O₃) exposure and blood pressure (BP) remains inconclusive. Given the scarcity of Chinese epidemiological data, more research on this association is of paramount importance, particularly among middle-aged and older Chinese populations.

Methods

This study involved 10,875 participants (median age: 60.0 years) in Xiamen, China, from 2013 to 2019, with 34,939 repeated BP measurements. Air pollutant exposure data, including O₃, particulate matter, nitrogen dioxide, sulfur dioxide, and carbon monoxide were derived from China High Air Pollutants and High-resolution Air Quality Reanalysis datasets using a k-nearest neighbor algorithm. The relationship between mixed air pollutant exposure and BP was evaluated using Bayesian kernel machine regression model. The effects of daily-specific O₃ exposure on BP were assessed by distributed lag models integrated into a linear mixed-effects framework. The mediating role of total cholesterol (TC), serum total bilirubin (STB), triglyceride (TG), and low-density lipoprotein (LDL) were examined using multilevel mediation analysis with a fully adjusted model.

Results

Mixed air pollutant exposure was positively correlated with BP, with O₃ being a predominant contributor exhibiting an inverse effect. O₃ exposure had immediate effects on pulse pressure (PP), while systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean arterial pressure (MAP) showed delayed responses, with 3-, 14-, and 8-day lags, respectively. During the study period of up to 30 days, each 10 μg/m³ increase in maximum daily 8-h average O₃ concentration was associated with reductions in SBP (β =  − 1.176 mm Hg), DBP (− 0.237 mm Hg), PP (β =  − 0.973 mm Hg), and MAP (β =  − 0.544 mm Hg). Stronger correlations were observed in the older participants (aged ≥ 65 years), overweight/obese individuals, smokers and alcohol consumers, and those with hypertension or type 2 diabetes mellitus. STB and LDL mediated these effects, while TC and TG played mitigating roles.

Conclusions

Short-term O₃ exposure is negatively associated with BP in middle-aged and older Chinese individuals. The findings provide preliminary evidence for the impact of O₃ exposure on BP regulation and underscore the urgent need to reassess public health policies in response to O₃ pollution.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-024-03783-4.

Background

Ozone (O₃) pollution has emerged as a critical environmental and public health concern. According to the State of Global Air Report, O₃ exposure contributed to 178 thousand premature deaths and 6.21 million disability-adjusted life-years worldwide in 2019 [1]. In China, O₃ pollution continues to be worsening, with a 10.4% increase in the annual average of the 90th percentile of the maximum daily 8-h average (MDA8) O₃ concentration across 337 cities since 2015 [2]. Epidemiological studies have linked short-term O₃ exposure to increased all-cause [3], respiratory [4], and cardiovascular mortality [5]. However, the health impact of short-term O₃ exposure significantly varies across populations with different characteristics [6].

China, with the world’s largest population and a rapidly aging demographic, faces significant challenges in managing blood pressure (BP) [7]. The aging population, with inherently compromised immune function, is particularly vulnerable to ambient air pollutants [8]. Despite these concerns, research on the BP impact of short-term O₃ exposure among middle-aged and older Chinese adults (≥ 45 years) remains limited. Previous studies often focused on individuals with chronic diseases, yielding inconsistent findings. For example, a U.S. panel study of older diabetic patients reported a decrease in systolic BP (SBP) following a 5-day O₃ exposure [9], while research on older Chinese individuals with chronic obstructive pulmonary disease showed a positive association between 7-day O₃ exposure and SBP [10]. These discrepancies suggest that chronic diseases may alter the relationship between O₃ exposure and BP. Furthermore, the variation in population characteristics complicates the generalizability of these results to the broader Chinese population.

The inconsistencies in previous studies may derive from small sample sizes [9–11] and single-measurement designs [12–15], which can introduce statistical bias and measurement errors, especially considering BP’s high intra-individual variability. Additionally, the multicollinearity of air pollutants and their delayed effects challenge traditional statistical models (e.g., generalized linear model, generalized additive model, and linear mixed-effects model) in capturing the temporal dynamics of health impacts post-exposure [16–18]. The scarcity of large-scale, repeated-measurement data underscores the need for robust evidence to elucidate the effects of short-term O₃ exposure on BP among the general middle-aged and older populations in China.

In this study, we aim to investigate the impact of air pollutants on BP indices, with a particular focus on O₃, using Bayesian kernel machine regression (BKMR) models. Additionally, we evaluate the daily-specific O₃ exposure effects on BP indices using a distributed lag model (DLM) integrated with a linear mixed model (LMM) over 7- and 30-day windows, capturing both cumulative and lag effects [19, 20]. Furthermore, we aim to discern the mediating role of clinical indicators in these associations. To our knowledge, this is the first large-scale, repeated-measurement investigation of the association between daily-specific O₃ exposure and BP within a general middle-aged and older population in China.

Methods

Study population

The participants for the study were selected from the Guankou Ageing Cohort Study (GACS), an ongoing prospective cohort situated in Xiamen, Fujian Province, China (Additional file 1: Fig. S1). Detailed procedures and protocols for GACS have been described in our earlier report [21]. Briefly, the study recruited individuals aged 45 years and above between 2013 and 2019. Baseline and annual follow-up visits were conducted to gather information including residential addresses, demographic characteristics, lifestyle factors, anti-hypertensive medication usage, and comorbidities through a structured questionnaire. Physical examination data, including anthropometric measurements and laboratory tests, were collected via an electronic medical record system. The study protocol was approved by the Medical Ethics Committee of the School of Medicine, Xiamen University (Approval Number: XDYX202208H04), and all participants provided written informed consent upon recruitment.

Assessment of BP and hypertension

BP was measured by trained nurses with an automated digital sphygmomanometer (Omron HEM-907, Japan) following standard procedures. Before measurements, participants were required to rest quietly for at least 10 min. Three consecutive BP readings were obtained from the left arm, with a 5-min interval between each measurement. In cases where the BP difference between any two of the three readings exceeded a 10-mm Hg range, a reexamination was scheduled for 3 days later. To reduce the impact of measurement variability, the mean of the three readings was used as the representative BP value. Pulse pressure (PP) and mean arterial pressure (MAP) were calculated using Eq. 1 and Eq. 2, respectively. The diagnosis of hypertension was made if SBP was ≥ 140 mm Hg and/or diastolic BP (DBP) was ≥ 90 mm Hg as documented in the electronic medical record system and/or if there was a recorded history of antihypertensive medication use in the questionnaire [22].

$$PP=SBP-DBP$$ 1

$$MAP=(SBP+2\ast DBP)/3$$ 2

Exposure assessment

The residential exposure data of ground-level MDA8 O₃, particulate matter with an aerodynamic diameter < 2.5 μm (PM_2.5), and particulate matter with an aerodynamic diameter < 10 μm (PM₁₀) (spatial resolution 1 km × 1 km) were obtained from China High Air Pollutants (CHAP) dataset [23, 24]. This dataset is generated through the integration of extensive data sources, including ground-based measurements, satellite remote sensing products, atmospheric reanalysis, and model simulations, which yields a high-quality dataset with a cross-validation coefficient of determination (CV-R²) of 0.87, a root-mean-square error (RMSE) of 17.10 μg/m³, and a mean absolute error (MAE) of 11.29 μg/m³ on a daily basis. The exposure data of sulfur dioxide (SO₂), carbon monoxide (CO), and nitrogen dioxide (NO₂) (spatial resolution 15 km × 15 km) were obtained from High-resolution Air Quality Reanalysis Dataset over China (CAQRA) [25], which is the first high-resolution air quality reanalysis dataset in China, demonstrating a good agreement with independent observations (R² = 0.74–0.86 and RMSE = 16.8–33.6) across various cities. Participants’ residential addresses were converted into latitude and longitude coordinates using the Baidu Map API (https://lbsyun.baidu.com/) and subsequently integrated into the CHAP and CAQRA datasets. To ensure the accuracy of exposure estimates, a k-nearest neighbor algorithm (k = 5) was applied by identifying the 5 nearest grid points to their geographic coordinates and computing a weighted average of the pollutant’s concentration at these points [26]. In addition, meteorological data including temperature (°C), relative humidity (%), air pressure (Pa), and wind speed (m/s) were directly obtained from the National Climatic Data Center (https://data.cma.cn/).

Covariates

Potential covariates were selected according to prior knowledge and literature review [27, 28]. Participants’ characteristics, including age (continuous variable), sex (male vs. female), marital status (unmarried vs. married, or divorced/widowed), body mass index (BMI), physical activity frequency (low vs. medium, or high), smoking status (current smoker vs. former smoker, or non-smoker), and alcohol consumption (yes vs. no), status of type 2 diabetes mellitus (T2DM, yes vs. no), and the data on antihypertensive treatment (yes vs. no), were obtained through face-to-face interviewing. Among these, smoking status was assessed by asking the question “What is your current smoking status?” The possible answers were “smoker,” “former smoker,” or “non-smoker.” Alcohol consumption status was evaluated with the question “Have you ever consumed alcohol?” Participants could answer “yes” or “no.” Given that physical activity can impact both the exposure and absorption of air pollutant, physical activity frequency was also considered as a potential covariate. It was calculated by multiplying the number of days per week spent on outdoor activities by the number of hours dedicated to these activities each day. The resulting values were then stratified into low, moderate, and high activity levels according to tertiles. Besides, average temperature, relative humidity (both as continuous variables), the season, and the day of week (DOW) were included for model adjustment.

Statistical analysis

Prior to analysis, missing values were imputed using the multiple imputation by chained equations (MICE) method [29]. To elucidate the joint effects of mixed air pollutant exposure and identified the specific pollutants contributing to BP changes, we initially conducted BKMR analysis with 50,000 interactions via a Markov chain Monte Carlo (MCMC) algorithm [30]. A hierarchical clustering method was utilized to discern the structure within the air pollution data before BKMR analysis, arranging the pollutants using a distance matrix derived from Pearson’s and Spearman’s correlation coefficients, which indicated the similarity between each pair of air pollutants. Considering that PM₁₀ and PM_2.5 may exert distinct physiological effects due to their differences in size and composition, we divided them into five groups: O₃ as group 1, PM_2.5 as group 2, PM₁₀ as group 3, CO and NO₂ as group 4, and SO₂ as group 5. We computed group posterior inclusion probabilities (groupPIP) to assess the importance of each pollutant group, and calculated conditional posterior inclusion probabilities (condPIP) to evaluate the relative contribution of each pollutant on BP indices. We considered a PIP cutoff of 0.5 as indicative of variable significance [31]. To mitigate the computational demand of the BKMR, we used a subset of the data, randomly selecting 10% of the original dataset, adhering to the approach described in previous study [32, 33].

To further dissect the day-specific effects of O₃ on BP, we employed DLM within LMM, using participant’s ID as random intercept. The lag period was set to 0 days, 1 day, 2 days, ⋯, and 29 days, respectively. Due to traditional DLM assumes a linear dose–response curve [16], we assessed the linearity through restricted cubic splines (RCS). Among various configurations, the RCS model with 3 degrees of freedom exhibited the lowest Akaike information criterion (AIC) and was thus chosen as the best-fitting model for our data (Additional file 2: Table S1). The DLM framework utilizes a cross-basis function for simultaneous estimation of the exposure–response and the lag-response relationships. Considering the model interpretability, data applicability on air pollutants and meteorological factors, and previous studies [34], natural splines with df of 3, 2, and 3 were applied to model the lag-response function for O₃, temperature, and relative humidity, respectively (Additional file 2: Table S2). The regression model was constructed using Eq. 3.

$$Y_{\mathit{ij}}={\beta }_0+{\beta }_{1}Ozone+{\beta }_2{X}_{1ij}+\cdots +{\beta }_n{X}_{\mathit{nij}}+{\xi }_j+{\text{e}}_{\mathit{ij}}$$ 3

where Y_ij is the BP measurement, β₀ is the overall intercept, and β₁ is the regression coefficient for individual O₃ exposure concentration. Ozone is a two-dimensional matrix of MDA8 O₃ concentration and lag time constructed by using DLM, β₂ … β_n are the regression coefficients of the covariates, and ξ_j is the random effect for the participant. j and i represent the study participant and the lag day, respectively, while e_ij indicates the residual error term.

The crude model included residential O₃ exposure concentration and the average temperature and relative humidity during the corresponding exposure period. Model 1 was adjusted for age, sex, BMI, and marital status based on the crude model. Model 2 was additionally adjusted for smoking status, alcohol consumption, exercise frequency, the use of antihypertensive treatment, and the status of T2DM based on model 1. Model 3, the fully adjusted model, was further adjusted for the season and the DOW based on Model 2.

Stratified analyses were conducted by the factors such as age, sex, BMI, smoking status, drinking habits, frequency of outdoor activities, and the diagnosis of T2DM and hypertension. In all analyses, effect coefficients of O₃ (β) with 95% confidence intervals (CI) were reported for every 10 μg/m³ increase in exposure concentration.

Additionally, we hypothesized that clinical indices, including hemoglobin (Hb), white blood cell (WBC), platelet (PLT), serum total bilirubin (STB), conjugated bilirubin (CB), blood urea nitrogen (BUN), uric acid (UA), low-density lipoprotein (LDL), high-density lipoprotein (HDL), total cholesterol (TC), and triglyceride (TG), might mediate the association between O₃ exposure and BP. To test this hypothesis, we performed multilevel mediation analyses employing fully adjusted models over 7- and 30-day windows [35]. The analysis encompassed the direct effect of exposure on the outcome (c’), the effect on the mediator (a), the mediator’s effect on the outcome (b), and the proportion of mediation, which were estimated using nonparametric bootstrapping with 1000 samples.

All analyses were performed using R software version 4.0.5 (R Foundation or Statistical Computing, Vienna, Austria) and involved packages such as “tidyverse,” “reshape2,” “dlnm,” “lmerTest,” “splines,” “ggplot2,” “pheatmap,” “corrplot,” and “bkmr.” Statistically significance was set at a two-tailed probability threshold of P < 0.05.

Sensitive analyses

To ensure the robustness of our findings, we conducted several sensitivity analyses. First, recognizing the influence of temperature, air pressure, and wind speed [36] on the generation, transport, and dispersion of O₃, we incorporated meteorological factors, including the maximum temperature, the maximum air pressure, and the average wind speed into models and re-examined the association between O₃ exposure and BP. Second, given the documented impact of other gaseous pollutants (SO₂, NO₂, and CO) and particulate matter (PM_2.5 and PM₁₀) on BP, we included them into the two-pollutant model to re-analyze the associations. Third, some previous studies also reported the effect of the day of year [37], we further adjusted for this covariate and assess its impact on the robustness of the associations. Fourth, considering that imputation of missing values might affect the associations, participants with complete data (n = 10,837) were incorporated and re-analyzed. Fifth, we also performed the regression analyses annually from 2013 to 2019 based on cross-sectional data to test the stability on the associations of O₃ and BP.

Results

Characteristics of participants

In this study, 14,937 volunteers participated in the standardized medical examination and questionnaire survey. As shown in Fig. 1, a total of 4062 participants were excluded from the study according to the pre-set criteria. Consequently, 10,875 participants with at least two medical records were eligible for the analyses, contributing a total of 34,939 visits. Each participant had an average of 3.21 observations, with the maximum number of measurements being 7. Missing values of Hb (n = 9), STB (n = 6), LDL (n = 11), HDL (n = 8), TC (n = 13), and TG (n = 17) were imputed using MICE method. Given the skewed distribution of continuous variables in this study, these variables were presented as medians (25th percentile; 75th percentile) (Table 1). The median age of study participants was 60.0 years (25th percentile: 57.0; 75th percentile: 67.0). Females constituted 58.34% of the participants. Among the participants, 5590 individuals (51.40%) had been diagnosed with hypertension. Notably, the median age of those with hypertension was 61.0 years (25th percentile: 57.0; 75th percentile: 67.0), while the median age of participants without hypertension was 60.0 years (25th percentile: 56.0; 75th percentile: 67.0).

Fig. 1.

Flow diagram of the study

Table 1.

Baseline characteristics of participants in the GACS

	Overall N = 10,875	Participants without hypertension n = 5285 (48.60%)	Participants with hypertension n = 5590 (51.40%)	P
Sex				0.014
Male	4531 (41.66%)	2139 (40.47%)	2392 (42.79%)
Female	6344 (58.34%)	3146 (59.53%)	3198 (57.21%)
Age (years)	60.0 (57.0, 67.0)	60.0 (56.0, 67.0)	61.0 (57.0, 67.0)	0.012
Marital status (n, %)				0.568
Unmarried	61 (0.56%)	29 (0.55%)	33 (0.59%)
Married	9121 (83.87%)	4452 (84.24%)	4669 (83.52%)
Divorced/widowed	1692 (15.56%)	804 (15.21%)	888 (15.89%)
Weight (kg)	59.0 (52.0, 65.5)	57.0 (51.5, 63.8)	60.00 (53.0, 68.0)	< 0.001
Body temperature (°C)	36.4 (36.3, 36.5)	36.4 (36.3, 36.5)	36.4 (36.3, 36.5)	< 0.001
Pulse	78 (73, 83)	78 (73, 81)	78 (73, 85)	< 0.001
Height (cm)	158.5 (153.0, 165.0)	158.5 (154.0, 164.0)	158.5 (153.0, 165.0)	0.134
BMI (kg/m2)	23.14 (21.30, 25.46)	22.43 (20.91, 24.56)	23.89 (21.78, 26.14)	< 0.001
WC (cm)	82.0 (77.0, 89.0)	80.0 (76.0, 86.0)	84.0 (78.0, 90.0)	< 0.001
SBP (mmHg)	136.0 (124.0, 151.0)	125.0 (118.0, 133.0)	150.0 (139.0, 161.0)	< 0.001
DBP (mmHg)	81.0 (74.0, 89.0)	76.0 (71.0, 82.0)	87.0 (80.0, 94.0)	< 0.001
PP (mmHg)	54.0 (46.0, 65.0)	49.0 (42.0, 55.0)	62.0 (52.0, 73.0)	< 0.001
MAP (mmHg)	100.0 (91.7, 109.0)	92.7 (87.0, 98.0)	108.0 (101.0, 115.3)	< 0.001
WBC (× 109/L)	6.01 (5.40, 6.94)	5.80 (5.30, 6.72)	6.20 (5.50, 7.15)	< 0.001
PLT (× 109/L)	220.00 (197.00, 245.00)	216.00 (197.00, 242.00)	224.00 (197.00, 249.00)	< 0.001
Hb (g/L)	136.00 (128.00, 144.00)	134.00 (127.00, 142.00)	138.00 (130.00, 146.00)	< 0.001
FPG (mmol/L)	5.10 (4.62, 5.80)	4.99 (4.60, 5.61)	5.26 (4.70, 5.96)	< 0.001
ALT (U/L)	22.70 (17.30, 31.20)	22.20 (16.90, 31.20)	23.00 (17.70, 31.20)	< 0.001
AST (U/L)	21.90 (18.80, 26.30)	21.90 (18.70, 27.00)	21.80 (18.90, 26.00)	0.225
ALB (g/L)	45.00 (43.20, 46.50)	44.60 (42.60, 46.10)	45.30 (43.80, 46.90)	< 0.001
STB (μmol/L)	11.10 (8.40, 14.10)	10.70 (8.10, 13.80)	11.40 (8.70, 14.50)	< 0.001
CB (μmol/L)	3.10 (2.20, 4.60)	3.20 (2.20, 4.90)	3.00 (2.10, 4.30)	< 0.001
SCr (μmol/L)	66.40 (54.70, 80.40)	67.00 (55.60, 80.10)	65.90 (53.90, 80.50)	0.446
BUN (μmol/L)	5.10 (4.30, 5.90)	5.20 (4.40, 5.90)	5.10 (4.20, 5.90)	< 0.001
TC (mmol/L)	5.08 (4.33, 5.77)	4.96 (4.16, 5.65)	5.18 (4.50, 5.87)	< 0.001
TG (mmol/L)	1.19 (0.82, 1.71)	1.05 (0.79, 1.52)	1.32 (0.92, 1.89)	< 0.001
LDL (mmol/L)	3.22 (2.72, 3.74)	3.16 (2.63, 3.65)	3.29 (2.82, 3.84)	< 0.001
HDL (mmol/L)	1.47 (1.34, 1.60)	1.46 (1.35, 1.58)	1.48 (1.34, 1.61)	0.002
UA (μmol/L)	363.40 (311.90, 424.90)	352.60 (302.10, 407.92)	376.70 (321.40, 440.70)	< 0.001
T2DM (n, %)				< 0.001
No	9524 (87.58%)	4752 (89.91%)	4772 (85.37%)
Yes	1351 (12.42%)	533 (10.09%)	818 (14.63%)
Exercise frequency (n, %)				< 0.001
Low	6440 (59.22%)	3265 (61.78%)	3175 (56.80%)
Medium	1543 (14.19%)	843 (15.95%)	700 (12.52%)
High	2892 (26.59%)	1177 (22.27%)	1715 (30.68%)
Smoking status (n, %)				< 0.001
Non-smoker	8542 (78.55%)	4238 (80.19%)	4304 (76.99%)
Former smoker	333 (3.06%)	123 (2.33%)	210 (3.76%)
Current smoker	2000 (18.39%)	924 (17.48%)	1076 (19.25%)
Drinking status (n, %)				< 0.001
Non drinker	9334 (85.83%)	4673 (88.42%)	4661 (83.38%)
Former drinker	870 (8.00%)	402 (7.61%)	468 (8.37%)
Current drinker	671 (6.17%)	210 (3.97%)	461 (8.25%)

Given the skewed distribution of continuous variables in this study, these variables were presented as medians (25th percentile; 75th percentile). Categorical variables were expressed as frequencies and percentages (n (%)). Continuous variables with skewed distributions were analyzed using the rank-sum test, while Pearson’s χ² tests were used to compare categorical values. The P values indicated whether the differences in the data between participants with and without hypertension were statistically significant

Abbreviations: WC waist circumference, BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, PP pulse pressure, MAP mean arterial pressure, WBC white blood cell (leukocyte), PLT platelet, Hb hemoglobin, FPG fasting plasm blood, ALT alanine aminotransferase, AST aspartate aminotransferase, ALB albumin, STB serum total bilirubin, CB conjugated bilirubin, UCB unconjugated bilirubin, SCr serum creatinine, BUN blood urea nitrogen, TC total cholesterol, TG triglyceride, LDL low-density lipoprotein, HDL high-density lipoprotein, UA uric acid, T2DM type 2 diabetes mellitus

In the total population, the median values of SBP, DBP, PP, and MAP were 150.0 mm Hg (25th percentile: 139.0; 75th percentile: 161.0), 87.0 mm Hg (25th percentile: 80.0; 75th percentile: 94.0), 54.0 mm Hg (25th percentile: 46.0; 75th percentile: 65.0), and 100.0 mm Hg (25th percentile: 91.7; 75th percentile: 109.0), respectively. All these BP indices were significantly higher in the hypertensive group compared to those without hypertension.

The effects of air pollutant mixture on BP indices

As shown in Additional file 1: Fig. S2, the air pollutant exposure of the participants exhibited decreasing trends from 2013 to 2019. The seasonal distribution patterns of PM₁₀, PM_2.5, SO₂, NO₂, and CO were similar, all following a “W” shape with higher levels in winter and lower levels in summer, while O₃ exhibited an opposite distribution pattern (Additional file 2: Table S3).

As shown in Additional file 1: Fig. S3, the number of extreme values for air pollutants were 2, 15, 11, 59, 267, and 31 for O₃, PM₁₀, PM_2.5, NO₂, SO₂, and CO, respectively. These extreme values were substituted with the lower bound (25th percentile − 1.5 × IQR) or the upper bound (75th percentile + 1.5 × IQR) if they were below or exceeded these thresholds [38]. Based on Pearson’s and Spearman’s correlation analyses, and given that PM₁₀ and PM_2.5 may exert distinct physiological effects due to differences in their size and composition, we divided the variables into 5 separate groups: O₃ as group 1, PM_2.5 as group 2, PM₁₀ as group 3, CO and NO₂ as group 4, and SO₂ as group 5 (Additional file 1: Fig. S4, Fig. S5A). When characterizing the overall effect of the air pollutant mixture on BP using the BKMR model, we observed increasing trends for all BP indices compared to their 50th percentile (Additional file 1: Fig. S5B). Specifically, BP indices showed increasing trends with the exposure concentrations of PM₁₀ and PM_2.5, while O₃ predominantly exerting negative effects (Additional file 1: Fig. S5E). The BP responses to other air pollutants were not significant. Combining with PIP values, O₃ and PM₁₀ were the primary contributors (Additional file 1: Fig. S5C ~ D), exerting negative and positive effects on BP, respectively.

Association of O₃ exposure and BP

Additional file 1: Fig. S6 shows exposure–response curves of the association between O₃ exposure and BP indices. These curves exhibited a nearly linear trend, confirming that the essential condition for applying the DLM has been adequately met.

Table 2 reveals the correlations between O₃ exposure and all BP indices. After full adjustment in model 3, every 10 μg/m³ increase in MDA8 O₃ concentration was negatively associated with SBP [β =  − 0.276 mm Hg (95% CI: − 0.432, − 0.119 mm Hg)], DBP [− 0.012 mm Hg (95% CI: − 0.099, 0.076 mm Hg)], PP [β =  − 0.278 mm Hg (95% CI: − 0.410, − 0.146 mm Hg)], and MAP [β =  − 0.095 mm Hg (95% CI: − 0.192, 0.002 mm Hg)] over the 7-day study period. The effects were more pronounced over the 30-day study period with SBP [β =  − 1.176 mm Hg (95% CI: − 1.308, − 1.045 mm Hg)], DBP [− 0.237 mm Hg (95% CI: − 0.310, − 0.163 mm Hg)], PP [β =  − 0.973 mm Hg (95% CI: − 1.083, − 0.863 mm Hg)], and MAP [β =  − 0.544 mm Hg (95% CI: − 0.626, − 0.463 mm Hg)].

Table 2.

The estimated effects of an increase of 10 μg/m³ in O₃ exposure on BP

	SBP		DBP		PP		MAP
	β (95% CI)	P	β (95% CI)	P	β (95% CI)	P	β (95% CI)	P
7-day period
Crude model	− 0.254 (− 0.409, − 0.099)	0.001	− 0.068 (− 0.155, 0.018)	0.126	− 0.191 (−0.322, − 0.060)	0.004	− 0.129 (−0.225, − 0.033)	0.008
Model 1	− 0.317 (− 0.473, − 0.162)	< 0.001	− 0.020 (− 0.106, 0.067)	0.649	− 0.304 (− 0.434, − 0.174)	< 0.001	− 0.118 (− 0.214, − 0.022)	0.016
Model 2	− 0.384 (− 0.540, − 0.227)	< 0.001	− 0.053 (− 0.140, 0.034)	0.232	− 0.349 (− 0.480, − 0.218)	< 0.001	− 0.158 (− 0.254, − 0.061)	0.001
Model 3	− 0.276 (− 0.432, − 0.119)	< 0.001	− 0.012 (− 0.099, 0.076)	0.787	− 0.278 (− 0.410, − 0.146)	< 0.001	− 0.095 (− 0.192, 0.002)	0.055
30-day period
Crude model	− 0.557 (− 0.673, − 0.440)	< 0.001	− 0.148 (− 0.212, − 0.083)	< 0.001	− 0.417 (− 0.516, − 0.319)	< 0.001	− 0.288 (− 0.360, − 0.216)	< 0.001
Model 1	− 0.704 (− 0.823, − 0.585)	< 0.001	− 0.035 (− 0.101, 0.031)	0.299	− 0.674 (− 0.773, − 0.575)	< 0.001	− 0.263 (− 0.337, − 0.189)	< 0.001
Model 2	− 0.839 (− 0.960, − 0.717)	< 0.001	− 0.101 (− 0.169, − 0.034)	0.004	− 0.768 (− 0.870, − 0.666)	< 0.001	− 0.343 (− 0.418, − 0.267)	< 0.001
Model 3	− 1.176 (− 1.308, − 1.045)	< 0.001	− 0.237 (− 0.310, − 0.163)	< 0.001	− 0.973 (− 1.083, − 0.863)	< 0.001	− 0.544 (− 0.626, − 0.463)	< 0.001

Multivariable-adjusted DLM incorporated in LMM was used to assess the associations of MDA8 O₃ exposure concentration with BP for 7- and 30-day windows. Crude model: Included MDA8 O₃ concentration, average temperature, and relative humidity during the exposure period. Model 1: Included variables of age, sex, BMI, and marriage status based on the crude model. Model 2: Further included variables of smoking status, alcohol consumption, exercise frequency, antihypertensive treatment, and status of T2DM based on model 1. Model 3: Further included variables of the season and the DOW based on model 2

Abbreviations: SBP systolic blood pressure, DBP diastolic blood pressure, PP pulse pressure, MAP mean arterial pressure, CI confidence interval, DLM distributed lag model, LMM linear mixed model, CI confidence interval, BMI body mass index, T2DM type 2 diabetes mellitus, DOW the day of the week

Figure 2 illustrated the daily effect of O₃ exposure on BP. The negative correlations between MDA8 O₃ concentration and BP indices initially strengthened and then gradually weakened, with such trends being particularly pronounced at higher exposure doses.

Fig. 2.

Effects of daily-specific O₃ exposure on SBP, DBP, PP, and MAP. All models were adjusted for age, sex, BMI, marital status, smoking and drinking habits, activity frequency, antihypertensive medication, the status of T2DM, average temperature, relative humidity, season, and DOW. Abbreviations: O₃, ozone; SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure; MAP, mean arterial pressure; BMI, body mass index; T2DM, type 2 diabetes mellitus; DOW, the day of the week

Figure 3 depicts the cumulative effects, highlighting a progressive strengthening of the negative impacts of O₃ on BP indices over time. Notably, there was no apparent lag effect for PP, suggesting that the influence of O₃ exposure on this indicator was immediate. On the other hand, SBP, DBP, and MAP presented lag phases of 3-, 14-, and 8-day post O₃ exposure, as detailed in Additional file 2: Table S4.

Fig. 3.

Cumulative daily-specific effects of O₃ exposure to SBP, DBP, PP, and MAP. DLM incorporated in LMM was used to calculate multivariable-adjusted β (95% CI) per 10 μg/m³ increment in MDA8 O₃ exposure concentration. All models were adjusted for age, sex, BMI, marital status, smoking and drinking habits, activity frequency, antihypertensive medication, the status of T2DM, average temperature, relative humidity, season, and DOW. The solid blue dots represent the effect size estimates, and the grey areas represent the 95% CI. Abbreviations: O₃, ozone; SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure; MAP, mean arterial pressure; CI, confidence interval; DLM, distributed lag model; LMM, linear mixed model; BMI, body mass index; T2DM, type 2 diabetes mellitus; DOW, the day of the week

The results of stratified analyses, as presented in Table 3 and Additional file 2: Table S5, further confirmed that the heterogeneity in population characteristics did not substantially alter the association between O₃ exposure and BP. More pronounced negative associations between O₃ exposure and BP were observe in specific subgroups, including older individuals (≥ 65 years), male, overweight/obese participants, former or current smokers and drinkers, and patients with hypertension or T2DM for 7- and 30-day windows.

Table 3.

Stratified analyses of the association between O₃ exposure and BP for periods of up to 30 days

Subgroups	SBP		DBP		PP		MAP
	β (95% CI)	P	β (95% CI)	P	β (95% CI)	P	β (95% CI)	P
All participants	− 1.176 (− 1.308, − 1.045)	< 0.001	− 0.237 (− 0.310, − 0.163)	< 0.001	− 0.973 (− 1.083, − 0.863)	< 0.001	− 0.544 (− 0.626, − 0.463)	< 0.001
Age
< 65 years	− 0.977 (− 1.144, − 0.808)	< 0.001	− 0.123 (− 0.209, − 0.036)	0.005	− 0.644 (− 0.765, − 0.524)	< 0.001	− 0.333 (− 0.426, − 0.240)	< 0.001
≥ 65 years	− 1.276 (− 1.489, − 1.063)	< 0.001	− 0.266 (− 0.366, − 0.165)	< 0.001	− 0.673 (− 0.834, − 0.512)	< 0.001	− 0.479 (− 0.587, − 0.370)	< 0.001
Sex
Male	− 1.331 (− 1.526, − 1.136)	< 0.001	− 0.189 (− 0.305, − 0.073)	0.001	− 1.161 (− 1.322, − 1.000)	< 0.001	− 0.568 (− 0.694, − 0.442)	< 0.001
Female	− 1.170 (− 1.346, − 0.994)	< 0.001	− 0.329 (− 0.424, − 0.235)	< 0.001	− 0.877 (− 1.026, − 0.727)	< 0.001	− 0.602 (− 0.708, − 0.495)	< 0.001
BMI
< 24 kg/m2	− 1.213 (− 1.386, − 1.040)	< 0.001	− 0.238 (− 0.333, − 0.142)	< 0.001	− 1.002 (− 1.148, − 0.856)	< 0.001	− 0.557 (− 0.663, − 0.451)	< 0.001
≥ 24 kg/m2	− 1.327 (− 1.531, − 1.124)	< 0.001	− 0.322 (− 0.438, − 0.206)	< 0.001	− 1.039 (− 1.210, − 0.868)	< 0.001	− 0.653 (− 0.781, − 0.526)	< 0.001
Hypertension
No	− 0.561 (− 0.708, − 0.414)	< 0.001	− 0.006 (− 0.102, 0.088)	0.903	− 0.558 (− 0.692, − 0.423)	< 0.001	− 0.191 (− 0.287, − 0.095)	< 0.001
Yes	− 1.466 (− 1.635, − 1.297)	< 0.001	− 0.316 (− 0.412, − 0.220)	< 0.001	− 1.158 (− 1.304, − 1.012)	< 0.001	− 0.697 (− 0.802, − 0.593)	< 0.001
T2DM
No	− 1.295 (− 1.438, − 1.153)	< 0.001	− 0.263 (− 0.343, − 0.183)	< 0.001	− 1.059 (− 1.179, − 0.939)	< 0.001	− 0.603 (− 0.691, − 0.515)	< 0.001
Yes	− 1.416 (− 1.756, − 1.076)	< 0.001	− 0.577 (− 0.769, − 0.385)	< 0.001	− 0.857 (− 1.141, − 0.573)	< 0.001	− 0.858 (− 1.070, − 0.645)	< 0.001
Smoking
Non-smoker	− 1.066 (− 1.218, − 0.914)	< 0.001	− 0.253 (− 0.337, − 0.170)	< 0.001	− 0.844 (− 0.972, − 0.716)	< 0.001	− 0.517 (− 0.611, − 0.424)	< 0.001
Former smoker	− 2.096 (− 2.890, − 1.302)	< 0.001	− 0.755 (− 1.209, − 0.300)	0.001	− 1.368 (− 2.034, − 0.701)	< 0.001	− 1.201 (− 1.699, − 0.702)	< 0.001
Current smoker	− 1.715 (− 2.007, − 1.424)	< 0.001	− 0.310 (− 0.478, − 0.141)	< 0.001	− 1.427 (− 1.664, − 1.191)	< 0.001	− 0.778 (− 0.964, − 0.592)	< 0.001
Drinking
Non-drinker	− 1.131 (− 1.275, − 0.987)	< 0.001	− 0.258 (− 0.338, − 0.179)	< 0.001	− 0.905 (− 1.026, − 0.784)	< 0.001	− 0.543 (− 0.632, − 0.455)	< 0.001
Former drinker	− 1.405 (− 1.882, − 0.929)	< 0.001	− 0.198 (− 0.486, 0.088)	0.178	− 1.221 (− 1.604, − 0.838)	< 0.001	− 0.596 (− 0.910, − 0.282)	< 0.001
Current drinker	− 1.950 (− 2.472, − 1.428)	< 0.001	− 0.308 (− 0.601, − 0.015)	0.039	− 1.645 (− 2.065, − 1.224)	< 0.001	− 0.855 (− 1.184, − 0.526)	< 0.001
Exercise frequency
Low	− 1.335 (− 1.522, − 1.148)	< 0.001	− 0.319 (− 0.422, − 0.216)	< 0.001	− 1.042 (− 1.199, − 0.884)	< 0.001	− 0.653 (− 0.768, − 0.539)	< 0.001
Median	− 1.730 (− 2.353, − 1.107)	< 0.001	− 0.332 (− 0.678, 0.014)	0.006	− 1.397 (− 1.885, − 0.908)	< 0.001	− 0.792 (− 1.187, − 0.397)	< 0.001
High	− 1.271 (− 1.487, − 1.054)	< 0.001	− 0.236 (− 0.358, − 0.114)	< 0.001	− 1.047 (− 1.228, − 0.865)	< 0.001	− 0.580 (− 0.715, − 0.445)	< 0.001
FPG
≤ 6.1 mmol/L	− 1.286 (− 1.434, − 1.137)	< 0.001	− 0.234 (− 0.317, − 0.151)	< 0.001	− 1.075 (− 1.200, − 0.950)	< 0.001	− 0.582 (− 0.675, − 0.490)	< 0.001
> 6.1 mmol/L	− 1.352 (− 1.645, − 1.059)	< 0.001	− 0.505 (− 0.668, − 0.342)	< 0.001	− 0.861 (− 1.106, − 0.616)	< 0.001	− 0.787 (− 0.968, − 0.606)	< 0.001
WBC
< 3.5 × 109/L	− 1.727 (− 3.339, − 0.115)	0.036	− 0.327 (− 1.200, 0.544)	0.463	− 1.303 (− 2.576, − 0.030)	0.044	− 0.755 (− 1.756, 0.245)	0.139
3.5 ~ 9.5 × 109/L	− 1.172 (− 1.306, − 1.038)	< 0.001	− 0.238 (− 0.313, − 0.162)	< 0.001	− 0.968 (− 1.081, − 0.855)	< 0.001	− 0.544 (− 0.627, − 0.461)	< 0.001
> 9.5 × 109/L	− 1.572 (− 2.368, − 0.776)	< 0.001	− 0.292 (− 0.721, 0.136)	0.182	− 1.278 (− 1.907, − 0.648)	< 0.001	− 0.721 (− 1.217, − 0.225)	0.004
TC
≤ 5.2 mmol/L	− 1.249 (− 1.457, − 1.042)	< 0.001	− 0.282 (− 0.401, − 0.163)	< 0.001	− 0.990 (− 1.163, − 0.817)	< 0.001	− 0.607 (− 0.738, − 0.476)	< 0.001
> 5.2 mmol/L	− 1.506 (− 1.690, − 1.322)	< 0.001	− 0.390 (− 0.491, − 0.289)	< 0.001	− 1.136 (− 1.292, − 0.980)	< 0.001	− 0.755 (− 0.868, − 0.643)	< 0.001
TG
≤ 1.7 mmol/L	− 1.321 (− 1.476, − 1.166)	< 0.001	− 0.225 (− 0.311, − 0.138)	< 0.001	− 1.120 (− 1.250, − 0.990)	< 0.001	− 0.584 (− 0.680, − 0.488)	< 0.001
> 1.7 mmol/L	− 1.365 (− 1.643, − 1.087)	< 0.001	− 0.453 (− 0.608, − 0.298)	< 0.001	− 0.927 (− 1.159, − 0.695)	< 0.001	− 0.761 (− 0.933, − 0.588)	< 0.001
LDL
≤ 3.4 mmol/L	− 1.545 (− 1.738, − 1.352)	< 0.001	− 0.298 (− 0.406, − 0.190)	< 0.001	− 1.276 (− 1.436, − 1.115)	< 0.001	− 0.712 (− 0.832, − 0.592)	< 0.001
> 3.4 mmol/L	− 1.155 (− 1.362, − 0.947)	< 0.001	− 0.281 (− 0.396, − 0.166)	< 0.001	− 0.882 (− 1.057, − 0.708)	< 0.001	− 0.569 (− 0.697, − 0.441)	< 0.001
HDL
≤ 0.9 mmol/L	− 1.065 (− 2.462, 0.331)	0.135	0.471 (− 0.288, 1.230)	0.224	− 1.489 (− 2.632, − 0.347)	0.011	− 0.027 (− 0.888, 0.832)	0.951
> 0.9 mmol/L	− 1.171 (− 1.305, − 1.036)	< 0.001	− 0.244 (− 0.319, − 0.169)	< 0.001	− 0.961 (− 1.074, − 0.848)	< 0.001	− 0.546 (− 0.629, − 0.463)	< 0.001

Multivariable-adjusted DLM combined with LMM was used to assess the cumulative effects of O₃ exposure on BP for periods of up to 30 days. The estimated BP changes for each 10 μg/m³ increase in MDA8 O₃ exposure were reported. Covariates of age, sex, marital status, BMI, smoking and drinking habits, physical activity frequency, the use of antihypertensive treatment, the status of T2DM, average temperature, relative humidity during the corresponding period, season, and DOW were included for adjustment

Abbreviations: O₃ ozone, SBP systolic blood pressure, DBP diastolic blood pressure, PP pulse pressure, MAP mean arterial pressure, CI confidence interval, BMI body mass index, T2DM type 2 diabetes mellitus, DOW the day of the week, FPG fast plasma glucose, WBC white blood cell, TC total cholesterol, TG triglyceride, LDL low-density lipoprotein, HDL high-density lipoprotein

Mediation analyses

To explore the potential mediation effects of clinical indices on the association between O₃ exposure and BP, we performed mediation analyses. We observed that TG for a 7-day window (mediated from − 9.2 to − 1.5%, Additional file 1: Fig. S7, Additional file 2: Table S6) as well as TG (mediated from − 8.4 to − 0.9%) and TC (mediated from − 1.4 to − 0.2%) (Fig. 4) for a 30-day window may exhibit potential suppressive effects. Conversely, 7-day STB (mediated from 4.2 to 6.8%) and 30-day LDL (mediated from 1.4 to 8.5%) partially contribute to the association between O₃ exposure and BP, highlighting their role as mediators. Despite these findings, the mediation effects accounted for only a small portion of the total effects, suggesting that their clinical significance may be limited.

Fig. 4.

Mediation analysis of the O₃ exposure, clinical indices and BP for periods of up to 30 days. The direct effect of exposure on the outcome (c’), the effect of exposure on the mediator (a), the effect of the mediator on the outcome (b), and the proportion of mediation were estimated using nonparametric bootstrapping based on 1000 bootstrap samples. Covariates of age, sex, marital status, BMI, smoking and drinking habits, physical activity frequency, the use of antihypertensive treatment, the status of T2DM, average temperature, and relative humidity during the corresponding period, as well as season and DOW, were included for adjustment. *P < 0.05, **P < 0.01, ***P < 0.001. Abbreviations: O₃, ozone; SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure; MAP, mean arterial pressure; BMI, body mass index; T2DM, type 2 diabetes mellitus; DOW, the day of the week; TC, total cholesterol; TG, triglyceride; LDL, low-density lipoprotein

Sensitive analysis

Meteorological factors, including the maximum temperature, the maximum air pressure, and the average wind speed, were incorporated into the model to re-evaluate the impact of O₃ exposure on BP. The results, as presented in Additional file 2: Table S7, demonstrated the stability of initial findings. In two-pollutant models, we adjusted for particulate matter and other gaseous pollutants and found that the effects of O₃ exposure on BP were consistent with the main analysis (Additional file 2: Table S8). The potential effects of the impact of day of year were considered and included in the multivariable-adjusted DLM. The results, depicted in Additional file 2: Table S9, indicated that the association between O₃ and BP did not experience significant changes after accounting for the influence of time trends. The association analyses using data without imputation showed similar results as major analyses (Additional file 2: Table S10). Finally, the analyses for each time point also demonstrated the negative associations between O₃ exposure and BP (Additional file 2: Table S11). These sensitivity analyses validated the consistency of the association across different conditions.

Discussion

Our findings indicated positive associations between short-term exposure to the air pollutant mixture and BP indices. The positive association was also reported among 221 participants with the mean age of 50 years old in Paris, France [39]. In our studied pollutants, PM₁₀ was the most significant factor causing an increase in BP. Several systematic reviews and meta-analyses also reported positive associations between short-term exposure to PM₁₀ and BP [40–43], which were consistent with our findings.

In this study, O₃ emerging as a major contributor that negatively affects BP, raising our research interests. While the respiratory effects of O₃ are well-documented, its cardiovascular implications, particularly on BP, are still under active investigation. As shown in Additional file 2: Table S12, previous studies on the relationship between ambient O₃ exposure and BP have been inconsistent, contributing to ongoing debate and research interest. Notably, there has been a research gap regarding the broader population impact, especially among middle-aged and older individuals. Our study addresses this by focusing on the effects of short-term O₃ exposure on BP within the general middle-aged and older population in China.

We observed the effect of O₃ on PP was immediate, while SBP, DBP, and MAP presented lag phases of 3-, 14-, and 8-day post O₃ exposure. These findings highlight the complex and time-dependent nature of the relationship between O₃ exposure and BP. The immediate impact on PP suggests a rapid physiological response to O₃, potentially indicating acute effects on vascular function or autonomic nervous system activity. The observed lag phases for SBP, DBP, and MAP suggest a more gradual and cumulative effect of O₃ exposure over time, which could be related to longer-term changes in vascular tone, inflammation, or other systemic responses.

The exposure duration is a key determinant in the BP response to O₃. Several studies align with our findings. For example, an acute exposure study conducted at an hourly level within a Korean population (N = 98,577) illustrated an initial increase in SBP, which then declined, culminating in a significantly negative correlation with SBP after 8 h of O₃ exposure [13]. Similarly, another study involving 4580 Iranian adults suggested that prolonged exposure to O₃ over 14 days could result in a decrease in SBP [42]. A more recent extensive cohort study in Taiwan (N = 9238) observed that O₃ exposure with a lag of 1 to 3 days was associated with decreased SBP and PP [14]. However, not all studies have yielded consistent results with us. For instance, a randomized crossover study with 87 U.S. healthy middle-aged and older volunteers detected no significant change in SBP and DBP after 3-h O₃ exposure with the concentrations of 137.42 or 235.58 μg/m³ [11]. A population-based cohort study among Taiwanese population (N = 7578) noted an increase in DBP following a 3-day O₃ exposure with the median concentration of 52.67 μg/m³, while SBP was unaffected [15]. Several systematic reviews and meta-analyses observed positive associations between O₃ exposure and BP, although these findings lacked statistical significance [6, 40]. The variability in the findings across different studies highlights the challenges in understanding the cardiovascular effects of O₃ exposure. Factors such as the specific population characteristics, the timing and duration of exposure, the levels of O₃, and the methods used to assess exposure and health outcomes (Additional file 2: Table S12) can all influence the observed associations. This complexity underscores the need for further research to elucidate the mechanisms underlying the relationship between O₃ exposure and BP as well as to identify the conditions under which these effects are most pronounced.

Our findings indicated that O₃ exposure might reduce BP over 7- and 30-day period, consistent with its use as a vasodilator in medical treatment [44]. O₃ inhalation might trigger receptors in the airways, possibly initiating an autonomic reflex that influences vascular tone regulation [45]. Animal studies have shown O₃ can activate pulmonary C-fibers, which in turn leads to a reduction in cardiac output and subsequent bronchial vasodilation [46]. Furthermore, a decrease in BP might suggest a reduction in cardiac contractility. Research on heart rate variability has indicated a minor elevation in high-frequency power, primarily influenced by vagal tone [47]. An increase in vagal tone, shifting the sympathovagal balance, might lead to decreased cardiac contractility or heart rate. This is supported by animal studies that have revealed significant reductions in cardiac fractional shortening following particle exposure, indicating reduced myocardial contractility [48]. While changes in BP related to O₃ exposure may be minor and negligible for healthy individuals, they can be significant for vulnerable populations.

Our findings suggest that certain demographic and health-related factors may exacerbate the impact of O₃ exposure on BP, indicating a potential differential susceptibility among these subgroups. The increased vulnerability of these groups underscores the importance of targeted interventions and precautions to mitigate the adverse effects of O₃ exposure, particularly in populations that are already at higher risk due to their age, lifestyle, or health status. In this study, a pronounced negative association between O₃ exposure on BP was observed in individuals aged over 65. Their blood vessels become less flexible and more rigid, which can lead to higher BP levels. In addition, we observed strong inverse associations in current or former smokers and drinkers. Numerous studies have demonstrated that tobacco and alcohol consumption can attenuate vascular elasticity [49, 50], which may explain our findings.

Chronic diseases also modify the effects of O₃ exposure on BP. Our study found that the negative changes in BP due to O₃ exposure were more pronounced in patients with hypertension and T2DM. Another study conducted among 70 older diabetes patients in the U.S. demonstrated that an interquartile increase in the 5-day O₃ exposure was associated with decreases of 4.0%, 2.0%, and 2.8% in SBP, DBP, and MAP [9], respectively, which is consistent with our findings. Therefore, public health policies should consider the cumulative and interactive effects of multiple risk factors, including air pollution, lifestyle choices, and chronic health conditions, to develop comprehensive approaches to protect cardiovascular health.

In mediation analysis, we found that O₃ exposure was associated with decreased STB and increased TC and TG. The observed associations could be indicative of the complex mechanisms through which O₃ exposure affects cardiovascular health. Recent molecular epidemiological studies have demonstrated that short-term O₃ exposure was positively associated with lipid metabolites, indicating its potential ability to disrupt lipid metabolism. Bilirubin, the end product of heme metabolism, offers cytoprotective benefits like antioxidant and anti-inflammatory effects [49–51] but also poses cytotoxic risks at high serum levels [52, 53]. STB is linked with deteriorating cardiovascular health markers and hypertension risk [21]. Moreover, the increase in TC and TG could reflect lipid metabolic responses to O₃, which are known to be involved in the pathogenesis of cardiovascular diseases. These results indicate that certain clinical indices may play a role in the pathway between O₃ exposure and BP, but the relatively small proportion of the total effect explained by these mediators suggests that other pathways may also be involved. Further research is needed to elucidate these additional mechanisms and to determine the clinical significance of these findings.

Our research has significant public health implications by providing a comprehensive examination of the effects of short-term O₃ exposure on BP. Understanding these temporal dynamics is crucial for interpreting the health impacts of O₃ exposure and for developing effective public health interventions. However, the study has several limitations. First, this study did not fully consider the impact of the “white coat” effect for limited participants [54], where the initial BP reading is relatively high. Although we used the mean of the three readings as the representative BP value, the effect may influence our research findings to some extent. Second, the estimation of air pollution exposure based solely on residential locations might lead to inaccurate assessments, as this approach does not account for individual time-activity patterns or personal behaviors [55]. Moreover, our analysis did not include indoor exposure to O₃, which could introduce some deviation in the findings. Third, despite efforts to adjust for various confounding factors, unmeasured covariates, such as socioeconomic status and education level, may still influence the results. Fourth, as an epidemiological study, it does not elucidate the molecular mechanism underlying BP responses to O₃ exposure.

Conclusions

Our study revealed a negative association between O₃ exposure and BP over 7- and 30-day windows among the Chinese middle-aged and older population. The effect on PP was rapid, while SBP, DBP, and MAP presented 3-, 14-, and 8-day delayed responses following O₃ exposure, respectively. More pronounced inverse associations were observed among specific subgroups, including older individuals (≥ 65 years), overweight and obese participants, former or current smokers and drinkers, and patients with hypertension or T2DM. The effect of O₃ exposure on BP was mainly mediated by STB and LDL, while TC and TG exhibited suppression effects. With the worsening of O₃ pollution in China and other global regions, our findings highlight the need to re-evaluate the health implications of O₃ exposure, especially among middle-aged and older adults.

Abbreviations

ACME

Average causal mediation effect

AIC

Akaike information criterion

ALB

Albumin

ALT

Alanine aminotransferase

AST

Aspartate aminotransferase

BKMR

Bayesian kernel machine regression

BMI

Body mass index

BP

Blood pressure

BUN

Blood urea nitrogen

CAQRA

High-resolution Air Quality Reanalysis Dataset over China

CB

Conjugated bilirubin

CHAP

China High Air Pollutants

CI

Confidence intervals

CO

Carbon monoxide

condPIP

Conditional posterior inclusion probabilities

CVD

Cardiovascular diseases

CV-R²

Cross-validation coefficient of determination

DBP

Diastolic blood pressure

DLM

Distributed lag model

DOW

Day of the week

GACS

Guankou Ageing Cohort Study

groupPIP

Group posterior inclusion probabilities

Hb

Hemoglobin

HDL

High-density lipoprotein

LDL

Low-density lipoprotein

LMM

Linear mixed model

MAE

Mean absolute error

MAP

Mean arterial pressure

MCMC

Markov chain Monte Carlo

MDA8

Maximum daily 8-h average

MICE

Multiple imputation by chained equations

NO₂

Nitrogen dioxide

O₃

Ozone

PLT

Platelet

PM_2.5

Particulate matter with a diameter ≤ 2.5 μm

PM₁₀

Particulate matter with a diameter ≤ 10 μm

PP

Pulse pressure

RCS

Restricted cubic splines

RMSE

Root-mean-square error

SBP

Systolic blood pressure

SCr

Serum creatinine

SO₂

Sulfur dioxide

STB

Serum total bilirubin

TC

Total cholesterol

TG

Triglyceride

T2DM

Type 2 diabetes mellitus

UA

Uric acid

UCB

Unconjugated bilirubin

WBC

White blood cell

WC

Waist circumference

Authors’ contributions

J.Z., Y.X.W., C.T. contributed to the conception and design of the study; C.T., J.P.Y., Z.H.L. developed the model code; J.Z., Y.Q.Z., X.F.X., H.X.J. contributed to acquiring funding; H.X.J., T.W., J.H.Y., S.Y.Z. contributed to acquiring the data and administering the project; C.T., J.P.Y., Z.H.L., D.B.G. contributed to writing the original draft and supervision; Z.H.L., H.Y.X. contributed to exposure data modelling; J.Z., Y.X.W. contributed to reviewing and editing. All authors read and approved the final manuscript.

Funding

Our study was supported by grants from the National Natural Science Foundation of China (22076157), Science and Technology Planning Project of Fujian Province (2023D020), and Natural Science Foundation of Fujian Province (2023J011642, 2024J01046).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

All procedures were approved by the Medical Ethics Committee of the School of Medicine, Xiamen University (Approval Number: XDYX202208H04). All participants provided written informed consent at recruitment. Participants voluntarily joined the study and were free to withdraw at any time.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.