Shenzhen is a major city located in the Pearl River Delta region of China, adjacent to Hong Kong. As one of the country’s 4 province-level municipalities, Shenzhen has a population of 17.68 million and a gross domestic product per capita of CNY ¥173,700 (US $24,446) as of 2021. The Shenzhen Center for Prehospital Care developed the 120 Emergency Medical Services system in 1994, which includes 73 emergency networks and 103 emergency stations. In 2022, the center received 2,278,232 all-cause emergency calls, an increase of 41.96% from the previous year, and had 295,295 emergency dispatches, an increase of 14.23% from the previous year. Shenzhen has a subtropical monsoon climate that is typical of southern areas of China. We recruited individuals aged >18 years who called the emergency ambulance to reach the hospital, and those with time records linked to transfer were eligible to be included during the period from 2013 to 2019. We used ambulance emergency dispatch data and environmental data between January 1, 2013, and December 31, 2019, from the Shenzhen Ambulance Emergency Centre and the National Environmental Monitor Station, which included patients’ self-reported medical claims, the time of AECs, prehospital diagnosis, demographic characteristics, and hourly air pollutant concentrations. Each call in the emergency data set was linked to a patient care report that was completed by the attending paramedics.

We used a time-stratified case-crossover study design to estimate the associations of PM pollutants (ie, PM_2.5 and PM₁₀) with all-cause and cause-specific AECs. Each patient was exposed to PM pollutants before the hour in which the AEC occurred, compared to control periods of the same individual when the AEC did not occur [17,18]. This study design allows for the elimination of potential confounding from all known and unknown time-invariant factors (ie, age, sex, ethnicity, behavioral factors, and socioeconomic status) and covariant factors that vary slowly, such as seasonality [19]. We determined that each case index hour was the hour in which the AEC occurred (ie, case period), and we used 3 or 4 control index hours to match this hour by the same hour of the day, day of the week, month, and year with the case index hour to control for long-term time trends [12]. For example, if the first AEC occurred at 6 AM on Monday, May 20, 2013, we would define 6 AM on Monday, May 20, 2013, as the case index hour and 6 AM on all other Mondays in May 2013 (May 6, 13, and 27) as the control index hours.

We obtained hourly concentrations of air pollutants (ie, PM_2.5 and PM₁₀) from the National Environmental Monitor Station during the study period. The station provides real-time data on criteria air pollutants across all nationally controlled monitoring stations operated by the China National Environmental Monitoring Centre with strict standard data quality control procedures. To facilitate communication, we first defined each IQR increase as the difference between the 25th and 75th percentile of air pollutant concentrations [20]. We also defined the extreme concentration (ie, PM_2.5 and PM₁₀) as the maximum concentration equal to the 97.5th percentile of concentrations over the study period. We calculated a population-weighted average of each IQR concentration increase and the maximum hourly concentration for each hour [21].

We combined patients’ self-reported medical claims and prehospital diagnosis as the final diagnosis. Our primary outcome was all-cause AECs, which we further classified into 3 cause-specific categories: AECs related to cardiovascular diseases, respiratory diseases, and reproductive illnesses. For instance, we classified hypertension, acute ischemic stroke, myocardial infarction, ischemic heart disease, acute coronary syndrome, and unstable angina as cardiovascular diseases–related AECs, whereas calls due to lung diseases, bronchitis, respiratory difficulties, and asthma were categorized as respiratory diseases–related AECs. Similarly, calls related to pregnancy, abortion, or other reproductive disorders were classified as reproductive illnesses–related AECs. Additionally, we created a separate category for AECs due to injury, poisoning, mental health disorders, endocrine disorders, and digestive disorders. We were careful to exclude any vague claims such as dizziness, abdominal pain, discomfort, unconsciousness, headache, and allergy while generating disease stratification. We excluded patients who had both missing self-reported claims and missing prehospital diagnoses. We extracted the relevant demographic information (ie, age, sex, and year) and admission date for each diagnosis. We calculated cause-specific AECs by 24 hours and all-cause AECs by age to reflect the hourly distribution of AECs.

We generated a well-established, distributed lag nonlinear model for nonlinear concentration-response and nonlinear lag-response functions [16]. We modeled the concentration-response function using a natural cubic B-spline with 1 or 2 knots to account for potential nonlinear relationships. We also modeled the lag-response function using a linear function with 1 or 2 knots placed on the log scale of lags up to 48 hours. We used conditional logistic regression to estimate odds ratios (ORs) with 95% CIs, examining the association of all-cause and cause-specific AECs with hourly air pollution concentration. The following covariates were adjusted in our main models: natural spline functions with 3 degrees of freedom for temperature and humidity, public holidays, and the day of the week. We converted ORs with 95% CIs to percentage changes in the risk of all-cause and cause-specific AECs associated with each IQR increase in air pollutant concentrations [20,22,23]. The following equations were used:

Percentage change IQR = (e^{β × IQR} – 1) × 100% (1)

Lower 95% CI = (e^{[β – 1.96 × SE] × IQR} – 1) × 100% (2)

Upper 95% CI = (e^{[β + 1.96× SE] × IQR} – 1) × 100% (3)

where β is the regression coefficient.

We conducted several stratified analyses by age (18-64 vs ≥65 years), sex (male vs female), season (warm vs cool), and the time of day (daytime vs nighttime) to examine potential effect modifications. The warm season ranged from March to October and the cool season ranged from November to February in Shenzhen. Nighttime is defined as being from 8:00 PM to 7:00 AM in the next day, and daytime is defined as being from 8:00 AM to 7:00 PM within one day [20]. Missing rates of PM air pollutants were less than 0.51%, and we filled in missing data by using the next or previous entry.

We conducted a number of sensitivity analyses to assess the robustness of our results. First, we examined the correlations between several air pollutants to estimate the multicollinearity. Second, we repeated the main analyses based on exposure to the extreme concentration, rather than using only each IQR increase, to observe the robustness of the exposure metric. Third, we used variable key modeling parameters, including modelling the concentration-response functions using a natural B-spline with 3 internal knots for all-cause AECs and 2 or 3 knots for cause-specific AECs and the lag-response function using a natural cubic B-spline with 4 knots for all-cause AECs and 2 or 3 knots for cause-specific AECs placed on the log scale of lags up to 48 hours.

All analyses were performed in R (version 4.2.1; R Foundation for Statistical Computing). The survival package was used for conditional logistic regression, and the dlnm package was used for the distributed lag nonlinear model.

Ethics approval and consent to participate in this project was approved by the Peking University Health Science Center Institutional Review Board (PUIRB-YS2023123). Informed consent was obtained from all participants prior to questionnaire administration.

We observed the associations of transient exposure to air pollutants with an increased risk of all-cause AECs, without any discernible threshold effects. The associations were stronger for PM_2.5 than for PM₁₀. In addition, the risks were apparent and statistically significant within 0-4 hours after exposure to the air pollutants. The excess relative risks showed substantial differences between the time-of-day strata.

Our findings were consistent with a robust body of previous studies demonstrating the positive associations of hourly air pollutants with all-cause and cause-specific AECs [15,16,23,25-30]. However, much of the existing evidence has focused on assessing the associations between cardiovascular diseases–related hospital admissions and exposures at the daily timescale through time-series analyses [31,32]. We have filled these knowledge gaps and generalized these results to the emergency medical service setting, as this study demonstrated that AECs may serve as a sensitive and timely marker of the adverse health effects of air pollution.

Given the use of different exposure metrics, time periods, study designs, and analytical strategies, our results cannot be compared directly across studies. For example, an analysis of Shenzhen AECs between January 18, 2013, and December 31, 2016, found that each 10 µg/m³ increase in PM_2.5 and PM₁₀ was associated with a 1.44% (95% CI, 0.70-2.19%) and 0.95% (95% CI, 0.39-1.51%) increase in hourly AECs over 5 hours, respectively [23]. In another study in Shenzhen [25], the authors observed positive associations of exposure to PM_2.5 with an increased risk of emergency department visits across different lag days using daily excessive concentration hours and the daily mean metric. In contrast, relatively weak effects were found for hourly peak PM_2.5. Rao et al [33] conducted a case-crossover study by using British Columbia Emergency Health Service data from 2010 to 2015 in Canada and showed a positive association between cardiovascular disease–related AECs and PM_2.5 exposure (OR 1.007, 95% CI 0.997-1.031), and there was a slight increase in the odds of respiratory disease–related AECs at 1.005 (95% CI 0.998, 1.013) over 48 hours. Another study conducted by Ai et al [15] between 2014 and 2016 found that each 10 µg/m³ increase in PM_2.5 and PM1₀ was associated with a 0.19% (95% CI 0.03-0.35%) and 0.13% (95% CI, 0.02-0.24%) increase of all-cause AECs, respectively, whereas no significant effects of PM_2.5 and PM₁₀ on cardiovascular morbidity were found.

We observed significant effects during the first 24 hours after exposure to the PM pollutants. Several human and animal studies have shown that acute exposure to PM may enhance thrombogenicity through various pathways, including platelet activation, oxidative stress, and the interplay between interleukin-6 and tissue factors [30,34,35]. A few studies have also identified the possible pathways, including endothelial dysfunction, inflammation, dyslipidemia, and autonomic and vascular dysfunction [33,36].

The stratified analyses showed stronger associations of air pollutants with cardiovascular diseases–related AECs in the nighttime than in the daytime, which is in line with existing evidence [22,37]. This may be due to low atmospheric pressure that always occurs during the night (from 1:00 AM to 5:00 AM). Moreover, we found several significant season differences in associations of PM air pollutants with respiratory diseases–related AECs, which could be explained by the virus or allergen being more active in the warm season. Our analysis observed stronger associations of all-cause AECs in the daytime, this could be explained by people often spending more time outdoors during the day and that the associations were estimated based on fixed-site monitors [20]. In addition, there were no associations of air pollutants with AECs stratified by sex in our study. These results were in line with previous literature [7,17], indicating that all patients, regardless of sex, seem to be at higher AEC risks after transient exposure to air pollution.

This study has several limitations. First, we used the individual weighted average of hourly air pollution concentrations as a proxy for personal exposure. Although there is some degree of exposure misclassification that may lead to the underestimation of associations, this limitation may not be avoidable in most time-series studies [22,38-40]. Second, we used prehospital diagnosis rather than medical diagnosis with specific International Classification of Disease codes because data on the final clinical diagnosis were not available. However, we defined our subtypes of AECs using the Medical Priority Dispatch System [16,41], which should not substantially bias our results. Third, we did not have detailed information on the locations of environmental monitoring stations and the proximity of roads, making it challenging to explore the possible effect modification of the associations. Fourth, we were unable to match the monitor station measurements to the closest patient because we were unable to acquire patients’ location information. Therefore, we used the average city-level PM concentration as the exposure for patients in this study. Although the impact of this limitation is likely minimal, we plan to gather and analyze this association in future studies [20].

This study has some strengths. First, the Shenzhen Ambulance Emergency Centre provided high-quality data covering most hospitals in Shenzhen. The large sample size and individual-level data allowed us to conduct comprehensive statistical analyses to maximize the validity of the study results. Second, the time-stratified case-crossover study design facilitated causal inference of the study results by adjusting for time-invariant confounding [18,40]. Third, we collected hourly data on air pollutants when all-cause or cause-specific AECs occurred, which characterized the subdaily time effects of exposure to AECs.

Air pollution–related AECs are considered as a public health problem. This study provides the distributions of all-cause and cause-specific AECs over 24 hours and the associations of each IQR increase with AEC risks. The adverse health effects of air pollution are preventable through a combination of reducing exposure, reducing susceptibility, and improving adaptive capacity. In the context of public health preparedness for air pollution, local government response systems should typically provide air pollution–related information to the public, develop strategies to reduce air pollution–related risks, and allocate emergency ambulance resources equitably. It is also important for hospitals and emergency centers to adapt their procedures to meet the increased demands associated with air pollution. For example, local health systems need not only to allocate more resources (ie, health care providers and medical facilities) to general emergency care but also to increase the capacity of health care providers to provide specialized emergency care (ie, cardiovascular and respiratory diseases) at night because of the stronger associations of PM air pollutants found during nighttime hours.

PM air pollutants were associated with a higher relative risk of all-cause AECs and cardiovascular diseases–, respiratory diseases–, and reproductive illnesses–related AECs. The risk of all-cause AECs increased consistently with increasing concentrations of PM air pollutants, showing a nearly linear relationship with no apparent thresholds. The results of this study may be valuable to air pollution attributable to the distribution of emergency resources and consistent air pollution control.