From February 2013 to December 2019, ALRI outpatient data for patients aged <14 years were obtained from Guangdong Second Provincial General Hospital, which is one of the largest tertiary hospitals in Guangzhou, China (18). This administrative database set up by the hospital regularly collects data including demographics, medical conditions, and diagnosis codes (19–21). The diagnoses were completed by attending physicians and further validated by trained medical coders. The causes of outpatient visits were defined using the International Classification of Diseases, Tenth Revision (ICD−10) as follows (20, 22): ALRI (J12–J18 and J20–J22), pneumonia (J12–J18), and bronchiolitis (J20–J21).

The daily concentrations of air pollutants, including PM_2.5, PM₁₀, nitrogen dioxide (NO₂), sulfur dioxide (SO₂), and ozone (O₃), were retrieved from 11 air monitoring stations in Guangzhou during the study period. The mean concentration of the air pollutants collected by the 11 air monitoring stations was used as the daily concentration of air pollutants. We used a linear interpolation approach (the “na. approx” function in “zoo” package in R) to impute missing data (0.78% of the total observation days). Considering the potential impact of weather on ALRI, we obtained daily meteorological data (mean temperature and relative humidity [RH]) from the National Weather Data Sharing System (http://data.cma.cn/).

We compared the associations between PM_2.5 and daily ALRI-related hospital admissions using three different exposure metrics: PM_2.5 DECH, hourly peak concentration, and daily mean concentration. The definitions of these three metrics are available elsewhere (13–16). Briefly, DECH was developed by Lin et al. (23) and defined as the daily total concentration hours above a specific concentration level. Based on the reference Air Quality Guidelines (daily mean of 15 μg/m³ for PM_2.5) formulated by the World Health Organization (WHO) (24), we calculated the DECH using the following formula (16):

where i is the hourly time of 1 day, C_i is the concentration of PM_2.5 at a given time point, and ΔC_i is the difference between C_i and the threshold concentration (15 μg/m³).

Another metric, PM_2.5 hourly peak concentration, was proposed to investigate the adverse effects caused by high levels of PM_2.5. PM_2.5 hourly peak concentration was defined as the maximum concentration of PM_2.5 during a 24-h period on an observation day. The PM_2.5 daily mean was the most commonly used definition of ambient PM_2.5 concentration in the literature.

Following the design of previous time-series studies in the field of air pollution epidemiology research (10, 11, 25–27), the association between PM_2.5 and ALRI-related hospital outpatient visits was estimated using generalized additive Poisson models. Public holidays, days of the week, and winter and summer vacations for students were controlled as categorical variables in the models. Temporal trends, temperature, and RH were adjusted for as smoothing splines. We also controlled for the number of doctors per day in the models. In line with prior studies (25–27), we chose six degrees of freedom (df) per year for temporal trends, six df for moving average temperature of the current day, and the previous 3 days (Temp03), and RH.

Considering the potentially delayed adverse effects of air pollution, different lag structures were assessed to examine potential lag effects. In the single-lag day models, we begin with the same day (lag0) up to a five-day lag (lag5) based on previous studies (11, 25). In the multi-day lag models, we considered the accumulated effects (moving averages for the current day and the previous one, two, and three days [lag01, lag02, and lag03]).

To investigate whether the health effects of PM_2.5 on ALRI, differed by sex, age group (age <5 vs. 5–14 years), and season (warm vs. cold), we performed subgroup analyses stratified by these factors. The warm and cold seasons are defined as the periods from April to September, and from October to March, respectively. In subgroup analyses stratified by season, we chose 3 dfs per year for temporal trends as each season covers only half of the year. We tested whether the differences between strata were significant by calculating the 95% CI, according to previous studies (10, 11).

We conducted a set of sensitivity analyses to check the robustness of the main results to alternative modeling strategies for temporal and meteorological factors, as well as multiple two-pollutant models. To consider the potential collinearity caused by the inclusion of two highly correlated variables in the same model, we examined the correlation between independent variables. If the correlation was > 0.80, they were not included in the same two-pollutant models.

The main results were first estimated by altering the df for the temporal trends and meteorological variables (df alternating from five to eight). Second, gaseous air pollutants (SO₂, NO₂, and O₃) were further adjusted in addition to PM air pollution using two-pollutant models (28, 29).

All statistical analyses and data visualization were conducted using R version 4.0.5. Statistical significance was set at P < 0.05.

In 2,058 days of observations, from February 2013 to December 2019, we identified 105,639 outpatient visits of patients with ALRI and aged <14 years, among which 35,310 (33.4%) and 68,218 (64.6%) were due to pneumonia and bronchiolitis, respectively. Table 1 presents the descriptive statistics of the daily outpatient visits, concentrations of air pollutants, and meteorological variables analyzed. The mean daily number of ALRI outpatient visits during the study period was 38 (standard deviation [SD]: 18), among which 13 (SD: 10) were pneumonia-related visits and 25 (SD: 11) were bronchiolitis visits. The daily mean of PM_2.5 during the study period was 35.3 μg/m³ (SD: 19.1), the mean of PM_2.5 hourly peak concentrations was 49.8 μg/m³ (SD: 27.2), and the mean of PM_2.5 DECH in 24 h was 443.0 μg/m³ (SD: 409.6).

Figure 1 shows the pairwise Pearson correlation coefficients for the air pollutants and meteorological factors. Ambient particulate matter air pollutants had Pearson correlation coefficients of over 0.9. The associations with SO₂ and NO₂ were moderately strong (Pearson correlation coefficients in the range of 0.6 and 0.8), while the absolute values of Pearson correlation coefficients for particulate matter and O₃, as well as meteorological factors, were <0.5, indicating a lower strength of linear correlation.

Table 2 and Figure 2 present the excessive risks (ER) and 95% CIs of ALRI, pneumonia, and bronchiolitis-related outpatient visits per interquartile range (IQR) increment in the three metrics of ambient PM_2.5 exposure (daily mean, DECH, and hourly peak) at lag03. An IQR increment in PM_2.5 DECH was consistently associated with the highest risk of outpatient visits for all three diseases [12.30% (95% CI: 9.49% to 15.18%) increase in ALRI, 16.32% (95% CI: 12.68% to 20.07%) increase in pneumonia, and 10.27% (95% CI: 7.09% to 13.55%) increase in bronchiolitis], followed by daily mean [11.20% (95% CI: 8.34% to 14.13%) increase in ALRI, 15.60% (95% CI: 11.81% to 19.52%) increase in pneumonia, and 9.50% (95% CI: 6.22% to 12.89%) increase in bronchiolitis] and hourly peak [9.73% (95% CI: 6.97% to 12.55%) increase in ALRI, 13.75% (95% CI: 10.11% to 17.50%) increase in pneumonia, and 8.09% (95% CI: 4.92% to 11.36%) increase in bronchiolitis].

To test the robustness of our findings to alternative models and lag periods of exposure, we conducted two sensitivity analyses. We conducted a series of two-pollutant models by further including gaseous air pollutants (SO₂, NO₂, and O₃), and the results (Table 2) were consistent with the main findings that PM_2.5 DECH was associated with the highest risk of ALRI-related outpatient visits, followed by the daily mean, and hourly peak. To test the robustness of the results to different lag periods of PM_2.5, we ran our main models for exposure measured at various lag periods (lag0 to lag5 and lag01 to lag 03). The ERs of ALRI-related outpatient visits were larger when PM_2.5 was measured as moving averages (lag01, lag02, and lag03). The trend that PM_2.5 DECH was associated with the highest risk of ALRI outpatient visits was generally consistent at different lag periods. To examine the consistency of the results for different dfs for the spline effects of temporal trends and temperature, we conducted the models using various dfs (Supplementary Table S1). The sensitivity analysis results showed that the estimates were larger at larger df, but the results were consistently larger for PM_2.5 DECH than for daily mean and hourly peak, and all ERs remained statistically significant.

The ERs and 95% CIs of ALRI-related outpatient visits for the three metrics of PM_2.5 by sex, age group, and season are shown in Table 3. We found that the associations between PM_2.5 and ALRI-related outpatient visits were significantly larger in the cold season (October to March) than in the warm season (April to September), and this trend was consistent among the three metrics. The associations were insignificant during the warm season in several models. Although the ERs of ALRI outpatient visits were higher among men than among women and larger among children aged between five and 14 years than among those under 5 years, the differences across sex and age subgroups were not statistically significant.