Fine and coarse particulate air pollution and hospital admissions for a wide range of respiratory diseases: a nationwide case-crossover study

Abstract

Background

The associations between fine and coarse particulate matter (PM_2.5 and PM_2.5–10) air pollution and hospital admissions for full-spectrum respiratory diseases were rarely investigated, especially for age-specific associations. We aim to estimate the age-specific associations of short-term exposures to PM_2.5 and PM_2.5–10 with hospital admissions for full-spectrum respiratory diseases in China.

Methods

We conducted an individual-level case-crossover study based on a nationwide hospital-based registry including 153 hospitals across 20 provincial regions in China in 2013–20. We applied conditional logistic regression models and distributed lag models to estimate the exposure- and lag-response associations.

Results

A total of 1 399 955 hospital admission records for various respiratory diseases were identified. The associations of PM_2.5 and PM_2.5–10 with total respiratory hospitalizations lasted for 4 days, and an interquartile range increase in PM_2.5 (34.5 μg/m³) and PM_2.5–10 (26.0 μg/m³) was associated with 1.73% [95% confidence interval (95% CI): 1.34%, 2.12%)] and 1.70% (95% CI: 1.31%, 2.10%) increases, respectively, in total respiratory hospitalizations over lag 0–4 days. Acute respiratory infections (i.e. pneumonia, bronchitis and bronchiolitis) were consistently associated with PM_2.5 or PM_2.5–10 exposure across different age groups. We found the disease spectrum varied by age, including rarely reported findings (i.e. acute laryngitis and tracheitis, and influenza) among children and well-established associations (i.e. chronic obstructive pulmonary disease, asthma, acute bronchitis and emphysema) among older populations. Besides, the associations were stronger in females, children and older populations.

Conclusions

This nationwide case-crossover study provides robust evidence that short-term exposure to both PM_2.5 and PM_2.5–10 was associated with increased hospital admissions for a wide range of respiratory diseases, and the spectra of respiratory diseases varied by age. Females, children and older populations were more susceptible.

Introduction

Respiratory disease is one of the top leading causes of morbidity and mortality worldwide. It was estimated that nearly 7.4% of the individuals worldwide (545 million) suffered from chronic respiratory diseases and 4 million people died prematurely owing to chronic respiratory diseases annually.¹ Ambient air pollution contributes to an immense disease burden for Chinese populations.^2,3 Particulate matter (PM) is a complex mixture of small particles and liquid droplets found in ambient air. PM with an aerodynamic diameter of up to 10 μm (PM₁₀) is inhalable and can be categorized as fine particulate matter (PM_2.5) and coarse particulate matter (PM_2.5–10).⁴

Previous epidemiological studies emphasize that PM is related to increased hospitalization and mortality for respiratory diseases,^5–8 but there is no consensus on the difference in the respiratory effects of PM size ranges. Additionally, the associations between PM_2.5 and PM_2.5–10 exposure and hospital admissions for a wide range of respiratory diseases has been rarely investigated. Moreover, investigations on the age-specific associations for respiratory diseases have been rather limited.^9,10 Given that the susceptibility to respiratory diseases varies in different age groups, investigation of the age-specific associations between PM_2.5–10 and PM_2.5 exposure and hospital admissions for the respiratory disease spectrum is indispensable. Besides, a majority of previous findings was obtained from ecological time-series studies and did not consider the cumulative respiratory effects of PM over days.^6,11

Therefore, with a nationwide disease registry in China, the present individual-level case-crossover investigation aimed to estimate the associations between PM_2.5 and PM_2.5–10 exposure and hospital admissions for full-spectrum respiratory diseases, with a long lag period considered. We also evaluated the age-specific respiratory disease spectra and examined various potential effect modifiers (i.e. sex, age, region and season) in stratified analyses.

Methods

Hospital admissions data

We obtained the records of hospitalizations for respiratory diseases from SuValue database in 2013–20. This dataset is a hospital-based registry including 153 hospitals across 20 provincial regions in China, and it was generated after extracting, validating and aggregating individual-level hospital admissions records directly from the hospital information system. Patients with a primary diagnosis of respiratory disease [International Classification of Diseases, 10th revision (ICD-10) codes J00–J99] were included in this study (Supplementary Table S1, available as Supplementary data at IJE online). We evaluated 30 specific diseases with hospital admissions records over 2000 to reduce the statistical uncertainty caused by small sample size. The respiratory diseases are categorized into acute upper respiratory infections (J00–J39), chronic lower respiratory diseases (J40–J47), lung diseases due to external agents (J60–J70), other respiratory diseases principally affecting the interstitium (J80–J84), suppurative and necrotic conditions of the lower respiratory tract (J85–J86), other diseases of the pleura (J90–J94) and other diseases of the respiratory system (J95–J99). There are 34 other specific diseases excluded from this analysis because of the very few records of hospitalizations (less than 2000). For each patient, the basic characteristics (including sex and age), the date of hospital admission and the address of the hospital were also extracted. This study was authorized by the institutional review board in the School of Public Health, Fudan University (IRB#2021–04-0889).

Environmental data

The daily mean concentration of criteria air pollutants [i.e. ozone (O₃), sulphur dioxide (SO₂), carbon monoxide (CO), PM_2.5, PM with an aerodynamic diameter of 10 μm or less (PM₁₀), and nitrogen dioxide (NO₂)] were obtained from the air quality monitoring station adjacent to the hospital’s address (7.2 ± 7.1 km) from China’s National Urban Air Quality Real-time Publishing Platform. This platform was operated by China's Ministry of Environmental Protection under strict data quality control procedures. We calculated the daily mean PM_2.5–10 concentrations by subtracting the daily mean concentration of PM_2.5 from PM₁₀. Daily meteorological data (i.e. temperature and relative humidity) were derived from the weather stations adjacent to the hospital’s address (10.6 ± 8.1 km) in the China Meteorological Data Sharing Service System.

Statistical analysis

For statistical analysis, we used an individual-level, time-stratified, case-crossover design to evaluate the associations between PM and hospital admissions for full-spectrum respiratory diseases. This design has been widely used in previous environmental epidemiological studies.^12,13 The case day was defined as the hospital admission date for each patient. For a case day of each patient, three (N = 854 701, 61.1%) or four (N = 545 254, 38.9%) control days in the same month with the same day of the week were selected.¹⁴ This patient also served as his/her own control to minimize the impacts of confounders that were unchanged within a month (e.g. sex, age, body mass index, diet etc.).

In this study, we applied conditional logistic regression models and the distributed lag model (DLM) to estimate the exposure- and lag-response associations between short-term PM exposure and hospital admissions for full-spectrum respiratory diseases. The DLM can flexibly estimate the linear/non-linear and lagged association between exposure and health outcome, through a cross-basis function defining the exposure-response and lag-response structures.¹⁵ According to previous findings on air pollution and hospital admissions,^11,16 we a priori assumed a linear exposure-response relationship and estimated the cumulative association of PM over multiple lag days with the DLM. Thereafter, we plotted the curves of exposure-response relationships for total respiratory diseases with the distributed lag non-linear model to validate the linear assumption. We empirically used a maximum lag period of 5 days to capture the lagged associations of PM. The maximum lag period was supported by our preliminary exploration (from lag 3 to 7 days) and previous studies.^16,17 Specifically, we estimated the cumulative associations of PM with respiratory diseases by building the cross-basis function of PM including a linear exposure-response association and a lag response curve using a natural cubic spline with four degrees of freedom (df). Then we plotted the exposure-response relationship using an exposure-response association with a quadratic B spline with 3 df and a lag response curve using a natural cubic spline with 4 df. We introduced a natural cubic spline function of the moving average temperature (df = 6) and relative humidity (df = 3) of the current day and the previous 3 days to account for the potential confounding effects of meteorological factors. An indicator variable for statutory holidays was also included in the main model. We used trimmed air pollution data (beyond the 1st and 99th percentiles) for statistical analysis to reduce the statistical uncertainty resulting from extreme outliers. In order to reduce the chance of false-positive findings, we also calculated the false-discovery rate (FDR) using the Benjamini–Hochberg method.

To estimate the age-specific associations with the full spectrum of respiratory diseases, we fitted separate models by age group [children (aged ≤17), young and middle-aged adults (age 18–64) and older people (aged ≥65)]. We also conducted stratified analyses by sex (male/female), season (warm/cool), and region (Northern/Southern China) to examine the potential effect modifications. We used a two-sample z test to evaluate the statistical differences between estimates in the stratified analyses. For the sensitivity analyses, we used two-pollutant models to modulate criteria air pollutants (PM_2.5 or PM_2.5–10, NO₂, SO₂, O₃ and CO) and to test the robustness of the estimated associations. We also conducted sensitivity analyses on the associations between PM and respiratory hospitalizations over lag 0, 1, 2, and 3 days. Additionally, we adjusted for temperature using distributed lag non-linear models (DLNM) with 4-day lag period and for meteorological variables using natural cubic splines with different dfs.

The statistical analyses in this study were performed using R (version 3.6.3). We used the dlnm package to fit DLM and the survival package to fit the conditional logistic regression models. The estimates are presented as percentage changes and 95% confidence intervals (95% CI) of hospital admissions for respiratory diseases associated with an interquartile range (IQR) increase in air pollutants. We used two-sided P-values, and P-value <0.05 was considered to be statistically significant.

Results

Descriptive results

During our study period from 2013 to 2020, we totally identified 1 399 955 hospital admission records with a primary diagnosis of respiratory diseases from 153 hospitals in China (Supplementary Figure S1, available as Supplementary data at IJE online). As shown in Supplementary Table S1 , 739 687 (52.8%) records were children (aged ≤17), 326 421 (23.3%) were young and middle-aged adults (aged 18–64) and 333 847 (23.8%) were seniors (aged ≥65 yrs). The top five respiratory diseases were pneumonia, other chronic obstructive pulmonary disease, acute bronchitis, acute upper respiratory infections of multiple and unspecified sites, and other respiratory disorders. We also summarized the hospital admission records for full-spectrum respiratory diseases (from J00 to J99) by age group (Supplementary Table S2, available as Supplementary data at IJE online). The descriptive statistics of air pollutants and meteorological conditions on the case days during the study period are shown in Table 1. Supplementary Table S3 (available as Supplementary data at IJE online) presents the coefficients of Spearman’s correlation among criteria air pollutant concentrations as well as meteorological conditions during our study period. PM_2.5 was in moderate correlation with CO (r_s = 0.58), NO₂ (r_s = 0.54) and PM_2.5–10 (r_s = 0.52).

Regression results

Figure 1 displays the overall lag patterns and cumulative non-linear exposure-response relationship curves of PM_2.5 and PM_2.5–10. The lag patterns indicated that the associations of both PM_2.5 and PM_2.5–10 with hospital admissions for total respiratory diseases occurred immediately (at lag 0 day), attenuated rapidly and then vanished after 4 days (Figure 1A). Thereafter, we calculated the cumulative effects of PM_2.5 and PM_2.5–10 over lags of 0 to 4 days. Figure 1B shows that respiratory hospitalizations increased with higher PM_2.5 or PM_2.5–10. The curves were almost linear, but the slope of PM_2.5 was steeper at concentrations <40 μg/m³ and the shape of PM_2.5–10 was steeper at concentrations <100 μg/m³. Specifically, for an IQR increase in PM_2.5 (34.5 μg/m³) and PM_2.5–10 (26.0 μg/m³), the hospital admissions of total respiratory diseases would increase 1.73% (95% CI: 1.34%, 2.12%) and 1.70% (95% CI: 1.31%, 2.10%) over lags of 0 to 4 days, respectively.

We also plotted the lag patterns and exposure-response relationship curves for the associations between PM_2.5 and PM_2.5–10 and hospital admissions for total respiratory diseases in children, young and middle-aged adults and older populations (Supplementary Figures S2 and S3, available as Supplementary data at IJE online). We found that the exposure-response relationship curves of PM_2.5 and PM_2.5–10 were consistently linear among children, young and middle-aged adults and older populations. The lag patterns differed among different age groups, specifically the lagged period in the child population (2 days) was relatively shorter than among young and middle-aged adults and older populations (3–4 days). To ensure the estimates are comparable across different age groups, we consistently used the same lag 0–4 days as the main lag period to evaluate the associations among different age groups.

Figures 2–4 showed the effect estimates on the full-spectrum respiratory diseases related to an IQR increase in PM_2.5 and PM_2.5–10 in children, young and middle-aged adults and older populations. The associations with acute and chronic respiratory infections (e.g. pneumonia, acute bronchitis, bronchiolitis and emphysema) and chronic obstructive pulmonary disease were largely consistent across all age subgroups. In children (aged ≤17), we found that PM_2.5 was associated with acute laryngitis and tracheitis, whereas PM_2.5–10 was associated with influenza and bacterial pneumonia. In young and middle-aged adults (aged 18–64), both PM_2.5 and PM_2.5–10 were associated with acute tonsillitis; PM_2.5 was associated with emphysema and bronchiectasis; and PM_2.5–10 was associated with increased hospital admissions for nose and nasal sinuses as well as pneumothorax. In the elderly (aged ≥65), PM_2.5 and PM_2.5–10 were both associated with chronic obstructive pulmonary disease, asthma and other chronic respiratory diseases; while PM_2.5–10 was associated with acute bronchitis and emphysema.

We also provided the effect estimates on the full-spectrum respiratory diseases related to an IQR increase in PM_2.5 and PM_2.5–10 in all age populations (Supplementary Figure S4). We found that the respiratory disease spectra differed between PM_2.5 and PM_2.5–10. In particular, both PM_2.5 and PM_2.5–10 exposures were related to increases in hospital admissions for acute respiratory infections (e.g. bacterial pneumonia, pneumonia, acute bronchitis and acute bronchiolitis), chronic respiratory diseases (e.g. chronic bronchitis, emphysema, asthma, and chronic obstructive pulmonary disease), and other diseases of the respiratory system (i.e.. other respiratory disorders). It is noteworthy that only PM_2.5 was associated with acute laryngitis, tracheitis and bronchiectasis, whereas only PM_2.5–10 was associated with influenza and pneumothorax. We found that most of the FDR results remained statistically significant.

Supplementary Tables S7 and S8 (available as Supplementary data at IJE online) display the percentage change of hospital admissions for total respiratory diseases associated with an IQR increase in PM_2.5 and PM_2.5–10, stratified by sex, age, region and season. The associations in females were stronger than those in males (P for difference = 0.014, P for difference = 0.013, respectively). The magnitude of the association of PM_2.5 and PM_2.5–10 in children was larger than that in young and middle-aged adults (P for difference = 0.026, P for difference = 0.005, respectively), whereas only the association of PM_2.5 in older people was stronger than that in young and middle-aged adults (P for difference = 0.002). Both the associations of PM_2.5 and PM_2.5–10 in Southern China were higher than those in Northern China (P for difference = 0.004, P for difference <0.001, respectively). Besides, the associations in the warm season were more pronounced than in the cool season (P for difference = 0.001, P for difference = 0.001, respectively).

In sensitivity analyses (Table 2), the associations of PM_2.5–10/PM_2.5 on total respiratory diseases were not apparently changed when controlling for SO₂, O₃ and CO. The adjustment of PM_2.5–10 only slightly decreased the magnitude of the association of PM_2.5, whereas the adjustment of PM_2.5 considerably reduced the magnitude of the association of PM_2.5–10. The estimated effects for PM_2.5–10 and PM_2.5 were both considerably decreased after NO₂ was controlled. We found that the cumulative effect of PM_2.5–10 and PM_2.5 (over lag 4 days) was greater than any less cumulative exposure considered (over lag0, 1, 2 and 3 days), indicating that a 4-day lag period could fully capture the lagged effect of PM (Supplementary Table S5, available as Supplementary data at IJE online). In addition, the results barely changed after adjusting for temperature using DLNM and meteorological variables using nature spline with different dfs (Supplementary Table S6, available as Supplementary data at IJE online).

Discussion

To our knowledge, our study is the first nationwide case-crossover study in China investigating the associations between PM_2.5 and PM_2.5–10 exposure and hospital admissions of full-spectrum respiratory diseases. We found that both PM_2.5 and PM_2.5–10 were associated with increased respiratory hospitalizations, which could last for 4 days. We found that the spectra of respiratory diseases varied by age. Besides, the associations were stronger in females, children and older populations. Females, children, the elderly and patients in Southern China and in the warm season were more susceptible to the respiratory effects of PM_2.5 and PM_2.5–10.

In the current study, we observed almost linear exposure-response relationship curves for the associations of PM_2.5 and PM_2.5–10 with increased hospital admissions for total respiratory diseases. The hospital admissions for total respiratory disease increased with higher PM exposure concentrations, which were broadly consistent with previous findings.^17–19 It has been well demonstrated that PM exposure, especially the deleterious constituents [e.g. black carbon, nitrate, ammonium, heavy metals and particle-bound polycyclic aromatic hydrocarbons (PAHs)] could promote the reactive oxygen species production in lung cells and increase oxidative stress,^20–22 which may consequently induce airway inflammation and contribute to the onset of respiratory diseases.²³ PM exposure may also induce telomere erosion, DNA methylation or related epigenetic effects, which are involved in the important mechanisms for respiratory diseases.^2,20 Besides, PM exposure may also impair the immune system and is clinically associated with respiratory exacerbations.^24,25

Most of the previous studies had used single-day or average-day exposure.^11,18 However, researchers indicated that this may lead to an underestimation of the PM-related health effects.^26,27 With the development of time scale (daily or hourly) in environmental monitoring, DLM was widely used to estimate the air pollution-related health effects in environmental epidemiological studies.^28,29 In this study, we used the DLM to fully capture the effects of PM_2.5/PM_2.5–10 over multiple lag days and estimated the cumulative effects over lags of 0 to 4 days according to the lag pattern for the associations with total respiratory hospitalizations. We estimated that an IQR increase in PM_2.5 was associated with 1.70% increase in total respiratory hospital admissions, which is higher than a case-crossover study conducted in 26 Chinese cities (0.90%, at lag 0 day).¹⁶ The difference could be attributed to the failure of considering the lagged effects of PM_2.5. The estimated increase in respiratory hospital admissions associated with PM_2.5–10 (1.73%) in our study was lower than that reported in a Medicare Cohort Air Pollution Study (MCAPS) from 1999 to 2010 in the USA (2.31%, at lag 0 day).¹¹ We found that the estimated effects for PM_2.5–10 and PM_2.5 were both considerably decreased after adjusting for NO₂, suggesting the potential confounding effect of NO₂. As a typical traffic-related air pollutant, NO₂ mainly originated from the same sources (traffic or fossil fuel combustion) as PM and correlated with PM, which was also observed in several previous studies.^30,31 A worldwide multiple countries/regions study and several epidemiological studies have found that short-term NO₂ exposure was associated with increased respiratory hospitalizations and mortality risk.^31–33

With a case-crossover study on the individual level in China, we observed that the respiratory disease spectra varied between PM_2.5 and PM_2.5–10, which may be attributed to the different physicochemical properties and biological mechanisms of PM_2.5 and PM_2.5–10. Specifically, we identified several respiratory diseases associated with both PM_2.5 and PM_2.5–10, including acute or chronic respiratory infection (e.g. pneumonia, acute bronchitis and bronchiolitis, chronic bronchitis) and chronic obstructive pulmonary disease, which have been demonstrated by numerous previous studies.^34–38 PM_2.5 and PM_2.5–10 may induce airway inflammation, ultimately prompting the onset of acute respiratory infections.³⁹ Upregulated generation of pro-inflammatory cytokines, oxidative stress and impairment of airway immunity may also result in increased susceptibility to respiratory pathogens and contribute to increased risks of developing respiratory diseases.^38,40 Moreover, we also observed that PM_2.5 and PM_2.5–10 exposures were related to increased hospital admissions for asthma and emphysema. The biological plausibility for these diseases was supported by several studies.^41,42

In addition to the respiratory diseases mentioned above, we identified several rarely reported respiratory diseases. Specifically, PM_2.5 was associated with higher hospital admissions for acute laryngitis, tracheitis and bronchiectasis, which were rarely investigated in previous studies.^43,44 PM_2.5 exposure could induce bacterial infection in the airways and may further contribute to the development of bronchiectasis.⁴⁵ However, a time-stratified case-crossover study in Australia failed to observe a statistically significant association between PM_2.5 and acute laryngitis,³⁹ which may be due to the very low concentrations of PM_2.5 (mean = 5.2 μg/m³). Besides, in the current study, PM_2.5–10 was associated with influenza and pneumothorax, which was also biologically supported by several studies.^46,47 Notably, the present study revealed different disease spectra between PM_2.5 and PM_2.5–10, which may be due to the varying sizes and compositions.⁴⁸ Briefly, PM_2.5–10 is primarily produced by processes like windblown dust, mechanical grinding and biological aerosols (e.g. bacteria, moulds and pollens),^8,48,49 and therefore is likely to impair the cellular defense and increase susceptibility to bacterial pathogens.³⁴ Otherwise, PM_2.5 is mainly generated by combustion processes containing toxic components^6,35 which can penetrate the bronchioles and deposit inside the alveoli, owing to their small size.¹¹

Our study indicated that the respiratory disease spectra associated with PM exposures varied in different age groups. For children, the associated hospital admissions were mainly acute respiratory infections (e.g. pneumonia, acute bronchitis and bronchiolitis), consistent with previous studies.^40,50,51 It is worth noting that we provided first-hand evidence for the relationship between PM_2.5–10 exposure and increased hospital admissions for influenza in children, whereas previous studies usually included influenza as a potential confounder in statistical analysis.^34,52 Acute tonsillitis was usually observed among children but almost ignored in young and middle-aged adults.^39,53 Our study identified higher hospital admission for acute tonsillitis in young and middle-aged adults. In addition, higher hospital admissions for disorders of the nose and nasal sinuses were associated with PM_2.5–10 exposure, which could be attributed to the larger size of PM_2.5–10 and the propensity to deposit in the upper airways.⁴⁸ Consistent with previous studies,^11,34,36 we also observed associations with chronic obstructive pulmonary disease, asthma and acute/chronic respiratory infection in the older population.

For stratified analyses, we observed stronger associations in females than males, indicating that females were more susceptible to both PM_2.5 and PM_2.5–10 exposures. The physiological difference by sex may contribute to the differential susceptibility in relation to sex.^12,19,38 Consistent with previous studies, this study found that children and older populations were more susceptible to PM_2.5 and PM_2.5–10 compared with young and middle-aged adults. Previous studies also support our finding,^18,19,38 which may be interpreted by the fact that the respiratory tract barrier and immunity of children are immature and easily infected, presenting clinical symptoms.^37,51 For older populations, there could be a higher prevalence of pre-existing respiratory disorders increasing their susceptibility.⁵⁴ The estimated associations were higher in Southern China, which may be due to differential factors in relation to chemical compositions and exposure patterns between the two regions.⁴⁹ The associations were more prominent in the warm season, which could be attributed to the fact that people typically spend more time outdoors warm seasons, hence increasing the likelihood of PM exposure.⁴⁹

Our study has several notable strengths. First, with a nationwide dataset, this individual-level case-crossover study has a higher ability to make causal inferences than previous time-series studies. Second, we estimated the associations between PM_2.5 and PM_2.5–10 exposure and hospital admissions for full-spectrum respiratory diseases, avoiding possible publication bias. Third, the wide range of air pollution concentrations in China makes it possible to explore the associations of respiratory hospital admissions in the full range of air pollution concentrations.

Several limitations of this study should be also acknowledged. First, as all environmental data were derived from fixed-site monitoring stations, measurement errors for environmental data were inevitable. However, this kind of exposure misclassification was liable in non-differential distribution and would typically cause underestimation of the associations.⁵⁵ Second, the inherent feature of this case-crossover design is self-matching, and could only explore the acute effects of PM on exacerbations of respiratory diseases rather than the chronic effects on the development of respiratory diseases. Third, the cases in this study were defined by the primary diagnosis, and the corresponding diagnostic, clinical classification or coding errors for respiratory diseases and potential selection bias were inevitable in such a nationwide study, but these may occur randomly and their influences may be reduced by the large sample size. Fourth, with large numbers of people and potential diseases, the chances of false-positive findings may increase. We used the FDR to account for this issue and the estimated association in our results should be cautiously interpreted.

Conclusion

In this large-scale, individual-level, case-crossover study, we provide robust evidence that short-term exposure to both PM_2.5 and PM_2.5–10 was related to increased hospital admissions for a wide range of respiratory diseases, including several diseases that were rarely reported previously. The exposure-response relationship curves were almost linear and the associations could last for 4 days. The spectra of respiratory diseases associated with PM_2.5 and PM_2.5–10 varied appreciably among different age groups, with greater vulnerability among females, children and older populations. Our findings may have important implications in designing tailored measures for the prevention and control of specific respiratory diseases sensitive to particulate matter exposures among different populations.

Ethics approval

This study was approved by the institutional review board in the School of Public Health, Fudan University (IRB#2021–04-0889).

Data availability

Data will be made available on request.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

JL: methodology, formal analysis, validation, writing—original draft, visualization. RC: methodology, formal analysis, validation, writing—original draft, visualization. CL: methodology, formal analysis, validation, writing—original draft, visualization. YZ: data curation, writing—review and editing. XX: data curation, writing—review and editing. YJ: data curation, writing—review and editing. SS: data curation, writing—review and editing. YG: data curation, writing—review and editing. HK: conceptualization, methodology, supervision, funding acquisition, resources, writing—review and editing. JX: methodology, supervision, resources, writing—review and editing. All authors contributed to the interpretation of data and the revision of the manuscript, and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (92043301), National Key Research and development Program (2022YFC3702701) and the Shanghai International Science and Technology Partnership Project (21230780200).

Conflict of interest

None declared.

References

Author notes