Daily mortality data and weekly counts of influenza-like illness (ILI) for the years 2014-2016 were collected from the Guangzhou Center for Disease Control and Prevention. The causes of death were coded according to the International Classification of Diseases, Tenth Revision. All-cause and nonaccidental mortality were defined using the codes A00-Z99 and A00-R99 (from the International Classification of Diseases, Tenth Revision), respectively. We also extracted data regarding deaths due to cardiovascular (I00-I99), respiratory (J00-J99), digestive (K00-K93), and genitourinary diseases (N00-N99), together with diseases of the nervous system (G00-G99) and endocrine system (D50-D89 and E00-E90). Data on deaths from external causes (V01-Y89) and neoplasms (C00-D48) were also collected. In addition, 26 subcategories within the previously mentioned 10 broad disease categories were also considered (Table S1 in Multimedia Appendix 1). In addition, daily counts of deaths were further categorized by age group (≤64, 65-74, 75-84, and ≥85 years) and gender.

Daily concentrations of PM₁ during the 2014-2016 period were collected from the monitoring stations of the Chinese Atmosphere Watch Network in Guangzhou [15,16]. Data for PM_2.5 and PM₁₀ were obtained from Guangzhou Bureau of Environmental Protection, together with data for other air pollutants, including ozone, sulfur dioxide, nitrogen dioxide, and carbon monoxide. We averaged the daily concentrations for each pollutant from 11 fixed-site monitoring stations in Guangzhou (Figure S1 in Multimedia Appendix 1). China Meteorological Data Service Center provided the daily meteorological data in Guangzhou, containing daily relative humidity (%); minimum, mean, and maximum temperatures (^oC); and air pressure (hPa).

This study involved only a secondary analysis of daily aggregated and deidentified data, and it is classified as exempt from institutional review board approval according to the Chinese legal documents on ethics review issued by the National Health Commission of the People’s Republic of China (document number: 4; 2023) [17].

We conducted a time-stratified case-crossover method to evaluate the impact of PMs (PM₁, PM_2.5, and PM₁₀) on mortality [11,18]. The following quasi-Poisson function accounting for overdispersion was used to assess the relationship between size-fractionated PMs and cause-specific mortality:

where Y_t is the observed number of daily deaths on day t; α is the model intercept; Strata_t denotes the time stratum variable used to control for seasonality and long-term trend; NS represents the natural cubic spline function, with 3 dfs for relative humidity (RH_t) and air pressure (PRE_t) and 6 dfs for mean temperature (TEMP_t); λ , γ, and η are the vectors of coefficients. β denotes the changes in mortality risk per every 10 μg/m³ increase in the concentrations of PM pollutants [3,5,6]. Given that influenza could potentially confound the association between air pollution and health [19], the daily average of ILI_t occurrence was also included in the model. Relative risk [RR=exp(β)] was estimated from the model and the impact of PMs was expressed as the percentage change [(RR-1) × 100%] in the daily number of mortality associated with a 10 μg/m³ increment in PMs.

Furthermore, a smoothing spline function with 3 dfs was used to evaluate the exposure-response relationship between PMs (PM₁, PM_2.5, and PM₁₀) and mortality risk. We also conducted a sensitivity analysis by changing the dfs (5-9) to test the robustness of the association. To investigate the lag patterns of PMs, we fitted the models using different single lags. Moving average approach was further applied to capture the cumulative lag effects of PMs.

To estimate the excess mortality burden caused by size-fractionated PMs with daily concentrations higher than the recommended target, we considered 5 target levels of PM_2.5 and PM₁₀ as per the World Health Organization’s (WHO) new air quality guidelines. The air quality guideline level and 4 interim targets for PM_2.5 are 15, 25, 37.5, 50, and 75 μg/m³, respectively, while for PM_10, they are 45, 50, 75, 100, and 150 μg/m³, respectively [20]. There are no officially announced target levels for PM₁ yet; however, given the high correlation between PM₁ and PM_2.5 and the fact that they are both secondary pollutants emitted from some common sources [21], we assumed the percentiles of the PM₁ target levels to be the same as those of the PM_2.5 target levels. We were able to identify the PM₁ target levels (10, 20, 30, 40, and 55 μg/m³) by locating these percentiles within the PM₁ distribution (Table S2 in Multimedia Appendix 1). The death burden due to PMs was calculated by combining the relative risk of PMs associated with their daily concentrations and the corresponding observed daily number of deaths. The total number of deaths attributable to each specific PM target was computed by summing the excess deaths when daily PM concentrations exceeded that target. The empirical CI (eCI) for the attributable deaths was estimated using Monte Carlo simulation with 1000 replications [5,22].

Stratification analyses were conducted to explore the potential modification effects of season (cold period: November-April; warm period: May-October) and demographical characteristics (gender and age) on the relationship between PMs and mortality. The z statistic was performed to test the statistical difference between the 2 relative risks obtained from the subgroup analyses [23,24].

The robustness of our main findings was examined by several analytical strategies. First, we changed the dfs for daily meteorological variables from 3 to 6. Second, we changed the dfs for the time variable from 3 to 9 per year. Finally, to test the confounding influence of other air pollutants, we performed the two-pollutant models by separately introducing a pair of air pollutants that are not highly correlated, thus avoiding collinearity (Spearman correlation coefficient <0.7) [3,25]. All the data analyses were conducted using the R software (version 4.0.1; R Core Team), and a 2-sided P value less than .05 was considered statistically significant.

In total, there were 146,459 all-cause deaths from 2014 to 2016 in Guangzhou. The number of deaths was 138,396 for nonaccidental causes, including 56,587 circulatory deaths, 21,395 respiratory deaths, 4587 deaths from digestive diseases, 1303 from nervous system diseases, 1930 from genitourinary diseases, 8052 from external causes, 5469 from endocrine diseases, and 41,709 from neoplasms. The average concentrations of daily PM₁, PM_2.5, and PM₁₀ were 28 (SD 13) μg/m³, 38 (SD 21) μg/m³, and 56 (SD 28) μg/m³, respectively (Table 1 and Table S1 In Multimedia Appendix 1). The correlation coefficients between PM₁, PM_2.5, PM₁₀, and other air pollutants were less than 0.7, except for nitrogen dioxide. Temperature and relative humidity were negatively correlated with PMs, while air pressure was positively correlated with PMs (Table S3 in Multimedia Appendix 1).

The lag pattern effects of PM₁, PM_2.5, and PM₁₀ on mortality risk manifested similar patterns, with estimates peaking at lag 1 or lag 2 for all-cause and nonaccidental mortality as well as for deaths from circulatory diseases, respiratory diseases, and neoplasms. The effects generally lasted for 4 days (Figure 1); however, the effects were statistically nonsignificant at different lag days for other diseases.

The association between size-fractionated PMs and mortality at lag days 0-3 were generally positive and monotonically increasing (Figure 2). Similar associations were obtained when using 5-9 dfs for the spline function (Figures S2-S4 in Multimedia Appendix 1).

For every 10 μg/m³ increment in 4-day average concentrations of PM₁, PM_2.5, and PM₁₀, the risk of all-cause mortality increased by 2.00% (95% CI 1.08%-2.92%), 1.54% (95% CI 0.93%-2.16%), and 1.38% (95% CI 0.95%-1.82%), respectively. The increments in nonaccidental mortality were similar across the 3 PMs (2.06%, 1.59%, and 1.41%). The largest effect estimates were observed for cardiovascular diseases (2.78%, 2.25%, and 1.85%) and respiratory diseases (2.61%, 2.53%, and 2.21%). The association between size-fractionated PMs and deaths from neoplasms were marginally significant, with effect estimates of 1.55% (95% CI 0.11%-3.01%), 0.69% (95% CI –0.28% to 1.67%), and 0.71% (95% CI 0.01%-1.40%; Figure 3 and Table S4 in Multimedia Appendix 1). The effect estimates of all size-fractionated PMs were statistically nonsignificant for deaths from nervous system diseases, digestive system diseases, genitourinary diseases, external causes, and endocrine diseases. For specific subcategories, significant effect estimates of PMs were detected among deaths due to chronic rheumatic heart diseases, hypertensive and cerebrovascular diseases, stroke (notably ischemic stroke), influenza, and pneumonia.

In the subgroup analyses by season, age, and gender, we found a slightly greater impact of PM₁ on mortality in hot months compared to cold ones. The mortality risk of PM₁ was higher among older people, particularly for those aged ≥85 years. The effect estimates of PM₁ were slightly stronger in female individuals compared to their male counterparts for overall and nonaccidental mortality and deaths from neoplasms. However, the opposite trend was observed for cardiovascular diseases and respiratory diseases (Table 2). A similar trend by seasons and individual characteristics was observed for PM_2.5 and PM₁₀ (Tables S5 and S6 in Multimedia Appendix 1). However, the differences in mortality risks of PM between subgroups (ie, season, age, and gender) were not statistically significant (P>.05).

As presented in Table 3, the excess all-cause deaths advanced by PM₁ were 7921 (95% eCI 4454-11,206), 6790 (95% eCI 3704-9611), 4899 (95% eCI 2696-6971), 3053 (95% eCI 1790-4393), and 1278 (95% eCI 711-1831), respectively, when the daily concentrations exceeded the target levels of 10, 20, 30, 40, and 55 μg/m³. For PM_2.5, the excess death burden was 8303 (95% eCI 5063-11,248), 7194 (95% eCI 4183-9833), 5504 (95% eCI 3245-7620), 3717 (95% eCI 2242-5171), and 1352 (95% eCI 840-1861), with daily concentrations exceeding the WHO targets (15, 25, 37.5, 50, and 75 μg/m³), respectively (Table S7 in Multimedia Appendix 1). For PM₁₀, 8326 (95% eCI 5980-10,690), 7565 (95% eCI 5130-9602), 4487 (95% eCI 3068-5843), 1991 (95% eCI 1360-2582), and 299 (95% eCI 209-384) deaths were caused, with daily concentrations exceeding the WHO targets (45, 50, 75, 100, and 150 μg/m³), respectively (Table S8 in Multimedia Appendix 1). Therefore, 5.4%, 5.7%, and 5.7% of all deaths in Guangzhou were preventable by reaching the lowest target levels of PM₁, PM_2.5, and PM₁₀, respectively_.

In the sensitivity analyses, when we used 3-6 dfs for relative humidity, mean temperature, and air pressure and spline function with 5-9 dfs per year for calendar days, the effect sizes remained stable. In the two-pollutant models, after separately including ozone, sulfur dioxide, and carbon monoxide in the main model, the effect estimations of PMs on mortality remained similar and statistically significant (Figures S5-S7 in Multimedia Appendix 1).

To the best of our knowledge, this is the first investigation to assess the associations between size-fractionated PM exposure and deaths from a wide range of causes within the same population. In comparison to PM₁₀ and PM_2.5, we observed a stronger association between PM₁ and mortality risk from cardiorespiratory diseases and neoplasms. Among more specific diseases, significant effect estimates of PM₁ were found among deaths due to hypertensive diseases, chronic rheumatic heart diseases, stroke (notably ischemic stroke), influenza, and pneumonia. Over 5% of all deaths were caused by PMs_, with the daily concentrations exceeding the target levels in the WHO’s air quality guidelines.

Consistent with previous research [8,26-28], we found that the health risk increased with the shrinkage in PM size, with PM₁ ranking the highest in effect estimates. For instance, Zhang et al [8] revealed that for every 10 μg/m³ increment in PM₁, PM_2.5, and PM₁₀, the hospital admissions due to cardiovascular diseases increased by 6.7%, 4.5%, and 3.4%, respectively. This gradient in health risk might be mainly attributed to the different particle sizes of PMs, which affect their deposition and absorption in the lungs [7,29]. As the particle size decreases, the surface area increases, leading to a greater catalytic effect on the generation of active oxygen. Smaller molecular size is also conducive to cell absorption, which may further cause the production of high reactive oxygen species, DNA damage, and an increase in interleukin 8 [30]. It is important to note that the relationships between size-fractionated PMs and cause-specific mortality were mostly monotonical in our study, suggesting no safe threshold for these PM pollutants and that any efforts to reduce the PM levels could achieve appreciable health benefits.

A large number of diseases were suscepitable to size-fractionated PMs in our study, most of them originating from cardiorespiratory systems. Previous studies mainly focused on the broad categories of diseases and found similar results—the mortality risk was highest for circulatory and respiratory diseases in general [9]. For the subcategories of causes of deaths, significant impact was detected among deaths due to hypertensive diseases, chronic rheumatic heart diseases, stroke (notably ischemic stroke), influenza, and pneumonia for all size-fractionated PMs. This highlights the importance of prioritizing resource allocations during exposure periods and raising awareness of self-protection among patients with these diseases.

It is interesting to note the positive association between PMs and cancer mortality. Attribution bias might partially explain this finding, that is, the incorrect recording of cause of death as cancer, rather than recording the actual causes, if there is a diagnosis of cancer in a person’s medical history. In other words, the risk of dying from comorbidities (eg, cardiorespiratory diseases and infection) in patients with cancer increases during highly polluted days, but the cause of death may be falsely assigned to cancer. This bias is almost inevitable, even in developed countries, where death registration data are fairly good [31]. Meanwhile, we cannot exclude the possibility that there is a true relationship between short-term exposure to PMs and cancer mortality. Indeed, there are studies showing the acute impact of PMs on cancer hospitalization and mortality [32,33]. However, we acknowledge that the underlying mechanism remains poorly characterized. Although it is unknown whether this elevated risk of death among patients with cancer is due to competing mortality or cancer-specific mortality, the message is clear—patients with cancer could be vulnerable when exposed to air pollution, and significant efforts are needed to raise awareness among these populations and provide the much-needed protective measures.

Furthermore, the association of size-fractionated PMs with mortality varied by season, age group, and gender. The mortality risks of PMs were stronger during warm seasons, which is in line with previous findings [6,9]. The sources of PMs varied at different temperature levels, and PM₁ may be present in more toxic forms in warmer seasons, potentially transmiting faster and staying longer in the air [34]. During warm seasons, people are prone to open windows and go outdoors more frequently [35], and the absorption of air pollutants may also be enhanced through the temperature regulation system (eg, by increasing sweating, minute ventilation, and cardiac output) [34]. People aged 85 years and older were the most vulnerable to PMs. With the rapid aging in Guangzhou, the death burden is set to increase if stricter controls on air pollution are unavailable in the coming years. The gender effect varies by different health outcomes. For instance, Hu et al [6] (2018) and Yin et al [9] (2020) found that the effect of PM₁ on all-cause mortality among female individuals was larger than that in their male counterparts. Yin et al [27] (2020) reported a stronger impact of PM₁ on male individuals for circulatory diseases. The varying impact of PMs between genders may be partly explained by the biological differences between male and female individuals, such as different gas-blood barrier permeability, particle deposition efficiency, and hormonal status [27]. Another reason may be due to between-gender demographic and behavioral differences, including occupational type, smoking, and lifestyle, which may modify the PM-mortality relationship [6].

Some limitations of this study need to be noted. First, there might be diagnostic errors, and therefore, misclassification in the specific causes of death. However, the influence is likely to be small in our study because several quality control measures are available in the Guangzhou death registration system to ensure the accuracy of death classifications [36]. Second, we estimated the death burden of size-fractionated PMs by using several recommended target levels as the reference. However, since no safe threshold was detected for all PMs, our calculations are largely conservative in terms of the health benefits that can be achieved by reducing PMs levels. Finally, as an ecological study, we were unable to explore the individual-level PM-mortality association or control for potential covariates at an individual level, such as smoking habits and occupational exposures. Therefore, the causal relationship cannot be tested in our study. Our study was mainly intended to generate hypotheses on their association, and further research is warranted to test and validate these findings.

Our study indicates that smaller particles are more hazardous to human health. Among specific diseases, significant effect estimates of PM₁ were found for deaths due to hypertensive diseases, chronic rheumatic heart diseases, stroke, influenza, and pneumonia. Our findings highlight the importance of size-based strategies in the control of PMs and management of their health impact.