Ambient fine particulate matter of diameter ≤ 2.5 μm and risk of hemorrhagic stroke: a systemic review and meta-analysis of cohort studies

Ambient fine particulate matter of 2.5 μm or less in diameter (PM2.5) of environment contamination is deemed as a risk factor of cerebrovascular diseases. Yet there is still no explicit evidence strongly supporting that PM2.5 with per unit increment can increase the risk of hemorrhagic stroke (HS). Literatures were searched from PubMed, Cochrane, and Embase. After the systemic review of relevant studies, random effects model was used to perform meta-analysis and to evaluate the association between PM2.5 and risk of HS. Seven cohort studies were finally included, involving more than 6 million people and 37,667 endpoint events (incidence or mortality of HS). Total scores of quality assessment were 50. Pooled hazard ratio (HR) for crude HRs was 1.13 (95%CI: 1.09–1.17) (CI for confidence interval). Pooled HR of subgroup analysis for current smoking with exposure to growing PM2.5 was 1.14 (95%CI: 0.92–2.15) and for never and former smoking was 1.04 (95%CI: 0.74–1.46). Ambient PM2.5 level is significantly associated with the risk of HS, which might be a potential risk factor of HS. Smoking does not further increase the risk of HS under exposure of PM2.5.


Introduction
Ambient air pollution is a major and significant environmental risk to the health of people in both cities and rural areas (https://www.who.int/). According to the data from the World Health Organization, 58% of premature deaths were related to outdoor air pollution. Especially, the fatal effects of air pollution were presented as ischemic heart diseases and strokes resulted from exposure to fine particulate matter of 2.5 μm or less in diameter (PM 2.5 ). Moreover, hemorrhagic stroke (HS) accounted for one-third of strokes. Therefore, PM 2.5 might be closely related to risk of HS. Recently, Sheng Yuan (Yuan et al. 2019) and other collaborators conducted a meta-analysis and concluded that the long-term exposure to PM 2.5 was an important risk factor for stroke. However, the relationship between risk of HS and ambient PM 2.5 exposure has not been accurately confirmed. We supposed that some risk factors of ischemic stroke such as smoking were not associated with risk of HS. Thus, we searched recent cohort studies from open medical database to conduct a metaanalysis and to elucidate the relationship between risk of HS and increase of PM 2.5 .

Inclusion and exclusion criteria
Articles from three electronic databases were pooled to complete the preparation of screening. After deleting the duplications, we established two initial criteria for inclusion and exclusion, respectively. Inclusion criteria included cohort studies, "stroke" and "particulate matter" found in title or abstract. Exclusion criteria: studies without cohort design, animals, or special people enrolled as subjective of studies, and no MeSH term in abstract. After completion of initial screening, final   ,832  ICD-10 I60,I61,I62,I690,I691,  I692, criterion was used to select studies for meta-analysis. Final inclusion criterion for full texts of article: "PM 2.5 " and "hemorrhagic stroke." Final exclusion criterion: absence of hazards ratio (HR) or "increase of PM 2.5 ." Screening of articles was conducted by one author (Kai Zhao) and examined and verified by another author (Jing Li). The third author (Mingfei Yang) would settle disagreement.

Data extraction
The data of first author's name, publication year, study region, exposure, period, total numbers of participants, gender ratio (percent of male in all of the participants), definition of endpoints, increase extent of PM 2.5 , and HR with 95% confidence interval (CI) were extracted. When there were different results of HR with 95%CI for different models adjusted for various covariates, data and relative information of all models were extracted.

Quality assessment
All enrolled studies were independently assessed by two authors (Kai Zhao and Yu Guo) according to Newcastle-Ottawa Quality Assessment Scale Cohort Studies (NOS). All eligible studies were assessed from three main aspects: representativeness of participants included, levels of ambient PM 2.5 , and other factors that could influence the outcome, objectivity, and accuracy of determination of endpoint.

Data standardization
There were various factors that were adjusted to different proportional hazard models. If the coefficient of each factor was supplied in full text, crude data would be restored by conversion formula. The data adjusted by the least factors would not be selected to perform data processing, unless there was no way to acquire crude data. Likewise, data adjusted by different factors would be processed in the same way.

Statistical analysis
The pooled HR and 95%CI was calculated by random effects model. I-square (I 2 ) was used to test the heterogeneity. Funnel plot asymmetry and Egger's regression were used for detecting publication bias. Sensitivity analysis was used to attenuate heterogeneity, in which each study was omitted one by one and the pooled HR of the rest studies was retrieved or the random effects model was switched to fixed effects model. Meta-analysis was completed through corresponding modules in Software for Statistics and Data Science (version 15.1; College Station, TX 77845 USA). All p values were twosided with a significant level at 0.05.

Results
Overall, 2557 publications were obtained from the literature search (780 from PubMed, 15 from Cochrane, and 1762 from Embase). After removing duplicates, 1975 articles were initially included. After excluding reviews, researches of animals, and titles and abstracts that were not conformed to the theme of "PM 2.5 " and "Stroke," 17 articles were retained for final screening. After excluding full texts without "PM 2.5 " and "Hemorrhagic Stroke" or other types of studies except for cohort studies, ultimately, 7 studies were included in this meta-analysis. Of these 7 studies, 6 studies Cai et al. 2018;Downward et al. 2018;Qiu et al. 2017;Shin et al. 2019;Noh et al. 2019) reported the HR between exposure to the circumstance contaminated by growing PM 2.5 and occurrence of HS or hemorrhagic cerebrovascular accident except that one study (Kim et al. 2018) mentioned the mortality caused by HS and used it as the endpoint. Detailed information of screening is exposed to Fig. 1. The characteristics of eligible studies are shown in Table 1.

Quality assessment for included studies
All the studies were assessed by NOS. Six studies were scored 7 points and one study was scored 8 points. All studies were scored in representativeness of the exposed cohort, ascertainment of exposure, comparability of cohorts on the basis of the design or analysis, assessment of outcome, and a long followup duration. Only one study ) was scored in adequacy of follow-up of cohorts. Detailed information of screening is shown in Table 2.

Bias from characteristics of population
Cohorts of the 7 studies selected average located residents for studies' participants which covered people in urban and rural regions, men and women, adults of all ages, and other aspects; thereinto, 3 studies (Downward et al.    (Kim et al. 2018) did not report the detailed information or characteristics of research population, and one study (Qiu et al. 2017) selected the elder whose age was over 60 years; moreover, these 2 studies were solely scored in reports of the non-exposed cohort.

Bias from covariate adjustment
Data processing for all 7 cohort studies was performed via proportional hazards models adjusted by different covariates. Mutual covariates were age (4 studies), gender (5 studies), education (4 studies), body mass index (4 studies), and smoking status (5 studies). Besides, 2 studies exhibited crude HR (Cai et al. 2018;Noh et al. 2019) and the data in a sole study (Kim et al. 2018) did not adjust any covariate.

Bias from definition of endpoint
All endpoints of included studies were defined according to items of International Classification of Diseases (ICD). Nevertheless, different editions and ranges of items in detail might result in bias of the meta-analysis. Three studies (Kim et al. 2018;Noh et al. 2019;Huang et al. 2019) only adopted content in ICD-10. Among them, one study (Kim et al. 2018) reported the most detailed items. Both ICD-9 and ICD-10 were used in 2 studies (Cai et al. 2018;Shin et al. 2019). Yet one study (Downward et al. 2018) did not mention editions or items of ICD and the last one (Qiu et al. 2017) only used ICD-9.
Increase of PM 2.5 and HS

Total outcome
The crude HRs or HRs adjusted for the least factors of all 7 studies were shown in Fig. 2a, including exposure to PM 2.5 increase and incidence or mortality of HS. The pooled HR for each 1.4-10 μg/m 3 increase in PM 2.5 was 1.10 (95%CI: 1.04-1.16), which indicated a positive association between exposure to growing PM 2.5 and incidence or mortality of HS. Yet the heterogeneity of these studies was significant (I 2 = 65.4%, p = 0.008). There was no publication bias according to Funnel plot and Egger's regression (p = 0.225, Fig. 2b). Therefore, to attenuate the heterogeneity, sensitivity analysis was performed. Crude HRs were omitted one by one, and pooled HRs of the rest studies were calculated. All the pooled HRs, 95%CI, I 2 , and p are shown in Table 3. The difference between I 2 was readily discernible. The heterogeneity did not disappear (p = 0.599) until the study ) would be deleted. This might be the HR of the study was from a proportional hazards models adjusted by the least covariates. After a comprehensive survey of covariates in other studies, recent immigrants, income quintile, urban/rural area, and northern/southern Ontario might be special covariates that led to the bias. As the study was omitted, the pooled HR was 1.13 (95%CI: 1.09-1.17).

Subgroup analysis
Adjusted HRs for the most factors of all 7 enrolled studies are shown in Fig. 3a. The pooled HR for each study was 1.16 (95%CI: 1.03-1.30), which indicated a positive association between exposure to growing PM 2.5 with different covariates and incidence or mortality of HS. Yet the heterogeneity of these studies was significant (I 2 = 87.9%, p = 0.000). Publication bias was not evident according to Funnel plot and Egger's regression (p = 0.401, Fig.. 3b). Therefore, sensitivity analysis was conducted to attenuate the heterogeneity. Adjusted HRs were omitted one by one, and pooled HRs of the rest studies were re-calculated. All the pooled HRs, 95%CI, I 2 , and p are shown in Table 4; however, the I 2 was not attenuated.
According to the definite fact that smoking could damage the vascular endothelium, smoking might aggravate the effect of PM 2.5 increment to brain vessels. Thus, subgroup analysis was performed. For individuals who never smoke and those  Fig. 4 a Subgroup analysis for smoking. b Sensitivity analysis for subgroup analysis by switching effects models who were former smokers versus individuals who were currently smoking, significant association between exposure to growing PM 2.5 and incidence or mortality of HS was found (never or former smoking, pooled HR = 1.04, 95%CI: 0.74-1.46; current smoking, HR = 1.41, 95%CI: 0.92-2.15). Subgroup analysis and its sensitivity analysis carried out through switching the random effects model to fixed effects model is shown in Fig. 4 a and b.

Discussion
PM 2.5 is a kind of the inhaled particulates that could enter the circulatory system via permeating alveolar epithelium and vascular endothelium. Therefore, PM 2.5 might be transmitted to the arteries in the brain and deposited on the surface of vascular endothelium, further leading to HS. Thus, effect of deposition or particulates trundle might result in inflammation and injury of vascular endothelium. Finally, the suffered arteries might have corrosion damaging the vessel wall. Chronic diseases  including angina, asthma, congestive heart failure, and diabetes are closely related to PM 2.5 . Many articles reported relationship between longterm exposure to PM 2.5 and increased risks of incident stroke (Atkinson et al. 2013;Lipsett et al. 2011;Ljungman et al. 2019;Lin et al. 2016;Korek et al. 2015;Hoffmann et al. 2015;Dirgawati et al. 2019) and ischemic heart disease (Carey et al. 2016;Hartiala et al. 2016;Katsoulis et al. 2014;Loop et al. 2018;Ueda et al. 2012;Villeneuve et al. 2015;Stockfelt et al. 2017). A case-control study (Qian et al. 2019) reported that fatal intracranial hemorrhage incidence was associated with PM 2.5 exposure and diabetes might increase the risk for intracerebral hemorrhage incidence in relation to PM 2.5 . In this meta-analysis, we found the evidence that exposure to different levels of ambient PM 2.5 per unit was related to risk of HS in cohort studies. Sheng Yuan et al. (Yuan et al. 2019) concluded that pooled HR for stroke and longterm exposure to 5 μg/m 3 increment of PM 2.5 was 1.20 (95%CI: 0.79-1.80; I 2 = 64.2%, p = 0.039) and the subgroup analysis outcomes of smoking status were 1.08 (95%CI: 1.03-1.13; I 2 = 12.3%, p = 0.334) for never smoking, 1.11 (95%CI: 1.01-1.22; I 2 = 0, p = 0.898) for former smoking, and 1.08 (95%CI: 0.94-1.25; I 2 = 0, p = 0.462) for current smoking. This meta-analysis included less literatures than previous studies and had un-unified PM 2.5 increment levels. Besides, due to limited data of included studies, the subgroups in this meta-analysis were divided into the current smoking subgroup and the subgroup of never smoking and former smoking.
In perspective of recommendations for further research about relationship between PM 2.5 and risk of HS, a prospective cohort study is the first choice of study design selection. In the environment exposed by participants, the different levels of PM 2.5 should preferably be obvious and balanced. All participants do not present the endpoint (such as HS) at the beginning period of a study. Exposure measurement is related to the sensitivity and accuracy of the outcome; therefore, the third scientific institutions and proper mathematical models would be used. As to adjustment for covariates, according to this meta-analysis, educated levels, classification of body mass index, and different age grade might be new researchable variables, which can be arranged to appropriate scales for subgroup analysis in the future. Definition of endpoints would be standardized and detailed according to professional items in the world; especially, adequacy of followup of cohorts will be manifested in articles.
As previously mentioned, there were several limitations of this systemic review. Only 7 articles from 3 open electronic databases during 2017 to 2019 were included in meta-analysis. Therefore, those not published in open platform were lost. Moreover, increment of PM 2.5 was not unified, which ranged from 1.4 μg/m 3 to 10 μg/m 3 . Numerous covariates were concealed in data synthesis, which might result in bias or heterogeneity of the metaanalysis and subgroup analysis.
Conclusion Ambient PM 2.5 level was significantly associated with the risk of HS, which might be a potential risk factor of HS. Additionally, under exposure of PM 2.5 , smoking does not further increase the risk of HS.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.