Acute effect of ambient fine particulate matter on heart rate variability: an updated systematic review and meta-analysis of panel studies

Background Decreased heart rate variability (HRV) is a predictor of autonomic system dysfunction, and is considered as a potential mechanism of increased risk of cardiovascular disease (CVD) induced by exposure to particulate matter less than 2.5 μm in diameter (PM2.5). Previous studies have suggested that exposure to PM2.5 may lead to decreased HRV levels, but the results remain inconsistent. Methods An updated systematic review and meta-analysis of panel studies till November 1, 2019 was conducted to evaluate the acute effect of exposure to ambient PM2.5 on HRV. We searched electronic databases (PubMed, Web of Science, and Embase) to identify panel studies reporting the associations between exposure to PM2.5 and the four indicators of HRV (standard deviation of all normal-to-normal intervals (SDNN), root mean square of successive differences in adjacent normal-to-normal intervals (rMSSD), high frequency power (HF), and low frequency power (LF)). Random-effects model was used to calculate the pooled effect estimates. Results A total of 33 panel studies were included in our meta-analysis, with 16 studies conducted in North America, 12 studies in Asia, and 5 studies in Europe. The pooled results showed a 10 μg/m3 increase in PM2.5 exposure which was significantly associated with a − 0.92% change in SDNN (95% confidence intervals (95%CI) − 1.26%, − 0.59%), − 1.47% change in rMSSD (95%CI − 2.17%, − 0.77%), − 2.17% change in HF (95%CI − 3.24%, − 1.10%), and − 1.52% change in LF (95%CI − 2.50%, − 0.54%), respectively. Overall, subgroup analysis suggested that short-term exposure to PM2.5 was associated with lower HRV levels in Asians, healthy population, and those aged ≥ 40 years. Conclusion Short-term exposure to PM2.5 was associated with decreased HRV levels. Future studies are warranted to clarity the exact mechanism of exposure to PM2.5 on the cardiovascular system through disturbance of autonomic nervous function. Supplementary Information The online version contains supplementary material available at 10.1186/s12199-020-00912-2.


Introduction
Cardiovascular disease (CVD) is the major cause of mortality worldwide, which contributed to 17.8 million deaths in 2017 [1]. In recent years, epidemiologic studies have shown that exposure to particulate matter less than 2.5 μm in diameter (PM 2.5 ) increases the risk of CVD [2,3], and even a short-term exposure to PM 2.5 may lead to acute cardiovascular events [4][5][6]. One of the potential mechanisms of PM 2.5 -related acute cardiovascular events is dysfunction of the autonomic nervous system, which is always assessed by the heart rate variability (HRV) levels [7][8][9]. HRV is regulated by parasympathetic autonomic activation including vagus nerve and sympathetic activation [10], and is usually assessed by time domain indicators (deviation of all normal-to-normal intervals (SDNN), the root mean square of successive differences between normal heartbeats (rMSSD), and frequency domain indicators (high frequency (HF) and low frequency (LF)). The reduction of any of those 4 indicators reflects a dysfunction of the autonomic nervous system [11,12].
In recent years, many panel-designed studies on the associations between exposure to PM 2. 5 and HRV have been published, which could provide direct evidence for acute health effects of exposure to PM 2.5 and its potential mechanisms [7,13]. However, the results remain inconsistent. Some studies reported negative associations between exposure to PM 2.5 and HRV, whereas others reported no association [5,[14][15][16]. For example, Wu S et al. examined the relationship between PM 2.5 exposure and HRV in 11 taxi drivers during the 2008 Olympic Games and found that SDNN and HF change by − 2.2% (95% confidence intervals (95%CI) − 3.8%, − 0.6%) and − 6.2% (95%CI − 10.7%, − 1.5%) with an interquartile range (IQR, 69.5 μg/m 3 ) increase of PM 2.5 exposure, respectively [17]. However, Bartell SM et al. evaluated relationship between exposure to PM 2.5 and HRV in 50 elderly people with coronary artery disease and did not found significant association in SDNN (percent change = − 0.92%, 95%CI − 3.79%, 1.95%) or rMSSD (percent change = − 0.26%, 95%CI − 4.74%, 4.22%) with an IQR (16.1 μg/m 3 ) increase of PM 2.5 exposure [5]. Wheeler A et al. evaluated the effects of exposure to PM 2.5 on HRV in 18 chronic obstructive pulmonary disease (COPD) patients and 12 myocardial infarction (MI) patients and observed a significant effect in COPD patients. However, no significant effects were found in MI patients [18]. The inconsistencies in results may be due to different participants, study designs, sample size, PM 2.5 exposure measurement, and so on.
Meta-analysis study can deal with inconsistent findings to evaluate a pooled effect estimates. Despite 2 metaanalysis studies on the associations between PM 2.5 exposure and HRV have been published [7,19], most of included studies were conducted in high-income countries. More than 7 panel-designed studies on the associations between PM 2.5 exposure and HRV have been published after the latest meta-analysis. Updated studies, especially those conducted in low-and middle-income countries, should be included in the meta-analysis to provide a more comprehensive evidence of effect of PM 2.5 exposure on HRV. We therefore conducted an updated systematic review and meta-analysis of paneldesigned studies to examine the acute health effects of exposure to PM 2.5 on HRV.

Literature search
Literature was searched in three online databases (PubMed, Embase, and Web of Science), with published date until November 1, 2019. Only panel-designed studies that examined the associations between ambient PM 2.5 and HRV levels were included. The search strategy was a combination of exposure and outcome including the following three main domains: (1) ambient PM 2.5 exposure; (2) HRV effects; and (3) panel-designed study [7]. The search strategy was shown in Additional file 1: Appendix A.
We first selected articles by reading titles and abstracts, and then read the full texts of the selected articles to determine whether they should be included in the meta-analysis. Reference lists of all the included studies were also manually searched. Literature was reviewed by two authors independently (ZP N and FF L). Conflicts between the two authors during article selection and data extraction were resolved by discussing with an arbitrator (H X).

Inclusion and exclusion criteria
Inclusion criteria for articles were (1) original peerreviewed human subjects research studies, (2) paneldesigned studies, (3) published in English, (4) quantitative assessment of outdoor (ambient) PM 2.5 exposure, and (5) reported to the percent change of HRV with per increment in PM 2.5 exposure [20]. We excluded studies if they were (1) toxicological studies, summaries or reviews; and (2) focused on indoor or occupational exposure.

Data extraction
Data were extracted from all eligible studies, including (1) study characteristics: first author, published year, study location, and period; (2) study population: sample size, mean age, the number of males and females, health status; (3) outcome assessment: HRV measures; (4) PM 2.5 measurement: PM 2.5 exposure assessment method, exposure-time-window, lag effects, mean and standard deviations (SD) of PM 2.5 concentrations, increment in PM 2.5 used in effect estimates; and (5) effect estimates of the association between PM 2.5 and HRV: percentage change and 95% CIs per 10 μg/m 3 , IQR or SD increment in PM 2.5 exposure.
Considering that short-term exposure to PM 2.5 may disturb the autonomic balance for only 1 day [21], and the frequently used exposure-time-window among panel designed studies were within 24 h, we therefore selected results by using PM 2.5 exposure of the current day and previous one day as the acute HRV effects of exposure to PM 2.5 . For studies provided multiple effect estimates, we selected the representative result based on the following criteria: (1) if a study reported multiple effect estimates of both fixedsites and individual monitors, we selected the results of individual monitors since individual monitors might reflect the real exposure to PM 2.5 more accurately; (2) if a study used more than one exposuretime-window, the effect estimate with exposure-timewindow nearing 24 h was selected; (3) for study that performed multiple subgroup (locations, health status), we chose the combined effect estimates if it was reported. Otherwise, the results of subgroup were treated as several separate studies; (4) for two or more studies from the same population, only the most recent one with the exposure-time-window nearing 24 h was selected [22]; and (5) for study that performed multiple statistical models, we extracted the results of full-adjusted models.

Quality assessment
The Newcastle-Ottawa Scale (NOS) was used for quality assessment of included studies. The NOS designed 8 items to assess the critical appraisal of the potential risk of bias. Total score of NOS ranged 0-9. Studies scored higher than or equal to 6 were regarded as high-quality, while those scored less than 6 as low-quality [22]. Two authors (ZP N and FF L) worked independently and inconsistencies in quality assessment were resolved through discussion.

Statistical analysis
As studies reported effect estimates change in different increments of PM 2.5 (percent change per 10 μg/ m 3 , IQR, or SD increase in PM 2.5 exposure), we first converted them into a standardized form (percent change per 10 μg/m 3 increase in PM 2.5 ) using the formula as follows: Standard error (SE) for each effect estimate was calculated by using the formula: (Upper limit -Lower limit)/ 3.92.
Heterogeneity among different studies was examined using Chi-square-based Cochran Q-statistics test and standard I 2 . Random-effects model was used to estimate the overall effect. Sources of heterogeneity were explored using subgroup analyses including location (North America, Europe or Asia), exposure assessment method (individual monitor, fixed site, or others), health status of participants (healthy population or patients), and age (< 40 years or ≥ 40 years). The main reason for dividing studies by age of 40 mainly based on the report of World Health Organization (WHO), which defined people aged 40 years or older as high risk of cardiovascular disease [23]. Besides, meta-regression was conducted to explore if heterogeneity was modified by potential modifier (location, PM 2.5 measurement, health status, age, and PM 2.5 levels). Subgroup analysis and meta-regression were not performed if the number of one subgroup was less than 5.
The publication bias of included studies was assessed by using funnel plot and Begg's and Egger's test. In addition, sensitivity analysis was performed by omitting one study at a time to evaluate if the omission of any study would change the significance of the pooled results. All these statistical analyses were conducted in Stata version 15.0 (StataCorp, College Station, TX, USA).
In total, 27 studies investigated the association of short-term exposure to PM 2.5 with rMSSD. Metaanalysis showed that a 10 μg/m 3 increase in exposure to PM 2.5 was associated with a − 1.47% change in rMSSD (95%CI − 2.17%, − 0.77%) (Fig. 3). Subgroup analysis by study location suggested that the adverse effect of PM 2.5 on rMSSD was similar among North American populations (percent change = − 1.64%, 95%CI − 2.89%, − 0.38%) and Asian populations (percent change = − 1.58%, 95%CI − 2.43%, − 0.73%). Short-term exposure to PM 2.5 was also associated with decreased rMSSD levels in Europeans, although the association was not statistically significant (percent change = − 0.90%, 95%CI − 2.47%, 0.67%). Furthermore, subgroup analyses by health status showed that the decrease level of rMSSD caused by PM 2.5 among healthy population (percent change = − 2.43%, 95%CI − 3.40%, − 1.45%) was larger than in patients (percent change = − 0.87%, 95% CI − 1.58%, − 0.77%). Subgroup analysis by age suggested that the decrease level of rMSSD caused by PM 2.5 among people aged    Studies performed more than one exposure-time-window or lag effects; however, the results were just presented as figures and the effect estimates were not reported. Linear mixed-effects models were all used statistical approach; HRV measures were all log-transformed; PM 2.5 levels were showed by the mean ± standard deviation (SD) or mean if it was not specified.  Table 2).
Heterogeneity, meta-regression analysis, publication bias, and sensitivity analysis Heterogeneity existed in all four measures of HRV (I 2 > 50%, P < 0.001). The meta-regression analysis identified that health status could explain the heterogeneity for rMSSD change associated with exposure of PM 2.5 (Table  S4, Figure S1, Figure S2). Funnel plots of PM 2.5 and rMSSD, HF and LF showed a slight asymmetry, but the P values of Begg's test and Egger's test were all greater than 0.05, demonstrating that publication bias were acceptable (Table S3, Figure S3). However, publication bias may exist among studies on SDNN, since the P value of Egger's test was less than 0.05. In sensitivity analyses, we found that the average percent changes in time (SDNN, rMSSD) and frequency domains (LF, HF) of HRV were all in the combined confidence interval, suggesting that the results of the meta-analysis were reliable and stable (Figure S4).
The results of subgroup analysis showed that percent changes of HRV for a 10 μg/m 3 PM 2.5 increase among Asians were larger than in North American populations and European populations, which may be attributed to the serious environmental pollution and different composition of particulate matters [50,51]. Previous studies have reported the air pollution levels in many Asian countries, such as China, were 10 times higher than that in Europe and North America [52]. Combined with the results of our study that the effect estimates of exposure to PM 2.5 on HRV in Asia were higher than that in Europe and Northern America, it may partly explain why the risk of PM 2.5 on cardiovascular diseases/mortality is higher in Asian countries [53,54]. We found that the decreased levels of rMSSD, HF, and LF were greater in healthy population than in patients, since patients may take anti-autonomic nerve dysfunction medication, such as benazepril, nimodipine, and thus weaken the effect of PM 2.5 on HRV response. In addition, subgroup analysis by age in our study showed a higher effect estimates in people aged over 40 compared with those aged under 40. Previous studies also reported that elders were more vulnerable to the PM 2.5 -related risk of decreased HRV levels and adverse cardiovascular events [55][56][57]. For example, the European Study of Cohorts for Air Pollution Effects (ESCAPE) project reported that participants aged over 60 years were more sensitive to PM 2.5 exposure than the younger participants [58].
Potential pathophysiological mechanisms included autonomic imbalance, increased oxidative stress, and inflammation, through which PM 2.5 may accelerate the development of CVD [59]. The dysfunction of the autonomic nervous system has been found as the major pathway that result in PM 2.5 -related adverse cardiovascular outcomes [7][8][9]. Both toxicology experiments and epidemiological studies have provided substantial evidence that PM 2.5 exposure would decrease HRV level, and then lead to autonomic nervous dysfunction, which subsequently increase the risk of CVD [21,[24][25][26]60]. For example, Chiarella SE et al. found that the levels of alveolar lavage fluid and plasma catecholamine in mice increased after inhaling PM 2.5 , with the activation of sympathetic nervous system [60]. Chen SY et al. examined the effects of short-term exposue to urban air pollution among 61 high-risk CVD subjects and found that PM 2.5 caused an immediate autonomic nervous dysfunction as well as long-term inflammatory and thrombotic responses [21].
Some limitations of our meta-analysis should be noticed. Firstly, subgroup analysis and meta-regression by location, PM 2.5 measurement, and age group did not explain the observed heterogeneity. Meta-regression suggested that heterogeneity among exposure to PM 2.5 and rMSSD changes may be explained by health status. Secondly, we could not assess the effect of lag more than 24 h of PM 2.5 exposure, because there were large differences on the reported lag effects of PM 2.5 exposure among different studies and many studies were not represented in the percent change of lag effects. Finally, we failed to explore gender disparities in associations between PM 2.5 exposure and HRV levels because most of the panel studies including our meta-analysis only reported the combined effects estimates both male and female participants and subgroup effects estimates were not represented.

Conclusion
Our study demonstrated that exposure to PM 2.5 was associated with decreased levels of HRV, suggesting that exposure to ambient PM 2.5 may increase CVD risk through the activation of autonomic nervous system. Further studies should be conducted to clarity the specific mechanism of exposure to PM 2.5 on health effects.