Introduction

Previous studies show that fine particulate matter (particulate matter with an aerodynamic diameter of 2.5 μm or less, PM2.5) exposure has short-term effects on the increase of cardiovascular disease morbidity and mortality (Kan et al. 2007; Peters et al. 2001; Brook et al. 2010) and the disturbed heart rate variability (HRV) is reported to be the major contributor to the trigging of cardiac electro-physiologic diseases (Tsuji et al. 1996; Thayer and Lane 2007).

Heterogeneous associations between PM2.5 and HRV indices are reported in previous studies, in which elderly individuals with compromised health status were chosen as subjects, and outdoor PM2.5 measurements based on the ambient stationary monitors were conducted as exposure evaluations (Chuang et al. 2007; Wheeler et al. 2006; Lipsett et al. 2006; Schwartz et al. 2005; Park et al. 2005; Gold et al. 2000; Zanobetti et al. 2010). However, until recently, the associations between PM2.5 and HRV of the healthy elderly subjects in the indoor and outdoor general environments have not been studied. Due to the disturbance from the health status and biased exposure evaluation, there is a great limitation in inferring the above associations from the current studies. It has been demonstrated that health conditions and drug mediations may greatly modify the associations between PM2.5 and HRV (de Hartog et al. 2009; Wheeler et al. 2006; Park et al. 2005). Besides, exposure evaluation based on the stationary monitoring station conducted in former studies is also reported to mask or disturb the effects of PM2.5 on HRV indices (Suh and Zanobetti 2010).

In this study, we conducted the personal PM2.5 exposure evaluation by micro-environment analysis method and evaluated the associations between PM2.5 and HRV indices in a panel of healthy elderly subjects. We also investigated different associations in indoor and outdoor environments.

Materials and methods

Study protocol

This study was conducted in the community located in the North-Western Beijing along the 4-ring road, which was surrounded by busy traffic roads. The ambient PM2.5 maintains at relatively high level and mainly comes from the traffic emission (Chan and Yao 2008).

Our study period was from August 2008 to September 2008 (summer in Beijing). The measurements were conducted on the weekdays (from Monday to Friday) in order to exclude the unusual activity on the weekends.

Panel participants’ recruitment

Forty-one elderly subjects (age > 50 years) living in the community were initially recruited through annual health checkups at the Community Health Center. Before the study period, everyone received a physical examination (including the blood pressure, X-ray chest perspective, height and weight measurement, and the examinations of blood markers) and completed a questionnaire. The subjects were included from the following criteria: (1) Non-current-smokers who have no home or occupational exposure to smoke, dusts, or fumes; (2) Suffering from no obesity (BMI ≤ 30 kg/m2), and no history of diabetes, hypertension or other chronic diseases, and common cardiopulmonary diseases (atrial fibrillation, second-degree heart block, left bundle branch block, paced rhythm, unstable angina, myocardial infarction, pneumonia, asthma, or chronic obstructive pulmonary disease); (3) Having normal hemogram indices. Thirty-nine subjects met our inclusive criteria, and finally 30 subjects were included in our panel (eight subjects did not receive the HRV measurement and one subject received Holter monitoring for less than 5 h, which is too short to be acceptable). Most of them are retirees and the scopes of their activities were usually within 1 km around the community. The study was approved by the Institutional Review Board of Peking University Health Science Center, and all the subjects signed an informed consent before the study.

Continuous ambulatory HRV monitoring

Participants were hooked up to the Holter monitors by the trained technicians. The participant’s skin was shaved if necessary, cleaned, and slightly abraded to ensure proper lead contact. Every subject wore the Holter Recorder for 24 h from the 9:00 am to 9:00 am the next day. On the testing day, the subjects were asked to follow their regular daily routine and complete a standard time-activity pattern diary, including: time intervals (30 min a segment), places (indoor or outdoor; 1/0), sporting (yes or no; 1/0), cooking (yes or no; 1/0), cleaning (yes or no; 1/0), sleeping (yes or no; 1/0), windows-open (yes or no; 1/0), etc. during the HRV recording.

The ambulatory electrocardiogram (ECG) was recorded using the Holter monitors (H12+, Ver 1.0, Motara Instrument Inc., Milwaukee, WI, USA). The power spectral density (HF and LF) was calculated in 5-min segments for the entire recording using the Fast Fourier Transform (FFT) with PC-based software (H-Scribe, Ver3.7, Motara Instrument Inc., Milwaukee, WI, USA). Two frequency indices were chosen by the power spectral density analysis as recommended by the Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology (Anonymous 1996; Stein et al. 1994): high-frequency component (HRV-HF; 0.15–0.40 Hz), low-frequency component (HRV-LF; 0.04–0.15 Hz), and the LFHFR, which is the ratio of LF/HF. Trained personnel performed all analyses. All recordings could either be accepted or rejected according to the recommended standard criteria (Anonymous 1996), and the 5-min recordings with less than 300 beats were excluded to ensure the quality of the HRV testing. Since sleep may greatly disturb the HRV modification (Chan et al. 2004; Chuang et al. 2005), we distinguished the wakeful and sleeping period according to the time-activity diary and used measurements when subjects were awake for data analysis.

Monitoring of indoor/outdoor pollutants and meteorological parameters

The real-time meteorological data and PM2.5 concentration (indoor and outdoor) were measured concurrently with the ambulatory HRV testing. As the subjects spent most of their wakeful period in the living room, the indoor monitors were conducted in the living room of the subjects who undergo the HRV test. The outdoor monitors were placed on the six-floor rooftop of the residential building (18 m in height) in the community with no emission sources nearby.

The temperature (Temp) and Relative Humidity (RH) (indoor/outdoor) were recorded by the HOBO Pro V2 Temp/RH (Onset Corp., Pocasset, MA, USA) every 1 min. The relative PM2.5 concentrations per minute were measured by a digital dust monitor (LD-3 K; Sibata Scientific Technology Inc., Tokyo, Japan), which was a portable monitor based on the light scattering principle by measuring the intensity of the laser beam scattered by fine particulates. Conversion coefficients were calculated based on the mass concentrations measured simultaneously using the filtration sampling method to convert relative concentrations to mass concentrations. We used collocated portable air samplers (MP-Σ300; Sibata Scientific Technology Inc., Tokyo, Japan) equipped with cascade impactors (ATPS-20H; Sibata Scientific Technology Inc., Tokyo, Japan) with a flow rate of 1.5 l/min (the cut-off points of aerodynamic diameter were 2.5 and 2.5–10 μm, respectively) to measure the mass concentrations while the LD-3 K monitoring worked simultaneously in the same place. After gravimetric analysis, the daily average PM2.5 mass concentration was used to calibrate the real-time mass concentration by the following formula:

$$ \begin{aligned} & {\text{Calibrated}}\,{\text{real}}\,{\text{time PM}}_{2.5} {\text{mass}}\,{\text{concentration}} \\ & \quad = {\text{real}}\,{\text{time}}\,{\text{PM}}_{2.5} {\text{relative}}\,{\text{concentration}} \\ & \quad \times \left( {\begin{array}{*{20}c} {{\text{measured}}\,{\text{daily}}\,{\text{average}}\,{\text{PM}}_{2.5} {\text{mass}}\,{\text{concentration}}} \hfill \\ { \div \,{\text{daily}}\,{\text{average}}\,{\text{real}}\,{\text{time}}\,{\text{PM}}_{2.5} {\text{concentration}}} \hfill \\ \end{array} } \right) \\ \end{aligned} $$

All the minute-average variables (PM2.5, Temp, and RH) were aggregated to 5-min average variables synchronized with the 5-min HRV indices testing.

Calculation of personal levels of exposure variables

The personal level of exposure variables was calculated by the micro-environment analysis method. We calculated the personal exposures of 5-min average of PM2.5, Temp, and RH indicated by the PM2.5p, Tp, and RHp using the following formula:

$$ X_{p} = \left| {place - 1} \right| \times X_{0} + place \times X_{i} $$

X p is the personal level of variables, place is the subject’s location during the 5-min interval (1/0, indoor/outdoor), and the X i , X 0 are the indoor and outdoor levels of variables, respectively.

Statistical methods

We used nonparametric test (Wilcoxon test) for the indoor-outdoor comparisons of the PM2.5 and HRV indices. The response variables HF, LF, and LFHFR were log10 transformed to improve the normality and stability. The mixed linear regression model was applied to estimate the effects of 5-min PM2.5 on simultaneous log10 transformed 5-min average HRV indices for its advantage of analyzing repeated measurements. The mixed model was fit and assumed that:

$$ Y_{it} = b_{0} + u_{i} + b_{1} x_{1} + \ldots + b_{p} x_{p} + \beta {\text{Pollutant}} + \varepsilon_{it} $$

Y it is the 5-min logarithm of HRV in subject i at time t, b 0 is the overall intercept, and u i is the separate random intercept for subject i. x 1 −x p are covariates measured simultaneously.

A random intercept for each subject, as well as the fixed covariates such as gender, age, BMI, time of day, hear rate, Tp, RHp, and activity were included in the model. We conducted the sensitivity analysis by analyzing two models to examine the potential instability from the covariates selection and the nonlinearity associations between meteorological data (temperature and RH) and HRV indices. In model 1, only PM2.5 was chosen as the fixed effect in the model analysis (with no covariates adjusted). Model 2 included the PM2.5 and the selected linear (gender, age, BMI, time of day, hear rate, Tp, RHp, and activity) and nonlinear terms (quadratic temperature/RH) covariates as fixed effects. The autoregressive-1 (AR-1) was specified as covariance structure in the model analysis. Statistical analyses were performed using the mixed procedure in the SAS (proc mixed in SAS version 9.1; SAS institute inc., Cary, NC, USA). Model selection was based on minimizing AIC.

Several different PM2.5 moving averages (15-, 30-min and 1-, 2-, 4-, 6-h) and the 5-min average PM2.5 at different lag time before every 5-min average HRV measurement were conducted for lag effect evaluation. We estimated the percent change in every HRV parameter for 10 μg/m3 increase of PM2.5 as [10(β × 10) − 1] × 100%, with 95% confidence intervals (CIs) {10[10 × (β ± 1.96 × SE)] − 1} × 100%, where β and SE were the estimated regression coefficient and its standard error (Wu et al. 2010).

Results

All the 30 subjects (12 men and 18 women) took part in the 24-h HRV measurements successfully. The demographic and healthy characteristics are shown in Table 1. The age of the subjects ranges from 51 to 73 years. Almost all the indices in Table 1 are within the normal reference range, which ensure the stable baseline of health status. The cholesterol and triglycerides levels (two additional indices) of 5 individuals were beyond the normal reference range (cholesterol, 3.38–5.72 mmol/l; triglycerides, 0.39–1.63 mmol/l).

Table 1 Personal characteristics of study subjects (n = 30)
Full size table

The Holter measurements in our study on average covered 22 h (11.8 h were in wakefulness), in which 650 5-min HRV and 335 5-min PM2.5 data were missed, and 3,265 data were used in the model analysis. Distributions of exposure variables in indoor and outdoor environments are shown in Table 2. The values are expressed by Median (1–99% percentile) because of the skewed distribution. The fluctuation of the 5-min average PM2.5 is wide, which ranges from 0.18 to 255.39 μg/m3. The median of 5-min average PM2.5 is higher when the subjects are indoors than when they are outdoors (45.58 vs. 37.52 μg/m3, Wilcoxon test, p = 0.001). Table 3 shows distributions of the 5-min average HRV frequency indices in different places (indoor/outdoor). After the exclusion of the noises, the samples of 5-min Holter recordings of the subjects in the model analysis were 116 ± 34. HF is higher when the subjects are indoors than outdoors (HF: median 231.0 vs. 154.0 msec2, p < 0.001), and LFHFR is much lower when indoors (LFHFR: median 1.29 vs. 1.59, p < 0.001).

Table 2 Distributions of exposure variables in the general environment
Full size table
Table 3 Distribution of 5-min average HRV frequency indices in the general environment
Full size table

We used two models to evaluate the effects of 5-min PM2.5 on the simultaneous HRV indices and found slight changes, which showed the stability and reliability of the associations between PM2.5 and HRV indices (shown in Table 4). In model 2, HF and LF increase, respectively by 1.30% (95% CI, 0.16–2.45%) and 1.34% (95% CI, 0.38–2.30%) per 10 μg/m3 increase of PM2.5 in the pooled data analysis (combining the indoor and outdoor data). If we separate the indoor and outdoor periods, 10 μg/m3 increases of PM2.5 associate with the 1.67% (95% CI, 0.38–2.97%) and 1.30% (95% CI, 0.21–2.39%) increases of HF and LF when subjects are indoors; however, no significant associations with HRV indices are found when the subjects are outdoors (p > 0.10). Figure 1 shows the lag effects in model 2 analysis. No prominent lag effects were found between 5-min average PM2.5 at different lag time and HRV indices. In the moving average analysis, HF shows no prominent lag effects, whereas LF showed stronger associations on the 6-h moving average PM2.5. Figure 2 shows the subject-specific effect estimates in model 2 of pooled analysis. For HF, 19 subject special estimated effects are positive, and 11 are negative; while 20 positive and 10 negative for LF. Most of the effect estimates for HF and LF are positive despite the heterogeneity.

Table 4 Estimated effects of 5-min PM2.5 average on 5-min HRV indices in general environments
Full size table
Fig. 1
figure 1

Percent changes of HRV indices per 10 μg/m3 increase of PM2.5 at different lag time intervals. Percent changes (95% CIs) of 5-min HRV frequency indices per 10 μg/m3 increase of PM2.5 at different lag time intervals (from 15 min to 6 h) in model 2 analysis. The square indicates the HRV changes of PM2.5 at different moving averages. The diamond indicates the HRV changes of 5-min average PM2.5 at different lag time

Full size image
Fig. 2
figure 2

Subject-specific effect estimates of HRV indices per 10 μg/m3 increase of 5-min average PM2.5. Subject-specific effect estimates (percent changes with 95% CIs) for 5-min HRV frequency indices per 10 μg/m3 increase of 5-min average PM2.5 in model 2 analysis. The x-axis indicates different subjects (subjects 1–12 are men; subjects 13–30 are women)

Full size image

Discussion

In this study, we measure the exposure to PM2.5 within several minutes and the changes in HRV frequency indices of healthy elderly subjects in general environments. Previous studies focused mostly on the elderly subjects with cardiovascular or pulmonary diseases; however, less attention has been paid to the healthy elderly subjects. We chose the healthy elderly subjects and excluded the confounders of health status and medicine mediations. The real-time micro-environment monitoring design of variables (PM2.5 and the meteorological data) in our study is based on the concurrent indoor/outdoor monitoring and the routine activity diary and may represent a substantial improvement over most studies only relying on central monitors.

Former studies mainly focused on health effects of ambient PM2.5, paying little attention to the PM2.5 in indoor environment. Since the elderly subjects spend most of their time indoors (78.0% of the wakeful period in our study was indoors), it is very important to pay more attention to indoor environment. Our study is the first investigation into the different effects of PM2.5 in the indoor and outdoor environments on HRV.

The frequency domain analysis used in our study has been recommended as a useful tool for achieving a more precise assessment of autonomic function and proved to be capable of describing the autonomic contribution to cardiac oscillation more accurately than time domain measures in short-term HRV recordings. High-Frequency spectral power reflects primarily parasympathetic influences, whereas Low-Frequency power has been shown to reflect both sympathetic and parasympathetic influences. LFHFR represents the relative balance between the sympathetic and parasympathetic nerves (Anonymous 1996; Rajendra Acharya et al. 2006). We chose the frequency domain analysis with focus on HF and LF power, which had also been applied successfully in prior air pollution studies (Holguin et al. 2003; Park et al. 2008).

Many previous epidemiological studies evaluating the associations between PM2.5 and HRV indices reported heterogeneous results. Some studies reported positive associations (Riediker et al. 2004; Magari et al. 2002; Gong et al. 2003), whereas others reported the negative (Schwartz et al. 2005; Gold et al. 2000; Pope et al. 2004) or zero association (Sullivan et al. 2005; Barclay et al. 2009). In our study, PM2.5 leads to the increase of HRV frequency indices of the healthy elderly subjects in pooled data analysis, which is consistent with some previous epidemiological and animal experimental studies (Godleski et al. 2000; Elder et al. 2007; Tankersley et al. 2004) (shown in Table 5). Studies on the mechanisms support the positive associations between PM2.5 exposure and HRV indices. The PM2.5 depositing in lung may stimulate the broncho-pulmonary C-fibers afferents and mediate the cardio-pulmonary parasympathetic reflex, which induce the shift toward predominance of parasympathetic stimulation indicted by the increased HRV indices (Coleridge and Coleridge 1984; Lee and Pisarri 2001).

Table 5 Short-term increased effects of fine particulate on hear rate variability—observational and animal studies
Full size table

We also noticed that some other studies reported negative associations between ambient PM2.5 and HRV indices. The reasons for the discrepancy may be as follows:

  1. 1.

    Many previous studies focusing on elderly participants with compromised health status reported decreased HRV (de Hartog et al. 2009; Wheeler et al. 2006; Lipsett et al. 2006; Chuang et al. 2005); however, we found no report about the study on the health elderly subjects in the general environments. The health status may be the most important modifier of the PM2.5 mediated HRV response. Wheeler et al. (2006) reported increased effects of PM2.5 on HRV for the panel of COPD subjects (n = 18), but no associations for the MI subjects panel (n = 12) (Wheeler et al. 2006). Therefore, the results in our study, which focused on the healthy subjects, cannot be compared directly with that of other published studies on subjects with compromised health. Our study provides the first evidences for the associations between PM2.5 and HRV in healthy elderly subjects.

  2. 2.

    The different analysis methods of HRV in air pollution studies (long-term 24-h average HRV indices measurements and short-term 5-min special protocol testing, etc.) may lead to different inferences of PM2.5 effect on HRV. The several minutes frequency domain indices are not perfectly synonymous to the 24-h measures (Anonymous 1996). Our study focused on the acute effects of PM2.5 several minutes exposure on HRV indices. To allow for the maximal comparability between exposure and outcome variables, we chose the 5-min short-term epoch analysis instead of the 24-h period measurements.

  3. 3.

    Different compositions of the PM2.5 may also be responsible for the heterogeneous associations. Magari et al. (2002) reported that the increases in the HRV were associated with lead and vanadium in the PM2.5, whereas no statistical associations were found with nickel concentrations in the 39 boilermakers (Magari et al. 2002). Riediker (2007) also reported that the proportion of different metal components and organic components induced different effects of PM2.5 on HRV indices in 9 male highway patrol troopers (Riediker 2007). Timonen et al. (2006) reported the negative, positive and zero associations between PM2.5 and HF, respectively, in subject panels of Helsinki Erfurt and Amsterdam and suggested that the sources of PM2.5 were the underlying modifiers (Timonen et al. 2006). Though no studies investigate the different PM2.5 compositions in Beijing and other study centers, different compositions may still be the underlying mechanisms, which need further studies.

Our study shows the positive associations between PM2.5 and HRV in indoor environment when separating the indoor and outdoor environments analysis. To our knowledge, no study has been conducted on the associations between PM2.5 and HRV in indoor environment. Since the subjects spent most of daily time indoors, the positive associations in the pooled data analysis may possibly come from the indoor environment. The associations between PM2.5 and HRV in indoor environment in our study cannot be compared directly with previous studies that observed the effects of ambient PM2.5 on HRV indices. In our study, the gas stove cooking and the re-suspension by indoor activity (cleaning, walking, etc.) are the major sources of indoor PM2.5, which is different from outdoor PM2.5 in sources and components (Pekey et al. 2010; Afshari et al. 2005).

No significant associations are found between PM2.5 and HRV of the healthy elderly subjects when the subjects are outdoors, which is consistent with some previous studies about ambient PM2.5 and HRV indices of the unhealthy elderly individuals (Sullivan et al. 2005; Barclay et al. 2009). The absence of the association in our study may be the result of the higher HRV levels of the healthy elderly subjects and the lower PM2.5 exposure level when the subjects are outdoors. We also noticed the negative estimates of coefficient in model 2 analysis of outdoor PM2.5 on HRV (p > 0.10). The insignificant associations may be caused by potential impreciseness in the exposure evaluation of outdoor PM2.5 conducted on the rooftop of the residential building and the smaller number of outdoor HRV measurements (n = 735).

On the other hand, the differing associations in indoor and outdoor environment for the subjects may be caused by the differences in PM2.5 concentrations and compositions and the different HRV baseline status. Our study shows the lower HF levels when the subjects are outdoors than indoors, which may be associated with the more exercise in outdoor environment (Arai et al. 1989). The detailed mechanisms still need further studies.

The dysfunction of the cardiac ANS, expressed both by the increased HRV indices (de Bruyne et al. 1999) and decreased HRV indices (Tsuji et al. 1996; Thayer and Lane 2007), are important mechanisms of cardiovascular diseases morbidity and mortality. Stone et al. (1999) have described two possible mechanisms in which PM2.5 increases cardiovascular risk by affecting HRV. They suggest that the adverse effects of PM2.5 can be mediated either through a sympathetic stress response which leads to the decreased HRV and tachyarrythmias or through a vagal (parasympathetic) response which leads to increased HF, increased HRV, and bradyarrythmias (Stone and Godleski 1999). Increased HRV (especially for the indices representing the parasympathetic activity) are reported to be associated with sinus node dysfunction (Bergfeldt et al. 1995) and the vagal (parasympathetic) response, which may predispose individuals to potentially fatal arrhythmias (Kasanuki et al. 1997) and paroxysmal atrial fibrillation (Bettoni and Zimmermann 2002).

Many previous studies find effects of particles on HRV within hours of exposure (Schwartz et al. 2005; Park et al. 2005), but it is less clear whether PM2.5 can exert such an acute effect of several minutes exposure on the autonomic nervous system. The 5-min concurrent and continuous monitoring of exposure and outcome variables in our study is suitable for the transient effects analysis. However, due to the great loss of valid data in the lag effects evaluation,the associations between PM2.5 and HRV in the lag effect evaluation cannot be compared directly with the associations between concurrent 5-min PM2.5 and 5-min HRV indices. Our study can only provide limited evidence for the lag effects of PM2.5 on healthy elderly subjects, which still need further studies. The heterogeneous effects of PM2.5 among subjects shown in the subject-specific evaluation may because of the gene polymorphisms (Ren et al. 2010).

Our study had some limitations. Although we had rather stringent inclusive criteria for the study subjects, they were not a strictly randomly selected sample group. In addition, the diary intervals were on a 30-min basis, which may also introduce potential misclassifications. The study was carried out during the Olympic Games in Beijing when the traffic was restricted strictly. The characteristics of the air PM2.5 in this period were lower concentrations and fewer traffic-related sources. The potential influences of the changed concentrations and components in the PM2.5 on HRV induces need further studies.

Conclusions

In conclusion, our findings provide the evidence that several minutes PM2.5 exposure increases HRV of the healthy elderly subjects, which may increase the cardiac risks. Prominent effect of PM2.5 on HRV is found when they are indoors, while the effect is not obvious in outdoor environment.