Traffic-Related Air Pollution and Acute Changes in Heart Rate Variability and Respiratory Function in Urban Cyclists

1Cycling is often promoted as a means of reducing traffic congestion and air pollution in urban areas. Indeed, exposure to traffic-related air pollution is known to contribute to adverse respiratory and cardiovascular outcomes, and even modest reductions in air pollution levels resulting from a shift from automobiles to bicycles may have important public health benefits. Nevertheless, individual commuter cyclists may still be exposed to increased levels of air pollution because they are often in close proximity to vehicle emissions. In particular, increased minute ventilation during cycling may result in substantial increases in inhaled doses of fine (≤ 2.5 μm in aerodynamic diameter) particulate matter (PM2.5) and ultrafine (≤ 0.1 μm in aerodynamic diameter) particles (UFPs) because vehicle emissions are a major source of these pollutants. Furthermore, although few studies have examined the acute health effects of air pollution exposures among cyclists, time spent cycling in traffic was associated with an increased risk of myocardial infarction in a case-crossover study conducted in Germany.

2In the present study we examined the relationship between traffic-related air pollutants and acute changes in heart rate variability (HRV). Decreased HRV has been associated with increased risks of cardiovascular morbidity and mortality, and in general, HRV provides information on the sympathetic- parasympathetic balance of cardiac autonomic modulation (; Task Force 1996). We also examined lung function and exhaled nitric oxide (FENO) and collected on-bicycle personal exposure measures to minimize potential exposure measurement error that may mask the relationship between traffic-related air pollutants and health outcomes.

3Participants and study design. Participants were recruited through advertisements posted across the city of Ottawa, Canada. Eligible participants were healthy nonsmoking men and women who were not exposed to tobacco smoke in the home and did not take medications for preexisting cardiovascular or respiratory conditions.

4Study days were scheduled on weekdays between 1030 hours and 1530 hours, and one to two subjects participated on each day. Cycling took place between approximately 1130 hours and 1230 hours, and study days for individual participants were scheduled at least 5 days apart. Participants were asked to take the same route and mode of transportation to and from the study site on each of their three visits. Once at the study site, participants were fitted with a digital Holter monitor and then rested quietly in a sitting position for 30 min. Questionnaires were completed to collect demographic information as well as information on medical history (e.g., allergies, recent medication use), recent illness, alcohol and caffeine consumption, and recent exposure to environmental tobacco smoke. Baseline HRV and respiratory measures were collected immediately after the 30-min rest period. After the rest period, participants were randomly assigned to cycle for 1 hr along a high-traffic route, along a low-traffic route, or indoors. Cycling routes are described in the Supplemental Material (http://dx.doi.org/10.1289/ehp.1003321). Follow-up health measures were collected immediately after cycling and every subsequent hour for 3 hr (i.e., 1–4 hr after the start of cycling). Participants rested quietly in a sitting position for the duration of the postcycling period. The Health Canada Research Ethics Board approved the study protocol, and written informed consent was obtained from each participant.

5Exposure monitoring. Real-time UFP, PM2.5, and black carbon (BC) data were collected using instruments mounted in panniers located on participants’ bicycles. Tubes attached to air intakes of each instrument were run along bicycle frames toward the handlebars in order to collect measurements as close as possible to participants’ breathing zones. Volatile organic compound (VOC) monitors were located on technician’s bicycles traveling immediately in front of participants along each route. One-hour average pollutant concentrations were calculated along each route and were used for the health analyses. In addition, 1-hr average ambient ozone (O3), nitrogen dioxide (NO2), and sulfur dioxide (SO2) data were collected from a fixed monitoring station in downtown Ottawa for the duration of each cycling period. A detailed description of exposure monitoring methods is provided in the Supplemental Material (http://dx.doi.org/10.1289/ehp.1003321).

6Heart rate variability. Electrocardiograms were continuously recorded for the duration of each study day (1030 hours to 1530 hours) using three-channel (seven-lead) digital Holter monitors (Seer Light Extend; GE Medical Systems Information Technologies, Inc., Milwaukee, WI, USA). Data were subsequently analyzed on a GE Medical Systems Information Technology workstation (MARS version 7.2). Time-domain [standard deviation of normal-to-normal (NN) intervals (SDNN), root mean square of successive differences (RMSSD) in adjacent NN intervals, and percentage of adjacent NN intervals differing by > 50 msec (pNN50)] and frequency-domain [low-frequency power (LF; 0.04–0.15 Hz) and high-frequency power (HF; 0.15–0.40 Hz)] measures of HRV were determined for the last 5 min of each segment of the study day: the rest period and 1–4 hr after the start of cycling. HF, RMSSD, and pNN50 reflect parasympathetic modulation of the heart, whereas SDNN reflects total power and LF reflects a mixture of both parasympathetic and sympathetic modulation (; Task Force 1996). The LF:HF ratio is thought to reflect the balance of sympathetic and parasympathetic modulation. Heart rate was also determined for each interval. All of the 5-min segments analyzed contained at least 200 beats, and 92% contained more than 300 beats. Five-minute segments used for HRV analyses occurred before the collection of spirometry and FENO measures during periods of quiet rest. Beat annotations were automatically assigned by the software and were verified and reviewed by trained technicians at the Arrhythmia Monitoring Center in Ottawa who were blind to exposure status. Beats and intervals were reviewed by trained technicians and only normal sinus beats were used. If the last 5 min of a segment was not suitable for use (< 90% valid recordings), the previous 5-min window was analyzed.

7Respiratory outcomes. KoKo Legend spirometers (nSpire Health, Longmont, CO, USA) were used to measure forced expiratory volume in 1 sec (FEV1), forced vital capacity (FVC), and forced expiratory flow at 25–75% of vital capacity (FEF25–75) according to criteria. Niox Mino instruments (Aerocrine, New Providence, NJ, USA) were used to measure FENO; all FENO measures were collected before spirometry procedures. All respiratory measures were collected by trained technicians, and each subject was assigned to the same instrument and technician on all three visits.

8Statistical analysis. Linear mixed-effects models were used to examine the relationship between traffic-related air pollutants and changes in health measures 1–4 hr after the start of cycling relative to baseline values measured before cycling. All models included a random intercept for subject to account for correlations between repeated measures collected from individual participants. Separate models were analyzed for each time period, and distributions for baseline changes in health outcomes were approximately normally distributed. Analyses were first conducted using cycling site as the primary exposure variable, with indicator variables used for the high- and low-traffic routes and the indoor site serving as the reference location. Single-pollutant models were then analyzed using continuous measures of mean pollutant levels during each cycling period (UFPs, BC, PM2.5, O3, NO2, and total VOCs). Covariates (age, sex, body mass index, allergies, asthma, ambient temperature during cycling, relative humidity during cycling, day of the week, average heart rate in the hour before HRV measures, average heart rate during cycling, and caffeine or alcohol consumption in the previous 24 hr) were examined one at a time, and only ambient temperature and average heart rate during cycling were retained in final models because these were the only factors that had a meaningful impact on model coefficients (change > 10%). None of the participants reported spending time in a smoking environment at home or elsewhere. Model fit was improved slightly when analyses were conducted using log-transformed data, but the direction of observed associations did not change, and the interpretation of our findings remained consistent. Therefore, nontransformed analyses are presented for ease of interpretation. Models were also examined using percent change in health outcome as the dependent variable, but model fit tended to be better for models using absolute changes, and thus these models are presented.

9Indoor data were excluded from models for O3 and NO2 because indoor measures were not available for these pollutants. In addition, analyses for total VOCs were conducted for all data combined as well as excluding indoor data because total indoor VOCs represent a different mixture than total outdoor VOCs. Analyses were not conducted for SO2 or carbon monoxide (CO) because pollutant levels were low and little variation was observed between study days. If an association was detected in single-pollutant models, two-pollutant models were analyzed to examine the stability of model coefficients with the addition of other air pollutants. For O3, NO2, and total VOCs, two-pollutant models were limited to outdoor data for reasons mentioned above. Separate HRV models were analyzed for UFPs, BC, and PM2.5 to include an interaction term for O3 because some findings suggest that O3 may modify the relationship between PM air pollution and HRV. Centered variables were used to generate interaction terms. Finally, analyses were conducted separately (i.e., stratified analyses) for the high- and low-traffic routes using multivariable linear regression models to evaluate potential confounding by unmeasured route-specific factors. Outcomes and covariates in these models were the same as those in mixed-effects models. Sensitivity analyses were conducted to evaluate the impact of potential outlying values on observed associations, and residual diagnostics were performed to verify model assumptions. All analyses were conducted using STATA (version 11; StataCorp LP, College Station, TX, USA). Model coefficients for specific air pollutants reflect interquartile range (IQR) increases (25th to 75th percentile) in air pollutant levels. An α-value of 0.05 indicated statistical significance.

10Forty-two participants took part in the study over 71 days between May and September 2010. Average ambient temperature and relative humidity values in Ottawa ranged from 16°C to 36°C and from 18% to 80%, respectively, during the study. Participants ranged in age from 19 to 58 years (mean = 35 years) and were predominantly Caucasian (95%), male (67%), and of healthy body weight (mean body mass index, 24.3 kg/m2; range, 19.2–41.1 kg/m2). Twenty-six participants reported having allergies, and 14 reported asthma diagnosis.

11In total, 118 of 126 possible routes were completed throughout the study period, with 38 participants completing all three cycling routes and four participants completing a single route (two low traffic, one high traffic, one indoors). Only 1 hr of Holter monitor data, 2 days of PM2.5 data, and 2 days of FENO data were lost to technician error or malfunctioning equipment during the study. Air pollution exposure data was complete for UFPs, BC, VOCs, and ambient O3, NO2, and SO2 (n = 118).

12Table 1 summarizes baseline health characteristics for study participants, which are comparable to measures reported in other studies of healthy adults. In addition, baseline values were similar before cycling in each separate location [for cycling location-specific baseline values, see Supplemental Material, Table 1 (http://dx.doi.org/10.1289/ehp.1003321)]. On average, mean UFP, BC, and PM2.5 concentrations were significantly greater for the high-traffic route than for the low-traffic and indoor locations (Figure 1; for numeric data, see Supplemental Material, Table 2). Average total VOC levels were significantly greater on the high-traffic route than on the low-traffic route, but indoor VOC levels exceeded those for both outdoor cycling locations. Ambient O3, NO2, and SO2 levels were generally low throughout the study period, and levels were not significantly different between high- and low-traffic cycling days (see Supplemental Material, Figure 1). Correlations between air pollutants at each cycling location were low to moderate, with the strongest correlation observed between PM2.5 and CO on the high-traffic route (r = 0.75; see Supplemental Material, Table 3). A single compound, limonene (a citrus-scented compound used in cleaning agents), was responsible for approximately 25% of total indoor VOC levels (see Supplemental Material, Table 4). As a result, we excluded indoor data from health analyses for total VOCs because total indoor VOCs represented a different mixture than total outdoor VOCs.

13Heart rate variability. Analysis by cycling site. HF power was decreased after cycling on the high-traffic route relative to indoors, with a significant decrease observed at the 4-hr time point (Table 2). LF:HF was significantly increased 2 and 3 hr after the start of cycling on the low-traffic route relative to indoors but not on the high-traffic route (Table 2). We identified one influential data point in the analysis of HF, as one participant experienced a large decrease at the 4-hr time point after cycling on the high-traffic route. When we removed this value, the direction of the observed association with high-traffic versus indoor cycling remained consistent but the magnitude decreased and was no longer statistically significant [β = –141; 95% confidence interval (CI), –302 to 20.6]. Cycling site was not associated with significant changes in other measures of HRV.

14Analysis by specific air pollutants. We observed consistent inverse relationships between an IQR increase in UFPs and HF, with significant decreases observed at the 2- and 4-hr time points (Table 3). UFPs were also inversely associated with pNN50, with a significant decrease observed 2 hr after the start of cycling (Table 3). We observed nonstatistically significant inverse associations between BC and HF, but these relationships disappeared when we included UFPs with BC in two-pollutant models, whereas coefficients for UFPs remained stable (data not shown). We identified one outlying data point in the analysis of UFPs and HF at the 2-hr time point, and when we removed this point the observed association was no longer statistically significant (β = –86.0 msec2; 95% CI, –201 to 28.9 msec2). We also identified one outlying data point in the analysis of UFPs and HF at the 4-hr time point, but the observed association remained statistically significant when we removed this point (β = –183 msec2; 95% CI, –307 to –59.4 msec2). Further adjustment for other air pollutants in two-pollutant models did not change the observed association between UFPs and HF or pNN50.

15IQR increases in ambient NO2 levels were associated with significantly increased LF:HF at the 2- and 3-hr time points (Table 4), and these relationships remained stable when adjusted for other air pollutants (data not shown). BC was also associated with a significant increase in LF:HF 3 hr after the start of cycling (Table 3), but this relationship decreased and was no longer statistically significant when we included NO2 in the model (data not shown). Ambient NO2 levels were associated with decreased SDNN 1–4 hr after the start of cycling, with a significant decrease observed at the 2-hr time point (Table 4). This relationship remained stable when we included other air pollutants in the model.

16We observed significant inverse relationships between O3 and RMSSD and pNN50 at the 3-hr time point (Table 4). Further adjustment for other air pollutants did not change these associations, and O3 did not appear to modify effects of UFPs, BC, or PM2.5 in the analysis of HRV (p > 0.05 for interaction terms; data not shown). Total VOC exposures during cycling were not associated with HRV (Table 4). In general, the observed associations remained consistent when we conducted the above analyses separately for the high- and low-traffic routes (data not shown).

17Respiratory outcomes. We did not observe strong relationships between traffic-related air pollution and acute changes in respiratory outcomes [see Supplemental Material, Tables 5–7 (http://dx.doi.org/10.1289/ehp.1003321)]. PM2.5 was associated with a small increase in FENO 2 hr after cycling (β = 1.1 ppb; 95% CI, 0.08–2.2 ppb), and UFPs were associated with increased FEF25–75 at the 1-hr time point (β = 191 mL; 95% CI, 10–371 mL). BC, O3, and total VOCs were not associated with significant changes in respiratory outcomes, but NO2 levels were associated with significantly increased FEV1 2 hr (β = 121 mL; 95% CI, 27–216 mL) and 3 hr (β = 129 mL; 95% CI, 23–234 mL) after the start of cycling.

18Our findings suggest that short-term exposure to traffic-related air pollution may contribute to changes in the autonomic regulation of the heart in the hours immediately after cycling. Specifically, UFP exposures were inversely associated with HF and pNN50, O3 was inversely associated with RMSSD and pNN50, and NO2 was inversely associated with SDNN and positively associated with LF:HF. In general, these findings suggest decreased parasympathetic modulation of the heart in response to traffic-related air pollution. Although some findings may have occurred by chance owing to multiple comparisons, consistency in the direction (i.e., reduced parasympathetic modulation) of observed associations for multiple HRV measures tends to suggest a true association. As in previous panel studies of healthy cyclists, we did not observe strong associations between traffic-related air pollution and acute changes in respiratory outcomes.

19We did not identify other studies examining the relationship between HRV and air pollution exposures among cyclists; however, 1- to 4-hr moving averages of personal UFP exposures were associated with decreased HF, LF, RMSSD, and SDNN in a panel of young adults and elderly subjects in Taiwan. Although our findings for UFPs and HF are consistent with these results, we did not observe significant associations between UFPs and LF, RMSSD, or SDNN; however, we observed consistent inverse relationships for RMSSD. One explanation for this discrepancy may be differences in PM composition, because this study specifically targeted exposures in traffic, whereas monitored exposures throughout normal daily activities. As in this study, reported associations between NO2 and LF:HF, but they did not observe a significant inverse relationship between UFPs and HF. Likewise, other studies have also failed to detect significant relationships between UFPs and HRV; however, non-statistically significant inverse relationships between UFPs and HF were reported in both of these studies. One explanation for this discrepancy may be increased exposure measurement error owing to the use of fixed-site UFP monitors in some of these studies. Alternatively, differences in population characteristics (i.e., healthy subjects vs. elderly subjects or those with cardiovascular disease) and UFP exposure averaging times may also contribute to discrepancies between studies.

20A number of studies have reported inverse associations between HRV and ambient PM2.5, BC (or elemental carbon), organic carbon (OC), NO2, and O3, particularly among elderly subjects and those with preexisting cardiovascular disease. Although our findings for NO2 and O3 are generally consistent with previous evidence suggesting decreased HRV with increased exposure, we did not observe a significant relationship between total VOCs (organic carbon) and HRV as previously reported. Specifically, recent studies have observed significant inverse associations between OC and HF, RMSSD, and pNN50 and between OC and SDNN, RMSSD, LF, and HF. However, both of these studies were conducted among patients with preexisting cardiovascular disease, so disease status may be an important effect modifier in the relationship between OC and HRV.

21We did not observe significant associations between BC or PM2.5 and HRV. This is consistent with previous findings from a controlled crossover study of diesel exhaust exposure but is in contrast to previous studies of traffic-related air pollution. However, time spent in traffic 2 hr before HRV measures was the strongest predictor of decreased HF and RMSSD in one of these studies, and our findings for HF and high-traffic cycling generally support this result. Other studies have also reported decreased HRV in response to short-term traffic-related exposures; thus, it seems plausible that decreased HRV may play a role in previously reported associations between short-term traffic exposures and onset of myocardial infarction.

22Although the use of personal exposure measures, real-life cycling conditions, and a crossover design were important strengths of this investigation, it is important to recognize several limitations. First, we could not control for personal exposures experienced en route to the study site, so it is possible that these exposures might have influenced the results. To address this concern, we asked participants to take the same route and mode of transportation to the site each day. In addition, the use of changes from baseline in health measures as the primary outcome likely minimized the impact of precycling exposures because outcomes measured at hourly intervals throughout the day were likely affected similarly by exposures en route to the study site. Second, we could not adjust for the effects of respiration on HRV, a factor that is particularly important for HF because this measure is sensitive to respiratory variation (Task Force 1996). However, we controlled study conditions as much as possible and asked participants to rest quietly in a sitting position for the duration of pre- and postcycling periods. Although we cannot rule out potential confounding by unmeasured respiration effects, such effects would have to be correlated with air pollution levels along cycling routes in order to confound the results. Third, our findings are based on a relatively small sample of subjects, and as a result, measures of association were often imprecise and may not be generalizable to the broader population. In addition, we had limited power to detect potentially important interactions between air pollutants such as those previously reported between PM air pollution and O3. Fourth, we did not examine the impact of traffic-related air pollution exposures beyond 4 hr after the start of cycling, and we did not have data on other cardiovascular measures such as ST-segment depression (indicative of cardiac ischemia) and deceleration capacity that have recently been associated with air pollution. Because we detected significant relationships 4 hr after the start of cycling, it seems plausible that the observed effects might have continued over longer time periods, and future studies should evaluate this possibility. Finally, personal exposure measures were not available for NO2 and O3, and we assigned each subject who participated on the same study day the same ambient concentration; therefore, exposure measurement error likely resulted in underestimation of effect estimates for these pollutants.

23To our knowledge, this is the first study to examine the relationship between traffic-related air pollution exposures and acute changes in HRV among cyclists. Although a number of studies have observed inverse associations between ambient air pollution and HRV in elderly populations and/or subjects with preexisting cardiovascular disease, our findings suggest that short-term exposure to traffic may have a significant impact on cardiac autonomic function in healthy adults. The clinical implications of this relationship are unclear because the health benefits of cycling might outweigh any potential negative impact that acute reductions in HRV may have in healthy adults. Indeed, altered HRV may be a single component of a more complex causal pathway that requires a series of other cofactors such as endothelial dysfunction and individual susceptibilities (i.e., genetic predisposition, health status) to ultimately result in cardiovascular morbidities associated with air pollution (, 2010). However, our findings suggest that, when possible, it may be prudent to select cycling routes that reduce exposure to traffic and to avoid cycling outdoors or to exercise indoors on days with elevated air pollution levels. In addition, the planning of new cycling routes/bicycle paths in urban areas should aim to minimize time spent in high-traffic areas in order to reduce exposures of recreational riders who may be more susceptible (e.g., elderly) to the acute cardiovascular health effects of traffic-related air pollution.