Study population. We recruited 76 participants who were referred by their cardiologist to the University of Rochester Cardiac Rehabilitation Center (CR Center) after having a recent coronary event (MI or unstable angina). We excluded participants with cardiomyopathy in the absence of coronary disease, coronary bypass grafting within the last 3 months, type 1 diabetes, chronic atrial fibrillation, anemia, left bundle branch block, presence of a prosthetic heart valve or pacemaker, regular use of amiodarone, and active smokers or nonsmokers living with an active smoker. Participants lived within 19 km of the particle monitoring site at the CR Center (median of 9 km) and within 21 km of the New York State Department of Environmental Conservation (NYS DEC) particle monitoring site (median of 9 km).
Study protocol. Each participant underwent a 10-week supervised exercise program (≤ 20 exercise sessions) at the CR Center from June 2006 to November 2009. At each visit, participants were in the CR Center for 30–60 min before exercising and then seated for ≥ 5 min for blood pressure and electrocardiogram (ECG) recording. They warmed up for 2–5 min, which included gentle stretching, and then exercised for 30–45 min using a bicycle, treadmill, or rowing machine. After a “cool down” period, the participants rested for 10 min. Exercise modality was determined by the patients as part of their clinical rehabilitation program and not as part of the study. The same modality was used throughout the 10-week study period. The 76 participants and the data that were collected at their 1,489 participant visits were used in all analyses.
At each visit, participants underwent 3-lead (modified V2, V5, and AVF) Holter ECG recordings (Burdick Altair-Disc holter recorder; Cardiac Science, Bothell, WA), which were analyzed using the Vision Premier Burdick Holter System (Cardiac Science) and custom-made programs at the University of Rochester Medical Center, which have been described previously in our air pollution studies (Bauer et al. 2008
; Cygankiewicz et al. 2008
). All study Holters were annotated first automatically by the commercial Holter scanning algorithm (Vision Premier Burdick Holter System) and then annotated by a trained technician using standard procedures. RR intervals were exported to a custom made heart rate variability (HRV) program that produced a set of HRV parameters. Heart rate turbulence (HRT) and deceleration capacity (DC) were analyzed using programs adopted from Bauer et al. (2006)
and from Schmidt et al. (1999)
. During both the preexercise resting period (~ 5 min seated) and for the entire recording (whole session was ~ 1–3 hr), we measured time domain HRV parameters including the mean normal-to-normal (NN) interval time between successive NN beats (MeanNN), the standard deviation of all NN beat intervals (SDNN), and the square root of the mean of the sum of squared differences between adjacent NN intervals (rMSSD). Short-term, preexercise, resting recordings provided information regarding HRV parameters unaffected by sympathetic stimuli during exercise, whereas the whole session recording (including the exercise session) reflected the overall behavior of heart rate and autonomic responses to daily conditions including the exercise. Based in part on Bigger et al. (1992)
, filtering criteria eliminated two RR intervals after premature ventricular or atrial beats. We did not apply preprocessing filtering to eliminate extreme values. We examined 5-min segments during the resting period to standardize conditions for all HRV and repolarization parameters, which required at least 200 beats for HRV analyses. As a postprocessing approach, we evaluated outliers and determined whether the values were valid or not based on intralab ranges developed during a prior study (Schneider et al. 2010
Across the whole session, we measured HRT and DC. HRT, a measure of baroreflex sensitivity (Bauer et al. 2008
; Cygankiewicz et al. 2008
) is characterized by a brief acceleration and subsequent deceleration of heart rate following a spontaneous premature ventricular contraction. HRT is described by two parameters: turbulence onset and turbulence slope, and is associated with increased risk of cardiac death (Bauer et al. 2008
). We focused on turbulence slope because previously this parameter was found to be more robust than turbulence onset in identifying participants with increased risk of cardiac events (Cygankiewicz et al. 2008
). Because only 657 of the 1,489 recordings (from 72 of the 76 participants) had at least one premature ventricular beat (a mean of 126 ventricular ectopic beats per recording), we performed HRT analyses on a subset of recordings. Otherwise, all other outcomes were measured in all recordings. DC is an additional measure of heart rate dynamics, which reflects the variability in heart rate during periods when the heart is slowing down, complementing information based on the other HRV and HRT parameters (Bauer et al. 2008
). Repolarization duration was analyzed using the QT interval duration, which was measured manually (i.e. technician evaluated 3 consecutive beats within each prespecified 2-min period from the beginning of the resting ECG in lead II, taking the average QT for each time-point), and corrected for heart rate (QTc) using Bazett’s formula (Bazett 1920
). In addition, we also measured the Tpeak–Tend (TpTe), a measure of late repolarization duration, which may reflect heterogeneity in repolarization (Szydlo et al. 2010
Blood pressure measurements were collected at each visit, and an atraumatically drawn venous blood sample was collected once weekly while the participant was resting and sitting before exercise. Blood pressure was measured by auscultation following a 5-min resting period with the arm supported at heart level. The blood pressure was measured three times as part of the research protocol, with an average used in the statistical analyses. Complete blood count, fibrinogen, and high sensitivity CRP analyses were measured in the Strong Memorial Hospital Clinical Laboratories (University of Rochester Medical Center, Rochester, NY). The study was approved by the Research Subjects Review Board of the University of Rochester, and informed written consent was obtained from all participants.
Air pollution and weather measurements.
Particle size distributions for ultrafine particles (UFP; 10–100 nm diameter) and for accumulation mode particles (AMP; 100–500 nm diameter) were measured using a wide range particle spectrometer (model 1000XP; MSP Corporation, Shoreview, MN) at the CR Center from June 2006 to November 2009. The measured distributions have been summarized by Wang et al. (2010)
. Although the size range of the AMP is different from the standard 100–1,000-nm definition, the majority of particles in this size range are smaller in mass and closer to the 100-nm cut-off value. Therefore, the use of the 100–500-nm size range to define AMP rather than 100–1,000 nm should result in minimal difference in AMP concentration. This monitoring facility in Rochester is approximately 1,500 m from an interstate highway beltway. Concentrations of PM2.5
were measured using a tapered element oscillating microbalance (ThermoFisher, Franklin, MA) at the NYS DEC site in Rochester (~ 1.2 km from the CR Center). Hourly temperature, relative humidity, and barometric pressure were also measured at this same site.
Statistical analysis. We estimated the difference in each outcome (preexercise resting period: MeanNN, SDNN, rMSSD, QTc, and TpTe; whole session: MeanNN, SDNN, rMSSD, HRT, and DC; preexercise measurement: CRP, fibrinogen, white blood cell count, diastolic blood pressure, and systolic blood pressure) associated with each interquartile range (IQR) increase in pollutant concentration (UFP, AMP, PM2.5). Data were analyzed using mixed models, with the participants entered as random effects (version 9.2; PROC MIXED; SAS Institute Inc., Cary, NC). All analyses controlled for visit number, calendar time since the beginning of the study for each participant, month of year, and hour of day. Before estimating effects of the pollutants, we performed some initial analyses to select the appropriate correlation structure. For most outcomes, the compound symmetry covariance structure outperformed other structures examined (autoregressive and spatial power) according to the AIC criterion. Thus this structure was used in all subsequent analyses. Before assessing the effects of the particulate pollutants of interest, we examined other possible confounders including temperature, barometric pressure, relative humidity, sulfur dioxide, carbon monoxide, and ozone. To ensure comparability of analyses for the different outcomes, the same statistical model was used for each. Because only temperature showed effects that were frequently statistically significant, strongest at lag 0, this variable was included in all analyses.
Before the analysis, we log-transformed UFP and AMP to reduce skewness. We estimated changes in each outcome associated with each of the pollutant measures in separate analyses, using pollutants averaged over the 24-hr period before the visit as well as a shorter lag period (lag hr 0–5) and longer lag periods (lag hr 24–47, 48–71, 72–05, 96–119). We present the change in each outcome [and its 95% confidence interval (CI)] associated with an IQR increase in pollutant concentration during the specified lag period.
To examine whether a change in an outcome (e.g., increased TpTe) associated with one pollutant (e.g., AMP lagged 24–47) was independent of a second pollutant (UFP or PM2.5) at the same lag time, we used the same model described above (same covariates and correlation structure as single-pollutant model) including both pollutants at the lag time of interest (e.g., AMP and UFP counts lagged 24–47 hr). We then compared the parameter estimates from the single- and two-pollutant models.
We examined residual plots as a check on model assumptions. Where indicated, analyses were repeated after log-transformation of the relevant outcome variables. Statistical significance was defined as p < 0.05.