The study was conducted in patients with CAD in Erfurt, Germany, between October 16th 2000 and April 27th 2001. The city, located 200 m above sea level and mainly surrounded by a 100 m high ridge except towards the North, had a population of approximately 200,000 at that time. These geographic conditions favor wintertime inversions, which result in elevated levels of ambient air pollution. Traffic, heating, energy production, and long-range transport are the major sources of ambient air pollution.
Study participants were recruited through a local cardiologist. They were required to be males aged 50 or more with physician-diagnosed ischemic heart disease, stable angina pectoris or prior MI (more than three months ago). Based on the study objectives, current smokers, individuals with pacemakers, bundle-branch block, type 1 diabetes, recent MI, bypass-surgery or balloon dilatation (less than three months ago) and patients on anti-coagulation therapy were excluded from participation. Of the 61 recruited subjects, 56 met the inclusion criteria for the ECG analyses. Three patients with bundle-branch blocks had to be excluded, as well as one who had constant arrhythmia and one who was not compliant. Written consent was obtained from each subject. The study protocol was approved by the German Ethics Commission "Bayerische Landesaerztekammer".
A total of 12 clinical visits were scheduled for each participant - one every two weeks on the same day at the same time of the day to control for weekly and diurnal patterns of the assessed parameters. Before the first examination a baseline questionnaire was administered to obtain demographic information, health status, pulmonary and cardiac symptoms, medication use, smoking history, exposure to environmental tobacco smoke (ETS), living conditions and indoor and outdoor sources of particulate air pollution. Each visit included a short interview to assess potential confounders and to collect information on current health status and changes in medication. Participants in addition kept a daily diary with information on times spent in smoke-filled rooms and times spent in traffic using a car, bus, tram or taxi.
ECGs were recorded with a 12-lead Mortara H12 recorder (Mortara Instrument, Milwaukee, USA) using a digital sampling rate of 180 samples/sec per channel. The short 20-minute ECG recordings included a 6-minute period of rest in supine position with spontaneous breathing. Every four weeks after the short ECG recordings had been made, the study subjects underwent 24-hour Holter monitoring and were advised to keep activity diaries, which included the time, duration and type of each activity during the monitoring period. Thus, during the six months of the protocol, each subject was supposed to have twelve short and six long Holter recordings paralleled by air pollution monitoring.
The ECG recordings were analyzed at the University of Rochester Medical Center, (Rochester, NY) for computing the HRV parameters and at the First Medical Clinic of the Munich University of Technology and German Heart Center Munich, Germany, for measuring DC. For the analysis, the first minute of the 6-minute period was discarded to avoid carry-over effects from previous minutes. For the 24-hour recordings, the entire recording was included in the analysis. Only normal sinus beats were used to calculate HRV - artifacts and ectopic beats were excluded after scanning and manual editing of the QRS complexes. Then, the tachograms (curves describing the length of the successive RR intervals across the analyzed period) were exported to be processed by computer algorithms providing both time and frequency domain HRV parameters. The list of HRV parameters included in this analysis was computed according to the current standard described in the recommendation of the Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology .
Average heart rate (HR) is expressed in number of cardiac beats per minutes such as 60/RR when RR intervals are measured in seconds. The time domain HRV parameters included SDNN, RMSSD and pNN50 where SDNN is the standard deviation of RR intervals on a given period for all normal-to-normal beats (NN), RMSSD is the root means square of successive differences in RR values, and pNN50 is the percentage of adjacent RR intervals which differ more than 50 ms.
The frequency domain HRV parameters used in the analysis were computed using the power spectral density method of the tachograms based on fast Fourier transformation. The power of the estimated spectrum was measured into the so-called low-frequency (LF, 0.04-1.5 Hz) and high-frequency (HF, 0.15-0.40 Hz) bandwidths. These measurements were normalized (LFn and HFn) with LFn = LF/(TP-VLF) and HFn = HF/(TP-VLF) where VLF represents the energy in the very-low-frequency bands (0.003-0.04 Hz) and TP the power of the total spectrum. LFn and HFn have no unit (n.u.).
For the 5-minute ECGs we analyzed HR, HF, LF and RMSSD. In association with air pollution, only the normalized frequency domain parameters HFn and LFn were used. For the 24-hour ECGs HR, SDNN, RMSSD and pNN50 were computed.
To assess DC from the 24-hour ECGs, the signal processing technique of PRSA was used to process sequences of NN intervals from pre-discharge Holter recordings. The technique provided separate characterization of deceleration-related and acceleration-related modulations, quantified by deceleration capacity and acceleration capacity. Details on the methodology can be found in Bauer et al. (2006). For the analysis with air pollution, only DC was used as Bauer et al. (2006)  showed that mechanisms which slow the heart down are clinically more important than those that speed it up and that DC was suitable as a screening method with high prognostic value.
Air pollution monitoring
The concentrations of ambient air pollutants were measured at a fixed monitoring site representing urban background levels. The Erfurt Aerosol Measurement Site and its equipment have been described in detail elsewhere . Briefly, we sampled the particle size distribution with a mobile aerosol spectrometer for particles of a size range between 0.01 μm and 2.5 μm. Hourly number concentrations for ultrafine particles (UFP, 0.01-0.1 μm) were calculated from the spectra. Moreover, hourly mass concentrations of particles with a size range between 0.01 μm and 2.5 μm (PM2.5) were calculated assuming spherical particles with an estimated mean density of 1.53 g/cm3. Elemental carbon (EC) and organic carbon (OC) were determined hourly from an ambient carbon monitor (5400, R&P, Inc., Albany, NY, USA) only after December 17th 2000. The continuous data on meteorological variables such as temperature, barometric pressure and relative humidity were collected from existing networks . Missing values for UFP and PM2.5, temperature and relative humidity were either replaced based on corrected parallel measurements (temperature, relative humidity) or imputed by semiparametric regression models based on data from other devices (UFP, PM2.5) . The semiparametric model allowed for the inclusion of a smooth function of time to account for temporal variations in the measurements. The Goodness of Fit values were 0.94 and 0.88 for the UFP regression model and for the PM2.5 imputations, respectively. Missing values for OC and EC were not imputed. To determine exposure to air pollutants prior to the recording of the short-term ECG individual 0-5, 6-11, 12-17, 18-23, 0-23 (lag 0), 24-47 (lag 1), 48-71 (lag 2), 72-95 (lag 3), 96-119 (lag 4) and 0-119 (5-day average) hour averages of air pollution parameters were calculated based on the starting time of the recordings. For the 24-hour Holter recordings the time at the end of the recording was taken to determine exposure to air pollutants concurrent and prior to the ECG-recording (lag 0 to lag 4). Average concentrations of the air pollutants were calculated when at least 2/3 of the hourly measurements were available.
Data were analyzed using the statistical package SAS Version 9.1 (SAS Institute Inc., Cary, NC, USA). A descriptive analysis of the characteristics of the study participants was based on data obtained through the baseline questionnaire. Mixed models with a random participant effect and covariance structure "compound symmetry" were used to analyze the association between air pollutants and ECG parameters taking into account the repeated measurements over time for each individual. Since measurements took place two weeks apart no adjustments for autocorrelation were necessary. RMSSD of the short - as well as of the long-term recordings was log-transformed since the residuals were otherwise not normally distributed. Non-parametric smooth functions (penalized splines) were used to explore the shape of the association between confounders such as trend or meteorological variables and the dependent variable. In addition, day of the week was also considered as a potential confounder. Model fit was based on minimizing the Akaike Information Criterion. Models were built for each ECG parameter separately. The selected confounder model for each ECG parameter can be found in the Additional File 1.
As a sensitivity analysis we re-analyzed the effects for PM2.5 and UFP using the same reduced time-period for which EC and OC were available. Moreover, we did a further sensitivity analysis by excluding participants who were taking anti-arrhythmic medication.
For effect modification, we considered body mass index (BMI < 30 kg/m2 vs. ≥30 kg/m2), smoking history (ex-smokers vs. never-smokers) and intake of beta-adrenergic receptor blockers (yes vs. no). Information from the daily diaries about times spent in smoke-filled rooms and times spent in traffic for the times during the ECG as well as the lagged times was coded binary (at least one hour vs. less than one hour) and used for interaction analysis of the long-term ECGs.
Effect estimates are presented as percent change from the mean (geometric mean for RMSSD, arithmetic mean for all other ECG-variables) together with 95%-confidence intervals based on an interquartile range (IQR: difference between the first and the third quartile) increase in air pollution concentration.