The evidence linking air pollution particles with cardiac morbidity and mortality continues to accumulate.1^–3 Susceptibility to these effects is not uniformly distributed and there is interest in subgroups that may be at elevated risk. One such group is people with chronic heart failure (CHF).4^–7 The mechanisms underlying these effects remain unclear. Several hypotheses have been proposed, including effects mediated by local and systemic inflammation, altered coagulation and neurally mediated or direct effects on cardiac rhythm.8 9 Some panel studies have considered specific pathophysiological endpoints, but in general such studies have included relatively few patients with overt cardiovascular disease or taking cardiac medications.10^–13 Thus, there is a case for studying the underlying mechanisms in individuals with established disease, who might be considered to be most susceptible.

This study was designed to investigate mechanistic hypotheses in patients with established heart failure. The aims were to study changes in cardiac rhythm, heart rate variability (HRV), markers of endothelial activation and inflammation, and parameters of coagulation, in relation to pollution. All of these, individually but not collectively, had been suspected or shown to respond to pollution. As well as using area background particle measurements, we estimated personal exposures of study participants. Concurrent personal and area background measurements of nitrogen oxides were obtained, as it has been suggested that nitrogen dioxide (NO₂) may act as a confounder for the effects of exposure to ultrafine particles.14

METHODS

Setting

The study was conducted in Aberdeen, Scotland, a coastal city. The dominant outdoor source of pollution is vehicular traffic, with contributions from ships, aeroplanes and distant transport. Subjects were recruited between January 2003 and July 2004, participating for up to 12 months. All lived within the city and participated from home, being visited by trained research nurses.

Study aims and design

The aims of the study were to investigate two separate adverse health associations with different measures of particulate air pollution: changes in heart rhythm and its autonomically controlled variability, systemic inflammation leading to endothelial activation, increased blood coagulability and red cell sequestration. Simultaneously, a proposed confounding effect of nitrogen oxides was to be investigated.

In a repeated-measures panel study we aimed to recruit 100 patients with documented heart failure (from any cause) and to study each six times at 2-monthly intervals over 1 year. For logistical reasons, the collection of data covered 2 years (January 2003–May 2005). Each individual was requested to keep a detailed activities diary over 3 consecutive days every 2 months and to wear a personal NO₂ monitor continuously over these six 3-day periods. On the third day of each period, the subject wore a 24 h Holter electrocardiogram (ECG) monitor and on the morning of the fourth day a 20 ml blood sample was taken.

Study participants

Subjects were recruited from hospital cardiology clinics. The study protocol was approved by the local ethics committee, and informed written consent was obtained from all participants. Subjects had a clinical diagnosis of stable CHF. There were no exclusions on the basis of underlying aetiology or therapy; patients with pacemakers or with atrial fibrillation were excluded, as were current smokers. Medications and New York Heart Association functional class were recorded at baseline and during each monitoring period. Subjects who completed more than three blocks were invited to attend for echocardiography. Of the 132 participants, 93 completed the study, providing data for all six 3-day monitoring blocks.

Air pollution and weather data

The study employed a combination of fixed-site/urban area daily background pollution measurements and estimates or measurements of personal exposure. Background area data were collected for the parameters listed below.

Particle number concentration

Continuous 24 h particle number concentration was measured by TSI 3934 Scanning Mobility Particle Sizer (SMPS), recording average number concentration in the size range 10–100 nm. When the SMPS required annual service, average number concentration was recorded by TSI P-TRAK ultrafine particle counter Model 8525 (TSI Instruments Ltd., High Wycombe, Buckinghamshire, UK). The two instruments were run in parallel for a period, to enable cross-calibration.

Ambient particle mass, nitric oxide, NO₂ and total oxides of nitrogen

Air pollution data for Aberdeen were obtained from the National Environmental Technology Centre website (the UK national air pollution agency) for all relevant days. Any day with fewer than 90% of the readings was excluded from the analyses. Note that these were measured in the city centre while particle number concentration (PNC) was measured 1 mile away at a background site.

Personal exposure assessments

Personal exposure to NO₂ was measured cumulatively over each 3-day monitoring period using NO₂ diffusion tubes supplied and analysed by Gradko International Ltd., UK. The subjects received a diffusion tube on the first day of each of the six block visits and wore it at all times except during sleep when it was placed on a bedside table.

Personal exposures to PNC and concentration of particulate matter with a median aerodynamic diameter of <2.5 micrometres (PM_2.5) were estimated using a model incorporating estimates for indoor and outdoor microenvironments, and the microenvironments and activities that have been identified as elevating a person’s particle exposures above the background, to calculate individual 24 h exposures.15 PM_2.5 was estimated from its relationship with concentration of particulate matter with a median aerodynamic diameter of <10 micrometres (PM₁₀). A daily activity diary was used to record time spent by each individual in various microenvironments. The mathematical expression of the model is:

E  =  ^I_iC_i/T*t_i

where:

E  =  total average exposure over time period T

C_i  =  PNC/PM_2.5 concentration in microenvironment i

t_i  =  number of minutes spent in microenvironment i

I  =  total number of microenvironments

T  =  total number of minutes (in this study  =  1440 (24 h))

Meteorological data

Daily local data for temperature, barometric pressure, relative humidity and wind speed for the city were sourced from the UK Meteorological Office. Daily averages and ranges were recorded.

Haematological and biochemical data

Blood was drawn at the end of each monitoring block and stored according to standardised procedures. Analysis was: red blood cell count (RBC), haemoglobin (Hb), haematocrit (Hct), white blood cell count (WBC) and platelet count by automated cell counter (Bayer Advia Haematology System; Bayer Corporation, Germany); plasma fibrinogen by Clauss clotting method, with mechanical detection of the clotting time; factor VIIc by one-stage clotting assay on ACL 3000R analyser (IL Laboratories, Warrington); von Willebrand Factor antigen by a validated in-house enzyme-linked immunosorbent assay (ELISA) using rabbit polyclonal antibodies (Dako UK Ltd, Ely, Cambridgeshire); soluble E-selectin by an ELISA kit (R & D Systems Ltd., Abingdon); plasma interleukin 6 (IL-6) by high-sensitivity ELISA (IDS Ltd., Boldon); high sensitivity C reactive protein (hs-CRP) in plasma by an automated, latex-enhanced immuno-turbidometric method (Technoclone Ltd, Dorking, Surrey) on the Cobas Mira (ABX, Shefford), adapted for low-level detection by increasing the sample volume twofold and diluting the standards to half their concentration; albumin by the Bromocresol Green method on the Cobas Mira; and D-dimer by VIDAS D-dimer new (DD2) immunoassay on the mini-VIDAS analyser (BioMérieux UK Ltd, Basingstoke).

Electrocardiographic data

Ambulatory 24 h ECG recording used digital solid state monitors (Syneflash; ELA medical, Sorin Group, St. Albans) attached to participants in their homes. A modified V5 and VF bipolar lead placement was used, and protocols were those described elsewhere (Marquette Medical Systems. 1996. ECG Lead Configurations: Series 8500 Ambulatory Tape Recorders Operator’s Manual, Milwaukee, Wisconson). ECGs were recorded digitally (sampling rate, 200 Hz/channel) on removable flash cards. The recordings were manually edited using PC-based software (Synescope Holter analysis software; ELA medical). Care was taken to ensure accurate classification of ectopic beats and artefacts. The frequency of ventricular and supraventricular extrasystoles was regarded as the primary outcome of interest. The number of extrasystoles, occurring in isolation, as couplets, and in runs was obtained for each 24 h recording period. The analysis also included other arrhythmias, heart rate trends and parameters of HRV including both time domain statistical and frequency domain spectral HRV analysis according to standard recommendations.16

Statistical methods

All subjects who completed more than one monitoring block were included. The relationships between the response variables and the environmental exposures were investigated using regression models. Potential confounding variables included age and daily fluctuating variables such as weather. Many models were fitted, including either random or fixed effects for subjects, with fixed effects for confounders, in various combinations. Models with all fixed effects were fitted using standard linear regression techniques, and those with random effects for subject were fitted using the residual maximum likelihood algorithm. All analyses were carried out using the GenStat package.17

Model assumptions were checked with graphical diagnostic tools, and for most of the response variables the assumptions were best met when the dependent variable was analysed on the logarithmic scale. All results presented are from analyses on the log scale.

It was also observed that, while some of the potential confounders were statistically significant (p<0.05) predictors of some responses, in no case did their inclusion materially alter the conclusions of the analysis, that is, they were not confounding an exposure–response relationship. For that reason, we show here only a fraction of the results, since the other models fitted did not modify or qualify the conclusions.

We have earlier shown results for the relationships between haematological responses and pollution measurements (fewer of both than in the present study) in elderly subjects from two other cities.18 In the interests of comparability, we report here results from a model analogous to that used in the earlier study. This was a mixed-regression model of the form:

Log(Response)  =  a + ν_i + b₁sin(2πt/365) + b₂cos(2πt/365) + b₃temp + b₄temp² + b₅E

where:

ν_i is a random term particular to each subject, assumed normally distributed

t is the number of days from the study start

the sin and cosine terms represent any annually fluctuating trend

E is whichever pollution exposure variable is being fitted

temp is the mean daily temperature and temp² its square.

In the earlier study we had used minimum daily temperature,18 but that variable was missing for a large number of subject-days in the current study, so mean temperature has been used, which is likely to have been well correlated with the minimum over the study period. The fitting procedure estimated the overall intercept a, the regression coefficients b₁ to b_5, and σ²_ν, the variance component for the random subjects term. For exposure variables available daily, we investigated the effects of including those for the current day, and at 1- and 2-day lags.

RESULTS

Subjects studied

We studied 132 patients with a diagnosis of heart failure (table 1). Subjects were fairly evenly distributed in terms of severity of heart failure symptoms. All were current non-smokers but almost 20% lived with a smoker and almost 50% were exposed to smoke regularly outside the home. The majority had gas central heating, and most had electric cookers. All subjects were on cardioprotective therapy, usually a combination of angiotensin-converting enzyme (ACE) inhibitors, beta blockers, diuretics, aspirin and statins.

Echocardiographic assessment was carried out in 91 subjects who had completed all six blocks of the study. Their mean age and sex distribution were similar to those of the whole study group. Ninety-three per cent had impaired left ventricular systolic function with an ejection fraction of <50% by echocardiography. Median plasma B-type natriuretic peptide (BNP), a marker of disease severity, was 100.5 ng/l with an interquartile range of 168.5 ng/l.

Summaries of data

Data on most variables were available for 639 patient blocks. Figure 1 shows the time course of measurements of oxides of nitrogen within the city during the study period. There are clear seasonal trends. Figure 2 shows the PM₁₀ and PNC measurements, showing less seasonality. Table 2 shows the summary data on the response, air pollution and meteorological variables. Episodes of ventricular and supraventricular tachycardia occurred only infrequently during the ECG recordings. Ventricular tachycardia was observed in 15 ECG recordings: one recording had 117 incidences but the other 14 had only a single episode. Only three episodes of supraventricular tachycardia were seen and these all occurred in a single recording. Regressions analyses were not performed for these parameters as they were too sparse for meaningful inclusion.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Time course of nitrogen oxides measured over the study period. NO, nitrogen oxide; NO₂, nitrogen dioxide.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Time course ofPM₁₀ (μg.m⁻³) and particle number concentrations (particles.cm⁻³) over the study period. PM₁₀, concentration of particulate matter with a median aerodynamic diameter of <10 micrometres.

Table 3 contains the matrix of pairwise correlation coefficients between pollution variables. Almost all were positive. The correlation between NO₂ and NO measurements was 0.83, but that between NO₂ and PM₁₀ was only 0.29. Modest correlations were noted between city NO₂/NO and particle number counts/traffic-related estimates, the latter being measured at a background site. The other stronger correlations were between variables that were derived from the same measurements. Measured personal NO₂ did not correlate with any other measure of pollution save total particle numbers. These low correlations imply that confounding between different exposure metrics was highly unlikely.

Results of regression analyses

Tables 4–8 contain results from 279 mixed-model regression analyses modelled for 31 response variables (on the logarithmic scale) against nine pollution variables. For brevity, the tables contain only the estimated regression coefficient for the named pollution variable, and omit the intercept, harmonic terms for seasonal variation, and the terms in temperature. Because of missing values the analyses were based on different numbers of patient blocks, between 506 and 563. Each entry includes the estimated regression coefficient and the 95% normal CI calculated from the estimated standard error of the coefficient (which allows for the actual number of subjects analysed). In general, models with lagged exposures did not give improved fits, and results from these are not shown.

For an individual analysis, statistical significance was taken at the 5% level, requiring a coefficient whose CI did not include 0; however, with 261 analyses to consider, it is likely that some would achieve this level of significance by chance. The expected number of such false positives is 13. As it turned out, we found 18 intervals whose bounds excluded 0 (highlighted in bold/italic font), with no apparent consistency in which predictor variable was involved or the direction of the implied effect across a range of hypothesised responses.

Changes in red blood counts did not occur in this population, the only suggestion being a fall in Hct in relation to personal NO₂ and indoor particle number counts. However, there was no consistency in the direction of change of these blood indices in relation to pollutant concentrations overall, and platelets, which fell in relation to PM₁₀ in our earlier study, showed a non-significant but positive association with PM₁₀ in this study.

There was no consistent response to different pollutants in WBC, IL-6 or CRP, and responses of white cells tended to be negative rather than the expected positive. The endothelial markers, von Willebrand Factor and E-selectin tended to be positive but coefficients were small and mostly not significant. Fibrinogen and d-dimer showed no consistent direction of coefficient. In contrast to the earlier study, factor VII tended to be positive.

HRV rose consistently in relation to pollution but few associations were significant even at the 5% level and the changes were small. The measures of short-term HRV, 24 h root mean square of successive differences in heart period (RMSSD) and percentage of absolute differences in successive intervals between normal to normal heart beat values in >50 milliseconds, showed no clear associations with particle numbers. There were no consistent associations between the other time and frequency domain parameters of HRV and pollution variables and no trends in the relations between pollution and any extrasystoles or arrhythmias.

DISCUSSION

This study was designed to investigate the potential mechanisms underlying the cardiovascular effects of air pollution in a group of patients thought to be at increased risk because of established heart failure. We found no consistent associations at any lag between the pollutants measured and a range of blood measures indicative of blood coagulation, inflammation, endothelial activation and red cell/platelet sequestration, and no evidence of any association with parameters of cardiac rhythm and HRV. The findings were unexpected and led us to consider potential explanations.

The study had adequate power and the methods of measurement and estimation of exposures were sufficiently robust to detect changes had they occurred. The major strength of this study was its repeated-measures design with subjects effectively acting as their own controls. While the pollution levels in Aberdeen were low, they did not differ much from those in our earlier study in Edinburgh and Belfast, in which we found changes in haematological parameters.18 Moreover, the weak correlations between individual pollutants and the amount of variation in them over time, together with the availability of individual indices of exposure, were sufficient for any strong effects of air pollution on the response variables we measured to have been detected.

As we were unable to replicate the findings of our previous study in healthy older people, we re-analysed the data from that study using the current methods and confirmed the earlier results. This leads us to conclude that there was a qualitative difference between either the subjects or the constituents of air pollution in the three cities. We cannot say whether pollution in Aberdeen differs from that in the other cities, but we think it is unlikely; all are east coast ports, in which sea salt makes a variable contribution, but in which the main pollution sources are vehicles. During the earlier study, coal burning was a feature of Belfast, but this did not seem to be associated with different responses from those in Edinburgh at the time. In the absence of any response to the other pollution variables in the present study and in view of the similarity of the three cities geographically, we think that differences in the subjects were responsible.

To our knowledge, this is the first panel study investigating pathophysiological endpoints in relation to ambient particulate air pollution and NO₂ in patients with stable heart failure on optimal therapy. While time series studies suggest that such patients may be more susceptible to pollution, the majority have recorded events using coding from death certificates or hospital discharge diagnostic categories. It is not clear what proportion of the events occurred in subjects who had a diagnosis of heart failure or were on treatment prior to the index admission or death. This issue has been partly addressed in a Korean study which utilised a case-crossover design to show an enhanced effect of ambient air pollutants on mortality in patients with a diagnosis of heart failure.5 However, data on medication were not available and the deaths occurred in the pre-beta blocker era. Similar results were apparent from a further study by Goldberg et al,4 but again this study pre-dated widespread use of neurohormonal therapies.

The major drugs taken by our subjects were beta blockers, statins, aspirin and ACE inhibitors or angiotensin receptor antagonists. There is evidence that beta-adrenergic antagonists,19 inhibitors of the rennin–angiotensin–aldosterone system,20 and statins21 are associated with increased HRV. In addition, drugs of the latter two classes have been associated with beneficial effects on endothelial function and inflammation.22 We therefore hypothesise that the widespread use of these drugs within our cohort attenuated the effects of air pollution on the response variables measured in this study. It was not powered to test the effect of individual medications, and the subgroups not taking these were too small for meaningful analysis.

The few panel studies examining endpoints similar to those in our study have included few patients with cardiac disease or taking medications. Indeed, some that have demonstrated effects on HRV have specifically excluded groups that might be considered particularly vulnerable, including subjects with heart failure.12 13

With few exceptions, most human studies relating air pollution exposure to haematological parameters including markers of endothelial activation, inflammation and coagulation have involved young healthy individuals. Overall the results have been somewhat inconsistent. A recent panel study in patients with established coronary artery disease demonstrated a significant association between CRP levels and particulate pollution.23 However, the authors concluded that markers of endothelial activation and parameters of coagulation were not consistently altered. They found evidence of effect modification in relation to statin use, positive associations between ambient pollutants and intercellular adhesion molecule 1 (ICAM-1) levels being largely driven by those not on lipid-lowering medications. Conversely, the associations with CRP appeared to be stronger in those taking statins.

Three studies have used ambulatory 24 h Holter recordings similar to ours and demonstrated associations.13 24 25 Both of the studies by Pope et al set out to monitor patients repeatedly during episodes of high and low pollution. The first involved seven patients. Overall the investigators found small but consistent negative associations between particulate pollution and both standard deviation of all normal to normal heart beat intervals over the entire duration of the recording (SDNN) and standard deviation of the 5-minute mean normal to normal heart beat intervals. The second study of 88 elderly patients found a negative association between PM_2.5 and both 24 h SDNN and RMSSD and suggested a positive association between PM_2.5 and CRP, largely driven by an outlier. The third study was conducted to investigate the effects of carbon monoxide in 36 patients with coronary artery disease.24 PM_2.5 was recorded as a potential confounder. A positive association between ambient carbon monoxide and SDNN was found in those subjects not taking beta blockers. There was evidence of effect modification with statin use but not with ACE inhibitors or calcium channel blockers. This study found no effect of PM_2.5.

Several panel studies using short-term recordings of HRV have reported evidence of effect modification with both medication use and comorbid conditions such as hypertension,11 26 27 diabetes,11 26 coronary heart disease,26 and prior myocardial infarction.28 However, the results have tended to be inconsistent and, with the exception of the recent study by Schwartz et al,29 few studies were designed to look at these aspects a priori. Beta blockers,10 26 calcium channel blockers26 and statins29 have been shown in some studies to attenuate the effects of air pollutants; while another study found no evidence of effect modification by beta blockers.28

There has also been a suggestion from some studies that drug therapy may modify the effects of air pollution on the development of myocardial ischaemia. The ULTRA study30 investigated the relationship between particulate air pollution exposure and ischaemic ST-segment depression during repeated sub-maximal exercise tests in patients with stable angina. Significant positive associations between both PM_2.5 levels and particle number concentrations and episodes of exercise-induced ST depression were more pronounced in patients who were not taking beta blockers. Another study that demonstrated a positive association between ozone exposure and the risk of acute myocardial infarction in subjects without a prior diagnosis of ischaemic heart disease, reported a lack of effect in patients with a personal history of coronary heart disease.31 The investigators speculate that this may relate to an effect of drug treatment. In contrast to these findings, a recent experimental study showed increased ischaemic effects on exercise and reduced release of tissue plasminogen activator during high-dose diesel exhaust (300 μg.m⁻³) exposure in patients with coronary heart disease on treatment.32

It seems intuitive that the effects of treatment for a previously known condition might be expected to ameliorate the predisposing effect of the condition itself. The inconsistency in effect modification described above may be related to the variable interaction between the underlying condition, the pollution concentrations and the countering effects of drug treatments. It is possible that these offsetting effects might be sufficient to cause a diminished effect which might not be detectable at everyday ambient levels, resulting in an effective threshold. This may be part of the reason for the lack of demonstrable association in our study and for the varying associations found in the literature.