A Case-Crossover Analysis of Out-of-Hospital Coronary Deaths and Air Pollution in Rome, Italy

Out-of-hospital coronary heart disease death, often used as a surrogate for sudden cardiac death, is a major public health problem, and accounts for a large proportion of the total mortality from cardiovascular diseases (1). Scientific and public health interest in health effects of airborne particulate matter is the result of epidemiologic studies conducted in the last decade, indicating both short-term (2, 3) and long-term (4) effects on mortality. Evidence of the relationship between air pollution and out-of-hospital coronary death is sparse. In the analysis of air pollution and daily mortality in Philadelphia, Schwartz (5) reported that deaths outside of hospitals, presumably predominantly coronary deaths, increased disproportionately on days with the highest particulate air pollution. In 10 U.S. cities, the effect of air pollution on daily deaths that occurred out of hospitals was substantially greater than the increase in deaths in hospitals (6). Two relatively small case-crossover studies conducted in the United States did not find a relationship between air pollutants, including fine particulate matter, and frequency of out-of-hospital sudden cardiac deaths (7, 8). On the other hand, a recent review by the American Heart Association highlighted the emerging evidence for the biological plausibility of the association between ambient air pollution concentrations in urban areas and cardiovascular disease exacerbation (9). Recent studies suggest that airborne particles may induce systemic inflammation, alter the blood thrombotic status, and induce dysrhythmia.

Among the pollutants in urban air, ultrafine particles (diameter < 0.1 µm), directly originating from incomplete combustion processes, may be particularly harmful from a biological point of view. They are capable of penetrating the pulmonary interstitium, causing interstitial inflammation and significant oxidative stress (10, 11), may pass into the blood circulation (12), and possibly have a prothrombotic effect (13, 14).

As part of the Health Effects of Air Pollution on Susceptible Subpopulations (HEAPSS) study, a European project on the effects of air pollution on myocardial infarction in five European cities, we evaluated the association of air pollutants-in particular, particle number concentration (PNC), a measure of ultrafine particles)-with out-of-hospital fatal coronary events in Rome. Some of the results of this study have been previously reported in the form of abstracts (15).

METHODS

Subject Data

Among the five cities of the Health Effects of Air Pollution on Susceptible Subpopulations study (Ausgburg, Barcelona, Helsinki, Rome, and Stockholm), data on out-of-hospital coronary deaths were available for Rome. Data were retrieved by record linkage from the Regional Register of Causes of Death and the Lazio Hospital Information System using individual fiscal codes. Persons over 35 years of age who died of ischemic heart disease (International Classification of Diseases, 9th edition [ICD-9], codes 410-414) during the period 1998-2000 were identified. We excluded deaths of nonresidents and deaths that occurred outside Rome, subjects who died within 28 d after admission for ischemic diseases (ICD-9: 410-414), all deaths occurring in a hospital after hospitalization for any cause (or 1 d after discharge), and subjects who had had a primary diagnosis of ICD-9: 410 (acute myocardial infarction) or any diagnosis of ICD-9: 412 (previous infarction) during the preceding 3-yr period. A total of 5,144 subjects were included.

We had individual information on sex, age, and any hospitalizations during the preceding 3 yr for the following: diabetes, hypertension, chronic obstructive pulmonary diseases (COPDs), angina, other ischemic heart disease, conduction disorders, dysrhythmias, and heart failure. Both primary diagnoses and secondary conditions were noted. See additional details in the online supplement.

Air Pollution Data

Data on paniculate matter with aerodynamic diameter less than 10 µm (PM^sub 10^), carbon monoxide (CO), nitrogen dioxide (NO^sub 2^), sulfur dioxide (SO^sub 2^), and ozone (O^sub 3^) were available. Daily mean values were considered for all pollutants except O^sub 3^, for which the daily maximum 8-h running mean during the warm season (April-September) was chosen. Data on PNC has been collected in Rome since April 2001 using optical particle counters (Condensation Particle Counter, CPC 3022A model; TSI, Inc., Shoreview, MN). Because ultrafine particles typically dominate ambient total particle number concentrations (PNC), it is a proxy for ultrafine particles. PNC daily data were estimated retrospectively for the study period (1998-2000) based on a regression model implemented for the PNC measurement period (April 2001-June 2002) that used PNC as the dependent variable and traditional air pollutants and meteorologic readings as explanatory variables. The overall performance of the model was very good, with an R^sup 2^ of 0.85. Additional details are described in the online supplement and elsewhere (16, 17).

Statistical Analysis

The association between pollutants and out-of-hospital coronary deaths was investigated with the case-crossover design (18). Control days were selected using the time-stratified approach (19): the period was divided into monthly intervals; all the same days of the week within that month (of the same year) were chosen. We evaluated the lag structure (0-3 d lags; lag 0 represents the air pollution level at the day of death, lag 1 is the level at the previous day, and so on) for five pollutants (PNC, PM^sub 10^, CO, NO^sub 2^, and O^sub 3^), and also considered the average of the same and previous day (lags 0-1). Conditional logistic regression analysis was fitted to the data to calculate percent increases of risk and 95% confidence intervals (CI). The logistic model included apparent temperature (20) on the index day, and its difference from the average of the values from the previous 3 d.

The effects of some potential effect modifiers (sex, age, comorbidities [defined as hospitalizations for the conditions indicated above during the preceding 3 yr]) were evaluated by introducing interaction terms into the conditional logistic model (only for the 0-1 lag). Percent increase of risk was calculated for an increment in exposure corresponding to the interquartile range of each pollutant. See additional details in the online supplement.

RESULTS

Table 1 summarizes the data regarding out-of-hospital coronary deaths. The mean age was 78 yr, and slightly more men than women were found among these subjects. The majority of the fatal events occurred in the oldest age group (> 74 yr). The most frequently registered comorbidities were hypertension (13%), angina and other ischemic heart disease (12%), and dysrhythmias (9%).

Table 2 displays levels of the environmental variables and Pearson correlation coefficients (r). There was a high correlation of estimated PNC with other pollutants, especially with CO (r = 0.89) and SO^sub 2^ (r = 0.68). PM^sub 10^ was moderately correlated with PNC (r = 0.38) and CO (r = 0.34). Furthermore, there was moderately high colinearity among many of the traditional pollutants, in particular between CO and SO^sub 2^ (r = 0.52), NO^sub 2^ and PM^sub 10^ (r = 0.45), and NO^sub 2^ and CO (r = 0.54).

Table 3 shows the results of the case-crossover analysis of the effect of air pollutants on out-of-hospital coronary deaths, for single-day lags (0 to 3) and for a cumulative lag (0-1). The highest and most significant values were found for PNC, PM^sub 10^, and CO. The air pollution effect was strongest on the event day (lag 0). The increase in risk of out-of-hospital coronary death associated with PNC at lag 0 was estimated at 7.6% per 27,790 particles/cm^sup 3^ (95% CI, 2.0-13.6%), and the estimate for PM^sub 10^, was 4.8% per 29.7 µg/m^sup 3^ (95% CI, 0.1-9.8%). The estimate of the CO effect fell between the others, and was also significant at lag O: 6.5% per 1.2 mg/m^sup 3^ (95% CI, 1.0-12.3%). For the other pollutants, the estimated effects were smaller and not statistically significant.

In the bipollutant analysis, the estimates for both pollutants decreased. The reduction was stronger for PM^sub 10^ than for PNC. At lag 0, we found 7.0% for PNC (95% CI, -0.2 to 14.7) and 1.1% for PM^sub 10^ (95% CI, -4.8 to 7.3). When PM^sub 10^ and CO were considered simultaneously, the effects were reduced to 2.1% (95% CI, -3.7 to 8.2) for PM^sub 10^ and to 5.1% (-1.8 to 12.4%) for CO (data not shown in the tables).

The estimated concentration-response functions of these pollutants on out-of-hospital coronary deaths are summarized in Figure 1. The increasing trend of the curves (clearly linear for PNC and CO; close to linear for PM^sub 10^) suggests that there is no threshold level below which these pollutants would be harmless.

In the case-crossover analysis by subgroups, no significant effect modification was detected (likelihood ratio test), although some tendencies emerged. Figure 2 illustrates the results for PNC; the findings were very similar for PM^sub 10^ and CO (data not shown). The age groups most clearly susceptible to the effects of air pollution were the 65-74- and > 75-yr groups. Hypertension and COPD were the only conditions for which there was some evidence of effect modification. However, even when the estimates for the two groups were very large, the effect modification was not statistically significant.

DISCUSSION

Our investigations concern two important novel areas of epidemiologic research: the relationship of air pollutants with out-of-hospital coronary death, and the role of ultrafine particles. The study indicates that fatal coronary events are associated with PNC, PM^sub 10^, and CO. The exposure-response relationship was linear. The effect appears stronger among the elderly and is not limited to a specific subgroup with a previously diagnosed comorbidity. When PNC and PM^sub 10^ were tested in a multipollutant model, PNC was more strongly associated with fatal coronary events.

Previous studies have found several indications of an effect of particles and carbon monoxide on acute hospitalization for cardiovascular conditions (21), but direct evidence that exposure to particulate air pollution is associated with increased risk of acute coronary events is limited and conflicting. A case-crossover study on 772 patients reported that the risk of acute myocardial infarction is related to the PM concentration in the 2 h before onset (22). A positive association was also found when considering exposure in the 24 h before onset. In Rome, from January 1995 to June 1997, hospitalizations for myocardial infarction increased in association with total suspended particulates (2.3% per 10 µg/m^sup 3^), NO^sub 2^ (2.6% per 10 µg/m^sup 3^), and CO (2.1% per 1 mg/m^sup 3^) (23). In San Paulo, Brazil (24), and in Atlanta, Georgia (25), emergency room visits for ischemic heart diseases were associated, within very short lag times, with traffic-derived pollutants such as CO and NO^sub 2^. In the American study, PM^sub 2.5^ and elemental and organic carbon had an effect, but PNC did not (however, a large quantity of PNC data were missing).

Two case-crossover studies performed in the Seattle, Washington, area, with 362 and 1,206 cases, respectively, failed to find an association between several air pollutants and out-ofhospital primary cardiac arrest (7, 8). Although the low power of these studies might have been an issue, Sullivan and colleagues (8) also speculated that particulate matter in the Seattle area (mainly driven by wood smoke) is relatively sparse in transition metals and sulfates, so the possible cardiovascular effect from traffic-related particles may be diluted. A recent case-crossover study conducted by the same authors in Seattle, including 5,793 confirmed cases of myocardial infarction, also found no consistent association between levels of fine particulate matter and risk of myocardial infarction (26). However, the stratified analysis by season (heating season with mainly wood smoke, and nonheating season with traffic-generated particles) did not reveal any effect modification.

The size of the effect of air pollution in our study should be viewed in the context of research evaluating the effects of PM^sub 10^ on total mortality. The American National Morbidity, Mortality, and Air Pollution Study and the European Air Pollution and Health: a European Approach studies have reported effect estimates of 0.2-0.6% per 10 µg/m^sup 3^ PM^sub 10^ (27). The corresponding estimate for out-of-hospital coronary death from our study is considerably higher, at 2.8% (per 10 µg/m^sup 3^ PM^sub 10^). We believe that this is an important finding, and clearly in contrast with the two previous negative reports on sudden death. The high estimate suggests that out-of-hospital coronary death, which, in Rome, represents approximately 8% of total natural mortality, might represent a rather specific effect of air pollution.

Toxicologic and epidemiologic studies suggest several possible pathways by which inhaled noxious particles could induce out-of-hospital coronary death, a fatal condition due to ventricular tachyarrhythmia, which can be triggered by several mechanisms, including ischemia. First, PM may increase the risk of sudden death by affecting autonomic control of the heart. Oxidative stress could alter the sympathetic and parasympathetic tone that affects heart rate and heart rate variability. Decreased heart rate variability has been associated with sudden cardiac death (28). Epidemiologic studies have associated air pollution with increased heart rate (29, 30) and decreased heart rate variability (31-33), both risk factors for severe arrhythmia and mortality. In a recent study by Park and coworkers (34), fine PM was associated with decreased heart rate variability, especially in patients with ischemic heart disease and hypertension. In a pilot study conducted in Boston among 100 patients with implanted cardioverter defibrillators, exposure to PM^sub 2.5^, as well as to pollutants that are closely related to traffic, such as NO^sub 2^ and CO, was associated with increased risk of defibrillator discharges (35). In a larger sample of 203 cardiac patients with implanted cardioverter defibrillators in Boston, Dockery and colleagues (36) recently reported that the risk of a ventricular tachyarrythmia was associated with exposure to air pollution, especially in patients with increased ventricular electrical instability. Second, oxidants can increase the level of blood coagulability and modify the adhesive properties of red blood cells, thus leading to increased risk of ischemic damage in individuals with poor coronary circulation (37). Nemmar and colleagues (14) found that the intratracheal instillation of diesel exhaust particles leads to a rapid activation of circulating blood platelets. In addition, a systemic reaction could occur with an acute phase response leading to increased synthesis of C-reactive protein, a known cardiovascular risk factor in healthy subjects (38). Finally, systemic microvascular dysfunction, which is usually associated with cardiovascular diseases, may be affected by PM exposure. A recent study indicated that endothelium-dependent arteriolar dilatation in the systemic microcirculation of rats is impaired by PM (39).

These suggested mechanisms may be of importance in interpreting our finding of a strong association between air pollution and sudden coronary death. One may speculate that the key mechanism of the PM-mortality association is severe arrhythmia, possibly among people with primary electrical abnormalities, and not taking drugs that block the effects of ß-adrenergic stimulation (40). However, there are insufficient data to profile those subjects who suddenly die of coronary disease (41).

The results of this study regarding effect modification by age, with a stronger effect suggested among the elderly, parallel results observed in other time series studies on air pollution and daily mortality (42, 43), and our results on hospitalization for myocardial infarction in Rome (23). There was a suggestion in our study that people with previous hypertension and COPD have a greater PNC effect than people without these conditions, although the power to detect a statistically significant effect modification was low. Several recent articles (34, 44, 45) have reported hypertension as a modifier of the effects of air pollution on heart rate variability. As Park and colleagues have suggested (34), hypertension is associated with lower heart rate variability, endothelial dysfunction, and higher levels of inflammatory responses induced by oxidative stress. People with hypertension may, therefore, be less able to cope with the effects of air pollution on the heart. A similar phenomenon may occur in subjects with COPD; both Sunyer and colleagues (46) and De Leon and colleagues (47) have already reported higher risks of circulatory death in subjects with COPD. In the long-term American Cancer Society study (4), PM^sub 2.5^ was strongly associated with mortality from ischemic heart disease, dysrhythmias, heart failure, and cardiac arrest (especially among smokers), but there was a paradoxically protective effect on COPD deaths. The findings of the American Cancer Society, as well as those of our own study, may be interpreted as indication that those subjects with COPD most susceptible to fine particles are prematurely removed by cardiovascular death. However, in our study, even the subgroup without any of the tested comorbidities showed a significant association with PNC and, overall, our results suggest that the effect of air pollution is not limited to a specific vulnerable population. Further research on this aspect is warranted.

Although we examined the effect of several pollutants, measures of PM^sub 2.5^ for the entire study period were not available. Data collected during 2001-2002 on PM^sub 2.5^ and other pollutants in Rome indicate that there is a relatively high correlation between PM^sub 2.5^ and both measured PNC (r = 0.65) and PM^sub 10^ (r = 0.61). A weak (Amsterdam and Helsinki) to moderate (Erfurt) correlation between PM^sub 2.5^ and PNC has been reported (48). The differences in correlations have been explained by the higher impact of more local sources on ultrafine particle concentrations than on PM^sub 2.5^. PNC data were not directly available for the study period, but were retrospectively estimated from one monitoring station from 2001-2002. The statistical model predicts PNC well, and our estimation procedure was certainly superior (in terms of R^sup 2^) to what has been done to estimate PM^sub 10^ or PM^sub 2.5^ in previous studies (49-55). However, PNC is derived from existing air pollutants, and it is highly correlated to some of them (Table 2). As such, this is not a true measure of PNC, but only a proxy. On one hand, retrospective estimation and the presence of only one monitor might have introduced error in the estimate of PNC exposure that is likely to be nondifferential and would, therefore, tend to decrease the estimated effect on mortality. On the other hand, estimated values from the imputation procedure have less variability than true values, and the coefficient derived from the logistic regression analysis using estimated PNC may be higher than would be expected had the actual measurements been taken. Moreover, the imputation can affect the variance estimate for the effect size. Because the other pollutants were measured directly and, hence, do not have this problem, the results of the comparison of the effect size and significance between PNC and the other pollutants should be interpreted with caution.

Aside from the problems related to the retrospective imputation of PNC, there are issues on exposure assessment of PNC in time series studies that should be considered (56, 57). As ultrafine particles are mostly produced by local traffic, whereas PM^sub 2.5^ generally is more dominated by long-range transport, a greater spatial heterogeneity of PNC and a closer relationship with distance from the roadway when compared with other PM size fractions should be expected. The use of central monitors may limit the ability to detect a real PNC effect when data on population average personal exposure are lacking. In fact, as described in the online supplement, the two PNC monitors in Rome had very different average values, depending on distance to traffic (values were 2.16 times higher at the traffic site than at the background site). However, the overall correlation between the two sites was high (r = 0.73), an indication that a correlation between daily variation in outdoor concentrations and population average personal exposure, at least outdoors, is also likely. Outdoor and in-vehicle exposures might be of special importance when the effects of traffic-related pollutants are considered (58).

The results of two-pollutant models are always difficult to interpret, especially when they include pollutants originating from the same sources. In the case of PNC and PM^sub 10^, the coefficient of PM^sub 10^ may be interpreted as the effect of particles whose source is not well correlated with PNC. As PNC is basically a surrogate for traffic particles, the coefficient of PM^sub 10^ in this twopollutant model may be interpreted as the effect of nontraffic particles. The situation of PM^sub 10^ and CO is somewhat similar, as CO is correlated with traffic particles, but not with transported particles. Overall, these two results suggest that traffic particles are more important for the effect on out-of-hospital coronary deaths. This is also supported by the evidence that traffic exposure may be linked to the onset of myocardial infarction (58).

Our definition of the outcome is based on routinely collected data. There is no other method to accurately enumerate primary coronary deaths unless a special register is present, as in the Seattle area (7). Our definition of out-of-hospital coronary deaths is similar to what has been used in other European and American studies (1, 59), and the method of data collection has the important advantage of covering the entire population. The case-crossover approach turned out to be useful in evaluating effect modification. However, the analysis by comorbidities relied on hospital discharge files, in which sensitivity may be suboptimal and specificity is usually high (60, 61). Comorbidities are identified by previous hospital admission in the past 3 yr. As this is a fairly short time period, the possibility of misclassification remains, and this misclassification may depend on the specific disease. However, we have attempted to minimize this problem by considering the conditions listed in both primary and secondary diagnoses. Moreover, it should be considered that we did not have information on relevant individual risk factors, notably smoking, whose importance as an effect modifier has been reemphasized in a recent study (4). Finally, information on medication use was not available, which is regrettable from an effect modification point of view, as medications may modulate susceptibility to air pollution.

In summary, we found a higher estimated risk of air pollution for out-of-hospital coronary deaths than has been found in previous studies on total mortality. The strongest effects were found for PNC and CO, both directly originating from traffic-related combustion sources. There is a suggestion of effect modification for hypertension and COPD, but the overall effect of air pollution is not limited to a specific subgroup, as defined by history of certain diseases. The present results need to be replicated in other locations.

Conflict of Interest Statement: None of the authors have a financial relationship with a commercial entity that has an interest in the subject of this manuscript.

© 2005 American Thoracic Society Provided by ProQuest LLC. All Rights Reserved.