|Home | About | Journals | Submit | Contact Us | Français|
Evidence on risk of cardiovascular (CVD) hospitalization associated with short-term exposure to outdoor carbon monoxide (CO), an air pollutant primarily generated by traffic, is inconsistent across studies. Uncertainties remain regarding the degree to which associations are attributable to other traffic pollutants and whether effects persist at low levels.
We conducted a multi-site time-series study to estimate risk of CVD hospitalization associated with short-term CO exposure in 126 U.S. urban counties from 1999–2005 for >9.3 million Medicare enrollees ≥65 years of age. We considered models with adjustment by other traffic-related pollutants: nitrogen dioxide (NO2), fine particles (PM2.5), and Elemental Carbon (EC).
We found a positive and statistically significant association between same day (L0) CO and increased risk of hospitalization for multiple CVD outcomes (ischemic heart disease, heart rhythm disturbances, heart failure, cerebrovascular disease, total CVD). The association remained positive and statistically significant, but was attenuated, with co-pollutant adjustment, especially NO2. A one part per million (ppm) increase in L0 daily 1-hour maximum CO was associated with a 0.96% (95% posterior interval 0.79, 1.12%) increase in risk of CVD admissions. With L0 NO2 adjustment, this estimate is 0.55% (0.36, 0.74%). The risk persisted at low CO levels <1 ppm.
We found evidence of an association between short-term exposure to ambient CO and risk of CVD hospitalizations, even at levels well below current U.S. health-based regulatory standards. This evidence indicates that exposure to current CO levels may still pose a public health threat, particularly for persons with CVD.
Carbon monoxide (CO) is a tasteless, odorless gaseous pollutant ubiquitous in the outdoor atmosphere, that is generated by combustion. Adverse health effects of CO exposure include death from asphyxiation at high exposure levels, and at lower levels impaired neuropsychological performance and risk for myocardial ischemia and rhythm disturbances in persons with cardiovascular disease (CVD).1, 2 The most definitive evidence on CO and CVD comes largely from controlled exposure studies, involving CO inhalation at concentrations to mimic exposures previously typical of urban environments.3 Findings of earlier epidemiological studies were mixed, and few recent epidemiological studies have been conducted.3 Little research has been done on the potential adverse health effects in humans from current ambient exposure to generally low CO levels.4
Evidence on risk of CVD hospitalizations associated with CO exposure5–10 is inconsistent across epidemiological studies,3, 4 perhaps due to their limited population sizes and the differing CVD causes considered. While a few multi-city CO studies have been conducted,7, 8 to the best of our knowledge a national-scale U.S. study of CO and CVD hospitalizations has not been performed. Further, because CO primarily results from traffic in most urban communities, risks associated with CO may be confounded or modified by other traffic-related air pollutants, such as nitrogen dioxide (NO2) and fine particles (PM2.5, particulate matter with aerodynamic diameter ≤ 2.5µm).3 Human clinical studies provide useful scientific evidence, but typically involve exposure to CO alone. Therefore studies that better reflect CO and its existence in urban air as a component of a complex mixture are needed.
Currently, U.S. outdoor CO levels are low and well below the U.S. Environmental Protection Agency’s (EPA’s) health-based National Ambient Air Quality Standard (NAAQS) (35 ppm for 1-hour daily maximum) in almost all areas.11 However, health risks of exposure to these low CO levels have not been addressed directly.3 Rather, experimental exposure studies have largely driven the standards. Should adverse health impacts occur at today’s low CO levels, the public health burden could be considerable.
We conducted a national multi-site time-series study to estimate the association between short-term exposure to outdoor CO and risk of CVD hospitalizations in a population of Medicare enrollees. We address key scientific questions regarding associations of CO with cause-specific CVD hospitalization categories; the shape of the exposure-response curve; the persistence of effects at low levels; and possible confounding by co-pollutants in the urban air pollution mixture, in particular those relating to traffic.
We constructed counts of emergency daily hospital admissions for Medicare enrollees ages ≥65 years for 1999–2005 for 126 U.S. urban counties with populations ≥200,000 and CO data for ≥75% of days in the study period. On average across the study period, >9.3 million subjects were included. Admission rates were based on the total number of Medicare enrollees for a county on a given day. Cause-specific CVD hospitalizations were considered based on primary diagnosis for International Classification of Disease Codes, 9th Revision (ICD-9): heart failure (ICD-9 428); heart rhythm disturbances (ICD-9 426–427); cerebrovascular events (ICD-9 430–438); ischemic heart disease (ICD-9 410–414, 429; and peripheral vascular disease (ICD-9 440–448). Total CVD admissions were calculated as the sum of these causes. We considered hospitalizations for injuries and other external causes (ICD-9 800–849) as a test outcome for which we anticipate no association with CO, an approach that has been used in earlier research.12, 13
Air pollution data were obtained from the U.S. EPA. The highest hourly value recorded for a given day (daily 1-hour maximum) was used as the exposure metric for CO, and daily averages for PM2.5, NO2, and EC. Because the co-pollutant data were applied as daily averages, we performed sensitivity analysis with daily (24-hour average) CO levels. On average across the communities, CO data were available for 99% of study days. Daily temperature and dew point temperature for each county were obtained from the National Climatic Data Center. Air pollution and weather data from multiple monitors within a county were averaged to generate county-level estimates.
Within each county, we first estimated risk of CVD hospitalization associated with CO, in a log-linear over-dispersed Poisson time-series model with variables to adjust for long-term trend, seasonality, day of the week, temperature and dew point temperature, and potentially different long-term trends for ages 65–74 and ≥75 years. Natural cubic splines were used to address non-linearity in variables for long-term trend, seasonality, and weather. Separate regression models were fit to data from each county. We conducted separate analyses for CO exposure concentrations for: 1) the same day as the hospitalization (L0); 2) the day before (L1); and two days previous (L2). For each analysis, we then combined the adjusted county-specific risk estimates across counties and estimated a national average with Bayesian hierarchical models, which accounts for within-county statistical error and between-county variability of the “true” relative risks (also called “heterogeneity”).14 The model produces a posterior probability distribution of the national average effect which we summarize by reporting the posterior mean. We used this statistical approach in previous multi-site time-series studies of particulate matter for mortality and morbidity.12, 13
Results are presented as the percent increase in risk of CVD hospitalization per 1 ppm increase in the daily 1-hour maximum CO. This increment (1 ppm) approximates the average interquartile range (IQR, difference between 75th and 25th percentiles) across all communities. Statistical significance was determined by the posterior probability of an estimate >0 being ≥0.95. We investigated whether results were robust to the degrees of freedom for control of seasonality and long-term trends.
Given the strong role of traffic as sources for CO, we investigated whether associations between CO and CVD hospitalizations were robust to adjustment by traffic-related pollutants PM2.5 or NO2 individually or by both in multi-pollutant models. We considered same day PM2.5 as previous studies found that same day (L0) PM2.5 concentration was associated with the largest risk of CVD admissions.13 We considered NO2 also on the same day (L0) as we found that this lag had the strongest effect in single lag models. In order to focus more on the traffic-related components of PM2.5 total mass, we also adjusted for the portion of PM2.5 that is Elemental Carbon (EC) based on EPA PM chemical component monitoring data.15
Because not all counties had data for all pollutants, we examined potential confounding by each co-pollutant by comparing the association between CO and CVD hospitalizations: (1) with co-pollutant adjustment; and (2) without co-pollutant adjustment, but only considering days for which co-pollutant data were available. For NO2 and PM2.5 we considered counties with co-pollutant data available for ≥75% of the study days, resulting in 92 counties for NO2 with data for this pollutant available an average 97.0% of the study days. For PM2.5, 67 counties were used with data available for an average 92.9% of days in the study period. There were 58 counties for both NO2 and PM2.5. Limited data are available for EC, which is not measured in all counties and is typically measured at a 1 in 6 day frequency. Using a criterion of at least 60 observations corresponding to approximately a year of measurements at a 1 in 6 day frequency, 81 counties were included in the EC analysis. Data were available for an average of 56.5% days in the study period. Due to the limited data for EC, this co-pollutant was not considered in conjunction with NO2 or PM2.5 total mass.
We performed sensitivity analysis to investigate whether the effect estimates for CO and hospital admissions were affected by monitor coverage based on the county’s area and number of CO monitors. A second-stage regression analysis was performed with county-specific health effect estimates as the outcome variable and the county-area per number of CO monitors for each county as the independent variable.
To investigate risk at low CO levels, we performed subset analyses using datasets including only days with CO levels below a specified value, s. We repeated the subset analysis for s values from 1 to 10 ppm in 1 ppm increments. Under the threshold hypothesis, no association would be observed in analysis of CO values at or below the threshold. We investigated the exposure-response curve as sensitivity analysis using a non-linear function of CO.16
This study was exempted by the institutional review board at Johns Hopkins Bloomberg School of Public Health. The authors have full access to the data and take responsibility for its integrity. All authors have read and agree to the manuscript as written.
Table 1 provides descriptive statistics on pollution and hospitalizations data. Ischemic heart disease and heart failure were the most common causes of CVD hospitalizations. While CO levels varied across communities, all counties had low levels during the study period. County-specific pollutant and health data summaries are available in Online Supplement in Tables S1 and S2. A plot of county-specific average CO levels versus CVD hospitalization rates is provided as Online Supplement Figure S1. The correlations between daily 1-hour maximum CO and other pollutants, on average across communities, were 0.26 for daily PM2.5 (67 counties, average of 2326 observation days per county), 0.56 for NO2 (92 counties, average of 2292 days/county), and 0.48 for EC (81 counties, average of 328 days/county). Correlations among co-pollutants are provided in the Online Supplement in Table S3.
On average across the 126 counties, the 1-hour maximum CO level was 1.6 ppm, well below the NAAQS (35 ppm). Figure 1a provides boxplots showing the distribution of the communities’ CO levels. We also described the change of CO levels with comparison of CO on adjacent days within cities (Figure 1b).
Table 2 summarizes estimated associations between CO and risk of CVD hospitalizations with adjustment by co-pollutants. Associations between CO and CVD hospitalizations remained positive and statistically significant with adjustment by any co-pollutant, although estimates were attenuated, especially with NO2 adjustment. Online Supplement Figure S2 provides county-specific and national estimates with and without co-pollutant adjustment. Associations between the co-pollutants (NO2, PM2.5, and EC) and CVD hospitalizations are provided in Online Supplement Table S4. Subsequent analyses were adjusted for L0 NO2 because of the change in CO estimates with adjustment by this pollutant.
Figure 2 shows the percent increase in risk of cause-specific and total CVD hospitalizations per 1 ppm increase in daily 1-hour maximum CO, adjusted by same day NO2. For most CVD outcomes, associations were positive and statistically significant for same day CO. Statistically significant associations were not observed between the outcome of accidents and injuries and CO at any lag.
Because daily values were used for co-pollutants, we performed sensitivity analysis with a 24-hour average metric of CO (Online Supplement in Table S5). Results were similar to those with the daily 1-hour maximum CO. Effect estimates were generally lowered with co-pollutant adjustment, especially NO2, but remained positive and statistically significant.
Sensitivity analysis found that results for the association between same day daily 1-hour maximum CO and total CVD hospitalizations were robust to degree of adjustment for seasonality and trend (Online Supplement Figure S3). Similarly, effect estimates for injury, showing no association, were robust to the degree of adjustment for trend and seasonality.
The average number of CO monitors per county was 2.3 (IQR = 1.9) and the monitor coverage (county-area per number of monitors) averaged 1614 km2 per monitor across all counties. The median correlation of the daily 1-hour maximum CO from monitor pairs within the same county was 0.66. An IQR increase in monitor coverage (1500 km2/monitor decrease) was associated with an 8% (95% interval −7, 24%) increase in same day daily 1-hour maximum CO effect estimates for CVD hospitalizations based on all 126 counties, and a 20% (−11, 52%) increase based on NO2 adjusted estimates for 92 counties. While the associations are not statistically significant, they are suggestive of higher effect estimates with more dense monitor coverage.
Figure 3 shows results of the subset analyses. For all values of s, from 1 to 10 ppm, associations between CO and CVD admissions adjusted by NO2 remained positive and statistically significant. When we restricted analysis to days with a 1-hour daily maximum CO less than the current U.S. EPA health-based standard (35 ppm), we found that a 1 ppm increase in CO is associated with a 0.55% (0.36%, 0.74%) increase in risk of CVD admissions. This estimate is identical to the estimated risk obtained with the entire data set.
We estimated an exposure-response curve allowing a non-linear relationship between CO and risk of CVD admissions (Online Supplement Figure S4). We found no evidence of departure for a linear exposure-response curve. These findings support the subset analysis and do not indicate a threshold level below which CO levels are not associated with the health response.
We performed a study of CO and CVD hospitalizations using data on >9.3 million Medicare enrollees in 126 U.S. urban counties over a 7-year period. Other recent time-series studies of CO and morbidity include a study of ten Canadian cities investigating congestive heart failure,5 and the two largest U.S.-based studies, which investigated congestive heart failure admissions in 7 communities10 and total CVD admissions in 8 communities.17 Mortality was examined in 19 European cities by the Air Pollution and Health: A European Approach (APHEA-2) investigators, finding independent associations of CO with total and CVD mortality.18 Ambient CO levels have also been associated with heart rate variability in a human exposure study of persons with coronary artery disease19 and other adverse health outcomes in multiple population-based studies, such as low birth weight20 and mortality.18, 21
Our results provide strong evidence of an association between outdoor CO concentrations and risk of CVD hospitalizations for an older population. The patterns of association by lag, with strongest effects on the same day, are consistent with CO kinetics and the likely mechanisms by which CO has adverse CVD effects. For the “control” category of injury, we found no evidence of an association, implying satisfactory adjustment for time-varying potential confounders.
Because the primary CO source in urban centers is traffic, we considered confounding by other traffic-related pollutants (NO2, PM2.5, EC). With adjustment for these co-pollutants, the association remained although it was attenuated, particularly with NO2 adjustment. We cannot exclude the possibility that the observed associations could reflect pollution from traffic emissions generally, which have been associated with cardiovascular endpoints22, 23 and not from CO specifically. CO levels were moderately correlated with NO2, also resulting from vehicle emissions among other sources.24 Our analyses estimate the effects of CO adjusted by other traffic-related pollutants. The total health impacts of air pollution from the traffic source, or from the air pollution mixture more broadly, are as yet unknown. Understanding the public health consequences of multiple pollutants is an area needing further research, as indicated by reports from the National Resource Council.25, 26
Because of the spatial distribution of ambient CO in urban areas, we anticipate potential misclassification of personal exposures for city-dwellers by our reliance on regulatory monitors. Extensive research has not been conducted on the relationship between personal exposure to CO and ambient measurements. Such work is needed, especially regarding which subpopulations (e.g., outdoor workers) may have differential exposure. A study of 56 subjects found little agreement between ambient and personal CO measurements.27 An examination of in-home CO distribution in Washington, DC found higher levels for those living in the metropolitan area than those in the suburbs.28 We did not find that the health effect estimates vary by the monitor coverage within a county although the results were suggestive of potential higher estimates with more dense monitor coverage. The exposure misclassification would tend to reduce estimates, implying that our results could underestimate risk of hospitalization associated with CO.
The present NAAQS for CO, promulgated in 1994, is based largely on results of studies involving exposure of volunteers with coronary heart disease to concentrations of CO sufficient to raise blood carboxyhemoglobin concentrations from the typical level of 1% in nonsmokers to as high as 6%.29, 30 The study participants had documented coronary heart disease, and study outcomes were clinically relevant indicators (increased arrhythmias,31, 32 time to ischemia or angina33, 34). Effects on indicators of myocardial ischemia were identified at carboxyhemoglobin concentrations as low as 2%34 and for arrhythmias at 6%.31 The Coburn-Forster-Kane equation, which describes the relationship between inhaled CO and blood carboxyhemoglobin, was used by EPA to calculate ambient exposures that could result in carboxyhemoglobin concentrations associated with adverse effects, although the equation’s original aim was to estimate the rate of endogenous CO production.35 In another study, carboxyhemoglobin measurements did not correlate with clinical status in CO poisoning.36
Our results are consistent with clinical and animal model studies finding that CO exposure can adversely affect cardiac function.4, 37, 38 Many of these studies investigated high CO concentrations, often elevated by cigarette smoking, which far exceed the ambient levels studied here; and an early review of this literature noted that few studies explored low CO levels.37 One critical finding of our research is the perhaps unexpectedly strong effect observed at current ambient levels (Table 1). A 1 ppm increase in CO concentration, approximating the IQR across communities, would correspond to an approximate 0.1–0.2% increase in blood carboxyhemoglobin on average. The maximum changes in CO concentration within city, based on comparison of CO levels on adjacent days, are at a level that would be expected to increase carboxyhemoglobin by about 1%, sufficient to increase the level from the baseline of 1% typical of nonsmokers to the 2% value at which effects were observed in some exposure studies (Figure 1b). Additionally, clinical CO studies, which used volunteers and involved an exercise protocol, may have underestimated potential susceptibility of persons with coronary heart disease.
While physiological responses to CO have been well studied, much of the scientific evidence involves concentrations that are quite high in relation to current U.S. ambient levels. The acute and lethal toxicity of CO at high levels is well documented, and human exposure studies have shown acute CO poisoning at very high concentrations,2 although debate still exists regarding biological mechanisms of CO toxicity.39–41 Additional information on physiological mechanisms continues to be gained. CO exposure causes adverse health response through binding to hemoglobin and subsequent lessened delivery of oxygen, as hemoglobin’s affinity for CO is over 200 times that of oxygen.38 Other mechanisms have also been hypothesized including detriment to myoglobin function, generation of reactive oxygen species, and interruption of the terminal oxidase of the electron transport chain.40, 42–45 Recent in vitro and animal model studies indicate that low levels of CO may have therapeutic effects on tissue such as antioxidative and anti-inflammatory responses.46–49
This study provides one of the first population-based investigations of the health effects of current, low ambient CO levels. We provided evidence that the association between CO and risk of cardiovascular hospitalizations persists at levels of daily 1-hour maximum CO <1 ppm and that daily excursions of CO concentration occur that are sufficient to elevate carboxyhemoglobin to a level at which adverse effects have been observed experimentally. Most of the U.S. is in regulatory compliance with health-based CO regulations (35 ppm for the daily 1-hour maximum).11, 30 Higher levels of CO exist in some regions including areas with developing transportation networks, such as urban Chinese centers, where CO primarily results from traffic.50 While much of the current research on health and traffic-related air pollution focuses on particulate matter, our study indicates that ambient CO and traffic may present a far larger health burden than previously suspected.
With growing traffic in many urban centers, the health impacts of traffic-related air pollution are a current public health concern. This national study of 126 U.S. urban counties from 1999 to 2005 examines whether exposure to carbon monoxide (CO) on the same day and previous few days increases risk of cardiovascular (CVD) hospitalizations for an older population based on >9.3 million Medicare enrollees ≥65 years of age. Our findings indicate that higher levels of CO exposure are associated with an increased risk of CVD hospitalizations on the same day for multiple cause-specific CVD outcomes (ischemic heart disease, heart rhythm disturbances, heart failure, cerebrovascular disease, and total CVD admissions). Although it is not possible to fully disentangle the effects of CO and of other air pollutants produced by traffic, the association between CO and CVD hospital admissions remained after adjustment for other traffic-related pollutants: nitrogen dioxide (NO2); fine particulate matter (PM2.5) total mass; and Elemental Carbon PM2.5. The risk persisted even at low CO levels <1 ppm, which are well below the current U.S. health-based regulatory standard. This study provides one of the first population-based investigations of the health effects of current, low ambient CO levels and indicates that exposure to current CO levels may still pose a threat to public health, particularly for persons with CVD. It adds to other research showing that air pollution harms the health of people with CVD.
Funding was provided by an Oak Ridge Institute for Science and Education (ORISE) Fellowship with the U.S. Environmental Protection Agency (EPA), EPA (RD-83241701), and by the NIH National Institute of Environmental Health Sciences (NIEHS) Outstanding New Environmental Scientist (ONES) Award (RO1 ES015028). The views expressed do not necessarily represent those of the funding agencies.
The authors have no conflicts of interest.