In this issue, Bell et al1 report on the role of air conditioning (AC) as an effect modifier of the associations between short-term differences in outdoor concentrations of particulate matter (PM) and either mortality or hospitalizations. The authors conclude that increased AC prevalence lowered fine PM (PM2.5) effects on cardiovascular hospital admissions.1 This modifying effect of AC is plausible, partly because we spend most of our time indoors. AC is obviously strongly linked to window closing, which in turn reduces house ventilation and indoor concentrations of pollutants that originate from outdoors. AC may also result in some filtering of pollutants, further reducing the concentrations indoors. The regression coefficients for pollutants in health models might then be expected to be smaller, in association with AC use, when exposure is estimated from outdoor pollution monitors. In fact, if effects were not smaller, one might question whether pollutant health effect estimates actually indicate the effects of the pollutants. Assessing the effect-modifying role of AC then serves in a sense as one gauge of the coherence of air pollutant effects.
The report by Bell et al is the latest in a series of studies in which attempts have been made to assess the modifying effect of AC on the mortality or hospitalization effects of outdoor PM or ozone.2–6 Because of the dramatic effect of window closing on reducing indoor ozone concentrations, one might expect AC to lessen the effects of outdoor ozone even more than for the effects of PM. None of these studies have individual-level information on AC use because they are all essentially time-series studies that used administrative data bases. By necessity, the evaluation of effect modification by AC in such time-series studies can be evaluated only from multicity studies in which effect modification is assessed in a second stage of the analysis (ie, a meta-regression) that assesses variance in the between-city estimates of pollutant effect. AC is specified in these second-stage models as individual-city (or county) prevalence of AC based on national housing survey data. Obviously these are ecologic analyses.
Bell et al also use a simple refinement to these analyses, which consists of comparing the modifying effect of AC in cities in which the pollutant of interest is higher in a specific season.1,2,4 For example, PM is higher in the summer in most eastern US cities, but higher in the winter in most western cities. Window opening is more frequent in the summer, and so the modifying effect of AC should be greater in cities with higher summertime PM concentrations.
There are several reasons why we might not be as confident as the authors in their conclusion:
1. Because AC is used as an ecologic measure, one must be especially alert to the possibility that the observed modifying effect is confounded. The geographic pattern of AC use in the United States is regional and could be associated with any number of regional features that might confound. For example, increased AC use is, obviously, associated with warmer temperature and higher humidity. To the extent that counties with warmer temperatures and higher humidity also have lower estimated effects of PM, spurious effect modification of AC could be observed. This possibility was not addressed. One might also consider possible confounding by socioeconomic status (SES), although ecologic indicators of SES were apparently not associated with AC use, making this concern less likely.
2. Even though the statistical power to address effect modification in the Bell et al study was substantially greater than in any preceding study of AC effect modification (with 84 counties and 168 counties being used, respectively, for the mortality and hospitalization effect modification analyses), the precision of the estimated modifying effects was relatively poor. Of the many strata in which effect modification was assessed, only those of cardiovascular hospitalizations and cardiovascular hospitalizations in counties with summertime-peaking PM2.5 demonstrated effect modification by central AC (as opposed to any AC), although effect modification was suggested in some other strata.
3. The greater effect modification in summertime-peaking PM2.5 counties was present only for hospitalizations; for mortality there was no indication that effect modification was greater in those counties. While it is true that power for detecting effect modification was greater for hospitalizations, arguing from coherence we would have expected at least a hint of an effect for mortality.
4. There was no effect modification by AC in the strata of respiratory hospitalizations. If AC reduces exposure, modification of pollutant effects would be expected for any health end point for which an effect of PM was identified. A previous study identified a modifying effect of AC on respiratory mortality that was larger than for cardiovascular mortality.4
What is needed to improve our understanding of the modifying effects of AC on the impact of outdoor exposure to air pollution? First, individual-level specification of AC use would help. Not only would this address some of the concerns pertaining to the use of an ecologic measure, but effect modification could be assessed in single-city time-series studies, or at least at the first level (ie, within-city) of multicity analyses. This could obviate the need for an exceedingly large number of cities or counties in a multicity effect modification analysis. While most time-series studies have used administrative data with no individual-level information, some time series studies have used a well-defined cohort that could theoretically be well-characterized with respect to AC use and potential confounding features.7 Second, up to this point, effect modification by AC has been addressed only in the context of short-term exposure. Effects of long-term exposure are equally if not more important. Because cohort studies are typically used for estimating long-term exposure effects, collecting individual-level information on AC use for assessing modification of long-term exposure effects should not be so difficult. Third, AC reduces indoor concentrations of some pollutants more effectively than others, and would be expected to preferentially reduce some PM size fractions and chemical components. These differences could be potentially exploited in studies involving multiple air pollutants or multiple PM components.
The attempt to identify modifying effects of AC is important. Effect modification by AC use can serve as a gauge of the coherence of observational air pollution health effects and, more importantly, could also have public health implications. However, public health policy decisions often involve trade-offs. As discussed by Bell et al, even if AC use reduced the health burden due to air pollution, it is not certain that increased AC use would have overall health benefit.
ABOUT THE AUTHOR
SVERRE VEDAL is Professor of Environmental and Occupational Health Sciences at the University of Washington School of Public Health. He serves on US EPA Clean Air Scientific Advisory Committee panels. His epidemiologic research currently focuses on air pollution sources and chemical components in producing adverse health effects, and on the modifying role of genetic makeup.
1. Bell ML, Ebisu K, Peng RD, Dominici F. Adverse health effects of particulate air pollution: modification by air conditioning. Epidemiology
2. Jannsen NAH, Schwartz J, Zanobetti A, Suh HH. Air conditioning and source-specific particles as modifiers of the effect of PM10
on hospital admissions for heart and lung disease. Environ Health Perspect
3. Bell ML, Dominici F. Effect modification by community characteristics on the short-term effect of ozone exposure and mortality in 98 U.S. communities. Am J Epidemiol
4. Franklin M, Zeka A, Schwartz J. Association between PM2.5
and all-cause and specific-cause mortality in 27 US communities. J Expo Sci Environ Epidemiol
5. Medina-Ramón M, Zanobetti A, Schwartz J. The effect of ozone and PM10
on hospital admissions for pneumonia ande chronic obstructive pulmonary disease: a national multicity study. Am J Epidemiol
6. Levy JI, Chemerysnki SM, Sarnat JA. Ozone exposure and mortality: an empiric Bayes metaregression analysis. Epidemiol
7. Sunyer J, Schwartz J, Tobías A, Macfarlane D, Garcia J, Antó JM. Patients with chronic obstructive pulmonary disease are at increased risk of death associated with urban particle air pollution: a case-crossover analysis. Am J Epidemiol