Content uploaded by Antonella Zanobetti
Author content
All content in this area was uploaded by Antonella Zanobetti
Content may be subject to copyright.
1 of 7
Journal of Urban Health: Bulletin of the New York Academy of Medicine,
doi:10.1093/jurban/jti043
The Author 2005. Published by Oxford University Press on behalf of the New York Academy of Medicine. All rights
reserved. For permissions, please e-mail: journals.permissions@oupjournals.org
Disparities by Race in Heat-Related Mortality in
Four US Cities: The Role of Air Conditioning
Prevalence
Marie S. O’Neill, Antonella Zanobetti, and Joel Schwartz
ABSTRACT Daily mortality is typically higher on hot days in urban areas, and certain
population groups experience disproportionate risk. Air conditioning (AC) has been
recommended to mitigate heat-related illness and death. We examined whether AC
prevalence explained differing heat-related mortality effects by race. Poisson regression
was used to model daily mortality in Chicago, Detroit, Minneapolis, and Pittsburgh.
Predictors included natural splines of time (to control seasonal patterns); mean daily
apparent temperature on the day of death, and averaged over lags 1–3; barometric
pressure; day of week; and a linear term for airborne particles. Separate, city-specific
models were fit to death counts stratified by race (Black or White) to derive the percent
change in mortality at 29 ºC, relative to 15 ºC (lag 0). Next, city-specific effects were
regressed on city- and race-specific AC prevalence. Combined effect estimates across all
cities were calculated using inverse variance-weighted averages. Prevalence of central
AC among Black households was less than half that among White households in all
four cities, and deaths among Blacks were more strongly associated with hot tempera-
tures. Central AC prevalence explained some of the differences in heat effects by race,
but room-unit AC did not. Efforts to reduce disparities in heat-related mortality should
consider access to AC.
KEYWORDS Air conditioning, Climate, Ethnic groups, Heat, Mortality, Socioeconomic
factors, Weather.
INTRODUCTION
Black race, lower educational attainment, age, death outside a hospital, and lack of
access to air conditioning (AC) modify associations between ambient temperatures
and daily mortality.
1–5
Because AC can protect people from heat-related
mortality
1,6
and access to AC and resources to use it may differ by race, we evalu-
ated whether AC played a role in previously reported racial disparities in heat-
related mortality.
5
Dr. O’Neill is with The Robert Wood Johnson Health & Society Scholars Program, Center for Social
Epidemiology and Population Health, University of Michigan, Ann Arbor, Michigan; Drs. Zanobetti and
Schwartz are with the Department of Environmental Health, Harvard School of Public Health, Boston,
Massachusetts; and Dr. Schwartz is with the Department of Epidemiology, Harvard School of Public
Health, Boston, Massachusetts.
Correspondence: Marie S. O’Neill, PhD, The Robert Wood Johnson Health & Society Scholars
Program, Center for Social Epidemiology and Population Health, University of Michigan, 1214 South
University Avenue, Room 249, Ann Arbor, MI 48104-2548. (E-mail: marieo@umich.edu)
2 of 7 O’NEILL ET AL.
METHODS
Data Sources
Mortality and environmental data were described previously.
5
Briefly, daily, nonin-
jury mortality counts were obtained for 1986–1993 and stratified by race (Blacks
and Whites). This analysis used data for the metropolitan area (defined by county)
of four cities: Chicago, Illinois; Detroit, Michigan; Minneapolis and St. Paul,
Minneapolis (“Minneapolis” in this article); and Pittsburgh, Pennsylvania. Here,
the term city refers to these metropolitan areas. We chose these cities because they
had an adequate number of hot days for estimating heat–mortality associations,
daily information on air pollution (which has been associated with daily
mortality
7
), and AC prevalence data. Between 1986 and 1993, 684,847 people died
in the four areas.
Weather data were from the airport station closest to each city. Apparent tem-
perature, an index of human discomfort, was calculated from ambient temperature
and dew point, as described previously.
5
Barometric pressure and concentrations of
particulate matter less than 10 microns in aerodynamic diameter (PM
10
) were also
covariates for this analysis. PM
10
can be inhaled into the lungs and has been linked
with several adverse health outcomes.
8
The American Housing Survey collects data on AC prevalence (central and one;
two; three or more room units) about every 4 years, sampling 4,800 or more housing
units per metropolitan area.
9
Between 1986 and 1993, two surveys were adminis-
tered in each of the four cities. City-specific AC prevalence was averaged from those
two surveys and applied over the entire study period. Prevalence was reported by
race/ethnicity (Black, Hispanic, and total). Because we lacked Hispanic mortality
data for the whole study period, we report AC prevalence statistics by Black house-
holds and “White/Other” (i.e., households not reported as being Black or Hispanic).
Statistical Methods
Robust Poisson regression was used to model daily death counts as the dependent
variable in models fit individually for each city. Independent variables included a
linear term for mean PM
10
on the day of death and the previous day and natural
cubic splines of mean daily barometric pressure, day of week, and day of study.
Because of the nonlinear dependence of mortality on temperature and previous
research showing differing lag structures for hot- and cold-weather mortality,
10
we
used two cubic splines to model temperature. The cold term was temperature aver-
aged over lags 1–3, and the heat term was temperature at lag 0. Other details of
model fitting criteria are described elsewhere.
5
City-specific regressions were fit to
daily death counts among Blacks and Whites, and a combined four-city effect esti-
mate was calculated for each race by using inverse variance-weighted averages.
To evaluate the effects of AC prevalence, we performed a meta-regression with
a model of the following form:
β
ij
=C
0
+
γ
Z
ij
+ε
ij
where
β
ij
is the coefficient for the effect of 29°C heat, at lag 0 in city i, stratum j
(Blacks or Whites), and
γ
Z
ij
is the AC prevalence in city i, stratum j. The expected
value of
ε
ij
is assumed zero, and the variance of
ε
ij
is represented as σ
ij
2
+σ, where σ
ij
2
is the estimated standard error of
β
ij
within city and stratum, and σ is the heterogeneity
DISPARITIES BY RACE IN HEAT-RELATED MORTALITY 3 of 7
in the β
ij
not explained by AC prevalence. The C
0
and γ are estimated with inverse
variance-weighted least squares regression, and between-city variance σs were esti-
mated with iterative maximum likelihood estimation.
11
A random effect for city
accounted for heterogeneity among the four cities.
RESULTS
Environmental variables for the four cities were reported previously.
5
They all had
comparable maximum mean daily apparent temperatures (range =34.3–36.9 °C).
Mean levels of particles and other meteorological variables were also similar. About
40% of those who died in Detroit during the study period were Black; the corre-
sponding percentages were 28 in Chicago; 11 in Pittsburgh; and 4 in Minneapolis,
consistent with city demographics (Table 1).
In all four cities, central AC prevalence in the “White/Other” households
was more than double the central AC prevalence in Black households (Table 2).
Pittsburgh had the lowest overall prevalence of central AC, at 25%, and Minneapolis
the highest, at 47%. There was less variation by race in room-unit AC prevalence
and no consistent pattern of differences by race.
Table 3 summarizes city-specific heat effects by race (also reported in O’Neill
et al.
5
) and the pooled effect across the four cities. In each city, heat-associated mor-
tality was higher among Blacks, and the pooled effect showed an effect among
Blacks over twice that among Whites.
TABLE 1. Demographics of four metropolitan areas, 1986–1993
Chicago Detroit Minneapolis Pittsburgh
Total population (million) 5.11 2.11 1.52 1.34
Black (%) 26 40 5 11
TABLE 2. Air conditioning (AC) prevalence by household race
9
Prevalence statistics are means of the two survey years.
*Reporting one or more room-unit air conditioners.
†All households in the survey not reported as Black or Hispanic.
AC (%)
City Years Population Central Room unit*
Chicago 1987, 1991 Total 41 35
Blacks 16 32
Whites/Other† 49 35
Detroit 1989, 1993 Total 35 26
Blacks 17 30
Whites/Other 39 25
Minneapolis 1989, 1993 Total 47 26
Blacks 21 19
Whites/Other 48 27
Pittsburgh 1986, 1990 Total 25 33
Blacks 10 46
Whites/Other 26 33
4 of 7 O’NEILL ET AL.
In the meta-regression, for each 10% increase in central AC prevalence, heat-
associated mortality, pooled across all four cities, dropped by 1.4% (95% CI =−0.1
to 2.9), a marginally significant result. The overall effect of heat on mortality (the
effect of heat in a city with a 0% prevalence of central AC) was a 10.2% increase
(95% CI =4.5–16.2). Applying the 1.4% estimated drop in mortality to the average
difference in AC prevalence between Blacks and White/Other across all four cities
(24%) suggests that differences in central AC prevalence explain no more than a
3.4% difference in heat-related mortality. Blacks had 5.3% higher heat-related
mortality than Whites (Table 2); therefore, as much as 64% of this disparity is
potentially attributable to central AC prevalence.
Excess heat-related mortality with 0% room-unit AC prevalence was estimated
with meta-regression at 2.5% (95% CI=−13.6 to 21.7). Each 10% increase in
room-unit AC prevalence was associated with a 0.95% increase in heat-associated
mortality (95% CI =−4.4 to 6.5).
Because there was less of a gradient in room-unit AC prevalence and no consis-
tent pattern of disparities by race, this variable had limited utility in this analysis.
Figure shows central AC prevalence plotted against the coefficients reflecting the
size of the association between heat and mortality by city and race. Higher heat and
mortality associations and lower central AC prevalence were seen for Blacks. Cen-
tral AC prevalence varied little among Black households across the cities.
DISCUSSION
Heat-related mortality in four US cities was reduced with increasing central AC
prevalence, and substantially higher effects of heat on mortality were observed
among Blacks compared with Whites. A large proportion of the disparity in heat-
related mortality may be due to differences in central AC prevalence. Room-unit
AC prevalence showed little effect on heat-related mortality and no consistent pat-
tern of disparities by race.
Several previous studies showed both Black race and lack of AC as indicating
vulnerability to heat-related health effects.
1–6
Heat-related mortality associations
were higher in areas with lower AC prevalence, even after adjusting for latitude.
4
Access to AC has been recommended as a key component of efforts to prevent heat-
related deaths.
12–14
Among 72,420 US residents, hot-weather death rates from 1980
to 1985 were 42% lower among people with central AC compared with people
with no AC, and AC benefits were highest for women, the elderly, people not in the
labor force, and those in dwellings of less than six rooms.
6
Comparing room-unit
AC with no AC, the effect was not significantly different from zero, except among
TABLE 3. Percent change in daily mortality and 95% confidence intervals (CIs) associated with
29 °C apparent temperature, pooled and city specific (1988–1993 for Chicago, 1986–1993 for
other cities)
Estimates are relative to 15 °C apparent temperature and control for barometric pressure, PM
10
, time trend,
day of week, and apparent temperature averaged over lags 1, 2, and 3 (heat effect is expressed for apparent
temperature lag 0).
Pooled Chicago Detroit Minneapolis Pittsburgh
Total mortality 4.6 (2.6–6.7) 4.5 (2.3–6.7) 7.5 (4.2–10.8) 2.4 (−2.1 to 7.1) 3.1 (−0.5 to 6.9)
Black 9.0 (5.3–12.8) 5.9 (2.0–9.9) 12.0 (6.8–17.4) 17.0 (−7.8 to 48.4) 12.5 (1.4–24.8)
White 3.7 (1.9–5.4) 4.1 (1.5–6.7) 5.5 (1.4–9.7) 2.3 (−2.4 to 7.1) 2.0 (−1.7 to 6.0)
DISPARITIES BY RACE IN HEAT-RELATED MORTALITY 5 of 7
people whose dwellings had one to three rooms, where room-unit AC was benefi-
cial.
6
An inverse association between expected risk of death at 30 °C and prevalence
of central AC, with 33% of the variation in heat-associated mortality explained by
AC prevalence, was seen in 12 US cities.
10
Central AC prevalence is likely correlated with other area socioeconomic char-
acteristics indicating vulnerability to mortality on extreme temperature days
15
;
therefore, the explanatory power of this variable probably reflects influences of
these other factors. During a 1995 heat wave in Chicago, social contacts, mobility,
affordability of electricity, and sense of personal security affected whether people
had adequate ventilation and cooling in their homes.
16
These factors differed across
small-scale geographic areas (neighborhoods). The AC prevalence statistics used for
this study were aggregated by city, limiting conclusions about smaller-scale neigh-
borhood characteristics. Because room-unit AC prevalence patterns differed by city
and race, this study design was not able to provide insights into whether these units
are beneficial at the population level, though intuition would suggest that they
would be. Studies examining individual-level data on AC use and ownership would
be required to further evaluate this question.
Although the number of deaths occurring in the four cities was substantial, con-
fidence intervals for the meta-regression results were wide. Additionally, Figure
shows substantial scatter in the relationship between central AC prevalence and
the heat and mortality effect. For heat-related mortality among Whites, Pittsburgh
10 20 30 40 50
5
1.001.050
.
00.0
10 20 30 40 50
5
1.001.050
.
00.0
Percent AC prevalence
tneiciffeocytilatrom/taeH
Pittsburgh
Chicago
Detroit
Minneapolis
Pittsburgh
Detroit
Chicago
Minneapolis
FIGURE. Coefficients for the relative risk of mortality on days at 29 °C apparent temperature com-
pared with days at 15 °C, by prevalence of central air conditioning (AC), race, and city. ▲, Whites
(and Whites/Others, for AC prevalence); ■, Blacks. Coefficients are from Poisson regression models
with covariates including barometric pressure, PM
10
, time trend, day of week, and apparent tem-
perature averaged over lags 1, 2, and 3 (heat effect is expressed for apparent temperature lag 0).
Data cover the period 1986–1993.
6 of 7 O’NEILL ET AL.
represented the greatest outlier in the group. The other three cities may be represen-
tative of US cities. Future analyses with more cities would help increase confidence
in the results.
In spite of these limitations, the findings of this study are consistent with previous
observations that central AC use is protective against heat-related mortality. They
also suggest that the strong racial disparities in heat-related mortality are partially
explained by central AC prevalence or other socioeconomic factors correlated with
central AC prevalence that differ by race. Outreach programs to reduce heat-
related mortality, which commonly include ensuring access to cool environments,
17
should take into account demographic patterns in AC prevalence to ensure equita-
ble protection.
ACKNOWLEDGEMENT
This work was supported in part by the National Institute of Environmental Health
Sciences (NIEHS), National Institutes of Health (NIH) (grant 2 T32 ES07069-24).
Its contents are solely the responsibility of the authors and do not necessarily repre-
sent the official views of NIEHS, NIH. (Additional funding sources include grants
NIEHS ES00002 and EPAR827353.)
REFERENCES
1. Semenza JC, Rubin CH, Falter KH, et al. Heat-related deaths during the July 1995 heat
wave in Chicago. N Engl J Med. 1996;335:84–90.
2. Whitman S, Good G, Donoghue ER, Benbow N, Shou W, Mou S. Mortality in Chicago
attributed to the July 1995 heat wave. Am J Public Health. 1997;87:1515–1518.
3. Greenberg JH, Bromberg J, Reed CM, Gustafson TL, Beauchamp RA. The epidemiology
of heat-related deaths, Texas – 1950, 1970–79, and 1980. Am J Public Health.
1983;73:805–807.
4. Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz JA. Temperature and mortal-
ity in 11 cities of the eastern United States. Am J Epidemiol. 2002;155:80–87.
5. O’Neill MS, Zanobetti A, Schwartz J. Modifiers of the temperature and mortality associ-
ation in seven U.S. cities. Am J Epidemiol. 2003;157:1074–1082.
6. Rogot E, Sorlie PD, Backlund E. Air-conditioning and mortality in hot weather. Am J
Epi emiol. 1992;136:106–116.
7. Stieb DM, Judek S, Burnett RT. Meta-analysis of time-series studies of air pollution and
mortality: effects of gases and particles and the influence of cause of death, age, and sea-
son. J Air Waste Manag Assoc. 2002;52:470–484.
8. Brunekreef B, Holgate ST. Air pollution and health. Lancet. 2002;360:1233–1242.
9. US Census Bureau. The American Housing Survey. Washington, DC: U.S. Department of
Commerce/U.S. Department of Housing and Urban Development; 1986–1993. Available
at: http://www.census.gov/prod/www/abs/h170.html. Accessed October, 2003.
10. Braga AL, Zanobetti A, Schwartz J. The time course of weather related deaths. Epidemi-
ology. 2001;12:662–667.
11. Berkey CS, Hoaglin DC, Mosteller F, Colditz GA. A random-effects regression model for
meta-analysis. Stat Med. 1995;14:395–411.
12. CDC. Heat-wave-related mortality – Milwaukee, Wisconsin, July 1995. MMWR Morb
Mortal Wkly Rep. 1996;45:505–507.
13. Semenza JC. Acute renal failure during heat waves. Am J Prev Med. 1999;17:97.
14. Semenza JC, McCullough JE, Flanders WD, McGeehin MA, Lumpkin JR. Excess hospi-
tal admissions during the July 1995 heat wave in Chicago. Am J Prev Med.
1999;16:269–277.
DISPARITIES BY RACE IN HEAT-RELATED MORTALITY 7 of 7
15. O’Neill MS. Air conditioning and heat-related health effects. Appl Environ Sci Public
Health. 2003;1:9–12.
16. Klinenberg E. Heat Wave: A Social Autopsy of Disaster in Chicago. Chicago, IL: University
of Chicago Press; 2002.
17. Smoyer-Tomic KE, Rainham DGC. Beating the heat: development and evaluation of a
Canadian hot weather health-response plan. Environ Health Perspect. 2001;109:1241–
1248.