Background: Short-term changes in temperature have been associated with cardiovascular deaths. This study examines changes in this association over time among the US elderly.
Methods: Daily cardiovascular mortality counts from 107 cities in the US National Morbidity and Mortality Air Pollution Study were regressed against daily temperature using the case-crossover method. Estimates were averaged by time and season using a meta-analysis.
Results: In summer 1987 the average increase in cardiovascular deaths due to a 10°F increase in temperature was 4.7%. By summer 2000, the risk with higher temperature had disappeared (−0.4%). In contrast, an increase in temperature in fall, winter and spring was associated with a decrease in deaths, and this decrease remained constant over time.
Conclusions: Heat-related cardiovascular deaths in the elderly have declined over time, probably due to increased use of air conditioning, while increased risks with cold-related temperature persist.
Cardiovascular disease rates change with temperature, and the direction of this change depends on location and season.1–3 In the winter increases in cardiovascular deaths are associated with decreases in temperature,3 while in the summer increases in deaths are associated with increases in temperature. These increases appear sharply during heat waves.4 This J-shaped risk suggests a comfort-zone outside of which physiological stresses produce circulatory problems. These temperature-related stresses are more dangerous in the elderly because the elderly are more likely to have pre-existing cardiovascular disease, including hypertension.1,5
The effects of temperature on health have been evident since at least the 1960s.6 An increased public awareness of the problem and improvements in the standard of living and health care should have led to a decrease in deaths over time.7 This study explored the evidence for such a decrease in the United States.
The analysis is based on data from the publicly available National Morbidity and Mortality Air Pollution Study (NMMAPS) study.8,9 This study monitored daily climatic conditions, air pollution levels, morbidity and mortality in 108 cities in the United States. Our analysis is restricted to the elderly because they are more sensitive to changes in temperature.1,5,10 Arkansas was excluded because of some missing data for dew-point temperature, leaving 107 cities with complete mortality and temperature data for 14 years (January 1987 to December 2000). Temperature data came from the National Climatic Data Center, and daily mortality counts came from the National Center for Health Statistics. More information on the cities used is available from the NMMAPS web site.9
The analysis had 2 stages: the association between temperature and counts of cardiovascular deaths was measured in each city, and then these estimates were combined to give an average effect. (A similar 2-stage design has been used in other analysis of the NMMAPS data.11) The analysis in each city used the case-crossover method.12 This method is similar to a case–control design because it compares the association between temperature and deaths in case (or index) days to nearby control (or referent) days. As each individual acts as their own control, time-independent confounders (eg, age) are controlled for by design. By choosing referent days that are close to the index day, the method also controls for trends and seasonality in cardiovascular deaths, and for trends in population size.12
A short delay between temperature exposure and cardiovascular disease onset was modeled using the mean temperature on the current day and the 6 preceding days. The optimal number of previous days was chosen using the mean Akaike Information Criteria (AIC) from all 107 cities.13 Results are shown for a 10°F increase in temperature.
Separate analyses were made in each season. This allowed a positive relation between increasing temperature and death in summer, and a negative relation in winter.1 To control for the effects of humidity, the same-day dew-point temperature was included. Day of the week was also included as an indicator variable to control for any weekly pattern in cardiovascular deaths.
A time-stratified case-crossover design was used,12 with months as the strata. Index days were every day in a month. The referents were the other days in the month, excluding the index day and 2 days either side of the index day (to reduce the correlation in exposure between the index and referent days). For example, an index day of January 9 had referent days of January 1 to January 6, and January 12 to January 31. Seasons were defined as: winter, December to February; spring, March to May; summer, June to August; and fall, September to November.
Heterogeneity among cities in the case-crossover estimates was calculated using the I-squared statistic.14 This statistic is on a scale of 0% (completely homogeneous) to 100% (extreme heterogeneity).
The case-crossover estimates across cities, years and seasons were combined using a Bayesian hierarchical model.15 Each city was given a random intercept and a random linear effect of time. A sensitivity analysis examined the random nonlinear (quadratic) effect of time. This change to the model was assessed using the Deviance Information Criteria (DIC).16 To examine differences in the effect of temperature by average climate, a hierarchy of geographical region was added to the model. The results from Anchorage and Honolulu were excluded from this sensitivity analysis, as they were not clearly a member of any region.
The case-crossover analysis was conducted using SAS (SAS Institute Inc., Cary, NC) and the Bayesian hierarchical model using WinBUGS (MRC Biostatistics Unit, Cambridge, UK). The Markov chain Monte Carlo (MCMC) analysis used a burn-in of 10,000 samples followed by a run of 50,000 thinned by 5. Convergence of the Markov chains was assessed using the Gelman-Rubin statistic.17
The highest risk of death after temperature exposure (according to the AIC) came at 0 to 4 days. The hierarchical model fit (according to the DIC) was not improved by using a random quadratic term for time, and so the results are based on modeling a linear change over time. The Markov chain Monte Carlo estimates showed good convergence in all 4 seasons.
Table 1 presents the estimated mean changes in cardiovascular deaths due to a 10°F increase in temperature (over the previous 0–4 days) at the start and end of the study (based on the hierarchical model). The largest change over time was in the effect of hot temperatures in summer. In summer 1987 cardiovascular deaths increased by 4.7% with a 10°F increase in temperature (95% posterior interval (PI) = 3.0% to 6.5%). By summer 2000 the risk was virtually zero (−0.4%; −3.2% to 2.5%). There was very little change over time during the other seasons, as shown by Figure 1. In all seasons the posterior intervals for the mean effect of temperature widened over time, partly due to a reduction in cardiovascular deaths over time. The annual percentage change in deaths (and 95% PI) were: for winter −0.06 (−0.23 to 0.10); for spring −0.06 (−0.24 to 0.11); for summer −0.39 (−0.65 to −0.13); and for fall −0.04 (−0.22 to 0.14).
Figure 2 shows that the decline in summer deaths over time varied by geographical region. The biggest declines were in the Northeast, Northwest, Industrial Midwest and Southern California; these were also the regions with the highest levels of temperature-related mortality in 1987. In the Southeast and Southwest regions, the higher mortality with summer heat changed little over time. In the Upper Midwest region, there was little evidence for temperature-related deaths in summer, and little change in this pattern over time. There was relatively little change over time in the effects of cold in winter, except that the mortality risk with cold temperatures got somewhat worse in the Industrial Midwest region and somewhat better in Southern California.
The I-squared values of between 21% and 31% in summer and winter indicate mild heterogeneity in the effects of temperature between the cities (Table 1). This heterogeneity is apparent in Figure 2, which shows variations in average temperature effects by region. This finding of mild heterogeneity is in contrast to a similar study that found a large disagreement in heat-related mortality among 12 US cities.18
The was a clear decrease in heat-related mortality in the elderly between 1987 and 2000. Related studies in the United States have found similar declines in heat-related all-cause mortality (all ages) from the 1960s to the 1990s,19 and from 1971 to 1997 in North Carolina.20 The authors of both studies suggested that the decline was due to increases in air conditioning use. Two studies of all-cause mortality in the United States have found that the effect of hot temperatures was associated with the level of air conditioning use in a city.18,21 Figures from the US Energy Information Administration for 1980 to 2001 show that air conditioning use has been steadily increasing in all areas of the United States.22 Although an increase in air conditioning is a plausible explanation of the decrease in heat-related cardiovascular deaths, it is confounded with other changes over time, such as improved health care.
While an increase in air conditioning over time may have affected heat-related cardiovascular deaths, nothing has changed the impact of cold temperatures on mortality. The mechanism by which cold temperatures lead to increased cardiovascular deaths is most likely via blood pressure.23 Susceptibility to cold-related mortality has been associated with race,24,25 education24 and female sex.3,25 The sex difference suggests either that clothing is an important modifier26 or that there is a biologic difference between the ability to thermoregulate. Body temperature is regulated by the hypothalamus neurons, which are directly influenced by estrogen through estrogen receptors. The associations with race and education suggest a socioeconomic effect, although results on the socioeconomic effect on cold-related deaths have been mixed. Of 2 large UK studies, one found an association between cold homes and increased risk of death,10 but another found that deprivation in an area was not related to risk of excess winter all-cause mortality.27
It is plausible that improvements over time in the standard of living (specifically housing quality and heating) would reduce the number of cold-related deaths. The results from this study suggest either that improvements in the US standard of living were insufficient or that such improvements are in fact not protective. Another possible pathway to protection of the elderly from low temperatures is more and better clothes in cold weather.26 The results shown here suggest that protective measures need to be taken not just in winter, but also in relatively cold days spring and fall. However, there is no direct evidence in the literature to support an intervention of increased clothing. Basu and Samet1 have provided guidelines for the future research into heat-related mortality. The results from the present study indicate that new studies of cold-related cardiovascular deaths are also needed. To date most research has analyzed temperature at a population level, using temperature measurements obtained from outdoor monitors. Although logistically more difficult, a cohort study that monitored temperature in subjects’ homes and collected details on subjects’ clothing would have much greater power to detect differences in risk related to individual and socioeconomic factors.
A successful intervention for cold-related mortality could have a substantial public health impact. Using the data analyzed here, the average difference in the number of cardiovascular deaths between winter and summer was 28 per 1000 of the elderly population, although some of this difference is due to other seasonal risk factors such as influenza.28
The results from Figure 2 are similar to a related study, which found the strongest association for heat-related cardiorespiratory mortality in the Southwest, and a negligible effect in the Upper Midwest.29 That study examined temperature and cardiorespiratory mortality in the elderly in the 20 largest metropolitan areas in the United States in 1992. The Upper Midwest had the best result of any region for heat-related mortality despite the study period including the 1995 Chicago heat wave.1 This study examined trends in temperature-related mortality, and thus extreme events were averaged out.
A related study in the United States showed a decline over time in the seasonality of cardiovascular disease.30 The study covered the years 1937 to 1991 and showed an approximate 2% per annum decline in the winter-summer ratio of all coronary heart disease deaths. The authors suggested the decline may be due to improvements in heating (indoor and vehicular) and air-conditioning. Declines in seasonality mortality, however, are not the same as declines due to temperature-related mortality. Seasonal estimates include the effects of other seasonal risk factors such as diet, physical activity and influenza.28 This study specifically examined the effects of temperature, and showed a substantial decline in summer-heat-related mortality after controlling for seasonality.
The strengths of the present study are that it covered a large number of cities in the United States, and so had power to look at changes over time in temperature-related deaths by season and geographical region. It used the case-crossover method to control for the large decrease over time in cardiovascular disease death rates. One limitation is that the delay between temperature exposure and death was fixed to 4 days, when it is possible that the delays varied by city and season. Also the study was able to examine only total cardiovascular deaths, while specific cardiovascular outcomes (eg, myocardial infarction) may be more sensitive to temperature.
The author thanks the Department of Biostatistics at the Johns Hopkins Bloomberg School of Public Health and the Health Effects Institute for making the National Morbidity and Mortality Air Pollution Study data publicly available.
1. Basu R, Samet JM. Relation between elevated ambient temperature and mortality: a review of the epidemiologic evidence. Epidemiol Rev
2. Braga ALF, Zanobetti A, Schwartz J. The effect of weather on respiratory and cardiovascular deaths in 12 U.S. cities. Environ Health Perspect
3. Barnett AG, Dobson AJ, McElduff P, et al. Cold periods and coronary events: an analysis of populations worldwide. J Epidemiol Community Health
4. Keatinge WR. Death in heat waves. BMJ
5. Keatinge WR, Donaldson GC. The impact of global warming on health and mortality. South Med J
6. Pell JP, Cobbe SM. Seasonal variations in coronary heart disease. Q J Med
7. Patz JA, McGeehin MA, Bernard SM, et al. The potential health impacts of climate variability and change for the United States: executive summary of the report of the health sector of the U.S. national assessment. Environ Health Perspect
8. Samet JM, Dominici F, Zeger SL, Schwartz J, Dockery DW. The National Morbidity, Mortality, and Air Pollution Study. I. Methods and Methodologic Issues
. Health Effects Institute; 2000.
10. Wilkinson P, Armstrong B, Landon M, et al. Cold comfort: The social and environmental determinants of excess winter deaths in England, 1986–1996.
Joseph Rowntree Foundation, York; 2001:N11.
11. Bell M, McDermott A, Zeger S, Samet J, Dominici F. Ozone and Short-term Mortality in 95 US Urban Communities. JAMA
12. Janes H, Sheppard L, Lumley T. Case-crossover analyses of air pollution exposure data: referent selection strategies and their implications for bias. Epidemiology
13. Dobson AJ. An Introduction to Generalized Linear Models.
2nd ed. Boca Raton, FL: Chapman & Hall/CRC; 2002.
14. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med
15. Gelman A, Carlin JB, Stern HS, Rubin DB. Bayesian Data Analysis.
2nd ed. Boca Raton, FL: Chapman & Hall/CRC; 2003.
16. Spiegelhalter DJ, Best NG, Carlin BP, van der Linde A. Bayesian measures of model complexity and fit (with discussion). J R Stat Soc Ser B Methodol
17. Brooks SP, Gelman A. General methods for monitoring convergence of iterative simulations. J Computat Graph Stat
18. Curriero F, Heiner K, Samet J, Zeger S, Strug L, Patz J. Temperature and mortality in 11 cities of the eastern United States. Am J Epidemiol
19. Davis RE, Knappenberger PC, Michaels PJ, Novicoff WM. Changing heat-related mortality in the United States. Environ Health Perspect
20. Donaldson GC, Keatinge WR, Näyhä S. Changes in summer temperature and heat-related mortality since 1971 in North Carolina, South Finland, and Southeast England. Environ Res
21. Braga ALF, Zanobetti A, Schwartz J. The time course of weather-related deaths. Epidemiology
23. Donaldson G, Robinson D, Allaway S. An analysis of arterial disease mortality and BUPA health screening data in men, in relation to outdoor temperature. Clin Sci
24. O'Neill MS, Zanobetti A, Schwartz J. Modifiers of the temperature and mortality association in seven US cities. Am J Epidemiol
25. Schwartz J. Who is sensitive to extremes of temperature? A case-only analysis. Epidemiology
26. The Eurowinter Group. Cold exposure and winter mortality from ischaemic heart disease, cerebrovascular disease, respiratory disease, and all causes in warm and cold regions of Europe. Lancet
27. Lawlor DA, Maxwell R, Wheeler BW. Rurality, deprivation, and excess winter mortality: an ecological study. J Epidemiol Community Health
28. Reichert TA, Simonsen L, Sharma A, Pardo SA, Fedson DS, Miller MA. Influenza and the winter increase in mortality in the United States, 1959–1999. Am J Epidemiol
29. Basu R, Dominici F, Samet JM. Temperature and mortality among the elderly in the United States. Epidemiology
30. Seretakis D, Lagiou P, Lipworth L, Signorello LB, Rothman KJ, Trichopoulos D. Changing seasonality of mortality from coronary heart disease. JAMA
© 2007 Lippincott Williams & Wilkins, Inc.