Effects of Heat Waves on Mortality

Effect Modification and Confounding by Air Pollutants

Abstract

Background: 

Heat waves and air pollution are both associated with increased mortality. Their joint effects are less well understood.

Methods: 

We explored the role of air pollution in modifying the effects of heat waves on mortality, within the EuroHEAT project. Daily mortality, meteorologic, and air pollution data from nine European cities for the years 1990–2004 were assembled. We defined heat waves by taking both intensity and duration into account. The city-specific effects of heat wave episodes were estimated using generalized estimating equation models, adjusting for potential confounders with and without inclusion of air pollutants (particles, ozone, nitrogen dioxide, sulphur dioxide, carbon monoxide). To investigate effect modification, we introduced an interaction term between heat waves and each single pollutant in the models. Random effects meta-analysis was used to summarize the city-specific results.

Results: 

The increase in the number of daily deaths during heat wave episodes was 54% higher on high ozone days compared with low, among people age 75–84 years. The heat wave effect on high PM₁₀ days was increased by 36% and 106% in the 75–84 year and 85+ year age groups, respectively. A similar pattern was observed for effects on cardiovascular mortality. Effect modification was less evident for respiratory mortality, although the heat wave effect itself was greater for this cause of death. The heat wave effect was smaller (15–30%) after adjustment for ozone or PM₁₀.

Conclusions: 

The heat wave effect on mortality was larger during high ozone or high PM₁₀ days. When assessing the effect of heat waves on mortality, lack of adjustment for ozone and especially PM₁₀ overestimates effect parameters. This bias has implications for public health policy.

Extensive scientific evidence has demonstrated the short-term effects of air pollutants, especially particulate matter, on health.^1–3 Furthermore, the association of ambient temperature with health outcomes has been documented.^4–6 Several studies have examined the health impact of extreme heat events (heat waves), and the 2003 heat wave in Western Europe in particular has received much attention.^7–12

Because both meteorologic variables and concentrations of air pollutants vary on a daily basis, it is reasonable to consider the possibility of confounding in short-term exposure studies of air pollution, meteorology, and health. It has been usual practice to adjust for meteorologic variables (mainly temperature and humidity) when analyzing the effects of air pollution. Adjustment for air pollutants when assessing temperature effects has been less common. Furthermore, effect modification of temperature health effects by air pollution is biologically plausible.^13,14 A few studies have investigated this issue, mainly in US cities or areas providing contrasting results.^14–28 Evidence of interaction between air pollution and high temperatures has been found in some urban areas during summer periods or during specific heat wave episodes.^{16,23,25–27} Furthermore, the joint effects of air pollution and temperature have been found to vary among cities according to local specific characteristics.^26,27 In other studies, pollutants did not appear to modify the temperature-mortality association.^19,21,28 Most of these studies analyzed data from one or more cities in one country or state.

Although most of the studies cited above have addressed joint effects of pollutants with temperature, few have focused on heat wave episodes. Heat waves are of special interest for public health because their frequency is expected to increase with climate change.²⁹

Within the EuroHEAT project, we studied the effects of heat waves on mortality in nine European cities across the continent, using a standardized methodology.³⁰ Here, we explore potential confounding and effect modification by air pollution of the effects of heat waves on mortality, a task that was set a priori, when the project was initiated.

METHODS

The EuroHEAT database includes daily counts of deaths and daily meteorologic and air pollution data from nine European cities (Athens, Barcelona, Budapest, London, Milan, Munich, Paris, Rome, and Valencia) for the years 1990–2004. For each city, data were assembled for daily number of deaths from all natural causes (International Classification of Diseases, Ninth Revision [ICD-9]: 1–799), cardiovascular diseases (CVDs; ICD-9: 390–459), respiratory causes (ICD-9: 460–519), and cerebrovascular causes (ICD-9: 430–438), stratified by sex and age groups (0–64, 65–74, 75–84, and 85+ years).

Daily concentrations of sulphur dioxide (SO₂; mean, 24 hours), PM₁₀ (mean, 24 hours), nitrogen dioxide (NO₂; mean, 24 hours), ozone (O₃, maximum 8-hour moving average), and carbon monoxide (CO; maximum 8-hour moving average) from fixed-site monitors within each city were averaged to provide daily pollutant concentrations.

The meteorologic data consisted of 3-hour air temperature (°C), dew point temperature (°C), wind speed (m/s), and barometric pressure (hPa). From these, the maximum 3-hour daily apparent temperature was calculated and used to define days with “heat waves,” as described in D’Ippoliti et al.³⁰ Briefly, a heat wave is defined as a period of at least 2 days with maximum apparent temperature exceeding the 90th percentile of the monthly city-specific distribution or a period of at least 2 days in which minimum temperature exceeds the 90th percentile of the minimum monthly distribution and maximum apparent temperature exceeds the median monthly value. The analysis was confined to the months of June, July, and August. City-specific estimates were obtained and were then combined using random effects meta-analysis.³¹

For the city-specific analyses, a generalized estimating equation modeling approach was applied.³² We assumed a Poisson distribution for the outcome variable (daily number of deaths) and specified a first-order autoregressive structure, on the assumption that the observations within each summer were correlated (an approach based on previous studies).^5,6 Furthermore, we assumed that observations in different summers were independent. Because the number of clusters (summers) was small and equal to the number of years in the study period, we used the model-based estimator of variance as recommended in the presence of few large clusters.³³

A dummy variable indicating heat wave days was the main exposure variable, and the following potential confounders were introduced as covariates: barometric pressure (linear term), wind speed (linear term), calendar month (set of dummy variables), day of the week (set of dummy variables), holiday (dummy variable), and time trend (quadratic term). Heat wave effects were estimated with and without adjustment for pollutants to assess the magnitude of confounding. In many locations, wind speed is correlated with air pollution; we therefore repeated the analysis without adjusting for wind speed. For the investigation of the modification of heat wave effects by air pollution, an interaction term between heat wave and pollution concentrations was introduced into the model. For better illustration of effect modification, the effect of heat waves is also reported during “high” and “low” air pollution days. High pollution days were defined as days with air pollution levels equivalent to the 75th percentile of the overall pollutant distribution across cities, and low pollution days were defined as days with air pollution levels equivalent to the 25th percentile of the same distribution. The values of the 25th and 75th percentiles were within the range of concentrations observed in all cities for all pollutants. Effect modification was also assessed by dichotomizing the duration of heat waves at the city-specific median.

RESULTS

Table 1 gives descriptive statistics for pollutants and meteorologic variables included in the study. In eTable 1 (https://links.lww.com/EDE/A660), the average daily number of deaths by cause and age group is shown. The average daily number of deaths from natural causes ranged from 15 in Valencia to 144 in London; 28–53% of all deaths were due to cardiovascular causes, and 3–14% were due to respiratory causes.

TABLE 1: Number of Heat Wave Days (Range Across Single Years During Study Period) and Descriptive Characteristics of Maximum Apparent Temperature and Various Air Pollutants During the Summer Months of the Study Period in the Nine Cities Participating in the EuroHEAT Project

Confounding

When we included pollutants in the models for all-cause, all-age mortality, the associations with heat waves were reduced (Table 2). The attenuation of heat wave effect estimates was largest with inclusion of PM₁₀ (~30%) and ozone (~15–25%), with little reduction after adding NO₂, SO₂, and CO. The results were similar when mortality was stratified by age (Figure 1).

TABLE 2: Percent Increase in the Total Daily Number of Deaths, All Ages, in Days With Heat Waves, With and Without Adjustment for Various Pollutantsa (Results from Random Effects Meta-analysis of Estimates from the Nine Cities Participating in the EuroHEAT Project)

FIGURE 1: Percent increase in the total daily number of deaths in days with heat waves, controlling for barometric pressure, wind speed, calendar month, day of the week, holiday, and time trend, with and without adjustment for various pollutants, by age group. Results from random effects meta-analysis of estimates from the nine cities participating in the EuroHEAT project.

Wind speed was associated with air pollutant concentrations, with moderately high correlation coefficients in some locations and low in others (range, 0.002–0.64). Excluding wind speed from the model had very little effect on heat wave and air pollutant effect estimates (data not shown). The wind speed effect on mortality was negligible.

Effect Modification

Table 3 shows the modification of heat wave effect estimates by ozone and PM₁₀ concentrations on total and cause-specific mortality for all ages, for all cities, and by geographic area. When estimates from all cities were combined, the effect of heat waves was larger on high ozone days for total and cardiovascular mortality, but without statistical evidence for interaction. Although the main effect of heat waves in the Mediterranean cities was larger, effect modification was much more evident in the North-Continental cities. A similar pattern was observed for the effect modification by PM₁₀: heat wave effects were stronger on high PM₁₀ days and more pronounced in North-Continental cities.

TABLE 3: Percent Increase in the Daily Number of Deaths for All Ages in Days With a Heat Wave and a “Low” or “High” Level of Ozone and PM10, by Cause of Deatha (Results from Random Effects Meta-analysis of Estimates from the Nine Cities Participating in the EuroHEAT Project)

Figures 2 and 3 show effect modification by ozone or PM₁₀ for each age group for all cities and separately for Mediterranean and North-Continental cities. For all causes of death, the effect of heat waves on mortality was larger among the elderly. For all cities combined, the strongest interaction between ozone and heat waves was for the age group 75–84 years, among whom the estimated increase in daily deaths during heat wave episodes was 54% higher on high compared with low ozone days. For PM₁₀, there was interaction with heat waves for both the 75–84 year and the 85+ year age groups; the corresponding increase in total daily number of deaths during heat wave episodes, on high versus low PM₁₀ days, was 36% and 106%, respectively. The pattern of associations was the same for CVD mortality as for total mortality. Modification pattern of the heat wave effects on respiratory mortality during low and high pollution days was not consistent, although the effect of heat waves on respiratory mortality is larger than on other causes of death. The pattern of associations in Mediterranean and North-Continental cities was comparable, although more pronounced in the latter.

FIGURE 2: Percent increase in the daily number of deaths in days with a heat wave and a “low” or “high” level of ozone, adjusting for barometric pressure, wind speed, calendar month, day of the week, holiday, and time trend, by cause of death and age group. Results from random effects meta-analysis in (A) all cities, (B) Mediterranean cities, and (C) North-Continental cities.

FIGURE 3: Percent increase in the daily number of deaths in days with a heat wave and a “low” or “high” level of PM₁₀, adjusting for barometric pressure, wind speed, calendar month, day of the week, holiday, and time trend, by cause of death and age group. Results from random effects meta-analysis in (A) all cities, (B) Mediterranean cities, and (C) North-Continental cities.

The number of deaths from cerebrovascular causes was small, and so we assessed interaction only for all ages (not by age groups). There was no statistically significant interaction, although the effects of a heat wave were somewhat larger on high ozone and high PM₁₀ days (data not shown).

Modification of the effect of heat waves by ozone and PM₁₀ was also assessed in relation to duration of the heat wave. There were increased heat wave effects with increased pollution levels for both long- and short-duration heat wave events (eTable 2, https://links.lww.com/EDE/A660), although these estimates are less stable. Interaction in both long- and short-duration heat waves was more evident in the North-Continental cities, whereas a less consistent pattern was found in the Mediterranean cities. Heat wave effects did not differ systematically by concentration of NO₂, SO₂, and CO (data not shown).

DISCUSSION

The EuroHEAT project is the first multicity project in Europe to assess heat wave effects on mortality using a common definition for heat wave and a standardized analysis protocol.³⁰ Our definition of heat waves took into account meteorologic variables, physiologic responses to heat, and duration and intensity of the heat wave. To our knowledge, this is the first attempt to estimate modification of heat wave (not temperature) effects on mortality by air pollutants.

Our results provide evidence of effect modification. Specifically, on high ozone and high PM₁₀ days, we observed stronger heat wave effects on total and cardiovascular mortality, especially among the elderly. These findings are consistent with those reported by Katsouyanni et al,¹⁶ who found evidence for synergy of very high temperatures and various air pollutants; by Sartor et al,³⁴ who found evidence of interaction between ozone and temperature at least in the highest temperature ranges; by Parodi et al,³⁵ who found a synergistic effect between ozone and temperature for CVD mortality, especially among the elderly; by Ren et al,²⁶ who also reported a positive interaction of ozone and temperature on CVD mortality; and by Ren et al,¹⁴ who identified a synergistic effect on heart rate variability during the summer period. In addition, our results are consistent with the positive PM-temperature interaction reported by Roberts,³⁶ Ren et al,²⁵ Qian et al,²³ and Stafoggia et al²²—all concerning total and CVD mortality in various parts of the world. Mohr et al²⁰ reported an interaction of temperature and elemental carbon on the number of pediatric emergency asthma visits. However, other studies have reported no synergistic effects: Hales et al²⁸ found no evidence of interaction between PM and temperature on mortality; Basu et al²¹ reported no interaction between several pollutants and temperature in association with mortality; and Zanobetti and Schwartz¹⁹ also found no interaction during the warm season.

Furthermore, we observed that even though heat wave effects were larger in the Mediterranean cities, effect modification was more pronounced in the North-Continental cities. Similarly, Ren et al²⁶ found that ozone positively modified the association between temperature and cardiovascular mortality in northern US cities, whereas the joint effect varied in southern cities. A possible explanation for inconsistency across studies could lie in geographic heterogeneity of the heat wave effect and in the underlying dynamics of susceptible populations at risk of dying from high temperature. In the Mediterranean cities, heat waves are more frequent and characterized by higher temperatures,³⁰ and therefore each episode decreases the pool of susceptible people. Under the hypothesis that heat-susceptible people are also vulnerable to air pollution, the dynamics of this group may reduce the potential for effect modification by a concurrent air pollution exposure during heat waves in Mediterranean cities. In the North-Continental cities where heat waves are rare events, the pool of susceptible people remains more constant over time (and may facilitate the detection of interaction), and a greater proportion of the population is affected (including both heat- and pollution-susceptible subgroups). This hypothesis could be tested in future studies. The geographic patterns may also be a fruitful topic of research, in that they may reflect differences in specific environmental or population characteristics or effective public health measures.

Studies cited above investigated the effects of temperature during either the whole year or in the warm/summer season, but not heat waves specifically. Some of these studies found effect modification by “very high temperatures.”^16,20,34 We assessed effect modification between heat waves and air pollutants, where the effect of the heat waves incorporates effects of high temperature. Because relatively few days can be classified as heat wave days, and because differences in the definition of a heat wave influence the estimation of the effect parameters, the power to detect robust associations is limited.^37,38 Although we know about the heat wave events with devastating consequences, such as the 1987 heat wave in Athens and the 2003 heat wave in Paris—both causing mortality increases of 500%^7,11—we have no a priori reason to think that the interaction with air pollutants is different. These major events are characterized by increased duration. Our results indicate that effect modification by pollution occurs in both long- and short-duration heat wave events. The enhancement of heat wave effects by high concentrations of ozone or PM₁₀ has consequences for public health policies that must target pollutant concentrations when extremely high temperatures are forecast.

There are various plausible explanations for a synergistic association between heat waves and air pollutants. Ozone and a proportion of particles (secondary particles) are generated by processes in the atmosphere in the presence of sunlight and primary emitted pollutants. Sunlight is associated with high temperatures, and so there is likely increased production of secondary pollutants during very hot days.³⁹ On hot days, emissions of pollutants may be further increased by behavioral changes (eg, inhabitants of cities may choose to use their possibly air-conditioned car more often). These differences in pollution sources could lead to pollutants with different characteristics (eg, more toxic). It is also possible that population exposure is better reflected in measurements from fixed-site monitors during hot days if windows are kept open for longer duration, although this may be modified by the presence of air conditioning. Exposure to extreme heat may also make subjects more susceptible to air pollutants by causing physiologic stress or exposure to high ozone or PM concentrations may make them more sensitive to the effects of heat.^13,14 Biologic mechanisms differ according to the pollutant. For example, ozone inhalation may adversely affect cardiovascular function through autonomic nervous system dysfunction.¹⁴

We also examined the possible confounding role of various pollutants by adjusting for them in the models assessing heat wave effects. The effects of heat waves on total mortality were reduced after adjustments for PM₁₀ (by ~30%); ozone (by ~15–25%); and NO₂, SO₂, and CO (by ~10%). O’Neill et al⁴⁰ reported similar results in an analysis of apparent temperature effects confounded by PM₁₀ and ozone. In contrast, Pattenden et al⁴¹ reported only slight confounding by black smoke, whereas Keatinge and Donaldson⁴² and Kassomenos et al⁴³ reported no confounding using SO₂, CO, and black smoke as pollution indicators, with various meteorologic variables. It should be noted that black smoke as a particle indicator reflects mainly primary particles, in contrast to PM₁₀, which includes primary and secondary particles. In addition, Rainham and Smoyer-Tomic⁴⁴ and Keatinge and Donaldson⁴⁵ reported no substantial confounding by ozone. The above articles assessed various meteorologic variables as exposure variables, but not heat waves. It has been shown that by simultaneously including both a temperature and a heat wave term in the model, the effect of heat waves could become null depending on the degree of flexibility applied to the modeling of the temperature terms.³⁸ Such an analysis is beyond the scope of the EuroHEAT project and also there is no reason to expect a change in the modification of heat/heat wave effects by air pollution.

Our study has some limitations. Data characterizing cities were not available to be considered as effect modifiers, and thus we were unable to adjust for lifestyle and housing conditions, which can modify the exposure of populations to heat waves. The differential use of air conditioning and possibly other socioeconomic characteristics may explain a large part of the differences across geographic areas. Our study had the methodological advantage of using standardized methods for the definition of exposure, outcome, and confounders across different cities in Europe.

In conclusion, we found that PM₁₀ and ozone modify the heat wave effects on total and CVD mortality. We also found that in the absence of adjustment for ozone and especially PM₁₀, the effect of heat waves on mortality was overestimated. The implications of these findings are important for policy measures; for example, when there is a heat wave forecast, additional measures to reduce air pollutant concentrations may be appropriate.

ACKNOWLEDGMENT

The EuroHEAT Project was coordinated by the WHO regional Office for Europe.