全球变化知识资源中心(globalchange): Vulnerability to the mortality effects of warm temperature in the districts of England and Wales

Warm temperatures adversely affect disease occurrence and death, in extreme conditions as well as when the temperature changes are more modest^{1, 2}. Therefore climate change, which is expected to affect both average temperatures and temperature variability, is likely to impact health even in temperate climates. Climate change risk assessment is enriched if there is information on vulnerability and resilience to effects of temperature. Some studies have analysed socio-demographic characteristics that make individuals vulnerable to adverse effects of temperature^{1, 2, 3, 4}. Less is known about community-level vulnerability. We used geo-coded mortality and environmental data and Bayesian spatial methods to conduct a national small-area analysis of the mortality effects of warm temperature for all 376 districts in England and Wales. In the most vulnerable districts, those in London and south/southeast England, odds of dying from cardiorespiratory causes increased by more than 10% for 1 °C warmer temperature, compared with virtually no effect in the most resilient districts, which were in the far north. A 2 °C warmer summer may result in 1,552 (95% credible interval 1,307–1,762) additional deaths, about one-half of which would occur in 95 districts. The findings enable risk and adaptation analyses to incorporate local vulnerability to warm temperature and to quantify inequality in its effects.

Events such as the 2003 European heatwave showed the need to identify people and communities vulnerable to the adverse effects of current weather as well as those of a changing climate^{5, 6}. Some studies have investigated whether the effects of temperature depend on individual sociodemographic characteristics, for example, their age or gender^{3, 4}. There has been less work on community-level vulnerability because studies tend to quantify, or at least pool and report, effects for whole countries or large geographical units. Community-level analysis is important for risk and adaptation assessment for a number of reasons: first, some determinants of vulnerability are related to community characteristics. Second, community-level analysis helps measure inequalities in effects, which are important above and beyond aggregate impacts. Third, if the units of analysis are administrative units such as districts, the results map directly to the scale of resource allocation and policy/programme implementation. As there are few local analyses of the effects of temperature, and none covering a whole country, assessments of the health effects of climate change have typically assumed that the observed relationships between temperature and health are transferrable across communities⁷.

Some studies have constructed indicators of community vulnerability to weather and climate change based on theoretical considerations^{8, 9, 10}, but have not assessed whether these indicators are associated with smaller or larger health effects. Some recent studies have analysed variations in health effects of temperature at finer spatial resolutions^{11, 12, 13, 14, 15, 16}. A few of these have focused on short-term episodes^{11, 12}; others were in small regions^{13, 14, 15, 16}, possibly limiting generalizability. To identify vulnerable and resilient communities, we analysed the effects of warm temperature on mortality in a national study with high spatial resolution.

We quantified the effects of warm temperature on mortality from cardiorespiratory causes for all 376 local authority districts in England and Wales for the period 2001–2010. Cardiorespiratory diseases form a parsimonious set of causes of death that have been consistently linked with increased temperature¹; they account for about half of all deaths in England and Wales. In sensitivity analysis, we used all non-injury deaths.

We adjusted for other factors that also vary daily and may affect cardiorespiratory mortality, including air pollution¹⁷ (particulate matter below 10 μm in aerodynamic diameter, PM₁₀, concentration in the main analysis; PM₁₀ and ozone in sensitivity analysis) and whether a case/control day was a national holiday as differences in behavior and provision of health service may affect mortality on holidays. We conducted separate analyses for men and women and by age group, because both the effects of temperature and their spatial patterns may differ by age and gender. The analysed age groups were <75, 75–84 and 85+ years. The youngest age group was not further divided because only 26% of cardiorespiratory deaths occurred below 75 years and only 4% below 55 years.

To examine vulnerability and resilience, we allowed the magnitude of effect to vary by district. We used a Bayesian spatial model, which uses the empirical similarity of the effects of temperature in neighbouring districts to borrow strength across districts. This approach balances between unstable district-specific estimates (due to the relatively small number of daily deaths in each district) and overly aggregated large-area estimates (for example, regional/national level) that mask local variation. The Bayesian framework estimates both the magnitude of effect and how confident we are about its differences from the national response. We conducted a number of sensitivity analyses, detailed in the Methods, to examine the robustness of our results to analytical choices.

After removing 205 deaths whose postcodes could not be matched to any district, there were 921,288 cardiorespiratory deaths in May–September of 2001–2010 (47% of all deaths in England and Wales over this period). Of these, 441,788 were among men and 479,500 among women. Of these deaths, 26% were below 75 years of age, 35% between 75 and 84 years and 39% in those aged 85 years or older. Nationally, women aged 85+ years were the most vulnerable to the effects of warm temperature, that is, had a larger response (Table 1), consistent with most previous analyses^{3, 4, 18, 19}. Above 75 years of age, women were more vulnerable to warm temperature than men, consistent with other studies of temperature³.

Table 1: National-level per cent increase in the odds of cardiorespiratory death for 1 °C increase in mean daily summer temperature above district-specific thresholds.

Full table

Data.

Death records were from national vital statistics held by the UK Small Area Health Statistics Unit. Age, postcode of residence at time of death and date of death were available for each record. We used deaths that occurred in May–September. In sensitivity analysis, we restricted the analysis to the warmest three months (June–August).

Daily temperatures at 5 km × 5 km resolution were from the UK Met Office with methods described elsewhere²⁶. In brief, the daily temperature in each grid is estimated based on inverse-distance-weighted interpolation of monitoring data, also accounting for latitude and longitude, elevation, coastal influence and proportion of urban land use. In the main analysis, we used mean daily temperature, calculated as the average of daily maximum and minimum temperatures. In sensitivity analysis, we used daily maximum temperature. We used the average of temperature on day of death and the preceding three days in the main analysis to allow for the possibility that the health effects may be cumulative^{18, 19, 22, 23}; this duration of averaging provides near-identical results to a more flexible model of lagged temperatures²⁴. In sensitivity analysis, we used alternative lag durations, including temperature on the day of death alone, which is found to have the strongest association with mortality^{1, 27}.

We calculated daily PM₁₀ concentrations by combining two data sources: annual average PM₁₀ levels in 1 km × 1 km grids from the UK government Department for Environment, Food and Rural Affairs (DEFRA) Ambient Air Quality Assessment and daily monitoring data for every day of the analysis period from DEFRA’s Automatic Urban and Rural Network. As PM₁₀ concentrations are correlated across monitors over time, we estimated location-specific daily PM₁₀ by scaling the gridded annual estimates so that they preserved the daily patterns of the data from monitors. We used an average of PM₁₀ on the day of death and on preceding two days based on evidence from previous studies²⁸, with sensitivity analysis around lag duration.

We assigned temperature and PM₁₀ on/preceding the day of death to each death record (and to control days as defined below) from the grid that contained the coordinates of the centroid of the decedent’s postcode of residence (average area covered by a postcode is <0.1 km²; median and 99th percentile of distances between centroid of each postcode and its nearest neighbour are 52 m and 760 m, respectively). Using high-resolution temperature data is advantageous to regional daily temperature because day-to-day temperature changes can be highly localized (Supplementary Fig. 7). Rural–urban status was based on the Office for National Statistics classification. Deprivation was measured using the Carstairs score, which combines indicators on unemployment, social class, crowding of housing and (lack of) vehicle ownership. These variables and green space (as the percentage of the Census ward)²⁹ were assigned to each death record through the coordinates of the centroid of the postcode of residence.

Statistical methods.

We used a time-stratified case-crossover design, commonly used for analysing effects of short-term exposures. Temperature on the day of death (case day), and as relevant preceding days, is compared with the temperature on control days on which the death did not occur³⁰. The case-crossover design naturally controls for potential confounding factors that are time-invariant or that vary slowly over time, for example, ethnicity, socio-economic status, smoking, healthcare and obesity. We used control days on the same day of the week as the case day, to automatically adjust for day of the week, and in the same calendar month to avoid the so-called overlap bias³¹.

The relationship between warm temperature and mortality is typically non-linear with virtually no effect below some threshold, and a dose-response at higher temperatures, confirmed in exploratory analysis of our data. Therefore, we analysed the temperature-mortality association using a piecewise linear model, as also used in previous national or regional analyses^{1, 3, 19, 23, 25}. We set the thresholds in two alternative approaches: district-specific thresholds (main analysis) and a common threshold for all districts (sensitivity analysis). The first approach implies that the hazardous effects of temperature start at a higher temperature where the long-term temperature is warmer. The second approach implies that hazardous effects begin at about the same temperature in all districts. We selected both types of threshold empirically, by using the Deviance Information Criterion. The district-specific thresholds were the 85th percentile of each district’s summer temperatures and the common threshold was 18 °C. District-specific thresholds ranged between 15.4 °C in northwest England and 19.9 °C in the City of London.

We used a Bayesian spatial model to borrow strength across districts. In this approach, the estimated effect in each district is influenced by its own data and by those of its neighbours. The extent to which neighbours influence one another depends on the variance of the estimated effects in each district and on the empirical similarity among neighbouring districts. We report the district-specific percentage change in the odds of mortality per 1 °C change in temperature, and the posterior probability that the estimated effect size is larger than the national average. We examined whether the proportional effect size itself is associated with characteristics of the community of residence by introducing these variables into the model as modifiers of the effect size. Detailed model specification is provided in Supplementary Methods.

All analyses were done in the software WinBUGS.


Email:
Passwd
验证:
	换一张
Have you forgotten your password? Stay signed in Log In

全球变化知识资源中心

Data.

Statistical methods.

用户服务

站点统计

相关链接

联系我们