https://orcid.org/0000-0003-4746-8236
Abstract
The complex role of urbanisation in heat-mortality risk has not been fully studied. Japan has experienced a rapid population increase and densification in metropolitan areas since the 2000s; we investigated the effects of population concentration in metropolitan areas on heat-mortality risk using nationwide data.
We collected time-series data for mortality and weather variables for all 47 prefectures in Japan (1980–2015). The prefectures were classified into three sub-areas based on population size: lowest (<1 500 000), intermediate (1 500 000 to 3 000 000), and highest (>3 000 000; i.e. metropolitan areas). Regional indicators associated with the population concentration of metropolitan areas were obtained.
Since the 2000s, the population concentration intensified in the metropolitan areas, with the highest heat-mortality risk in prefectures with the highest population. Higher population density and apartment % as well as lower forest area and medical services were associated with higher heat-mortality risk; these associations have generally become stronger since the 2000s.
Population concentration in metropolitan areas intensified interregional disparities in demography, living environments, and medical services in Japan; these disparities were associated with higher heat-mortality risk. Our results can contribute to policies to reduce vulnerability to high temperatures.
We investigated the effects of population concentration in metropolitan areas on the heat-mortality risk in Japan, using nationwide time-series data covering all 47 prefectures from 1980–2015.
Since the late 1990s, non-metropolitan areas have suffered rapid depopulation due to economic degenerations, and non-metropolitan people have migrated to metropolitan areas for work.
During the 2000s, the heat-mortality risk also showed a time trend that might be associated with the population shift. Before the 2000s, metropolitan areas (prefecture over 3 000 000) did not show a higher heat-mortality risk than non-metropolitan areas; however, heat-mortality risk has become higher in metropolitan areas since the 2000s; this pattern has generally become stronger over time.
In addition, the characteristics of metropolitan areas (higher population density and apartment %, and lower green space and access to medical services) were associated with heat-mortality risk; these associations were generally more evident in the 2000s than in previous periods.
Introduction
Heat-mortality risk is frequently reported to be higher in cities.1,2 This is attributed in part to urban heat island effects1 that could be compounded by the population concentration together with rising temperatures from climate change. However, other studies indicate higher vulnerability in non-urban areas. These studies suggest that non-climate vulnerability factors, such as ageing and air conditioner prevalence, could explain urban/non-urban disparities in heat-mortality risk.3–5
Most studies quantifying associations between heat vulnerability and urban/non-urban areas2,3,6 did not address the complexity of population concentration in urban areas (i.e. urbanisation). Urbanisation is associated with factors that reduce heat-mortality risk, such as higher income and better health;7 however, the associated overcrowding is related to factors that increase heat-mortality risk, such as urban heat islands, air quality deterioration, income inequality, and housing problems.8,9 Further, the characteristics of population concentration in urban areas may differ by area (developing or developed areas) and period (growth or recession periods). Therefore, studies of the health effects of population concentration in urban areas should consider the complicated characteristics of urbanisation that underlie the social contexts of the study areas.
Since the 1980s, Japan has experienced two major population shifts to metropolitan areas. The first took place in the bubble period (mid-1980s to early 1990s), when there was a decline in manufacturing in non-urban areas due to a stronger yen and the growth of metropolitan areas, along with developments in the service and financial industries. Thus, during the first shift, people were drawn to the metropolitan areas for employment and lifestyle reasons.10,11 The second population shift to metropolitan areas began in the late 1990s, along with a decade-long economic recession after the collapse of the real estate bubble. This shift was a result of a deterioration in rural economic status from population decline, together with slumps in the manufacturing and construction industries.10–12 Unlike with the previous shift, it was difficult to say that employment status in metropolitan areas in this period was desirable owing to aftereffects of the bubble collapse, the global economic recession, and a consequent increase in temporary employment.11,13,14 Thus, during the second shift, people were driven from non-urban areas by exacerbating economic conditions and employment capacity.10,11 The population disparities between metropolitan and regional areas have heightened and become a critical social issue in recent years. Furthermore, despite consequences for the temporal and spatial distributions of heat vulnerability, the effects of this population shift on heat vulnerability have not, to our knowledge, been examined.
We investigated the effects of population concentration within metropolitan areas on vulnerability to heat-related mortality (hereafter, heat-mortality risk) in Japan, using nationwide time-series data covering all 47 prefectures over the period of 1980–2015. Using a two-stage analytic approach, our study analysed changes in population and heat-mortality risk over time in both metropolitan and non-metropolitan areas and examined whether differences in risk between metropolitan and non-metropolitan areas occurred. In addition, we aimed to discern the regional indicators that can partly explain differences in heat-mortality risk among prefectures.
Methods
Sub-areas classified by population
We averaged the population for each prefecture during the study period and used the averages to divide all prefectures into three sub-areas: lowest population (21 prefectures; population under 1 500 000), intermediate population (16 prefectures; population from 1 500 000 to 3 000 000), and highest population (10 prefectures; population over 3 000 000; i.e. metropolitan areas). This stratification was used as an approximation of the urbanisation level of prefectures. Due to high variations in area size among the prefectures, we determined that the absolute population size is a more suitable urbanisation indicator compared to other potential indicators (e.g. population density, the percentage of developed area, etc.) that could be affected by the area size of prefectures. Additionally, the terminology ‘population concentration’ in the study indicates an increase in the absolute population number. The population data were collected from the Statistics Bureau of Japan (URL: https://www.e-stat.go.jp/).
Weather and mortality data
We collected daily time-series data on all-cause mortality and weather variables for each of the prefectures in Japan for the period of 1980–2015. The data were restricted to the summer, identified as the four warmest months (June to September). Daily mean temperatures (°C) and daily mean relative humidity (%) were provided by the Japan Meteorological Agency. For each prefecture, these weather variables were measured from a single monitor station in the capital city of each prefecture. All-cause mortality data were obtained from the Ministry of Health, Labour and Welfare in Japan and stratified by age (0–64 and 65 y+) and sex.
Regional indicators
In order to explain the difference in heat-mortality risk by sub-areas, we collected data on six regional indicators (at the prefecture-level) covering demographic, living environment, and emergency medical services, which are associated with population size (Pearson’s correlation coefficient > 0.3). All indicators were collected from the Statistics Bureau of Japan (URL: https://www.e-stat.go.jp/). All indicators were recalculated as the average values for each sub-period (i.e. one decade; details described in statistical analysis), and the average values were used as explanatory variables in the meta-regression.
First, we collected annual prefecture-level demography data, including population density (persons per 1 km2 of an inhabitable area) and ageing index (the ratio of the number of persons ≥ 65 years to the number of persons ≤ 14 years) for 1980–2015.
Second, we obtained annual prefecture-level living environment indicators from the Statistics Bureau of Japan, including % of apartments (to total number of houses; conducted every 5 years) for 1983–2013 and % of forest area (to the total area; conducted in March 1980, 1984, 1990, 2000, 2009, and 2014) for 1980–2014.
Third, we obtained annual prefecture-level emergency medical services indicators, including the number of hospitals and clinics providing emergency medical services per 100 000 persons (hereafter, the number of emergency medical centres) for 1981–2014 (conducted every 3 years) and the number of ambulances per 100 000 persons (hereafter, the number of ambulances) for 1980–2015.
Confounders
We considered several confounding variables that could affect the associations between regional indicators and heat-mortality risk. First, although it showed low association with population size (Pearson’s correlation coefficient < 0.1), the number of air-conditioners per 1000 households with two or more persons (conducted every 5 years for 1984–2014) was collected to consider the effects of air conditioning on heat-mortality risk, which has been suggested as a major factor influencing heat-mortality risk in recent decades.15 Second, to consider socioeconomic characteristics, average monthly income per household (1980–2015) and unemployment rate (conducted every 5 years for 1980–2015) were collected. Third, the average and range of daily mean summer temperatures (calculated from the weather data) were also included to consider climatic characteristics. All confounding variables were recalculated as the average values for each sub-period, and these were used as confounders in the meta-regression.
Statistical analyses
We divided the study period into 10-year, overlapping time frames with the starting year increased by one (i.e. a moving period) to examine the change in heat-mortality risk over time. In other words, a total of 27 serial sub-period datasets was considered for each prefecture: the first sub-period is 1980–1989, the second is 1981–1990, and so on, with the last sub-period 2006–2015. For each sub-period, we applied a two-stage analysis to calculate heat-mortality risks and to examine associations between the heat-mortality risk and regional indicators. We conducted the two-stage analysis for the total population as well as separately by age (< 65 and ≥ 65 years) and sex. We used R statistical software (version 3.5.3) for all statistical analyses.
In the first stage, we estimated the lag-cumulative summer temperature-mortality association for each prefecture using a generalised linear model with quasi-Poisson distribution and a distributed lag non-linear model.16 We modelled the cross-basis for the summer temperature-mortality association; a quadratic B-spline with two internal knots (placed at the 50th and 90th percentiles of the prefecture-specific summer temperature distribution for each sub-period) for the exposure-response relationship and a natural cubic B-spline with an intercept and two internal knots placed at equally spaced values on the log scale for the lag-response relationship. Ten-day lag periods were selected to capture delayed effects of heat. Relative humidity was adjusted by using a distributed lag model with a natural cubic B-spline for the lag response with an intercept and two equally spaced knots on the log scale. These choices of modelling specifications were based on previous studies.15,17,18 Temporal trends within a season were controlled by using a natural cubic B-spline of day of the season with equally spaced knots and four degrees of freedom (df). In addition, long-term trends were controlled using a natural cubic B-spline of time with equally spaced knots and 1 df per 10 years. Day of the week was also adjusted as an indicator variable.
In the second stage, we conducted a meta-analysis with a random intercept to obtain the prefecture-specific best linear unbiased predictor (BLUP) and the pooled temperature-mortality associations for each sub-area. First, using the BLUPs, we identified the minimum mortality temperature for each prefecture. Then, we obtained pooled estimates using a meta-regression with sub-area indicator variables as a meta-predictor. These analytic frameworks were based on a previous study.19 We repeated all procedures of the two-stage analysis for each sub-period. We finally defined the heat-mortality risk based on the relative risk (RR) of mortality for the 99th percentile of the summer temperature distribution versus the minimum mortality temperature for each sub-period.15
Relationships between regional indicators and heat-mortality risk
In order to explain the temporal and spatial differences in the heat-relation mortality risk according to the population size, we applied meta-regression with regional indicators. For each indicator and each sub-period, we applied separate meta-regression models. All confounders were adjusted in all meta-regression models. The associations were presented as the percentage change in heat-mortality risk per unit change in each meta-variable.
As an additional analysis, we applied ridge regression to examine whether each relationship between regional indicators and heat-mortality risk exists after all other indicators are considered. Detailed procedures regarding ridge regression are reported in the Supplementary Data (2. Ridge regression), available as Supplementary data at IJE online.
Sensitivity analysis
Several sensitivity analyses were performed to examine whether our results are consistent to the modelling specifications, definition of heat-mortality risk, potential outliers, and alternative green space indicator. Details of the sensitivity analysis are included in the Supplementary Data (3. Sensitivity analysis), available as Supplementary data at IJE online.
Results
Figure 1A presents the distributions of the average population during the study period. Most of the prefectures with the highest population were in the string of industrialized and urbanized areas along the Pacific coast between Tokyo and Fukuoka including Aichi and Osaka megalopolis known as the Pacific Belt Zone. Figure 1B displays the prefecture-specific average summer temperature during the same period. Supplementary Table S1, available as Supplementary data at IJE online shows prefecture-specific and national summary statistics for weather and mortality data.
Distributions of population and average summer temperature. Spatial distributions of (a) average population and (b) average of summer temperature during the entire study period (1980–2015)
Table 1 shows descriptive information on the regional indicators and the confounders during the study period by sub-areas. Averages of population density and apartment % were the highest, while averages of the forest area % and medical service indicators were lowest in the highest population areas. Supplementary Table S2, available as Supplementary data at IJE online displays the correlations between the regional indicators and population, and all correlation coefficients were estimated over 0.35.
Summary statistics of regional indicators and confounders in Japan (1980–2015). Values: mean (range)
| Lowest population | Intermediate population | Highest population | Total | |
|---|---|---|---|---|
| Regional indicators | ||||
| Demographya | ||||
| Population density | 770.9 (376.3, 1647.2) | 879.1 (499.1, 2283.1) | 3344.6 (257.0, 8901.1) | 1355.3 (257.0, 8901.1) |
| Ageing index | 1.3 (0.6, 1.6) | 1.2 (1.0, 1.4) | 1.0 (0.9, 1.2) | 1.2 (0.6, 1.6) |
| Living environmentb | ||||
| % of apartments | 20.3 (13.2, 45.5) | 22.5 (15.8, 35.5) | 42.8 (25.5, 65.5) | 25.8 (13.2, 65.5) |
| % of forest area | 67.9 (44.5, 83.2) | 64.3 (32.1, 79.8) | 45.9 (30.9, 68.2) | 62.0 (30.9, 83.2) |
| Emergency medical servicec | ||||
| Number of emergency medical centers | 5.0 (1.7, 9.0) | 4.8 (2.8, 6.7) | 3.6 (2.6, 5.8) | 4.6 (1.7, 9.0) |
| Number of ambulances | 5.6 (3.5, 9.1) | 5.1 (3.5, 6.6) | 3.3 (2.1, 5.8) | 5 (2.1, 9.1) |
| Confounder | ||||
| Number of air-conditioner (per 1000 households) | 1833.6 (467.6, 2677.1) | 1811.4 (723.4, 2495.1) | 1906.5 (138.1, 2417.6) | 1841.5 (138.1, 2677.1) |
| Monthly income per household (Yen) | 513.8 (378.3, 614.4) | 514.1 (448.7, 568.2) | 509.3 (455.3, 571.8) | 512.9 (378.3, 614.4) |
| Unemployment rate (%) | 4.4 (2.9, 9.0) | 4 (3.0, 4.7) | 4.5 (3.4, 5.9) | 4.2 (2.9, 9.0) |
| Lowest population | Intermediate population | Highest population | Total | |
|---|---|---|---|---|
| Regional indicators | ||||
| Demographya | ||||
| Population density | 770.9 (376.3, 1647.2) | 879.1 (499.1, 2283.1) | 3344.6 (257.0, 8901.1) | 1355.3 (257.0, 8901.1) |
| Ageing index | 1.3 (0.6, 1.6) | 1.2 (1.0, 1.4) | 1.0 (0.9, 1.2) | 1.2 (0.6, 1.6) |
| Living environmentb | ||||
| % of apartments | 20.3 (13.2, 45.5) | 22.5 (15.8, 35.5) | 42.8 (25.5, 65.5) | 25.8 (13.2, 65.5) |
| % of forest area | 67.9 (44.5, 83.2) | 64.3 (32.1, 79.8) | 45.9 (30.9, 68.2) | 62.0 (30.9, 83.2) |
| Emergency medical servicec | ||||
| Number of emergency medical centers | 5.0 (1.7, 9.0) | 4.8 (2.8, 6.7) | 3.6 (2.6, 5.8) | 4.6 (1.7, 9.0) |
| Number of ambulances | 5.6 (3.5, 9.1) | 5.1 (3.5, 6.6) | 3.3 (2.1, 5.8) | 5 (2.1, 9.1) |
| Confounder | ||||
| Number of air-conditioner (per 1000 households) | 1833.6 (467.6, 2677.1) | 1811.4 (723.4, 2495.1) | 1906.5 (138.1, 2417.6) | 1841.5 (138.1, 2677.1) |
| Monthly income per household (Yen) | 513.8 (378.3, 614.4) | 514.1 (448.7, 568.2) | 509.3 (455.3, 571.8) | 512.9 (378.3, 614.4) |
| Unemployment rate (%) | 4.4 (2.9, 9.0) | 4 (3.0, 4.7) | 4.5 (3.4, 5.9) | 4.2 (2.9, 9.0) |
Demographic variables: Population density: persons per 1 km2; Ageing index: % of the number of persons ≥ 65 years to the number of persons ≤ 14 years.
Living environment variables: % of apartments (annual values for the % to the total number of houses) and % of forest area (annual values for the % to the total areas).
Emergency medical service variables: The number of Emergency medical centers/Ambulances per 100 000 persons, individually.
Summary statistics of regional indicators and confounders in Japan (1980–2015). Values: mean (range)
| Lowest population | Intermediate population | Highest population | Total | |
|---|---|---|---|---|
| Regional indicators | ||||
| Demographya | ||||
| Population density | 770.9 (376.3, 1647.2) | 879.1 (499.1, 2283.1) | 3344.6 (257.0, 8901.1) | 1355.3 (257.0, 8901.1) |
| Ageing index | 1.3 (0.6, 1.6) | 1.2 (1.0, 1.4) | 1.0 (0.9, 1.2) | 1.2 (0.6, 1.6) |
| Living environmentb | ||||
| % of apartments | 20.3 (13.2, 45.5) | 22.5 (15.8, 35.5) | 42.8 (25.5, 65.5) | 25.8 (13.2, 65.5) |
| % of forest area | 67.9 (44.5, 83.2) | 64.3 (32.1, 79.8) | 45.9 (30.9, 68.2) | 62.0 (30.9, 83.2) |
| Emergency medical servicec | ||||
| Number of emergency medical centers | 5.0 (1.7, 9.0) | 4.8 (2.8, 6.7) | 3.6 (2.6, 5.8) | 4.6 (1.7, 9.0) |
| Number of ambulances | 5.6 (3.5, 9.1) | 5.1 (3.5, 6.6) | 3.3 (2.1, 5.8) | 5 (2.1, 9.1) |
| Confounder | ||||
| Number of air-conditioner (per 1000 households) | 1833.6 (467.6, 2677.1) | 1811.4 (723.4, 2495.1) | 1906.5 (138.1, 2417.6) | 1841.5 (138.1, 2677.1) |
| Monthly income per household (Yen) | 513.8 (378.3, 614.4) | 514.1 (448.7, 568.2) | 509.3 (455.3, 571.8) | 512.9 (378.3, 614.4) |
| Unemployment rate (%) | 4.4 (2.9, 9.0) | 4 (3.0, 4.7) | 4.5 (3.4, 5.9) | 4.2 (2.9, 9.0) |
| Lowest population | Intermediate population | Highest population | Total | |
|---|---|---|---|---|
| Regional indicators | ||||
| Demographya | ||||
| Population density | 770.9 (376.3, 1647.2) | 879.1 (499.1, 2283.1) | 3344.6 (257.0, 8901.1) | 1355.3 (257.0, 8901.1) |
| Ageing index | 1.3 (0.6, 1.6) | 1.2 (1.0, 1.4) | 1.0 (0.9, 1.2) | 1.2 (0.6, 1.6) |
| Living environmentb | ||||
| % of apartments | 20.3 (13.2, 45.5) | 22.5 (15.8, 35.5) | 42.8 (25.5, 65.5) | 25.8 (13.2, 65.5) |
| % of forest area | 67.9 (44.5, 83.2) | 64.3 (32.1, 79.8) | 45.9 (30.9, 68.2) | 62.0 (30.9, 83.2) |
| Emergency medical servicec | ||||
| Number of emergency medical centers | 5.0 (1.7, 9.0) | 4.8 (2.8, 6.7) | 3.6 (2.6, 5.8) | 4.6 (1.7, 9.0) |
| Number of ambulances | 5.6 (3.5, 9.1) | 5.1 (3.5, 6.6) | 3.3 (2.1, 5.8) | 5 (2.1, 9.1) |
| Confounder | ||||
| Number of air-conditioner (per 1000 households) | 1833.6 (467.6, 2677.1) | 1811.4 (723.4, 2495.1) | 1906.5 (138.1, 2417.6) | 1841.5 (138.1, 2677.1) |
| Monthly income per household (Yen) | 513.8 (378.3, 614.4) | 514.1 (448.7, 568.2) | 509.3 (455.3, 571.8) | 512.9 (378.3, 614.4) |
| Unemployment rate (%) | 4.4 (2.9, 9.0) | 4 (3.0, 4.7) | 4.5 (3.4, 5.9) | 4.2 (2.9, 9.0) |
Demographic variables: Population density: persons per 1 km2; Ageing index: % of the number of persons ≥ 65 years to the number of persons ≤ 14 years.
Living environment variables: % of apartments (annual values for the % to the total number of houses) and % of forest area (annual values for the % to the total areas).
Emergency medical service variables: The number of Emergency medical centers/Ambulances per 100 000 persons, individually.
Figure 2A shows the population percentages (to the total nationwide population for each time period) during the study period for each sub-area. The highest population areas showed a constant increase in the population percentage throughout the whole study period, whereas areas with the lowest and intermediated population sizes showed decreasing trends since the 2000s. Figure 2B presents temporal changes in the population percentages from the baseline period. Population concentration in the highest population areas was more pronounced in the 2000s. The changes in population during the study period for each sub-area are displayed in Supplementary Figure S1, available as Supplementary data at IJE online, which shows that an increase in population only appeared in the highest population area around 2000, while depopulation were seen in most of the other sub-areas in the same period.
Time-trends of the population change and heat-related mortality risk for each sub-area (1980–2015). (A) Percentage of the population during the study period (percentage of the average population for each sub-area compared to the average of the total population at each sub-period). (B) Changes in the percentage of the population during the study period: (difference between the corresponding sub-period and the baseline period compared to the baseline sub-period 1980–1989). (C) Time-varying heat-related mortality risk for all sub-period. All prefectures were divided into three population levels based on the average population during 1980–2015: >3 000 000 (highest population), 1 500 000–3 000 000 (intermediate population), and <1 500 000 people (lowest population). Results are presented at the midpoint for each sub-period (10-year subsets)
Figure 2C shows the temporal changes in the pooled heat-mortality risk for the three sub-areas during the study periods. Until the end of the 1990s, all sub-areas showed a decreasing trend in heat-mortality risk over time. For the 1980s, we did not observe distinguishable differences in heat-mortality risk among sub-areas; however, the highest population areas showed the highest heat-mortality risk since the 1990s. The risk differences between the highest and other sub-areas has become more evident since the mid-2000s, with the corresponding p-values near or less than 0.05 (Supplementary Figure S4, available as Supplementary data at IJE online). Supplementary Figure S5, available as Supplementary data at IJE online shows the temporal trends of the 99th percentile of the summer temperature that increased more dramatically in the intermediate and highest population areas than in the lowest population areas during the study period. Supplementary Figure S6, available as Supplementary data at IJE online displays the temporal trends of the confounders. The age and sex-stratified results for Figure 2C are displayed in Supplementary Figure S7, available as Supplementary data at IJE online. The results shown in Figure 4 were generally consistent across all sub-populations.
Figure 3 displays the temporal changes in average values of regional indicators compared with the baseline sub-period. Unlike other areas showing a decreasing trend after the mid-1990s, the highest population areas showed a consistent increment in average population density. Although the percentage of apartment/forest area has increased/decreased during the study period for all sub-areas, the margin of change was most prominent in the highest population areas. The average number of emergency medical centres decreased gradually for all areas since the 1990s; the decrease was highest in areas with the highest and moderate population sizes. Additionally, although the number of ambulances increased over time for all sub-areas, the increase was lowest in the highest population areas. Furthermore, for most of the regional indicators, the disparities among sub-areas (i.e. standard deviation) tended to increase over time and have intensified since the late 2000s (Supplementary Figure S8, available as Supplementary data at IJE online).
Time trend of regional indicators (1980–2015). Changes in averages of regional indicators (difference between the corresponding sub-period and the baseline period) compared with those of 1980–1989 (baseline sub-period). All prefectures were divided into three population levels based on the average population during 1980–2015: >3 000 000 (highest population), 1 500 000–3 000 000 (intermediate population), and <1 500 000 people (lowest population). Results are presented at the midpoint for each sub-period (10-year subsets). The numbers of emergency medical services and the number of ambulances: Both are the number per 100 000 persons
Figure 4 displays the time-varying associations between the regional indicators and heat-mortality risk during the study period. For most of the indicators, the associations with heat-mortality risk have become more pronounced since the 2000s (lower confidence intervals > or near 1). Since the mid- or late 1980s, higher values for population density and apartment % were associated with higher heat-mortality risk, and these associations have become more prominent since the 2000s. The negative associations between forest area % and heat-mortality risk were also slightly more evident for the 2000s. In addition, negative associations between heat-mortality risk and values of the emergency medical services have become pronounced since the 2000s. Table 2 shows the associations between regional indicators and heat-mortality risk in the last sub-period (2006–2015), which was estimated by the meta-regressions. The age and sex-stratified results corresponding to Figure 4 are displayed in the Supplementary Material (Supplementary Figure S9 and S10), available as Supplementary data at IJE online. The associations shown in Figure 4 were generally consistent across all age-groups (although most of the estimated percentage changes were greater in people aged 0–64), whereas they were relatively stronger in males than in females.
Time-varying association between regional indicators and heat-mortality risk (1980–2015). The associations are presented as the percentage (%) change in heat-mortality risk per unit change in each meta-variable, except for population density (per 1000 persons increase). Results are presented at the midpoint for each sub-period (10-year subsets). The numbers of emergency medical services and the number of ambulances: Both are the number per 100 000 persons.
Association between urbanization indicators and heat-related mortality risk in the latest sub-period (2006–2015) for the total population. Results are expressed as percentage change (95% confidence interval) in heat-related mortality risk per 1-unit increase in each meta-variable, except for population density (per 1000 persons increase). Each meta-variable is an average value of each regional indicator measured during the latest sub-period (2006–2015)
| Total | People Aged 65 y+ | People Aged 0-64 | Male | Female | |
|---|---|---|---|---|---|
| Demographya | |||||
| Population density | 1.17 (0.09, 2.27) | 1.02 (–0.05, 2.10) | 1.46 (–0.33, 3.29) | 1.97 (0.91, 3.04) | 0.26 (-1.33, 1.88) |
| Ageing index | –4.40 (–12.71, 4.71) | –3.55 (–12.08, 5.80) | –10.85 (–25.95, 7.33) | –8.08 (–17.88, 2.89) | −1.73 (-13.54, 11.69) |
| Living environmentb | |||||
| % of apartments | 0.18 (0.00, 0.35) | 0.14 (–0.04, 0.32) | 0.35 (0.05, 0.64) | 0.31 (0.13, 0.49) | 0.07 (-0.19, 0.32) |
| % of forest area | –0.13 (–0.27, 0.02) | –0.08 (–0.23, 0.06) | –0.36 (–0.62, –0.09) | –0.23 (–0.40, –0.06) | −0.05 (-0.26, 0.16) |
| Emergency medical servicec | |||||
| Number of emergency medical centers | –1.54 (–2.95, –0.11) | –1.96 (–3.67, –0.22) | –5.14 (–8.41, –1.76) | –2.30 (–4.42, –0.13) | −2.67 (-5.03, -0.24) |
| Number of ambulances | –1.64 (–3.02, –0.24) | –1.29 (–2.72, 0.16) | –3.62 (–6.40, –0.76) | –2.29 (–3.95, –0.59) | −1.02 (-3.10, 1.10) |
| Total | People Aged 65 y+ | People Aged 0-64 | Male | Female | |
|---|---|---|---|---|---|
| Demographya | |||||
| Population density | 1.17 (0.09, 2.27) | 1.02 (–0.05, 2.10) | 1.46 (–0.33, 3.29) | 1.97 (0.91, 3.04) | 0.26 (-1.33, 1.88) |
| Ageing index | –4.40 (–12.71, 4.71) | –3.55 (–12.08, 5.80) | –10.85 (–25.95, 7.33) | –8.08 (–17.88, 2.89) | −1.73 (-13.54, 11.69) |
| Living environmentb | |||||
| % of apartments | 0.18 (0.00, 0.35) | 0.14 (–0.04, 0.32) | 0.35 (0.05, 0.64) | 0.31 (0.13, 0.49) | 0.07 (-0.19, 0.32) |
| % of forest area | –0.13 (–0.27, 0.02) | –0.08 (–0.23, 0.06) | –0.36 (–0.62, –0.09) | –0.23 (–0.40, –0.06) | −0.05 (-0.26, 0.16) |
| Emergency medical servicec | |||||
| Number of emergency medical centers | –1.54 (–2.95, –0.11) | –1.96 (–3.67, –0.22) | –5.14 (–8.41, –1.76) | –2.30 (–4.42, –0.13) | −2.67 (-5.03, -0.24) |
| Number of ambulances | –1.64 (–3.02, –0.24) | –1.29 (–2.72, 0.16) | –3.62 (–6.40, –0.76) | –2.29 (–3.95, –0.59) | −1.02 (-3.10, 1.10) |
Demographic variables: Population density: persons per 1 km2; Ageing index: % of the number of persons ≥ 65 years to the number of persons ≤ 14 years.
Living environment variables: % of apartments (annual values for the % to the total number of houses) and % of forest area (annual values for the % to the total areas).
Emergency medical service variables: The number of Emergency medical centers / Ambulances per 100 000 persons, individually.
Association between urbanization indicators and heat-related mortality risk in the latest sub-period (2006–2015) for the total population. Results are expressed as percentage change (95% confidence interval) in heat-related mortality risk per 1-unit increase in each meta-variable, except for population density (per 1000 persons increase). Each meta-variable is an average value of each regional indicator measured during the latest sub-period (2006–2015)
| Total | People Aged 65 y+ | People Aged 0-64 | Male | Female | |
|---|---|---|---|---|---|
| Demographya | |||||
| Population density | 1.17 (0.09, 2.27) | 1.02 (–0.05, 2.10) | 1.46 (–0.33, 3.29) | 1.97 (0.91, 3.04) | 0.26 (-1.33, 1.88) |
| Ageing index | –4.40 (–12.71, 4.71) | –3.55 (–12.08, 5.80) | –10.85 (–25.95, 7.33) | –8.08 (–17.88, 2.89) | −1.73 (-13.54, 11.69) |
| Living environmentb | |||||
| % of apartments | 0.18 (0.00, 0.35) | 0.14 (–0.04, 0.32) | 0.35 (0.05, 0.64) | 0.31 (0.13, 0.49) | 0.07 (-0.19, 0.32) |
| % of forest area | –0.13 (–0.27, 0.02) | –0.08 (–0.23, 0.06) | –0.36 (–0.62, –0.09) | –0.23 (–0.40, –0.06) | −0.05 (-0.26, 0.16) |
| Emergency medical servicec | |||||
| Number of emergency medical centers | –1.54 (–2.95, –0.11) | –1.96 (–3.67, –0.22) | –5.14 (–8.41, –1.76) | –2.30 (–4.42, –0.13) | −2.67 (-5.03, -0.24) |
| Number of ambulances | –1.64 (–3.02, –0.24) | –1.29 (–2.72, 0.16) | –3.62 (–6.40, –0.76) | –2.29 (–3.95, –0.59) | −1.02 (-3.10, 1.10) |
| Total | People Aged 65 y+ | People Aged 0-64 | Male | Female | |
|---|---|---|---|---|---|
| Demographya | |||||
| Population density | 1.17 (0.09, 2.27) | 1.02 (–0.05, 2.10) | 1.46 (–0.33, 3.29) | 1.97 (0.91, 3.04) | 0.26 (-1.33, 1.88) |
| Ageing index | –4.40 (–12.71, 4.71) | –3.55 (–12.08, 5.80) | –10.85 (–25.95, 7.33) | –8.08 (–17.88, 2.89) | −1.73 (-13.54, 11.69) |
| Living environmentb | |||||
| % of apartments | 0.18 (0.00, 0.35) | 0.14 (–0.04, 0.32) | 0.35 (0.05, 0.64) | 0.31 (0.13, 0.49) | 0.07 (-0.19, 0.32) |
| % of forest area | –0.13 (–0.27, 0.02) | –0.08 (–0.23, 0.06) | –0.36 (–0.62, –0.09) | –0.23 (–0.40, –0.06) | −0.05 (-0.26, 0.16) |
| Emergency medical servicec | |||||
| Number of emergency medical centers | –1.54 (–2.95, –0.11) | –1.96 (–3.67, –0.22) | –5.14 (–8.41, –1.76) | –2.30 (–4.42, –0.13) | −2.67 (-5.03, -0.24) |
| Number of ambulances | –1.64 (–3.02, –0.24) | –1.29 (–2.72, 0.16) | –3.62 (–6.40, –0.76) | –2.29 (–3.95, –0.59) | −1.02 (-3.10, 1.10) |
Demographic variables: Population density: persons per 1 km2; Ageing index: % of the number of persons ≥ 65 years to the number of persons ≤ 14 years.
Living environment variables: % of apartments (annual values for the % to the total number of houses) and % of forest area (annual values for the % to the total areas).
Emergency medical service variables: The number of Emergency medical centers / Ambulances per 100 000 persons, individually.
Ridge regression results are displayed in Supplementary Table S3, available as Supplementary data at IJE online. Most of the associations estimated by the meta-regressions were generally consistent with those from the ridge analysis. Based on the size of percentage change, population density and the number of emergency medical centres showed the strongest associations with heat-mortality risk, and the % of apartments and number of ambulances were also associated with heat-mortality risk.
Based on the sensitivity analyses (Supplementary Figures S2–S3, Supplementary Tables S4–S6, available as Supplementary data at IJE online), the main results of this study are generally robust for the modelling specifications, potential outliers, and substitution variable for green space. In addition, Supplementary Figure S2, available as Supplementary data at IJE online H show that the difference in heat-mortality risk among sub-areas was more pronounced for extreme heat (99th percentile of the summer temperature) compared to mild heat (95th percentile).
Discussion
We investigated the association between population concentration in metropolitan areas and heat-mortality risk in Japan during the period of 1980–2015. Since the 1990s, the heat-mortality risk was generally higher in the highest population areas; this trend became more prominent during the 2000s as population concentration in metropolitan areas and depopulation in non-metropolitan areas intensified. Regional imbalances in demography, living environments, and medical services were generally most pronounced in the 2000s. During the same period, higher values for population density and apartment %, and lower values for forest area % and emergency medical services were associated with higher heat-mortality risk.
We found a positive association between population concentration in metropolitan areas and heat-mortality risk in Japan, especially in the 2000s. Effects of this population increase on heat-mortality risk requires further exploration. Previous studies reported that a higher population and population density might contribute to an increase in the heat island phenomenon because of a larger number of urban infrastructures and vehicles as well as high energy consumptions,20–22 and have suggested the heat island phenomenon as a major factor for the the higher heat vulnerability in urban areas than in suburban or rural areas.1,23 Our result shows a greater increase in extreme summer temperatures in the highest and intermediate population areas than in the lowest population areas (Supplementary Figure S5, available as Supplementary data at IJE online), and this gap has generally become greater since the 2000s. We also found that the population increase in metropolitan areas and related changes in residential type and lower availability of medical services were associated with an increase in vulnerability to heat. This finding suggests that socioeconomic changes accompanying population change could also be important factors in heat-mortality risk assessments. On the other hand, similar to findings in a Japanese study,19 we found no association between an ageing index and heat-mortality risk. This pattern may be related to improved health status in the elderly population;24 however, this needs further study.
Our results showed that living environments and emergency medical services were associated with heat-mortality risk in Japan, and the associations became more pronounced in the 2000s than in the 1980-90s. The higher heat-mortality risk for apartment dwellers can be explained in part by social isolation.25,26 Further, a previous study reported that the effects of social isolation on heat vulnerability were pronounced in elderly males living in a communal house, who generally have more minimal social networks than females.26 In addition, consistent with previous studies,25,27 our results showed that more green space was associated with lower heat-mortality risk. Previous studies reported that green space can reduce surface temperatures28,29 and provide positive effects on health by encouraging physical and social activities.30,31 With the proportion of forests constantly decreasing in metropolitan areas in Japan, our result provide important implications for urban green space planning. Our results also reveal that lower accessibility to emergency medical services was substantially associated with higher heat-mortality risk. Although Japan has the highest levels of medical facilities and health care quality among developed countries,32,33 we postulate that the degeneration of emergency medical accessibility related to the population concentration in metropolitan areas had adverse impacts on heat-mortality risk.
We observed that increases in population and apartment % and a decrease in forest area % in metropolitan areas have been more prominent in the 2000s. These changes are largely related to the population concentration in metropolitan areas following the depopulation and economic stagnation in non-metropolitan areas; however, they are also associated with the urban regeneration policy implemented during the same period. After suffering a decade-long economic recession after the collapse of the real estate bubble in the early 1990s, the Japanese government implemented an urban revitalisation policy (known as Urban Renaissance) in the 2000s as part of its economic rehabilitation policy,14 centred on Tokyo and other metropolitan areas. This policy was helpful in revitalising the old cities and urban economies,34 while at the same time, the policy has been criticised for accelerating population concentration in the metropolitan areas (especially in the Tokyo area) and amplifying intra and interregional inequalities in developments, housing, and social environments.12,34 Although further study is required, our results suggest the potential disadvantages of the urban regeneration policy for heat vulnerability.
The association between regional indicators and heat-mortality risk differed by age groups in Japan. We conjecture that this may be associated with the outdoor and economic activities of younger people: younger people might spend time and work (e.g. construction) outside more than older people who tend to spend more time at home, and/or many younger people tend to work in the business district in urban areas, leading to a more intensive heat island effect. In addition, repetitive exposure to extremely high temperatures may increase the risk of heat emergencies,35 such as heat stroke, which could lead to high sensitivity to emergency medical services in younger people. Furthermore, the association between regional indicators and heat-mortality risk was more pronounced in men. We hypothesise that men’s higher levels of outdoor activity (labour force participation % in Japan, 2015: 65.8% for men, 47.0% for women according to the Statistics Bureau of Japan) might be associated with this gender gap; however, this hypothesis requires further research.
In the present study, the higher heat-mortality risk in metropolitan areas was observed during the late study period (the 1990s-2000s), and associations between heat-mortality risk and regional indicators have also become stronger over time. We speculate that this phenomenon is confined to developed countries: as economic development and urbanisation have continued, average levels of income and health may have improved in all areas in Japan. Therefore, heat vulnerability factors may also have changed from rural characteristics (e.g. higher ageing and lower income) to urban characteristics (e.g. residential environments and lower accessibility of emergency medical services). Although further research is needed to define this hypothesis, our findings may support previous study results that showed higher heat vulnerability in more urban areas in developed countries,2,36 and may provide plausible explanations for this phenomenon.
The study had several limitations. First, the spatial unit (prefecture) used has a relatively large resolution to reflect the population concentration of smaller administrative areas. In addition, because there was only a single weather monitoring station for each prefecture, we have to acknowledge the possibility of exposure misclassification. Second, although we conducted ridge regressions to estimate the associations between regional indicators and heat-mortality risk after considering the multicollinearity among the indicators, we were not able to consider the distribution of true effect sizes that could be considered in the random-effect meta-regression models. Finally, although the unemployment rate and average income per household were considered as confounders, we could not fully cover poverty or inequalities between and within prefectures owing to data limitations. Further studies should consider suitable indicators (e.g. poverty gap, or GINI index) that can measure the absolute and relative levels of regional inequality.
Conclusion
Our study provides new insight into heat-related vulnerability. To the best of our knowledge, this is the first study to investigate the effect of population concentration in the metropolitan areas on heat vulnerability in Japan, which suffered population decline and rapid ageing in rural areas during recent decades, and our results provide epidemiological evidence to inform the modification of current and future area-specific heat action plans that explicitly incorporate consideration of climate change. Additionally, our findings may provide implications for public health and population policymakers in developed countries as well as for low- and middle-income countries that experience or will experience rapid urban concentration.
Supplementary data
Supplementary data are available at IJE online.
Ethics approval
Not required. The dataset used in this study was completely anonymous without any personal information.
Funding
Ho Kim is supported by the Global Research Lab (#K21004000001-10A0500-0710) through the National Research Foundation of Korea and by the Korea Ministry of Environment via the ‘Climate Change Correspondence Program’ (2014001310007). Yoonhee Kim is supported by the JSPS KAKENHI Grant Number JP19K17104 in Japan. No funding bodies had any role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Data availability
The data that support the findings of this study are available on request from the corresponding author.
Conflict of interest: None declared.
