Skip to main content

Table 1 Characteristics and results of studies analysing temporal changes in temperature related mortality

From: Changes in population susceptibility to heat and cold over time: assessing adaptation to climate change

Study Location time period population Exposure(s) and outcomes General modelling approach and methods to assess change in susceptibility over time Results: changes in (RR) of heat/cold related mortality (HRM, CRM) over time (all CI/PIs and significance are for 5 % level unless stated otherwise)
Bobb et al. 2014 [37] 105 US cities
All ages & age stratified
Heat (only summer months)
All-cause mortality & CVD / Respiratory mortality
Time series regression (daily series) model. Control for time varying factors. Estimated excess heat related deaths for each year (1987 and 2005 results compared). Each year allowed a separate coefficient for daily temperature. Heat related deaths per 1000 deaths (all cities):51 (95 % PI: 42,61) in 1987 compared to 19 (95 % PI: 12,27) in 2005. Decline observed for all ages & significant for heat related respiratory & CVD mortality. Cities with larger increases in AC had larger decreases in mortality (not significant).
Petkova et al. 2014 [36] New York (US)
1900–1948 & 1973–2006
All ages & age stratified
Heat (only summer months)
All-cause mortality
Time series regression (daily series). Control for time varying factors.
Modelled risk of mortality at 29 °C vs 22 °C for each decade. Decadal averages of RR at 29 °C vs 22 °C compared. Used random effects meta-regression, including linear term for decade.
Decrease in RR at 29 °C vs 22 °C of 4.6 % (2.4,6.7) per decade (all ages)
>65 years: highest initial risk and most decline in RR over time. Also found a change in lag structure over time - harvesting effect more prevalent in earlier part of century.
Astrom et al. 2013 [39] Stockholm, Sweden
All ages & stratified
by age and sex
Heat and cold ‘extremes’ (Defined in model 1 as above/below the 98th percentile for entire period)
Daily mortality
Time series regression (daily series). Control for time varying factors.
Examined trend in RR of mortality at extremes of temperature over time of mortality at 98th percentiles of temperature compared to mortality at average temperatures.
Significant decline in mortality risk for elderly and combined age categories for heat but non-significant for cold. Patterns similar for men & women
Significant declining trend in temperature related mortality risk for 0-14 s for hot and cold. In last decades, upward trend in the heat risk for the 15–64 age group observed.
Ha et al. 2013 [38] Seoul, S. Korea
1993–2009 (1994 excluded: extreme HW)
All ages & age stratified
All-cause mortality (excluding accidental deaths) and CVD mortality
Time series regression (daily series). Linear threshold model to estimate quantitative effects. Control for time varying factors.
Compared results from two periods (1993 and 1995–2000, and 2001–2009). Used common threshold throughout study period.
% increase in all-cause mortality per 1 °C increase in temperature above threshold (changes not significant):
All-cause mortality (pattern similar for >65s)
1990s 4.73 % (all ages) 2000s 6.05 % (all ages)
CVD mortality (pattern similar for >65s)
1990s 8.69 % (all ages) and 2000s (all ages) 5.27 %
Matzarakis et al. 2011 [40] Vienna, Austria
All ages
Heat (Physiological Equivalent Temperature (PET))
All-cause mortality
Time series analysis (daily series). Modelled daily excess mortalities, calculated as deviations from average annual mortality.
Linear regressions fitted to mortality rates per 10000 to give % change in heat related mortality per decade (1970–2007) for given ranges of PET.
% change per decade from 1970 to 2007 in mortality:
PET range <29 °C - 0.15 %: ( reported not significant)
PET range 29-35 °C −0.83 % (−0.68,-0.97)
PET range 35-41 °C −0.96 % (−0.77,-1.16)
PET range > =41 °C −1.32 % ( not significant - low numbers)
Christidis et al. 2010 [41] England and wales
All ages
Heat and cold
All-cause mortality
Daily excess HRM/CRM obtained by comparing to the average mortality within a 3 °C ‘comfort zone’. Compared: 1.yearly regression slopes (1976–2005) 2.Change in HRM/CRM obtained using regression slopes from different time periods (1976 compared to 2005) to demonstrate no adaptation or early adaptation. Slope of regression lines for heat and cold related mortality risk (SE) decreased in magnitude over time. CRM decreased by 85 deaths/million/year from 1976–2005. “No adaptation” scenario (1976 regression slope) CRM reduction less 47 deaths/million/year. HRM increased by 0.7 deaths/ million/ year. “No adaptation” scenario (1976 slope) HRM increased more (by 1.6 deaths/million/year).
Ekamper, 2009 [42] Zeeland, Holland
All ages & age stratified
Heat and cold
All-cause mortality
Times series analysis (daily series)
Compare: a) regression co-efficient from model between 25 year periods (thresholds allowed to vary between time periods) b) MMT value in each 25 year time period analysed.
Regression coefficients for HRM reported as decreasing over time (no test for significance). Pattern unclear for cold.
Found shift in MMT to higher temperatures in later time periods analysed: MMT slightly below 15 °C for 1855–1897 and around 17 °C for 1905–1929 and 1930-1954
Barnett, 2007 [43] 107 US cities
‘Elderly’ (age range not given)
Increases in temperature in both summer and winter (effects of heat & cold)
CVD mortality
Case-crossover design
Time stratified
Compare the % increase in of cardiovascular deaths per 10 °F increase in temperature within a given season and across the time period 1987–2000.
% increase in risk per 10 °F rise in temperature
summer Winter
1987 4.7 % (3.0, 6.5 %) 1987–4.2 (−5.1,-3.2)
2000–0.4 % (−3.2,2.5) 2000–4.9 (−6.8,-3.1)
Variation between geographic regions (e.g. biggest declines in heat risk in NW, NE, Industrial MW and California)
Carson et al. 2006 [44] London (UK)
All ages
Heat and cold
All-cause mortality and CVD and respiratory mortality
Time series regression (weekly series). Linear hockey stick model. Controlled for time varying factors. Threshold set at 15 °C. Compared a)decadal RR for heat and cold related mortality b)proportion of deaths attributable to heat/cold. RR (for heat related mortality above threshold) and % attributable deaths: increased between 1910 and 1937 then decreased for last 2 time points.
Davies et al. 2003 [46] 28 major US cities
Age standardised population
Heat only
All-cause mortality
Time series analysis (daily series) using HRM: daily mortality anomalies estimated using median mortality for given month as a baseline. Analysed daily fluctuations in excess mortality with temperature variation. Compared decadal HRM. Threshold varied by decade. Mean decadal HRM in standard population of 1 million for all cities declined over time. 12 cities showed no evidence of threshold AT above which heat related mortality begins to appear in the 1990s. Most decline in 1980s in the South in NE cities. Seattle and Washington show increased HRM in latest decades compared to the 1960s.
Donaldson et al. 2003 [45] North Carolina (NC), South East England (SEE)
South Finland (SF)
Age: > 55 yrs
Heat only
All-cause mortality
Time series analysis using HRM (daily mortalities at daily temperatures exceeding a 3 °C threshold band, minus daily mortalities in that 3 °C band for the given month. Summed to give annual heat related mortality)
Compared a) Change in temperature at which minimum mortality occurs (MMT) b) Change in excess heat related mortality per 10^6 between 1971 and 1997.
Changes in MMT (between 1971 and 1996):
Increase in MMT significant for NC and SEE but not for SF
Change between 1971–1996 in HRM per 10^6 population (unadjusted for age & sex and adjusted):
NC 228 in 1971 decreased to 16 in 1996. Change of 212 (59,365). Adjusted change 552 (significant)
SF 382 in 1971 decreased to 99 in 1996. Change of 282 (66–500). Adjusted decrease 414 (significant)
SEE 111 in 1971 decreased to 16 in 1996. Change of 2.1 (−119, 114). Adjusted decrease 53 (significant)