Shifts in the seasonal distribution of deaths in Australia, 19682007
Charmian M. Bennett & Keith B. G. Dear &Anthony J. McMichael
Received: 2 December 2012 /Revised: 7 March 2013 /Accepted: 9 March 2013# ISB 2013
Abstract Studies in temperate countries have shown thatboth hot weather in summer and cold weather in winterincrease short-term (daily) mortality. The gradual warming,decade on decade, that Australia has experienced since the1960s, might therefore be expected to have differentiallyaffected mortality in the two seasons, and thus indicate anearly impact of climate change on human health. Failure todetect such a signal would challenge the widespread assump-tion that the effect of weather on mortality implies a similareffect of a change from the present to projected future climate.We examine the ratio of summer to winter deaths against abackground of rising average annual temperatures over fourdecades: the ratio has increased from 0.71 to 0.86 since 1968.The same trend, albeit of varying strength, is evident in allstates of Australia, in four age groups (aged 55 years andabove) and in both sexes. Analysis of cause-specific mortalitysuggests that the change has so far been driven more byreduced winter mortality than by increased summer mortality.Furthermore, comparisons of this seasonal mortality ratiocalculated in the warmest subsets of seasons in each decade,with that calculated in the coldest seasons, show that particu-larly warm annual conditions, which mimic the expectedtemperatures of future climate change, increase the likelihoodof higher ratios (approaching 1:1). Overall, our results indicatethat gradual climate change, as well as short-term weathervariations, affect patterns of mortality.
Keywords Temperature . Mortality . Climate change .
Season . Poisson modelling . Mortality ratio
Climate change-related increases in temperature are likely tohave significant impacts on health. The adverse impacts oftemperature extremes (heatwaves and cold snaps) on mor-tality are already well known, with distinct increases in mortal-ity rates (Gosling et al. 2009; Basu and Samet 2002; Bambricket al. 2008) (Fig. 1). However there has been little opportunityto examine how longer-term, more subtle changes in climate,such as the observed trend (over decades) of increasinglywarmer summers and warmer winters in Australia, might affectthe seasonal distribution of mortality. Our aim was to determinewhether this multi-decadal warming trend was associated witha concurrent shift in the seasonal distribution of all-cause andcause-specific mortality across Australia since 1968.
It is well established that the relationship between weatherand mortality differs by season (Davis et al. 2004; Guest et al.1999; McGregor et al. 2004; Gemmell et al. 2000; Lerchl1998; Wilkinson et al. 2004; Davis et al. 2003b), though moststudies of seasonal changes have focussed only on the effectsof cold on mortality. Few studies have considered trends inthis seasonal variation. Lerchl (1998) found that while lowtemperatures had a stronger impact than heat on mortality inGermany, the impact of cold has been declining over the last~50 years, probably related to increased use of central heatingand health system improvements. Carson and colleagues(Carson et al. 2006) found that the ratio of winter:non-winterdeaths in London shifted over different time periods since1900, but with no consistent trend.
In many parts of the world, including Australia, there is aclear seasonal distribution in mortality, typically entailinghigher mortality rates in winter and summer than in spring
C. M. Bennett (*) :K. B. G. Dear :A. J. McMichaelNational Centre for Epidemiology and Population Health, TheAustralian National University, Canberra, ACT 0200, Australiae-mail: email@example.com
K. B. G. Deare-mail: firstname.lastname@example.org
A. J. McMichaele-mail: email@example.com
Int J BiometeorolDOI 10.1007/s00484-013-0663-x
and autumn. On a daily basis, there is a stronger associationbetween mortality and heat than between mortality and cold,as evidenced by large surges in mortality associated withextremely hot days and extended hot periods (heatwaves)that are not matched by equivalent surges in mortality on verycold days (Kalkstein and Greene 1997). Nevertheless, in mosttemperate climates, winter mortality rates are typically signifi-cantly higher than those in summer due to the winter seasonalimpacts of infectious diseases (especially influenza) and theexacerbation of respiratory and cardiovascular diseases (Daviset al. 2004). In Australia, a clear seasonal signal is observed inall-cause mortality, with higher mortality in winter than insummer (Guest et al. 1999). Total summer deaths from 1968to 2007 in Australia were 80.0 % of the number of winterdeaths. The greater number of winter season deaths is largelyexplained by infectious disease transmission peaks during win-ter and the exacerbation of chronic diseases, especially cardio-vascular and respiratory conditions (Cameron et al. 1985).
Climate change projections consistently indicate that aver-age temperatures in many places will rise, resulting in bothwarmer summers (more very hot days) and warmer winters(fewer very cold days). The frequency of temperature extremesin Australia appears to be changing in a similar manner, withmore extremely hot days and fewer extremely cold days ob-served every decade since 1960. The average number of recordhot days in summer has increased every decade, from ~10 daysper summer in the decade 19601969 to ~23 days per summerin the decade 20002009, whilst over the same period, thenumber of record cold days in winter has decreased everydecade, from ~21 days in the decade 19601969 to ~10 daysin the decade 20002009 (Bureau of Meteorology and CSIRO2010). Therefore, it is expected that the dominance of wintermortality will decline as summer mortality rates rise.
A shift in the seasonal distribution of deaths is consistentwith the hypothesised effect of a warming climate, andsupport for this hypothesis can be strengthened by exploringvariations in the trend by place, person and disease class, tosee whether these variations are also as might be expected.Our research examined the seasonal distribution of mortal-ity, including by age, sex, cause of death and location, tovalidate this hypothesis in the Australian context.
We examined the ratio of summer-to-winter deaths from1968 to 2008, for the whole of Australia, using all-causeand cause-specific deaths in those aged 55 years and above.
All-cause mortality data for those aged 55 and over wereobtained from the Australian Bureau of Statistics for alleight states and territories. Four age groups were used inthe analysis: 5564 years, 6574 years, 7584 years, and85+ years.
Cause-specific mortality was identified using the firstthree digits of the International Classification of Diseases(Australian Modification) (ICD-AM) mortality codes. Thestudy period spanned versions 8, 9 and 10 of the ICD-AM,so diagnosis codes were matched across ICD versions toensure consistency. The ICD codes used in this study arelisted in Table 1. Respiratory and cardiovascular diseaseshave been widely linked to seasonal variations in tempera-ture as well as to shorter-term temperature extremes (such asheat waves and winter storms) (Analitis et al. 2008; Basu
Fig. 1 Association betweentemperature (3-day average)and all-cause mortality in adultsaged 45+ in Brisbane, 19902007. The curve is a 6dfgeneralised additive model,adjusted for long-term trend andannual cycles. Relative risk isexpressed relative to theminimum, which is at about21 C
Int J Biometeorol
and Samet 2002; Gosling et al. 2009). Renal disease wasalso included as recent studies have suggested that heat-related dehydration and hyperthermia are associated withrenal dysfunction, especially in the elderly (Nitschke et al.2011; Hansen et al. 2008).Winter was defined as the three coldest months (June,
July and August), and Summer as the three warmestmonths (December, January and February). The availabledata included 40 winters from 1968 to 2007, and 39 sum-mers from 1968/69 to 2006/07.
Analyses were performed using Stata12SE (Statacorp,College Station, TX).
Poisson regression models of the number of deaths perseason were used. The ratio of summer to winter mortalitywas modelled as a 2-level (1df) covariate, and the trend inthis ratio over time was captured as the interaction of thisfactor with year, treated as continuous. Year k was codedas the calendar year (k=19682007) for the 40 winterseasons (mortality in June, July and August combined),and k=1968.52006.5 for the 39 intervening summer sea-sons. For example, total mortality in the summer seasonDecember 1968, January 1969 and February 1969 wasconsidered to occur at time year=1968.5.
The model was adjusted for age group (four levels), sex,and location (eight states and territories).
We defined demographic setting to represent the 64possible combinations of location, age and sex. Given ademographic setting i, season j (1=summer and 2=winter),and year k (1968, 1968.5, , 2006.5, 2007), our basemodel for seasonal deaths Yijk was
logE Yijk aij bi k cj k
The intercepts aij account for population denominators,but also allow for differential overall mortality by demo-graphic setting (i) in each season (j). Thus this term implic-itly includes the main effects of age, sex, location and
season and also all pairwise and higher interactions betweenthese four factors. The bi terms represent linear setting-specific trends in mortality, to capture demographic driftsand general trends in society and local health care. The cjterms capture any overall differences in these trends