Forensic Science, Medicine, and Pathology

, Volume 9, Issue 1, pp 31–35

Is sudden death random or is it in the weather?

Authors

  • Christopher Bierton
    • School of MedicineUniversity of Adelaide
  • Kara Cashman
    • Data Management and Analysis Centre, Discipline of Public HealthUniversity of Adelaide
    • School of MedicineUniversity of Adelaide
    • Forensic Science South AustraliaUniversity of Adelaide
Original Article

DOI: 10.1007/s12024-012-9380-8

Cite this article as:
Bierton, C., Cashman, K. & Langlois, N.E.I. Forensic Sci Med Pathol (2013) 9: 31. doi:10.1007/s12024-012-9380-8

Abstract

It has been suggested that the weather may promote some types of death; this study sought to determine if types of death in the region around Adelaide, South Australia, occur in non-random clusters and in relationship to the weather. A Poisson model was used to determine if the occurrence of types of death were random. An exploratory analysis was performed for each death type to see if there was a relationship to weather variables using data supplied by the Bureau of Meteorology. Cases examined at Forensic Science South Australia from 1 January 2008 to 31 December 2009 were reviewed. It was ascertained that cardiovascular deaths were distributed non-randomly; there was statistical evidence to suggest that deaths from ischemic heart disease, pulmonary embolus and drug toxicity had non-random occurrence. Maximum temperatures and increases in temperatures correlated with deaths from natural causes, cardiovascular disease, ischemic heart disease and pulmonary thromboembolus; lower hours of sunlight were statistically significant for deaths due to pulmonary thromboembolus. The distribution pattern of deaths resulting from motor vehicle collision did not fit the Poisson (random) model with variation through the week also being present. Non-random clusters of deaths do occur and weather events, such as increase in temperature, are associated with some types of death. However, analysis indicates that the weather is not responsible for all clustering. With regards to motor vehicle collision deaths temporal variation may be related to social factors, such as holiday periods. Further investigation may assist with health resource planning.

Keywords

WeatherRisk factorsMortalityNatural death

Introduction

Casual observations by those who perform post-mortem examinations for medico-legal purposes have lead to the suggestion that some types of death appear to occur in groups within short time intervals. However, these ‘temporal clusters’ may be perceived rather than true as the human brain tries to recognize patterns in random events [1]. Nonetheless, it has been suggested that the weather may be a factor in triggering some types of death [27], which could provide a reason for temporal clusters to form. To investigate if the occurrence of sudden death has a non-random pattern that appears influenced by climatic conditions, cases reported to the State Coroner’s Office for South Australia that had undergone post-mortem examination were examined in conjunction with meteorological data.

The Coronial service for the population of the state of South Australia (approximately 1.6 million people, which is around 7.4 % of the total population of Australia) is provided by the State Coroner’s Office and the mortuary in the Forensic Science building based in Adelaide (the capital of the State). South Australia has an area of nearly 1 million square kilometers (comprising nearly 13 % of the total area of Australia). However, the majority of the population is present in the southeastern coastal region—predominantly within and around the capital, Adelaide. Thus the region of study was limited to the region bounded by weather stations at Parafield, Adelaide airport, Edinburgh, Mt Lofty, Noarlunga, Mount Terrible, Kuitpo Forest, Hindmarsh Island and Murray Bridge. The Adelaide observation center (at Kent Town) provided the most intact range of observations and so this was selected as the reference for all weather data (which was provided by Bureau of Meteorology).

The Coroner’s Act for South Australia provides for the investigation of sudden death and unexplained deaths, which will include deaths from natural causes, suicides, drug-related deaths and deaths resulting from motor vehicle collision. As noted above, all autopsies directed by the State Coroner are performed at Forensic Science, South Australia. The autopsy data is collated on the CaseMan system, which allows retrieval of the manner of death [8] as determined by the pathologist, the cause of death, place of death and date of death. Autopsies performed on deaths resulting from all natural causes, deaths due to cardiovascular disease, deaths from ischemic heart disease, death resulting from pulmonary thromboembolus, suicides, drug-related deaths and deaths resulting from motor vehicle collision in the study region from 1st January 2008 to 31st December 2009 were investigated to determine if there were non-random clusters and if there was a relationship to weather events.

Methods

Post-mortem examinations performed at Forensic Science SA from 1 January 2008 to 31 December 2009 were identified from the CaseMan system. From this the date of death, cause of death, manner of death and place of death were obtained with the age and sex of the deceased. When required this was supplemented by review of the autopsy report. Cases were excluded when the date of death was uncertain (the latter being mostly cases with decomposition), if the place of death was not recorded, or when death had occurred following more than 24 h in a hospital.

The death types were categorized as from suicide, drug toxicity, natural causes [9], cardiovascular disease, ischemic heart disease, pulmonary thromboembolism [10], motor vehicle collision (all transportation [11] excluding train related) and other. Suicides were identified using data derived from a study of suicides in South Australia [12]. All deaths that had been attributed to toxic effects of drugs (including alcohol), but which had not been identified as suicides, were regarded as drug related. Cardiovascular deaths included all deaths related to hypertensive and atherosclerotic disease of the cardiovascular system (such as ruptured dissection of the aorta, ruptured abdominal aortic aneurysm as well as ischemic heart disease, but not pulmonary thromboembolism). Ischemic heart disease encompassed any case related to deficient blood supply to the heart (e.g., coronary artery narrowing acute or chronic; myocardial infarction) [13]. Motor vehicle collision related included all driver, passenger and pedestrian deaths related to motor vehicles (car, truck and motorcycle), but excluded train related collisions.

Daily weather data was obtained from the Australian Meteorology Bureau for all weather stations in South Australia. For the study, recordings from the weather station at Kent Town were used. For each day from 1 January 2008 to 31 December 2009 the maximum temperature, minimum temperature, precipitation (before 9am), hours of bright sunlight and direction of maximum wind gust were recorded. A database of deaths in Adelaide and surrounding region bounded by weather stations at Parafield, Adelaide airport, Edinburgh, Mt Lofty, Noarlunga, Mount Terrible, Kuitpo Forest, Hindmarsh Island and Murray Bridge was compiled with corresponding weather data.

A Poisson model with calculation of Deviance and Pearson Chi-Square values was utilized to determine if the occurrence of types of death appeared random or non-random. Once the Poisson models had been determined, an exploratory analysis was performed for each death type individually to see if there were any causative weather variables (regardless of the outcome of the Poison model analysis). Weather data (as above) were recorded on the date of the event and for the previous 5 days. Each of these variables (except for the direction of maximum wind direction) were analyzed on the day of the event, on the day of the event compared to the previous day and on the day of the event compared to the average of the previous 5 days. Direction of maximum wind direction was analyzed for the day of the event and on the day of the event compared with the previous day. Additional analyses were made in regards to whether there had been any precipitation on the day of the event or in the previous 5 days. To determine the relationship between death and weather variables, a time-stratified case crossover design was used. In this analysis, days on which deaths occurred were compared to surrounding days to determine if a death was more likely to occur with particular weather factors.

Results

In the region and period analyzed there were 233 cases of suicide, 746 natural deaths, (of which 53 deaths resulted from pulmonary thromboembolus and 474 deaths were attributed to cardiovascular disease that included 345 deaths from ischemic heart disease), 112 drug (including alcohol) related deaths, and 111 deaths resulting from a motor vehicle collision. This comprised a total of 1,202 cases from 2,778 total Coronial post-mortem cases over the study period of 2 years.

The distribution pattern of deaths resulting from motor vehicle collision did not fit the Poisson (random) model with deviance and Pearson Chi squared values being significant (p = 0.004 and p = 0.021 respectively). There were 13 weeks in which the observed number of deaths was significantly different (higher) than the predicted number in the 2 years (105 weeks) of the study period. Furthermore, the distribution of the number of deaths was non-random through the week, with significantly less cases on Tuesday (p = 0.024), Wednesday (p = 0.009) and Friday (p = 0.003) compared to Saturday (set as reference).

There was no significant deviation from the Poison model for all natural disease deaths (deviance p = 0.17 and Pearson Chi Squared p = 0.26). However, there was statistical evidence that the distribution of cardiovascular deaths was not random (deviance p = 0.017 and Pearson Chi Squared p = 0.027); there were 10 weeks when the observed numbers of death differed significantly (nine higher and one lower) from the predicted value. There was some evidence that the distribution of ischemic heart disease deaths was not random with a significant deviance (p = 0.03) and 4 weeks when the observed number of deaths differed significantly from the expected number (all being greater), but the Pearson Chi squared was not significant (p = 0.125). The Poison model did not adequately predict the distribution of deaths from pulmonary thromboembolus as the Pearson Chi Squared model had a p value of 0.0367 and there were 11 weeks when there were significantly more deaths than would be predicted, but it can only be suggested the distribution of deaths was non-random as the deviance had a p value of 0.193.

There was no evidence from the Poisson model that suicides were non-randomly distributed over the study period. However, there was some evidence suggesting that the number of drug related deaths per week were not random as the deviance model fit statistic was significant (p = 0.031) and there were 3 weeks in which the number of deaths differs significantly from the expected number (by being higher), but the Pearson Chi Squared p value was not significant (0.151).

The maximum temperature on the day of the event and maximum temperature on the day of the event minus the average maximum temperature for the previous 5 days to the event were statistically significant for natural, cardiovascular disease and ischemic heart disease deaths, with increases in temperature correlating with increased deaths (Table 1).
Table 1

Relationship of weather variables to cause of death

 

Hazard ratio

95 % hazard ratio confidence limits

p value

Natural disease

Maximum temperature on day of event

1.021

1.005–1.038

0.012

Maximum temp on the day of event minus average maximum for previous 5 days

1.019

1.003–1.035

0.018

Minimum temperature on day of event

1.024

1.001–1.047

0.041

Cardiovascular disease

Maximum temperature on day of event

1.032

1.012–1.053

0.002

Maximum temp on the day of event minus average maximum for previous 5 days

1.026

1.006–1.046

0.009

Minimum temperature on day of event

1.037

1.010–1.066

0.008

Minimum temp on the day of event minus average minimum for previous 5 days

1.028

1.002–1.055

0.036

Ischemic heart disease

Maximum temperature on day of event

1.045

1.021–1.070

<0.001

Maximum temp on the day of event minus average maximum for previous 5 days

1.039

1.016–1.063

<0.001

Minimum temperature on day of event

1.047

1.015–1.081

0.004

Minimum temp on the day of event minus average minimum for previous 5 days

1.040

1.008–1.071

0.012

Pulmonary thromboembolus

Minimum temperature on day of event

1.104

1.012–1.204

0.026

Hours of bright sunlight on day of event

0.871

0.805–0.942

<0.001

Hours of bright sunlight on day of event minus average for previous 5 days

0.898

0.836–0.966

0.004

The minimum temperature on the day of the event was statistically significant for natural, cardiovascular disease, ischemic heart disease and deaths due to pulmonary thromboembolus, but the minimum temperature on the day of the event minus the average minimum temperature for the previous 5 days to the event was statistically significant only for ischemic heart disease deaths. Again, deaths were associated with an increase in temperature (Table 1).

The amount of hours of bright sunlight on the day of the event and the amount of hours of bright sunlight on the day of the event minus the average amount of hours of bright sunlight for the previous 5 days to the event were statistically significant for deaths due to pulmonary thromboembolus, with lower sunlight values being associated with deaths (Table 1).

The exploratory analysis revealed no statistically significant weather variables for suicides or death resulting from motor vehicle collision.

Discussion

The results of this study confirm the impression that deaths due to certain causes (cardiovascular disease, ischemic heart disease, pulmonary thromboembolus and motor vehicle collision) do occur in clusters that would not be expected from chance alone. The data was also analyzed to examine if weather influenced the occurrence of deaths; this was performed on all cases regardless of the outcome of the Poisson analysis. The investigation was undertaken to ascertain if there was a relationship to the weather, which could provide a factor for the formation of non-random temporal groups of death (however, this approach cannot indicate if weather is the cause for the non-random clusters). Notwithstanding the possible influence of climate on death, analysis indicates that the weather is not responsible for all clustering. With regards to motor vehicle collision related deaths, the temporal variation may be related to social factors, which could explain the differences through the days of the week; public events and holidays may be responsible for clusters [14, 15].

Natural deaths included all cardiovascular disease and pulmonary thromboembolus deaths, while cardiovascular disease deaths included all the ischemic heart disease deaths. This may explain why natural deaths, cardiovascular disease and ischemic heart disease deaths have the similar weather variables that are statistically significant as most cardiovascular disease deaths in this study resulted from ischemic heart disease and the majority of natural deaths were a consequence of cardiovascular disease. However, the Poison model indicates the distribution of natural deaths is random, which could be interpreted as indicating that other factors (possibly relating to non-cardiovascular and pulmonary thromboembolus causes of death) outweigh the influence of the weather. The Poison model indicated that deaths from cardiovascular disease did not appear entirely random in their distribution. Published studies investigating the relationship between death and weather in the Northern hemisphere have reported an association between cold weather and a risk of cardiovascular disease deaths [46, 16]. However, an association with increases in temperature that accords with the findings of this study has been noted by others [17], including a study based in New Zealand [18].

There appears to be a relationship between environmental extremes of temperature, such as heat waves, and death. A study of hospital admissions revealed a borderline significant increase in mortality [19]. However, there was a marked increase of mortuary admissions to Forensic Science SA during that period with heat listed as a contributory factor in 58 deaths. Although this period was included in the study cohort, its duration of 2 weeks combined with omission of cases due to uncertainty regarding date of death means there is unlikely to be significant skewing of results. The reason behind the statistically significant association of higher temperature with natural deaths, which appears related to increased cardiovascular causes (including ischemic heart disease) in this study is unclear. A link between air temperature and pollution has been suggested [18], but may not be supported [20].

The observation of a statistically significant link between lower hours of sunlight and deaths from pulmonary thromboembolus is difficult to explain. Nonetheless, associations between the incidence of pulmonary thromboembolus and climatic conditions of air pressure [21, 22], humidity [22, 23], and rainfall [23] have been reported. This suggests there may be a true relationship between the weather and pulmonary thromboembolus that requires further elucidation.

A relationship between weather and suicide was expected from published studies [3, 24]; however, none was detected in this study, which also accords with the Poison analysis that indicates the distribution of deaths from suicide is random. However, it has been observed that a seasonal variation of suicide rates occurs only in temperate climate zones [25] and there may be a lag effect [24]. Hence effects may not be apparent due to the latitude of the region of investigation (in this case Southern against Northern hemispheres) and the method of investigation; e.g., if a sufficient period around the death is not assessed.

In considering the results it must be remembered that these are not all the deaths that occurred in SA during the set period, but only sudden deaths (natural and non-natural) investigated by post-mortem examination at the direction of the State Coroner. Nonetheless, this study confirms the role of the autopsy and Coronial system in public health surveillance [26]. From these coronial cases there were excluded cases, predominantly regarding uncertainty of the date of death. Analysis of the dataset indicates that clusters of some types of deaths occur not by chance alone. Although social factors may be the predominant reason in some death types (such as deaths resulting from motor vehicle collision), weather may exert an influence on other death types. Although effects on temperature on factors such as air pollution may provide a reason for the apparent association between deaths from cardiovascular causes and increasing temperature [27], there may be other mechanisms that remain to be elucidated. Further analysis of another dataset for South Australia is required and further studies around Australia would address the lack of studies relating to the Southern hemisphere. The results may provide useful information for the planning of the provision of health care, particularly with regard to vulnerable groups [5], that may be required as a result of prevailing meteorological conditions.

Key Points

  1. 1.

    Non-random clusters of sudden death do occur

     
  2. 2.

    There appears to be a statistical association between some forms of sudden death (e.g: from cardiovascular disease) and the weather: for example, sudden death from cardiovascular disease is associated with high maximal temperature and increasing ambient temperature.

     
  3. 3.

    Some clusters are not apparently associated with the weather, but may relate to social factors (e.g., day of the week).

     
  4. 4.

    The mechanism or mechanisms that could underlie an association between weather conditions and sudden death events is not clear.

     

Copyright information

© Springer Science+Business Media, LLC 2012