Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
IMPACTS OF WEATHER AND CLIMATE ON MORTALITY AND SELF-HARM
IN FINLAND
REIJA RUUHELA
CONTRIBUTIONS147
FINNISH METEOROLOGICAL INSTITUTE
CONTRIBUTIONS
No. 147
Impacts of Weather and Climate on Mortality and Self-harm
in Finland
Reija Ruuhela
Department of Physics
Faculty of Science
University of Helsinki
Helsinki, Finland
ACADEMIC DISSERTATION in meteorology
To be presented, with the permission of the Faculty of Science of the University of Helsinki,
for public criticism in auditorium LS2, A129 at Chemicum (A.I. Virtasen Aukio 1, Helsinki)
on November 21, 2018, at 12 noon.
Finnish Meteorological Institute
Helsinki, 2018
Supervisors Docent Ari Venäläinen
Weather and Climate Change Impact Research
Finnish Meteorological Institute, Finland
Professor Heikki Järvinen
Department of Physics
University of Helsinki, Finland
Reviewers Adjunct Professor Tiina M. Ikäheimo
Center for Environmental and Respiratory Health Research
University of Oulu, Finland
Dr. Daniel Oudin Åström
Occupational and Environmental Medicine
Umeå University, Sweden
Custos Professor Heikki Järvinen
Department of Physics
University of Helsinki, Finland
Opponent Dr. Marco Morabito
Institute of Biometeorology
National Research Council, Italy
ISBN 978-952-336-059-4 (paperback)
ISBN 978-952-336-060-0 (pdf)
ISSN 0782-6117
Erweko
Helsinki 2018
Whoever would study medicine aright must learn of the following subjects.
First he must consider the effect of each of the seasons of the year and the
differences between them. Secondly he must study the warm and the cold
winds, both those which are common to every country and those peculiar
to a particular locality.
Hippocrates, ‘‘Airs, Waters, Places’’,
ca. 400 B.C.E.
Series title, number and report code of publication
Published by Finnish Meteorological Institute Finnish Meteorological Institute
(Erik Palménin aukio 1), P.O. Box 503 Contributions 147, FMI-CONT 147
FIN-00101 Helsinki, Finland Date
October, 2018
Author Reija Ruuhela
Title Impacts of weather and climate on mortality and self-harm in Finland
Abstract
Human beings are able to adapt to their climatic normal conditions, but weather extremes may pose a substantial
health risks. The aims of this dissertation were to model the dependence of all-cause mortality on thermal
conditions in Finland, and to assess changes in the relationship over the decades and regional differences in the
relationships between hospital districts. Another aim was to assess impacts of weather and climate on committed
and attempted suicides Finland. Various methods were applied in these studies.
Time series of all-cause mortality in three age groups in Helsinki-Uusimaa hospital district were made stationary,
and thus the weather impacts were comparable over the 43-year long time period regardless of changes in
population and life-expectancy. The increase in relative mortality due to hot extreme was more than due to cold
extreme, when compared to seasonally varying expected mortality. However, a decrease in relative mortality in
extreme temperatures over the decades was found in all age-groups, even among the 75 years and older, indicating
decreased sensitivity to thermal stress in the population.
Regional differences in temperature‒mortality relationships between 21 hospital districts were studied using
distributed lag non-linear models (DLNM), and the differences were assessed by meta-regression with selected
climatic and sociodemographic covariates. Regional differences in the relationships were not statistically
significant indicating that the same temperature-mortality relationship can be applied in different parts of the
country. On the other hand, the meta-regression suggested that the morbidity indices and population in the hospital
districts could explain part of the small heterogeneity in the temperature‒mortality relationships.
The study on committed suicides on the basis of 33-year long time series showed a significant negative association
between suicides and solar radiation in the period from November to March, thus, a lack of solar radiation would
increase suicide risk in winter. Men appeared to be more sensitive to variation in solar radiation than women. The
study on weather dependence of attempted suicides in Helsinki on the basis of two shorter periods showed another
difference between genders. The risk of suicide attempts of men increased with decreasing atmospheric pressure,
while the risk of suicide attempts of women increased with increasing pressure.
The outcomes of this thesis can be utilized e.g. in preparedness to weather extremes in health sector and in further
studies on impacts of climate change on human health.
Publishing unit Finnish Meteorological Institute
Classification (UDC) Keywords
551.586 Biometeorology. Bioclimatology temperature-related mortality, suicides, suicide
551.583 Natural variations of climate. Climate change attempts, health impacts of weather and climate
614.873 Exposure to temperature variations, cold, heat
614.875 Exposure to non-ionizing radiations
ISSN and series title
0782-6117 Finnish Meteorological Institute Contributions
ISBN Language Pages
978-952-336-059-4 (paperback); 978-952-336-060-0 (pdf) English 62
Julkaisun sarja, numero ja raporttikoodi
Julkaisija Ilmatieteen laitos, (Erik Palménin aukio 1) Finnish Meteorological Insititute
PL 503, 00101 Helsinki Contributions 147, FMI-CONT 147
Julkaisuaika
Lokakuu, 2018
____________________________________________________________________________________________________
Tekijä Reija Ruuhela
Nimeke Sään ja ilmaston vaikutukset kuolleisuuteen ja itsetuhoisuuteen Suomessa
Tiivistelmä
Ihmiset sopeutuvat ilmastonsa tyypillisiin olosuhteisiin, mutta äärevät säätilanteet voivat aiheuttaa merkittäviä
terveysriskejä. Tämän väitöstyön tavoitteena oli mallittaa kuolleisuuden riippuvuutta lämpöolosuhteista Suomessa
ja arvioida siinä tapahtuneita muutoksia vuosikymmenien kuluessa sekä selvittää mahdollisia alueellisia eroja
sairaanhoitopiireittäin. Lisäksi tavoitteena oli arvioida sään ja ilmaston vaikutuksia itsemurhiin Suomessa ja
itsemurhayrityksiin Helsingissä. Tutkimuksissa sovellettiin useita erilaisia menetelmiä.
Helsingin ja Uudenmaan sairaanhoitopiirin kuolleisuuden 43 vuoden mittaiset aikasarjat kolmessa ikäryhmässä
muunnettiin stationaarisiksi, jolloin lämpötilan vaikutukset kuolleisuuteen olivat vertailukelpoisia väkiluvussa ja
eliniässä tapahtuneista muutoksista huolimatta. Kun kuolleisuusriskiä verrattiin vuodenajan odotusarvioihin,
kuumuuden aiheuttama kuolleisuuden nousu oli suurempi kuin kylmyyden. Kuolleisuusriski äärilämpötiloissa on
kuitenkin alentunut vuosikymmenien aikana kaikissa ikäryhmissä ‒ jopa yli 75-vuotiaiden keskuudessa, mikä
viittaa siihen, että väestön herkkyys äärilämpötiloille on pienentynyt.
Kuolleisuuden lämpötilariippuvuuden eroja 21 sairaanhoitopiirin välillä tutkittiin käyttäen viiveellisiä
epälineaarisia malleja (DLNM), ja eroja arvioitiin meta-regression avulla käyttäen kovariaatteina valittuja
ilmastollisia ja sosiodemografisia tekijöitä. Tutkimuksen mukaan alueelliset erot eivät olleet tilastollisesti
merkitseviä, joten samaa kuolleisuuden lämpötilariippuvuutta voidaan käyttää eri puolilla Suomea. Toisaalta meta-
regressio antoi viitteitä siitä, että sairaanhoitopiirien sairastavuusindeksit ja väestömäärät voivat selittää pientä
heterogeenisyyttä kuolleisuuden lämpötilariippuvuudessa.
Tarkasteltaessa 33 vuoden ajanjaksoa Suomessa toteutuneiden itsemurhien ja globaalisäteilyn välillä oli
negatiivinen korrelaatio marras-maaliskuun välisellä jaksolla, eli auringonvalon vähäisyys lisäisi itsemurhien
riskiä. Miehet olivat herkempiä auringonsäteilyn vaihtelulle kuin naiset. Tutkimus Helsingin itsemurhayritysten
sääriippuvuudesta kahdella lyhemmällä jaksolla toi esiin toisen sukupuolten välisen eron. Miesten
itsemurhayritysten riski kasvoi matalapaineen vallitessa, kun naisten itsemurha-yritysten riski kasvoi ilmanpaineen
myötä.
Näitä tutkimustuloksia voidaan hyödyntää muun muassa terveydenhuollon varautumisessa sään ääri-ilmiöihin ja
tutkimuksissa ilmastonmuutoksen vaikutuksista ihmisten terveyteen.
Julkaisijayksikkö Ilmatieteen laitos
Luokitus (UDK) Asiasanat
551.586 kuolleisuuden lämpötilariippuvuus, itsemurhat,
551.583 itsemurhayritykset, sään ja ilmaston
614.873 terveysvaikutukset
614.875
ISSN ja avainnimike
0782-6117 Finnish Meteorological Institute Contributions
ISBN Kieli Sivumäärä
978-952-336-059-4 (nid); 978-952-336-060-0 (pdf) englanti 62
Preface
My interest in health impacts of weather and climate originates around year 2000, when Ministry of
Health and Social Affairs made an initiative to the Finnish Meteorological Institute (FMI) to develop
weather service for pedestrians in order to reduce and prevent winter-time slipping injuries. During that
work my original ambivalence on the health impacts of weather and climate turned into enthusiasm ‒ I
realized how important and wide, yet underrated, a topic it was. Ever since I’ve been looking for new
possibilities to do human biometeorological research. I am grateful to Prof. Timo Partonen, Prof.
Timothy Carter and Docent Kirsti Jylhä for their interest to include health-impact studies into their
projects and, in practice, arranging possibilities for me to conduct this research. I am also grateful to the
Vilho, Yrjö and Kalle Väisälä Foundation for providing me a grant to collate research findings into this
dissertation and putting the outcomes in a wider context of current scientific knowledge on the health
impacts of weather and climate. Without this grant, I doubt if I would ever have had the possibility to
find enough time to finalize the dissertation. My supervisors, Prof. Heikki Järvinen and Docent Ari
Venäläinen helped me in setting the structure for my thesis and encouraging me to finalize it even though
my path has never been straightforward.
Making health-impact research has been and is challenging but also very rewarding. Throughout the
process I have learned so many new interesting things. I like to thank especially Prof. Andreas
Matzarakis, who greatly helped by sharing German biometeorological expertise with me, and made my
visit in the Research Center of Human Biometeorology at the German Meteorological Service (DWD)
in the spring of 2016 both useful and fun. In Finland I have had good collaboration with many experts
in the health sector: I would like to thank all my co-authors from the National Institute for Health and
Welfare (THL) with special thanks to Laura Hiltunen, Timo Partonen, Timo Lanki and Pekka Tiittanen
for their patience in explaining to me the complex issues related to health and epidemiological research.
My co-authors in FMI – Ari Venäläinen, Pentti Pirinen, Kirsti Jylhä and Otto Hyvärinen – I would like
to thank for following me to “unknown” and helping me to clarify both aims and results from the
meteorological point view and supporting me in my moments of doubt. I am grateful to Curtis Wood
for checking my English language and the interesting discussions it lead to and Sanna Luhtala for
valuable advices for both Finnish and English expressions. Finally, I would like to thank all my
colleagues at the FMI, especially in our unit – a good workplace boosts innovative thinking and easy-
going collaboration is a joy in our daily affairs.
However important and meaningful the work is, family and friends are far more important and carry
longer than any career. My parents have always encouraged me for further education and have showed
a good example on how curiosity to life maintains a good quality of life. My son, Tarek, has been
keeping my mind young by challenging me continuously with new ideas and viewpoints. Enjoyable and
refreshing leisure time with friends provides also inspiration for the scientific work. Thank you being in
my life.
Helsinki, October 2018
Reija Ruuhela
9
Contents
List of original publications ..................................................................................................... 11
1. Introduction and objectives .................................................................................................. 13
2. Background .......................................................................................................................... 15
2.1. Human adaptation to variation of climate and amount of light ..................................... 15
2.2. Weather sensitivity ........................................................................................................ 16
2.3. Human responses to variation in thermal conditions .................................................... 18
2.4. Temperature dependence of mortality ........................................................................... 19
2.5. Mental health and climate ............................................................................................. 22
2.6. Impacts of climate change on temperature‒related mortality and mental health .......... 24
3. Material and Methods ........................................................................................................... 27
3.1. All-cause deaths ............................................................................................................ 27
3.2. Suicides and suicide attempts ........................................................................................ 28
3.3. Meteorological data ....................................................................................................... 28
3.4. Methods ......................................................................................................................... 29
3.4.1. Assessing temperature‒mortality relationship ....................................................... 29
3.4.2. Baseline mortality definitions ................................................................................ 31
3.4.3. Assessing impacts of weather and climate on suicides and attempted suicides ..... 33
4. Summary of results ............................................................................................................... 35
4.1. Heatwaves in 1972 and 2010 ........................................................................................ 35
4.2. Thermal environment and all-cause mortality .............................................................. 36
4.2.1. Changes in sensitivity to thermal stress over decades ............................................ 36
4.2.2. Regional differences in the temperature‒mortality relationship ............................ 39
4.3. Impacts of weather and climate on suicides and suicide attempts ................................ 42
4.3.1. Deaths from suicide in Finland, 1971-2003 ........................................................... 42
4.3.2. Suicide attempts in Helsinki ................................................................................... 43
5. Discussion ............................................................................................................................ 44
5.1. All-cause mortality ........................................................................................................ 44
5.2. Suicides and attempted suicides .................................................................................... 46
6. Summary .............................................................................................................................. 49
References ................................................................................................................................ 52
10
11
List of original publications
I Ruuhela R, Jylhä K, Lanki T, Tiittanen P, Matzarakis A, 2017. Biometeorological
Assessment of Mortality Related to Extreme Temperatures in Helsinki Region, Finland,
1972–2014. Int. J. Environ. Res. Public Health 14(8), 944. doi:10.3390/ijerph14080944.
II Ruuhela R, Hyvärinen O, Jylhä K, 2018. Regional Assessment of Temperature-
Related Mortality in Finland. Int. J. Environ. Res. Public Health 15(3), 406.
doi:10.3390/ijerph15030406.
III Ruuhela R, Hiltunen L, Venäläinen A, Pirinen P and Partonen T, 2009. Climate
impact on suicides rates in Finland from 1971 to 2003. Int J Biometeorol 53:167-175.
doi:10.1007/s00484-008-0200-5.
IV Hiltunen L, Ruuhela R, Ostamo A, Lönnqvist J, Suominen K and Partonen T, 2012.
Atmospheric pressure and suicide attempts in Helsinki, Finland. Int J Biometeorol
56:6, 1045-1053. doi: 10.1007/s00484-011-0518-2.
In articles I – II the author conducted all the analysis and wrote the articles taking into account
the comments of the co-authors. In article III the author conducted the analysis and wrote most
of the article. In article IV the author conducted synoptic analysis and preliminary statistical
analysis for more detailed analysis and participated in writing the article as an expert in
meteorology.
12
13
1. Introduction and objectives
Impacts of weather and climate on human health have been considered important since
Hippocrates’ times in antiquities, but have been underrated in the 20th century when fast
progress in modern medicine has taken place. The awareness on adverse health impacts of
weather and climate started to increase especially after the heatwave in 2003, that caused about
70 000 extra deaths in Europe (Robine et al., 2008). In the future, heatwaves will become longer
and more intensive due to climate change, while cold stress is expected to decrease because of
climate change (Smith et al., 2014).
Depressive disorders are a major public health problem worldwide (WHO, 2017a). Weather
and climate also affect the mood of people, and seasonal affective disorder is a recognized
mental condition especially in high-latitude areas (Magnusson, 2000). Seasonality of suicides
with a peak in late spring or early summer is well known but the reasons for this seasonal pattern
are not fully understood. Typically it has been connected with increasing amount of sunshine
in spring.
This thesis focuses on direct health impacts of weather and climate in Finland via different
mechanisms. Both heat stress and cold stress lead to physiological responses that may
contribute to mortality. The impacts of heat and cold stress were studied by comparing daily
climate data and all-cause mortality in hospital districts.
The aims were to:
Model the relationship between mortality and thermal environment using various
modelling methods and biometeorological indices for thermal comfort.
Study changes in the mortality related to temperature extremes over decades, 1972–
2014, in various age groups.
Assess regional differences in temperature-related mortality in Finland.
The mental health impacts of weather and climate were studied by comparing climate factors
and deaths from suicide in Finland over three decades (1971–2003), and weather impacts on
suicide attempts in Helsinki for two shorter periods.
The aims related to the mental health impacts were to:
14
Explore the role of meteorological variables in the variation of number of deaths from
suicides in Finland.
Study the role of weather as a trigger for suicide attempts in Helsinki.
In Finland the research and awareness on health impacts of weather and climate has been quite
limited – notwithstanding the studies of a few forerunners. This thesis aims also for raising
awareness about health impacts of weather and climate and enhance the basis for further multi-
disciplinary research collaboration. Furthermore, themes of this dissertation are relevant in
planning climate change adaptation measures in the health sector. An important motivation for
health impact studies is a need to understand how climate change will affect health risks
together with the socio-demographic changes, such as aging population and urbanization, that
will take place in Finland in the future.
15
2. Background
The scientific articles included in this thesis focus on weather and climate dependence of
mortality and self-harm. There are numerous studies on the relationships between mortality and
temperature and less on weather-related morbidity and the impacts of weather on symptoms of
the chronically ill (Oudin Åström et al., 2011; Bunker et al., 2016). However, weather-related
morbidity and disease exacerbation of weather-sensitive individuals may lower the quality of
life and increase need for health services, and thus research on weather and climate impacts on
morbidity is relevant. Therefore this Background chapter begins with a wider discussion on the
adaptation of people to climatic conditions, and generic weather sensitivity, followed by more
detailed discussion on relationships between meteorological factors and mortality and mental
health.
2.1. Human adaptation to variation of climate and amount of light
Human beings have successfully adapted to various thermal environments and spread to live in
almost all climatic zones on the globe. Adaptation to climatic conditions of the living
environment can take place through various behavioural (housing, clothing) and physiological
processes on different time scales.
Genetic adaptation takes place over a long period of time. Skin colorization and basal metabolic
rate (energy consumption at rest) are typical examples of genetic adaptation to regional climatic
conditions (Jablonski and Chaplin, 2000; Leonard et al., 2002; Hancock et al., 2011). The best
cold-adapted people are found in Arctic areas and the best heat-adapted people in tropical areas.
Seasonal acclimatization takes a few weeks in the beginning of the hot or cold season (Koppe
and Jendritzky, 2005; De Freitas and Grigorieva, 2015). The physiological changes related to
this short-term acclimatization are not permanent, and they take place annually or when people
move to stay in new climate zone.
People have developed sophisticated biological mechanisms, circadian clocks, to adapt to daily
(circadian) and seasonal variation in light (Coomans et al., 2015; Honma, 2018). The
hypothalamic suprachiasmatic nucleus in the brain acts as the principal circadian clock and
synchronizes the functions of other clocks in peripheral tissues. Disturbances in the circadian
rhythms have been suggested to be associated with both physical and mental health problems
such as cardiovascular, metabolic and neuropsychiatric diseases (Hastings et al., 2003;
16
Karatsoreos, 2015; Preußner and Heyd, 2016). However, it is not clear up to date if circadian
disruption is a symptom or reason for diseases. Recently Liberman et al. (2018) presented a
model to investigate mechanisms how circadian clock genes lead to mood and circadian
disorders.
2.2. Weather sensitivity
In spite of human’s adaptation capability, weather and climate affect human health and quality
of life. The impacts depend on exposure to particular weather types and relevant individual
characteristics such as age, body size and composition, fitness, health, medication and
behaviour. Possibilities to reduce the exposure and recover from adverse impacts depend also
on a range of socioeconomic factors such as education, income, social isolation, health care
provision and urban design (Carter et al., 2016; Benmarhnia et al., 2015).
Weather and climate have some impact on all people, at least on the behaviour and outdoor
activities, but some individuals are more sensitive to weather than others. Generic weather
sensitivity, meteorosensitivity, has been studied mainly through surveys, where people
themselves report the impacts of weather on their well-being and their weather-dependent
symptoms. In surveys conducted in Germany and Canada, about 55% and 61% of the
population (Von Mackensen et al., 2005) reported that weather affects their well-being. Among
the most frequent weather-related symptoms that were listed were e.g. headache and migraine,
sleep disturbances, fatigue, pain in joints and depression. Most common weather types that
cause problems for well-being are related to cold or stormy weather. Typically women are more
sensitive to weather than men and weather sensitivity increases with age. According to a
German survey, within the group of people who consider themselves weather sensitive 76%
had chronical illnesses (Koppe et al., 2013).
The chronically ill are sensitive to different weather variables. The majority of studies report
on the significant relationship between temperature and total or cause-specific morbidity
including cardio-vascular and pulmonary diseases and diabetes, as well as emergency transport
and hospital admissions (Ye et al., 2012). A high risk of epileptic seizures has been associated
with low atmospheric pressure and high relative air humidity, whereas high ambient
temperatures seem to decrease seizure risk (Rakers et al., 2017). Strokes have been associated
with short-term changes is both high and low temperatures (Lian et al., 2015). A statistically
17
significant association has not been found between strokes and atmospheric pressure or
humidity (Cao et al., 2016).
Quality of life and mental health have a clear seasonality. For example Jia and Lubetkin (2009)
investigated the general Health Related Quality of Life (HRQoL) among the U.S. population
and found out that physical HRQoL was best during the summer and worst during the winter,
but the worst mental health occurred during the spring and fall. Grimaldi et al. (2008) found out
that in a Finnish population lower HRQoL and more severe mental ill-being were associated
with greater seasonal changes in mood and behaviour and poor illumination experienced
indoors.
In Finland, surveys on impacts of hot and cold on symptoms and thermal sensations have been
conducted in context of national FINRISK-surveys. Almost half of the subjects (46%) reported
some kind of cold-related symptoms (Näyhä et al., 2011); particularly people with pre-existing
medical conditions experience cardiovascular, respiratory or musculoskeletal symptoms in the
cold. About 80% of adult population perceive heat-related general cardiorespiratory or
psychiatric symptoms even during a normal summer, and 26°C was on average considered as
hot. Most symptoms increase by age, except feelings of thirst declines with age, and the
symptoms are more prevalent in women than men (Näyhä et al., 2014). The prevalence of heat-
related cardiorespiratory symptoms was 12% varying between 3% and 60% depending on
several factors. Heat-related cardiorespiratory symptoms are most common among people with
pre-existing lung or cardiovascular disease, agricultural workers, unemployed, pensioners, and
people having only basic education (Näyhä et al., 2017).
Many of the weather-related symptoms cannot be correlated to any single meteorological
variable, but rather to a weather type, which poses challenge to study methods. Furthermore, a
weakness of survey-based research method is that even though a big share of the variation of
the symptoms would be explained by weather variability, the causal pathways cannot always
be well explained. The best-understood responses to weather conditions are the consequences
of exposure to heat or cold.
18
2.3. Human responses to variation in thermal conditions
Physiological responses to the heat and cold stress can be described with the human energy
balance model. The model development has started in early twentieth century, but current
scientific understanding on thermal comfort is firmly based on the studies conducted in sixties
and seventies, and a classic work of Fanger (1970).
The energy balance model is based on thermoregulation of humans: the body aims to maintain
a core temperature of about 37°C, which is vital for the function of internal organs. If the subject
is exposed to cold stress, the blood vessels contract, and the blood circulation in the outer parts
of the body is reduced and skin temperature decreases. The body begins to react to cold
exposure also by e.g. shivering in order to start producing heat through muscular work.
Conversely, when the subject is exposed to heat stress, the blood vessels enlarge, and the blood
circulation in outer parts of the body increases in order to transfer excess heat from the core of
the body. Excess energy is also used for sweating and evaporation of the sweat. Severe heat or
cold stress can lead to severe symptoms or even death.
Thermal stress or thermal comfort of humans depends not only on ambient temperature but also
on humidity, wind speed and radiation. The energy exchange between environment and the
human body can be described with a heat budget model as follows (Jendritzky and de Dear,
2008):
M – W – [QH(Ta,v) + Q*(Tmrt,v)] – [QL(e,v) + QSW(e,v)] – QRe(Ta,e) ± S = 0, (1)
where M is metabolic rate of the subject, W is mechanical work of the subject, S is the storage
term describing the change in heat content of the body. Other energy terms in the equation are
also functions of meteorological factors. QH(Ta,v) describes the turbulent flux of sensible heat
that depend on air temperature (Ta) and wind velocity (v). Q*(Tmrt,v) is radiation budget, where
the mean radiant temperature (Tmrt) describes the temperature as a consequence of short- and
long-wave radiation fluxes. Turbulent fluxes of latent heat by diffusion of water vapour (QL)
and sweat evaporation (QSW) depend on water vapour pressure (e) and wind velocity (v). Energy
loss in respiration (QRe) depends on air temperature and water vapour pressure.
The complex energy transfer between the core and outer parts of the body and the insulation of
clothing are not explicitly mentioned in this equation while putting the emphasis of
meteorological factors, but these factors are essential and embedded into the energy terms
19
(Parson, 2003). The energy exchange between the environment and human body depend also
on individual characteristics such as age, weight, gender, health and acclimatization, and
environmental factors e.g. in the built environment (Rupp et al., 2015). The role of beige and
brown fat in individual differences and thermoregulation in cold may be substantial. Brown
adipose tissue (BAT) contributes to energy expenditure and cold-induced thermogenesis in
humans, and thus, impacts on the responses of humans to cold exposure (Kiefer, 2017;
Yoneshiro et al., 2016).
There are only a few rational thermal indices i.e. that are based on human energy balance. The
state-of-the-art in the field is UTCI (Universal Thermal Climate Index, Blazejczyk et al., 2012;
Blazejczyk et al., 2013). However, e.g. in comparative study of Morabito et al. (2014), it was
found that UTCI did not perform as well as expected in health impact studies. Yet, they
concluded that thermal indices, which include air temperature, humidity, wind speed and solar
radiation, correlate best with health impacts. In Article I of this thesis, daily mortality is
compared to values of Physiologically Equivalent Temperature, PET (Höppe 1993; 1999). PET
is also based on an energy budget and its values are given in degrees Celsius. PET-values in the
selected thermal environment are equal to similar thermal conditions indoors described by
temperature. Since it is impossible to calculate the heat stress at an individual level in a larger
population, the PET-values are typically calculated using standard values for light activity
(work metabolism 80W) and thermal resistance of clothing (0.9), which could be described as
clothes that people wear on cool summer day. The meteorological input variables need to be
transferred to the height of typical person: mass centre (1.1m) (Höppe, 1999; Matzarakis et al.,
1999).
Earlier, hundreds of different thermal indices have been developed for assessing indoor or
outdoor thermal comfort (e.g. review of Taleghani et al., 2013). Quite often in research on
health impacts of weather and climate, only temperature or simple empirical indices are used.
Simple indices are typically combinations of two or more meteorological variables.
2.4. Temperature dependence of mortality
The heatwave of 2003 in western and central Europe caused about 70 000 excess deaths (Robine
et al., 2008) and the heatwave of 2010 over Russia about 55 000 excess deaths (Barriopedro et
al., 2011). Both of these heatwaves extended also to Finland. According to Kollanus and Lanki
(2014), the number of non-accidental extra deaths was over 200 in 2003 and more than 300 in
20
2010. The impacts were most severe among the elderly, aged 75 years and older, as the daily
mortality increased on average by 21%. An exceptional heatwave took place in Finland in 1972
causing about 800 extra deaths. After that, remarkable heatwaves with increased mortality
occurred also e.g. in 1973, 1978, 1988, 1995 and 1997 (Näyhä, 2005).
The relationship between thermal environment and mortality is widely studied and well
understood. The temperature‒mortality relationship can be described as U-, V- or J-shaped
(Armstrong, 2006; Kovats and Hajat, 2008; Gosling et al., 2009). The mortality increases
towards both extremes of the temperature distribution, and the range of the minimum mortality
temperature (MMT) varies according to the latitude and geographical area (Basu and Samet,
2002; Curriero et al., 2002). People living in cold climates are more sensitive to hot weather
and are better acclimatized to cold conditions than people living in warmer climates. This is
reflected in the shape of the temperature‒mortality relationships.
The MMT describes the optimal thermal conditions that the population is acclimatized to and
it is lower in cool climates than in warm climates. The MMT is typically found around 75th
percentile of the annual daily mean temperature distribution, varying between the 66th and 80th
percentiles (Guo et al 2014). In Finland the mortality is lowest when the daily mean temperature
is at the range of 12–17°C (Figure 1) while in Mediterranean countries the same is true at 22–
25 °C (Keatinge et al., 2000; Näyhä, 2005; Näyhä, 2007). The MMT may vary even within a
country like reported e.g. by Tobias et al. (2017) from Spain and Curriero et al. (2002) in the
USA. A small difference in MMT between southern and northern Finland was also suggested
by Keatinge et al. (2000).
Otherwise, there are contradicting results on the climate dependence of the shape of
temperature–mortality relationships, thus in steepness of the slopes around minimum mortality.
For instance, Curriero et al. (2002) and Gasparrini et al. (2012) found a clear association of the
temperature‒mortality relationship with latitude in US cities, while climate-zone dependence
has not been found in studies of Keatinge et al. (2002) or Guo et al. (2014). Meta-analyses
across European and US cities (Baccini et al., 2008; D’Ippoliti et al., 2008; Medina-Ramón and
Schwartz, 2007) have indicated heterogeneity in the relationships. The risk varies by
community and country and differences in vulnerability and sensitivity of the population to
temperature extremes depend also on non-climatic environmental and socioeconomic factors
such as level of urbanization, buildings, share of elderly, income, education, lifestyles, access
to health care and social structures (Hondula et al., 2015; Bao et al., 2015; Carter et al., 2016).
21
Especially the elderly and people with pre-existing medical conditions such as cardiovascular
or respiratory diseases, diabetes or chronic mental illnesses are found to be vulnerable to
temperature extremes (Kovats and Hajat, 2008; Kollanus and Lanki, 2014).
The impacts of thermal stress depend also on the exposure time and they may appear with a
delay. The impacts of cold on mortality are more complex than the impacts of hot due to
different causal pathways. Heat stress leads to enlarging skin vessels and sweating which
increase cardiac work and blood viscosity, and also the risk of thrombosis. Cold stress causes
constriction of skin vessels which increases blood pressure and the risk of thrombosis.
Furthermore, breathing cold air increases the risk of respiratory infections which are associated
to increase in mortality (Näyhä, 2005). The impacts of hot weather on mortality typically appear
on the same day and last for a couple of days, whilst the increase in cold-related mortality can
be found with a delay varying from days up to weeks (Anderson and Bell, 2009; Rocklöv and
Forsberg, 2008; Yu et al., 2012). Most of the temperature-related mortality burden is
attributable to cold—even in tropical and sub-tropical areas (Gasparrini et al., 2015a). On the
other hand, recently e.g. Ebi and Mills (2013) have questioned the assumption that the seasonal
variation with higher mortality in winter than in summer would be attributable only to
temperature. Higher winter mortality may also be related to other factors that vary seasonally
such as influenza epidemics and solar radiation.
Prolonged heatwave or cold spell seems to cause an additional increase in mortality time series.
However, according to Guo et al. (2017) no added heatwave effect on mortality was found in
most of the studied countries. The excess mortality during the heatwaves is rather a cumulative
effect of hot days.
Short-term mortality displacement, so called harvesting effect, causes difficulties in assessing
overall attribution of heat and cold stress on mortality. The number of deaths increases in the
beginning of the heatwave and after a few days number of deaths may decrease even below the
baseline level. This harvesting effect may vary from one location to another. E.g. according to
comparative study of Hajat et al. (2005) the heat-related short-term mortality displacement was
higher in London than in Sao Paulo or Delhi. However, epidemiological studies have suggested
that regardless of mortality displacement, the thermal stress increases the number of deaths in
annual level. Based on data from 12 countries, Armstrong et al. (2017) concluded that most of
the deaths associated acutely with heat and cold extremes shortened lives by at least one year.
In Finland, Kollanus and Lanki (2014) did not find a decrease in mortality in the months
22
following the heatwaves in 2003 and 2010, either. Thus, the excess deaths indicated life
shortening rather than short-term mortality displacement.
Exposure to heat or cold stress may vary remarkably even in a small area due to mesoscale or
microscale climatic variation. In densely populated areas the urban heat island (UHI) may cause
an additional heat stress during heatwaves, and affect the spatial distribution of relative risk of
mortality (e.g. Taylor et al., 2015; Ketterer and Matzarakis, 2015). Especially in urban areas,
poor-air-quality episodes may take place simultaneously with heat waves or cold spells causing
an additional health burden. As a consequence of the heatwave 2010 almost 11 000 excess
deaths took place in Moscow. It has been assessed that statistically interactions between high
temperatures and air pollution from wildfires contributed to more than 2000 deaths during the
heatwave (Shaposnikov et al., 2014).
2.5. Mental health and climate
According to WHO depressive disorders are ranked globally as the single largest contributor to
non-fatal health loss (WHO, 2017a). A climate-dependent type of depressive disorder, seasonal
affective disorder (SAD), is characterized by the onset of a depressive illness during the winter
months, when there is less natural sunlight, and improved mood during the spring and summer.
SAD symptoms have been significantly associated with sunshine hours of the same and
previous week, and global radiation of the previous week (Sarran et al., 2017). On a monthly
level, more SAD symptoms have been associated with higher precipitation in the same and
previous month, and on the other hand, people had less depressive symptoms in areas with
sunnier conditions (O’Hare et al., 2016). There are also hypotheses that over-activated brown
adipose tissue might induce disrupted thermoregulation and disrupted circadian rhythm, and
might contribute to lowered mood and pronounced depressive behaviors (Partonen, 2012).
Patients with affective disorders are at higher risk to commit suicide than general population
(Bostwick and Pankratz, 2000).
In Finland almost 40% of population experience some changes in mood and behaviour routinely
during wintertime, while about 9% report actual winter depression symptoms (Grimaldi et al.
2009b). Seasonal affective disorders are linked also to physical health problems. People tend to
be physically less active and gain weight, which may lead also to health problems such as
metabolic syndrome (Grimaldi et al. 2009a).
23
Suicides accounted for 1.4% of all deaths worldwide, making it the 17th leading cause of death
in 2015 with incidence of 10.7/100,000 (WHO, 2017b). In eastern European countries the
incidence is highest and South-America lowest. In Finland over 700 persons committed suicide
in 2015 which means incidence of 13.3/100,000, which is higher than in EU-countries on
average. The median age of men who committed suicide was 48 years and of women 51 years.
About 10% of suicides were committed by younger than 25 years (OSF, 2015a).
The seasonal variation of suicides is well-known and reported all over the world both in
northern and southern hemisphere, and also in Finland (e.g. Christodoulou et al., 2012; Woo et
al., 2012; Hakko et al., 1998, Partonen et al., 2004). The suicide rate is typically highest in late
spring or early summer, and smallest in winter. Some studies have reported a secondary peak
in women suicides in autumn. Seasonality is stronger in suicides of men than women. Different
suicide methods have different seasonality (Räsänen et al. 2002) and suicides with violent
methods – hanging, shooting and jumping from high – have stronger seasonality than non-
violent methods such as poisoning. Based on earlier diagnoses, seasonality in suicides is found
with people suffering from severe depression, alcoholism, schizophrenia or other mental health
illness. However, according to the review of Ajdacid-Gross et al. (2010) the seasonality in
suicides has diminished in western countries on the basis of long time series of suicides. The
seasonality of attempted suicides is not as clear as with deaths from suicide. However, a peak
in spring and early summer is also the most typical seasonal pattern in attempted suicides
(Coimbra et al., 2016). In Finland the incidence of attempted suicide was found to be lowest in
December and highest in April (Haukka et al., 2008).
Up to date causal pathways of this seasonality are not fully understood but it has been associated
with a rapid increase in sunshine in springtime (Petridou et al., 2002; Partonen et al., 2004). In
an Austrian study a positive correlation was found between suicides and sunshine hours up to
10 days prior to suicides while more sunshine 14 to 60 days earlier was associated with lower
suicide rate (Vyssoki et al., 2014). Lambert et al. (2002) found that the rate of production of
serotonin by the brain was related to the prevailing duration of bright sunlight, and thus would
explain the seasonality of mood and seasonal affective disorder.
According to review of Thompson et al. (2018) high ambient temperatures have a range of
mental health effects with strongest evidence for increasing suicide risk. For instance, daily
mean temperature has been positively associated with suicides in East Asian countries (Kim et
al., 2016), while in the United States no correlation was found between temperature and suicide
24
rates (Dixon et al., 2007). However, possibly due to varying study methods there is little
consensus on impacts of weather factors on suicide rate other than seasonal variation, and
higher suicide risk has not been associated to specific weather conditions so far (Deisenhammer,
2003).
2.6. Impacts of climate change on temperature‒related mortality and mental health
Climate change is projected to increase heat-related mortality especially due to more intense
heatwaves and to decrease cold-related mortality due to fewer cold extremes (Smith et al.,
2014). Therefore the seasonal pattern in mortality is also expected to be gradually altered.
Ballester et al. (2011) assessed that in Europe the rise in heat-related mortality would start to
compensate for the reduction of deaths from cold during the second half of the century. A wide
multi-country study (Gasparrini et al., 2017) on projected temperature-related mortality under
climate change scenarios, assuming no adaptation or population changes, concluded that in
temperate climates such as in northern Europe a large decrease in cold-related excess deaths
would lead to marginally negative or null net change in mortality, and cold-related mortality
would remain higher than hot-related mortality even in high emission scenario. In warmer
regions a sharp increase in heat-related mortality would lead also to large net increases in
mortality, up to 12.5% (−4.7 to 28.1) in central America by the end of the century. However,
there is still ongoing discussion if climate change will substantially decrease cold-related
mortality in future because excess winter mortality is not entirely attributable to cold (e.g. Ebi
and Mills, 2013). For instance Staddon et al. (2014) also concluded that temperature
dependence of winter mortality has disappeared in the UK in recent decades and therefore
winter mortality would not decrease because of future climate change.
Only few studies have included scenarios for acclimatization and changes in sensitivity or
demographic changes (Huang et al., 2011) into projections on mortality in future climates. In a
European-wide multi-country study (Kendrovsky et al., 2017) projected changes in population
were included in an assessment of attributable heat-related deaths in two climate change
scenarios. The outcomes indicated an excess of almost 47 000 and 120 000 attributable deaths
per year by the end of the century under the Representative Concentration Pathways (RCP) 4.5
and 8.5 scenarios respectively. For Finland additional attributable deaths per year were almost
90 and about 290 respectively by the end of the century.
25
The long-term adaptation of population to a gradually warming climate and changes in
sensitivity of people to extreme temperatures cause substantial uncertainty in projected changes
in mortality. Based on past data, several studies have shown a decrease in population
susceptibility to heat over time, but a similar decrease in susceptibility to cold is not found
(Arbuthnott et al., 2016). For instance in Stockholm the relative mortality risk associated with
heat extremes has decreased but a similar decrease in cold-related mortality risk was not found
(Oudin Åström et al., 2013). Furthermore, the minimum mortality temperature has increased
over the course of the 20th century, suggesting that autonomous adaptation has taken place in
the Swedish population (Oudin Åström et al., 2016). In the study of Donaldson et al. (2003) a
decrease in heat-related mortality since 1971 was found in three climatically diverse regions
including southern Finland.
Muthers et al. (2010) concluded that heat-related mortality would increase significantly in
Vienna by the end of the 21st century also in an approach where long-term adaptation was
included. Ballester et al. (2011) suggested that if societies effectively adapt to a warmer climate,
at the end of the century the total mortality due to thermal stress might decrease in Europe,
when increases in heat-related mortality, decreases in cold-related mortality and acclimatization
are taken into account. The discussion of how to include adaptation and changes in sensitivity
into assessments of temperature-related mortality under climate change scenarios has only just
started. The method substantially affects the outcomes and the uncertainties of the outcomes,
but both shift in threshold (adaptation) and reduction in slopes (sensitivity) should be included
in climate change impact assessments (Gosling et al., 2017).
In high-latitude countries people presumably can adapt to gradually changing average thermal
conditions, and the health risks will be attributable to hot and cold extremes of the future
climate. In countries with hot climate already now, there is a risk that limits of human
thermoregulation will be exceeded. According to Mora et al. (2017) about 30% of the world’s
population is currently exposed to climatic conditions exceeding this deadly threshold for at
least 20 days a year. By 2100, this percentage is projected to increase to 48% in the scenario
with low greenhouse gas emissions and to 74% in a high emission scenario. Technical solutions
such as air-conditioning would be vital for adaptation in such climatic conditions, otherwise
migration to cooler climate would be necessary.
Studies on impacts of climate change on mental health and specifically on suicides are less than
on physical health. In causal pathways of impacts on mental health, the emphasis has been on
26
post-traumatic stress disorders as a consequence of weather-related hazards. Indirect impacts
such as mental exhaustion due to heat stress, decreasing wellbeing in communities due to
climate change impacts, and related anxiety or concerns have also arisen (Berry et al., 2010;
Doherty and Clayton, 2011; Trombley et al., 2017). Impacts of decreasing solar radiation on
seasonally affective disorders in high-latitude regions has been mentioned in few articles
without deeper studies (Berry et al., 2010). Based on analysis on relationships between suicides
and temperature variations, Williams et al. (2015) concluded that it is very difficult to predict
how climate change will affect the risk of suicide. Impacts of changes in solar radiation on
suicides under climate change scenarios were not considered.
27
3. Material and Methods
3.1. All-cause deaths
For research on the relationship between mortality and thermal conditions, daily number of all-
cause deaths and annual population in 21 hospital districts in Finland from 1971 to 2015 were
obtained from Statistics Finland. The data included the number of all-aged deaths and the
number of deaths in two age groups: 65–74 years and 75 years or older. The daily number of
deaths in additional age groups such as younger than 65 years were calculated from these data.
Daily population values were interpolated from the annual population values for each age
group. The life-expectancy in Finland over the study period has increased markedly. In 1980,
the median age at death was 68.2 years for men and 75.4 years for women. In 2015, the
corresponding values were 76.8 and 85.3 years, respectively (OSF, 2015b).
Study I concentrates on changes in the temperature‒mortality relationship in the most populated
Helsinki–Uusimaa hospital district and the whole time series of mortality by age group were
utilized in that study. Study II on regional differences in temperature‒mortality relationships
across hospital districts is based on all-aged number of deaths in a shorter period, 2000–2014.
Finland is a sparsely populated country and the characteristics of hospital districts vary
substantially. The highest population, about 1.5 million is in the Helsinki–Uusimaa hospital
district. In five hospital districts the population is less than 0.1 million and in the rest of the
hospital districts the population varies between 0.1 and 0.5 million. Daily number of deaths in
the Helsinki–Uusimaa hospital district varied between 11 and 57 with a median of 30 deaths
per day during our study period. In the smallest hospital district the median of daily deaths was
less than five while in most of the hospital districts the median of daily deaths was about 10
with maximum around 20 deaths per day. Typically the share of deaths in age group 75+ was
more than 50% of the all-aged deaths, while the share of elderly (75+) in the population varied
from 5% to 10% in hospital districts, and was highest in eastern Finland. Morbidity indices in
hospital districts were also used as covariates to explain potential differences in mortality‒
temperature relationships between hospital districts. Morbidity indices describe the generic
population health status and in the indices the weight of prevalence of selected disease groups
are based on their significance for mortality, disability, quality of life and health-care costs in
the population (THL’s morbidity index, 2016). The morbidity indices in hospital districts varied
between 66 and 147 and were highest in eastern Finland.
28
3.2. Suicides and suicide attempts
The daily number of deaths from suicide for the period 1969–2003 were obtained from Statistics
Finland for study III. From the total 43 393 suicides, 33 993 were men and 9400 women. Data
on attempted suicides are not collected systematically. Therefore data that were originally
collected for other studies (Billie-Brahe et al., 1995) were used in study IV on attempted
suicides in Helsinki. The para-suicide data of men and women included two separate periods:
the first period was from 1 January 1989 to 31 July 1990 (19 months) and second period from
15 January 1997 to 14 January 1998 (12 months). Altogether 3945 suicide attempts were made
during these two periods, about half of them were men.
3.3. Meteorological data
Various meteorological datasets were used in the studies. Selection of data was based on the
research questions and prior understanding on potentially relevant meteorological factors that
may have impact on health impact in question.
In study I, that concentrates on mortality in the Helsinki–Uusimaa hospital district, two
indicators for exposure to thermal stress were used, namely the daily mean value of PET
(Physiologically Equivalent Temperature) in Helsinki–Vantaa weather station and spatially
averaged daily mean temperature (Tavg) in the hospital district. Station-wise synoptic
temperature, relative humidity, wind speed and global radiation data were used as input to
calculate PET for the study period 1971–2014, and PET daily values we calculated from these
3-hourly data. The calculations were made with the RayMan model (Matzarakis et al. 2007;
Matzarakis et al. 2010). Values of Tavg were derived from the gridded 10 km × 10 km dataset
of the Finnish Meteorological Institute (Aalto et al. 2013).
In study II about regional differences in temperature‒mortality relationships in Finland, the
number of deaths were compared to spatially averaged daily mean temperatures (Tavg) in the
hospital districts in the period 2000–2014.
In study III, deaths from suicide in Finland were compared with global radiation, sunshine
hours, temperature and precipitation in the time period 1971–2003. Temperature and
precipitation were spatially averaged values based on the gridded database, but for solar
radiation station-wise data from Jokioinen in south-western Finland was used instead, because
29
gridded data were not available. About 80% of suicides in Finland take place in southern and
central part of the country.
In study IV, suicide attempts in Helsinki were compared to daily mean, maximum and minimum
temperatures, daily precipitation, global solar radiation, sunshine hours, and atmospheric
pressure from Helsinki–Kaisaniemi weather station. Furthermore, the deviations of temperature
and global solar radiation from their climatic normal values from the period 1971–2000 were
also considered. A descriptive analysis of weather types on peak days of attempted suicides
were based partly on synoptic weather charts, as well.
3.4. Methods
3.4.1. Assessing temperature‒mortality relationship
Study I aimed to assess changes in temperature‒mortality relationship 43-year-long study
period in the Helsinki–Uusimaa hospital district. Demographic changes were taken into account
by calculating daily mortality values as number of deaths / 100,000 inhabitants for each age
group. Expected mortality was used as a baseline mortality: daily expected mortality values
were calculated from the mortality data applying Gaussian smoothing with a filter of 365 days
for the whole study period. The time series of expected mortality include the seasonal cycle but
day-to-day variability is smoothed out. Relative mortality is then the deviation of mortality from
the expected mortality as a percentage. The time series of relative mortality are stationary,
which makes it possible to compare impacts of thermal extremes over the decades regardless
of demographic changes or lengthening life-time. The smoothing method shortens the mortality
time series from both ends and comparisons of relative mortality to the meteorological data
were made for the time period 1972–2014. Figure 1 clarifies this method that was developed
by Koppe and Jendritzky (2005).
The impacts of the thermal conditions on the relative mortality were studied by comparing the
daily mean values of PET and Tavg with the mean value of relative mortality in the same and
following day, thus applying a 1-day lag. Generalized additive model (GAM) was applied to
visualize the relationships and the analysis was done for the whole study period and separately
for two 21 year-long sub-periods, 1972–1992 and 1994–2014. The R-package “mgcv” (Wood,
2016) was used for modelling.
Quantitatively the relationships between relative mortality and the thermal indices were
calculated in 12 percentile categories of the PET and Tavg frequency distributions. Then the
30
mean values of relative mortality with 95% confidence intervals were calculated in these
percentile categories were calculated for the age groups: all ages, <65, 65–74, ≥ 75 years. In
order to assess potential changes in the relationships between relative mortality and the thermal
indices during the study period, we applied linear regression to explore linear time trends of the
mean relative mortality in each percentile category. Furthermore, the changes in relative
mortality were concretized by calculating relative mortality mean values in the percentile
categories in the two sub-periods: 1972–1992 and 1994–2014 using the percentile categories
that were defined from the whole study period. The statistical significance of differences in the
relative mortality between the sub-periods were tested by a Welch Two Sample t-test and the
Shapiro–Wilk test was used to check the normality of the distributions.
The dependence between mortality and thermal conditions varies depending on the time
window. Therefore, the relationships were calculate also for longer time windows using 7- and
14-day averages of relative mortality and thermal indices without lag considerations.
The main aim of study II was to assess regional differences is temperature‒mortality
relationships across 21 hospital districts in Finland. The research was conducted using daily all-
aged mortality data in the 15-yearlong study period, 2000–2014 and the exposure to thermal
stress was described by spatially averaged daily mean temperatures in hospital districts,
calculated from gridded temperature data.
In modelling the relationship between the daily deaths and mean temperatures in the hospital
districts, different versions of distributed lag non-linear model (DLNM) were applied
(Gasparrini et al., 2010; Gasparrini, 2011; Gasparrini and Leone, 2014). The daily number of
deaths follow quasi-Poisson distribution and the general model definition is as follows:
g(µt) = α + s(xt; β) + ∑ ℎ𝐽𝑗=1 j(cti; γj) (2)
where g is a log link function of the expectation µt ≡ E(Yt), with Yt being the time series daily
mortality counts in hospital district, α is an intercept, s(xt;β) is an exposure–response function
to temperature (xt) defined by β and it is chosen as quadratic B-spline defined by internal knots.
The cross-basis matrix of coefficients also describes lag effects of temperature, defined by knots
for lag on a logarithmic scale. Delayed effects of temperature on mortality were studied with a
lag of up to 25 days. Confounding factors (cti) were day of the week and time from the beginning
of the time series. Day of the week is modelled as a categorical variable and elapsed time as a
31
natural cubic spline with 7 degrees of freedom (df) per year to control seasonal variation and
long-term trend.
Simple model versions without delayed temperature impacts (lag = 0) and with 4 internal knots
for temperature distribution, were first applied for each hospital district (HD) separately without
confounding factors. After this first-stage modelling, meta-regression analysis (Gasparrini et
al., 2012) was conducted to assess if there is heterogeneity, thus significant differences, in
temperature‒mortality relationship across the hospital districts. In this method the effects of
random variation on the relationships is reduced by calculating best linear unbiased predictions
(BLUP) for the relationship. These BLUP estimates converge the HD-specific relationships
towards a pooled, averaged exposure–response relationship. A Cochran Q test and I2 were used
to study heterogeneity across the BLUP estimates of the relationships in hospital districts.
Climatological mean temperature, ranges of daily mean temperatures, morbidity indices,
population, and share of elderly (75 years and older) in the hospital districts were used as
covariates to explain potential heterogeneity in the temperature‒mortality relationship in the
hospital districts. An LR test and Wald test were applied to study the statistical significance.
Because temperature ranges deviate between the hospital districts, the meta-regression was
done on both absolute and relative temperature scales.
For studying the effects of heat stress and cold stress on mortality with a long delay, we applied
a more complex DLNM version with lag up to 25 days. Based on the lowest Akaike Information
Criteria (AIC) value, the best models in hospital districts had on average three internal knots
for temperature and two knots for lag, and these fixed numbers of knots were used in this
complex modelling version in all hospital districts. The R packages dlnm (Gasparrini, 2011)
and mvmeta (Gasparrini et al., 2012) were used while conducting these studies.
3.4.2. Baseline mortality definitions
The shape of temperature‒mortality relationship varies also according to the chosen definition
for baseline mortality and how seasonal variation is controlled. Different baseline mortality
definitions are used in studies I and II. The consequences of methodological differences needs
to be understood when interpreting the modelled temperature‒mortality relationships, and they
are demonstrated in Figure 1.
32
In study I the baseline mortality is seasonally varying expected mortality, which is higher in
winter than in summer. The deviation of the observed mortality from its seasonal expected
value, relative mortality, is then used in modelling the temperature‒mortality relationship, as
described is 3.4.1. In study II the temperature‒mortality relationship is modelled using mortality
at the minimum mortality temperature (MMT) as a reference, and relative risk for mortality is
then number of deaths at given temperature/ number of deaths at MMT. Different baseline
definitions lead to differences in mortality risks in the cold thermal range, while the risks in the
warm range are fairly similar and in study I the modelled increase in mortality due to cold stress
appears to be smaller than in the study II.
Figure 1. The definition for the baseline mortality effects on the shape of the temperature‒
mortality relationship. In study I (upper row) the baseline mortality is seasonally varying
“expected mortality”, while in study II (lower row) the baseline is the mortality at the
minimum mortality temperature (MMT, vertical line in the graph).
MMT
33
3.4.3. Assessing impacts of weather and climate on suicides and attempted suicides
The suicide rates and meteorological variables in the time period 1971–2003 were compared in
various time windows ranging from monthly to annual level (Study III). For each time window
the cumulative values of suicide rates (all, men and women), global radiation, sunshine hours
and precipitation were calculated as the sum of daily values for the period in question, whereas
temperature was averaged over the period. Delayed climate impacts were not considered in this
study.
Both simple and multivariate linear regression analyses were performed. Linear univariate
regression models were calculated using the suicide rate as the dependent variable and the
measured global radiation, sunshine hours, average temperature or precipitation as the
explanatory variable. For the multivariate models global radiation, average temperature and
precipitation were included stepwise as the explanatory variables.
The regression analyses were conducted both to the data that included long-term trends, and to
the residual data after the trends were removed in order to differentiate impacts of long-term
variations and trends (years) and short-term variations in weather variables. The trends were
filtered by fitting linear trends both to the suicide and weather data for two sub-periods, i.e., the
periods of increasing suicide rates from 1971 to 1990 and of decreasing suicide rates from 1991
to 2003. The regression analyses were also conducted separately to these sub-periods in order
to see if the impacts of meteorological factors differ between the time periods of increasing and
decreasing suicide trends.
Daily number of suicide attempts in Helsinki during the two shorter study periods (Study IV)
were an average two for both women and men, varying between zero and nine for women and
between zero and eight for men, and the frequency distributions followed Poisson distribution.
Based on weather data and synoptic charts, descriptive case studies were made in the beginning
in order to assess if certain weather types prevailed on cluster days with high number of suicide
attempts, and days without self-harm. The criteria for a cluster day was defined as the
probability for the number of suicide attempts being less than 0.01 according to a Poisson
distribution. Altogether 17 days fulfilled the cluster day criteria, either for men, women or both
sexes together. The outcomes of these preliminary, descriptive analyses were tested using the
34
chi-squared test. Similar descriptive analyses were made for cluster weeks and low-self-harm
weeks in order to control for day-of-the-week effect.
Poisson regression was used to study statistical association between weather variables and
suicide attempts in the whole datasets. The analysis was conducted for women and men
separately using weather parameters as independent explanatory variables and daily number of
suicide attempts as dependent variable. Daily mean temperature was taken as deviation from its
normal value, and global radiation as a proportion of its normal value in the period 1971–2000.
Furthermore, Poisson regression was performed separately on violent suicide attempts, which
are defined by the method of the suicide attempt, such as hanging or shooting.
35
4. Summary of results
4.1. Heatwaves in 1972 and 2010
Heat-related mortality is important also in Finland, but intense heatwaves do not take place
every summer. In Figure 2 the increases in mortality in the context of two remarkable heatwaves
are presented. In the data used in these impact studies in the period from 1971–2015 the highest
increase in mortality was found in the heatwave 1972. In context of exceptional the heatwave
in 2010, a new heat record was achieved. As a sign of changes in sensitivity of the population
to heat stress, the increase in mortality was less than the heatwave 1972. However, the heatwave
in 1972 lasted longer than the heatwave 2010 and the impacts are therefore not directly
comparable.
Figure 2. Daily relative mortality (all-aged, all-cause) and mean temperature (spatial average
over Finland) in summers 1972 (upper) and 2010 (lower).
36
In 1972 the heatwave started in late June and lasted for about two weeks. The highest maximum
temperature was 33.6°C and the heatwave was most intensive in eastern and northern Finland.
The peak in relative mortality was 58% and according to Näyhä (2005) during the heatwave
1972 over 800 extra deaths took place. The heatwave 2010 consisted of two shorter periods in
the middle and end of July. The highest measured maximum temperature was 37.2°C and
especially people in southern and eastern Finland were exposed to heat stress. The peak in
relative mortality was 34% and according to Kollanus and Lanki (2014) the heatwaves in 2010
caused over 300 non-accidental extra deaths.
4.2. Thermal environment and all-cause mortality
4.2.1. Changes in sensitivity to thermal stress over decades
The increase of the relative mortality in the hot extreme end of the thermal distributions is more
than in the cold extreme. The increase in relative mortality appears above the 95th percentile of
the thermal distribution and is highest above the 99th percentile among those 75 years and older.
On the other hand, from the 43-year-long time series of mortality and meteorological data we
found a statistically significant decrease in relative mortality in upper percentiles of the thermal
distribution, indicating decreasing sensitivity to heat stress over the decades. In Figure 3 the
relationship between relative mortality in Helsinki–Uusimaa hospital district and PET in
Helsinki–Vantaa airport for all-aged and elderly (≥ 75 years) for the two 21-year sub-periods.
The decrease in sensitivity to the hot extreme was statistically significant in upper percentiles:
for instance in the highest percentile category (>99th percentile) the all-aged relative mortality
decreased from 18% (1972–1992) to 9% (1994–2014). Among the elderly the decrease was
from 21% to11 %.
37
Figure 3. Relationships between relative mortality in the Helsinki–Uusimaa hospital district
and daily mean value of PET in Helsinki–Vantaa weather station in the periods 1972–1992
and 1994–2014 for all-aged and elderly, ≥ 75 years. (Adopted from study I Figure 2.)
In addition to the daily mean value of PET at Helsinki–Vantaa weather station, following other
thermal indicators were also used: daily max and min values of PET based on synoptic data in
Helsinki–Vantaa, station-wise daily mean, max and min temperatures in Helsinki–Vantaa and
spatially averaged daily mean, max and min temperatures over the Helsinki–Uusimaa hospital
district. Even though quantitatively the relationships between relative mortality and the thermal
indicators varied, the shapes of the relationships were fairly similar. In study I the relationship
38
between mortality and daily mean values of PET and spatially averaged daily mean temperature
Tavg were reported. In Figure 4 an example of the impact of thermal indicator on the shape of
the relationship between mortality and thermal environment is demonstrated.
Figure 4. Relative mortality in Helsinki–Uusimaa hospital district as a function of PET daily
mean value in Helsinki–Vantaa weather station and spatially averaged daily mean temperature
in Helsinki–Uusimaa hospital district. (Adopted from study I, Figure 2 and Figure 3.)
In the Tavg scale, the impact of heat stress appeared to be somewhat stronger than in the PET-
scale and the increase in mortality appeared to be also steeper in Tavg. This difference is partly
due to the fact that the PET-scale is wider than spatially averaged temperature, because of the
role of solar radiation in summer and wind in winter in contributing to PET-values. On the other
hand the distribution of spatial average of temperature over hospital district is narrower than
station-wise temperature distribution.
Both PET and spatially averaged temperature are feasible indicators in modelling the
relationship between mortality and thermal environment – depending on the scope of the study
and availability of the data. Based on this conclusion, spatially averaged temperature was used
as indicator for thermal stress in study II.
39
4.2.2. Regional differences in the temperature‒mortality relationship
The simple DLNM modelling without lagged effect and without controlling seasonality
produced U-shaped relationships in all hospital districts, except for the two least-populated
ones, which have populations of less than 50 000. The first-stage modelling showed large
differences and variation in the temperature–mortality relationships across hospital districts,
especially on the cold thermal range, but the BLUP estimation converges the relationships
towards the pooled, average relationship and the differences between hospital districts almost
disappear (Figure 5).
According to the meta-analysis there was no statistically significant heterogeneity in mortality–
temperature relationships among the hospital districts on an absolute temperature scale. Thus,
based on the meta-analysis the same mortality–temperature relationship can be applied in all
parts of the country. However, on the relative temperature scale, 21% of variation in the
relationships between hospital districts would be explained by heterogeneity. According to the
Wald test, morbidity index and population in the hospital districts would explain heterogeneity
on a statistically significant level, but the LR tests did not support these findings. Climatological
factors – climatological mean temperature and range of daily mean temperature in the hospital
districts – did not explain considerably the small heterogeneity.
On the basis of the pooled temperature–mortality relationships, the increase in relative risk (RR)
of mortality at a daily mean temperature of 24°C was 1.16 (1.12–1.20; CI 95%) when compared
to mortality at 14°C, which is the minimum mortality temperature (MMT) of the pooled
relationship. On the cold side, at a daily mean temperature of −20°C, RR was 1.14 (1.12–1.16;
CI 95%). The MMT was found at the 79th percentile on the relative scale of daily mean
temperature.
40
a) b)
c) d)
Figure 5. Temperature–mortality relationships based on the model without lag. a) Hospital
district (HD)-specific models (dash curves) with the pooled relationship (solid curve), b) best
linear unbiased predictions (BLUP) estimates for hospital districts, c) pooled, average
relationship (solid curve) with 95% confidence interval (shaded), relative risk (RR) reference
at T = 14°C, which is the minimum mortality temperature d) Meta-analysis on relative
temperature scale with morbidity index of hospital districts as covariate. (Adopted from Study
II, Figure 2 and Figure 3.)
The impacts of heat stress appeared typically on the same day and lasted for a few days while
the impacts of cold stress appeared after a few days but lasted for several days or even weeks.
Modelling with more complex DLNM with long 25-day lagged effects suggests that including
delayed impacts with a long lag may double or triple the estimated overall mortality risk
compared to the outcomes of a simple model (Table 2 in Study II). Figure 6 visualizes delayed
impacts of heat and cold in the Helsinki−Uusimaa hospital district.
41
a) b)
c) d)
Figure 6. Example of DLNM with 25-day delayed temperature effect on relative risk of
mortality (RR) in Helsinki−Uusimaa hospital district. a) 3D-visualizaton on RR as a function
of daily mean temperatures and lag. b) Overall temperature−mortality relationship when
impacts are aggregated over the 25-day period. c) Lagged temperature effects at Tavg = 24°C,
representing heat stress and d) Tavg = −20°C, representing cold stress. (Adopted from Study
II, Figure 4 and Figure 5.)
42
4.3. Impacts of weather and climate on suicides and suicide attempts
4.3.1. Deaths from suicide in Finland, 1971-2003
A key finding of this study was that from meteorological variables, solar radiation explained
best the inter-annual variability in suicide rates in Finland, and men appeared to be more
sensitive to lack of solar radiation than women (Figure 7). In simple regression, global solar
radiation was statistically significantly correlated with suicide rate at the inter-annual level
(negative correlation, R2=0.23). Temperature or precipitation alone could not explain variations
in suicide rates, but using them in multiple regression together with global radiation increased
the explanatory power (R2=0.32). In a multiple regression, temperature had also statistically
significant positive partial correlation with female suicide rate.
Figure 7. Annual suicide rates (upper panel) in Finland and annual global solar radiation and
sunshine hours in Jokioinen weather station in 1971−2003. (Adopted from Study III, Figure 1).
43
In the regression analysis, at monthly and seasonal levels only a few statistically significant
correlations between suicides rates and meteorological variables were found. In order to
identify the time window in which lack of solar radiation had best correlation with suicide rates,
the relationship was calculated in various cumulative time windows starting from November,
which can be considered the beginning of the darkest period in Finland. The strongest negative
correlation was found in a five month period from November to March, when (lack of) solar
radiation explained 40% of variation in male suicide rate. For women the correlation was
substantially smaller, only 14% of the variation of female suicide rate was explained by solar
radiation in this 5-month time window.
4.3.2. Suicide attempts in Helsinki
Based on Poisson regression, the daily number of suicide attempts in Helsinki were correlated
with atmospheric pressure but not with solar radiation, air temperature or precipitation. Small,
but statistically significant, correlations between daily mean atmospheric pressure and the
number of suicide attempts were opposite for women and men. For women the correlation was
positive with rate ratio, RR=1.005/hPa (p=0.026) and for men the correlation was negative with
RR=0.995/hPa (p=0.016). In the cases of violent methods of suicide attempt, the correlation
was stronger for men with RR=0.983/hPa (p<0.001), but for women not significant. The
Poisson regression was repeated by using a few-day average of the atmospheric pressure as an
explanatory variable. For women the correlation was found to be statistically significant for two
days and for men for one day before the suicide attempt.
Descriptive case studies on cluster days of suicides attempts indicated that low-pressure
situations with cloudy weather with rain or snow typically prevailed on male cluster day, but in
spring the cluster days were in high atmospheric pressure situations with sunny weather. Female
cluster days were days with high pressure and cloudy weather. No day-of-the-week effect was
found on cluster days.
Days free from suicides or suicide attempts can partly be explained by a harvesting effect, thus
short-term displacement in suicides and their attempts due to earlier higher incidences. Those
days were excluded from the case studies. For men “protective” weather conditions could be
described as near average atmospheric pressure with varying meteorological variables
otherwise. For women “protective” conditions appeared to be warmer than usual for the season
and cloudy weather in context of low pressure situations. Monday is the most probable day
without suicides or suicide attempts.
44
5. Discussion
5.1. All-cause mortality
The studies of this thesis increase the understanding of temperature-related mortality in Finland
and provide relevant information for further development of Heat-Health Warning System and
cold weather warnings. Heatwaves in Finland will become more intense and longer due to
climate change (Kim, 2017). Temperature‒mortality relationships produced in this study in the
current climate can be used as a baseline for further studies on climate change impacts on
mortality. Both studies indicate that heat-related mortality is substantial even in northern
countries like Finland. Seemingly contradictory outcomes concerning the importance of cold-
related mortality are due to different definitions for baseline mortality. In study I the increase
in mortality was compared to seasonally varying expected mortality while in study II the
increase in mortality was compared to minimum mortality, which was found at daily mean
temperature of 14°C. The baseline mortality used in study I is applicable especially in research
concentrating on impacts of extreme temperatures, while the baseline mortality used in study II
is applicable in research on generic temperature‒mortality relationships.
Study I, in which baseline mortality includes seasonal variation, indicates that increases in
relative mortality are greater in hot than in cold extreme conditions, and people in Finland are
less accustomed to high than low temperatures. This finding supports the globally demonstrated
phenomenon that people adapt to their climatic conditions. The increase in relative mortality
appears above the 95th percentile of the thermal distribution and is highest above the 99th
percentile among elderly, 75 years and older. However, the dependence of mortality on extreme
temperatures has weakened during the 43-year-long study period, even among the age group
75 years and older. Thus, the study shows that the sensitivity of the Finnish population to
temperature extremes has decreased in recent decades. It was beyond the scope of the study to
explain reasons behind the decreasing sensitivity. Potential explanations can be physiological
or behavioural changes of people or infrastructural changes in society, such as improved public
health and lengthening of life expectancy.
The calculation of PET values requires sub-daily meteorological data – temperature, humidity,
wind speed and solar radiation – and spatial distribution of PET values is affected remarkably
by topography and the built environment. The study I shows that temperature may give results
that in many cases are good enough for studies related to impacts of weather and climate on
mortality. Especially in climate change impact studies it may be adequate to use only
45
temperature scenarios without considering projections for all the input variables that are needed
for calculations of PET. The use of gridded temperature data instead of station-wise
meteorological data extends the possibilities to study weather impacts on mortality in sparsely-
populated larger areas, which is the case in most parts of Finland.
Study II showed that modelling temperature‒mortality relationships at the hospital-district level
gave realistic, U-shaped relationships even in sparsely populated areas, in hospital districts with
population less than 50 000. The shapes of hospital-district-specific first-stage relationship
varied substantially but the best linear unbiased prediction (BLUP) reduced the variability in
the relationships and statistically significant differences in the temperature‒mortality
relationships across hospital districts were not found. Thus, based on the meta-analysis the same
mortality–temperature relationship can be applied in all parts of the country.
There are substantial regional differences in mortality and morbidity in Finland (THL, 2016;
Figure 8). People in southern and western parts of the country are healthier and mortality lower
than in eastern and northern parts of the country. The hypothesis for the study II was that there
would be regional differences in temperature‒mortality relationships and acclimatization of
population in Finland, but this hypothesis could not be confirmed. However, the meta-
regression on the relative temperature scale suggests that morbidity index and population in the
hospital districts might explain some of the small regional heterogeneity of the temperature–
mortality association. In future studies this could be further examined by using more relevant
Figure 8. Mortality (1/100,000; left) and morbidity index (right) in hospital districts in 2014.
(THL, Sotkanet)
46
health indicators such as incidences of weather-sensitive chronic diseases (e.g. cardiovascular
and respiratory diseases) as additional explaining variables. Climatological factors –
climatological mean temperature and range of daily mean temperature – did not explain
considerably the small heterogeneity.
Seasonality in mortality is well-known, however, the reasons behind this seasonality are not
self-evident. Based on the studies of this thesis I tend to support the conclusion of Ebi and Mills
(2013) that higher mortality during winter is not related to temperature variation only but also
to other seasonally varying meteorological and behavioural factors, and influenza epidemics.
Furthermore people spend most of their time indoors. Based on a population study in Finland,
people spent only 4% of their total time under cold exposure (Mäkinen et al., 2006). A
climatologically interesting candidate to explain excess mortality during winter might be lack
of solar radiation and vitamin D concentration, which typically varies seasonally. In a meta-
analysis of Rush et al. (2013) vitamin D status was inversely associated with all-cause mortality.
That meta-analysis included also the study of Virtanen et al. (2011) from Finland.
Sensitivity to temperature extremes has decreased over decades, but based on that result from
the past times series we cannot conclude that this trend would continue also in the future. In
climate change impact studies various scenarios for development in sensitivity and
acclimatization should also be considered and applied together with different climate change
scenarios. People spend most of their time indoors. Therefore the studies, in which the exposure
to thermal stress is based only on outdoor thermal conditions, do not provide accurate
information on the real exposure to heat or cold stress and further studies are needed to assess
relationship between indoor and outdoor thermal conditions. Furthermore, urbanization is
expected to continue in the future, and a big share of the population will be exposure to thermal
stress, that is also modified by the urban heat island effect. Further holistic, multi-disciplinary
research is needed to better explain and predict future temperature-related mortality.
5.2. Suicides and attempted suicides
In 1970s the Finnish health authorities became concerned about the increasing suicide trend
and in 1986 a nationwide suicide-prevention program was launched, which lead to decreasing
suicide trend after 1990 (Beskow et al. 1999, Hakko et al. 1998). The main finding of study III
is that the variation in annual solar radiation explains part (about 20%) of the inter-annual
variation in suicides. Furthermore, this finding also suggests that both increasing suicide trend
47
until 1990 and decreasing trend after that could partly be explained by long-term variation in
solar radiation (see also Figure 7).
Lack of solar radiation was associated with higher suicide risk especially during a five month
period from November to March. Thus, the darker the winter the higher the suicide mortality.
This raises another concern related to climate change. In Finland the amount of solar radiation
is expected to decrease in winter, in the high-emission RCP8.5 forcing scenario the projected
change is −17% to +2% by middle of the century (Ruosteenoja et al., 2016). The decrease in
solar radiation will be exacerbated by a shortening snow-cover period, which will reduce the
amount of reflected shortwave radiation from the ground. The worst scenario is that the positive
development in suicide prevention and decreasing suicide trend will come to a halt because of
darker winters. Since e.g. seasonal affective disorders depend on solar radiation, future studies
need to be extended more widely to the impacts on mental health. Furthermore, the impacts of
various solar-radiation spectral ranges on mental health need to be better understood.
The outcome of this study may seem to contradict the earlier studies that associate quickly
increasing solar radiation with higher suicide incidence in spring and early summer. However,
season and length of the time windows explain these different outcomes. Correlations between
deaths from suicide and meteorological variables indicated that the lack of solar radiation
increased suicide risk especially in winter. It has been suggested that sunshine acts like an
antidepressant that first increase the suicide risk and later improves the mood (Christodoulou et
al., 2012). In future studies the impact of solar radiation could be studied by using more
advanced modelling methods that would take into account both cumulative lack of solar
radiation in winter as a factor that increases the number of people at suicide risk, and quickly
increasing amount of solar radiation as trigger for committing suicide in spring. As suggested
also by Vyssoki et al. (2014), sunshine may have a bimodal effect on suicidal behaviour: more
sunshine may increase suicide risk in short exposure time scale but decrease suicide risks in
longer exposure time scale.
Another interesting outcome of study III is that climatic conditions have impact on suicide risks
for both genders, but men and women react differently. Men appeared to be more sensitive than
women to variation in solar radiation. Furthermore, temperature was associated more with
suicide risk of women than of men. A later study (Hiltunen et al., 2014) also concluded that
daily temperatures and large increase in temperature in a few-day period, as can often happen
in spring, may contribute to higher suicide risk. These gender differences might be related to
48
differences in circadian rhythms and thermoregulation between sexes, but requires further
studies.
The main findings of study IV, on relationship between suicide attempts and weather, were that
atmospheric pressure predicted the risk of suicides attempts, and the impact of atmospheric
pressure was opposite on men and women. Low pressure conditions were associated with higher
risk for suicide attempts of men, but with lower risk for suicide attempts of women. The other
meteorological variables – solar radiation, temperature and precipitation – did not have
statistically significant correlation with the suicide attempts in Helsinki. However, the
robustness of these outcomes are limited due to short time series and small samples.
Understanding a causal pathway explaining this relationship needs further studies. If the
atmospheric pressure as such has direct impacts on the suicide attempt risks, and when the
impacts on men and women are opposite, the reason might even be hormonal and e.g. the levels
of some sex-specific hormones have been found to respond to atmospheric pressure (El-
Migdadi et al., 2000). On the other hand, atmospheric pressure might be an indicator describing
the weather type. The descriptive meteorological case studies indicated that the impacts of
weather types on para-suicides may also vary seasonally, but due to short datasets of suicide
attempts a firm conclusion cannot be made.
These studies have increased the understanding on relationships between meteorological
variables and deaths from suicide and suicide attempts, but only piecemeal, and further
comprehensive, multidisciplinary research is needed to understand how individual, societal and
environmental factors together impact on mental health and suicide risks. Associations between
meteorological factors and deaths from suicides and suicide attempts appear seemingly
contradictory. Partly these differences are related to differences in time spans used in the studies
and a small sample size of suicide attempts is a limitation for the study. However, there may be
also other factors explaining the differences, since for instance the differences in gender ratios
between deaths from suicides and suicides attempts are not fully understood. The impacts of
weather and climate on suicides and suicide attempts are small compared to the other risk
factors, but they may contribute by increasing the probability or affect timing of self-harm.
However, the applicability of these outcomes could also be studied in clinical work when
assessing risk of suicidal patients.
49
6. Summary
Human beings are able to adapt to their climatic normal conditions, but despite this capability,
weather extremes especially may pose a substantial health risk. Changing climate raises
questions on how the weather-related health risks will change and to what extent people can
adapt to continuously changing climate and extremes of the future climate. This thesis on
weather dependence of all-cause mortality and suicides and suicide attempts in the present
climate in Finland gives a basis for further studies on impacts of climate change on human
health in Finland, and in general in northern-latitude countries.
Human thermoregulation aims to maintain the core temperature of the body as constant.
Thermoregulation forms the basis of a human energy balance model that describes the heat
exchange between body and environment. Ambient temperature, humidity, wind speed and
radiation balance are essential meteorological factors in this energy exchange. Both heat and
cold stress may pose severe risk to health and contribute to premature death. Elderly and people
with pre-existing chronic diseases such as cardiovascular and respiratory diseases are most
vulnerable to temperature extremes.
In this thesis different methods were used to study dependence of mortality on thermal
conditions. The method to calculate relative mortality makes the long mortality time series
stationary and, thus, comparable over the decades regardless of changes in population and life-
expectancy. This method was applied to study changes in sensitivity of Finnish population to
temperature extremes over a 43-year-long period. Hospital-district-specific temperature‒
mortality relationships were studied using complex, distributed lag non-linear models and
differences in these relationships were assessed by meta-regression with selected climatic and
sociodemographic covariates.
The studies of this thesis show that heat-related mortality is remarkable also in a northern
country like Finland. The increase of the relative mortality in the hot extreme of the thermal
distributions is more than in the cold extreme. On the other hand, based on the analysis of the
43-year-long time series, a statistically significant decrease in relative mortality in the upper
percentiles of the thermal distribution was found indicating decreasing sensitivity to thermal
stress in the Finnish population over the decades.
Regional differences in temperature–mortality relationships were not found at the statistically
significant level on the absolute temperature scale, thus, the same pooled temperature–mortality
50
relationship can be applied in different parts of the country in e.g. climate change impact studies
in future. On the other hand, the meta-analysis on a relative temperature scale indicated that
morbidity indices and population in hospital districts could explain the small heterogeneity in
the relationships ‒ a characteristics that might be worthy of deeper studies. Further studies are
also needed to assess temperature-related mortality in areas where urban heat island effect is
clear.
Both station-wise PET (physiologically equivalent temperature) and spatially-averaged
temperature over larger areas like hospital districts are feasible indicators in modelling the
dependence of mortality on the thermal environment. The indicator can be selected depending
on the scope of the study and availability of the data.
The study on deaths from suicide on the basis of a 33-year-long time series from Finland
showed a significant association between suicide rates and global solar radiation especially
during winter in the period from November to March, suggesting that a lack of solar radiation
during the winter in higher-latitude regions increases the risk of suicide mortality. The study
showed also that men are more sensitive to variation in solar radiation than women. On the
other hand, women seemed to be more sensitive than men to variation in temperature, but the
temperature dependence of suicides was not as strong as the dependence on solar radiation. The
study of weather dependence of attempted suicides in Helsinki on the basis of two shorter
periods showed another interesting difference between genders. The risk of suicide attempts of
men increased with decreasing atmospheric pressure, while the risk of suicide attempts of
women increased with increasing pressure.
Further studies on self-harm should be conducted by using more sophisticated modelling
methods such as distributed lag non-linear models in order to understand better the impact of
solar radiation. Based on studies of this thesis, lack of solar radiation in winter may increase
the number of people at suicide risk, while – based on literature – quickly increasing amount of
solar radiation in spring may act as a trigger for committing suicide.
The impacts of decreasing global solar radiation on suicide risks due to climate change need to
be studied. Future winters in Finland will become milder with increasing cloudiness and
precipitation and shorter snow-cover period. These together with decreasing solar radiation may
lead to wider adverse mental health impacts in Finland through seasonally affective disorders,
SAD.
51
The outcomes of this thesis can be used in communicating the impacts of weather and climate
on both physical and mental health. They can also be applied in the health sector for instance
to improve preparedness for heatwaves and cold spells, and planning long-term adaptation
measures to climate change.
52
References
Aalto, J.; Pirinen, P.; Heikkinen, J.; Venäläinen, A. Spatial interpolation of monthly climate data for
Finland: Comparing the performance of kriging and generalized additive models. Theor. Appl. Climatol.
2013, 112, 99–111, doi:10.1007/s00704-012-0716-9.
Ajdacic-Gross, V.; Bopp, M.; Ring, M.; Gutzwiller, F.; Rossler, W. Seasonality in suicide - A review and
search of new concepts for explaining the heterogeneous phenomena. Soc. Sci. Med. 2010, 71, 657–666,
doi:10.1016/j.socscimed.2010.05.030.
Anderson, B. G.; Bell, M. L. Weather-Related Mortality. Epidemiology 2009, 20, 205–213,
doi:10.1097/EDE.0b013e318190ee08.
Arbuthnott, K.; Hajat, S.; Heaviside, C.; Vardoulakis, S. Changes in population susceptibility to heat and
cold over time: assessing adaptation to climate change. Environ. Heal. 2016, 15, doi:10.1186/s12940-016-
0102-7.
Armstrong, B. Models for the Relationship between Ambient Temperature and Daily Mortality. Source
Epidemiol. 2006, 17, 624–631, doi:10.1097/01.ede.0000239732.50999.8f.
Armstrong, B.; Bell, M. L.; Sousa, M. De; Stagliorio, Z.; Guo, Y. L.; Guo, Y.; Goodman, P.; Hashizume,
M.; Honda, Y.; Kim, H.; Lavigne, E.; Michelozzi, P.; Hilario, P.; Saldiva, N.; Schwartz, J.; Scortichini, M.;
Sera, F.; Tobias, A. Longer-Term Impact of High and Low Temperature on Mortality : An International
Study to Clarify Length of Mortality Displacement. Environ. Health Perspect. 2017, doi:10.1289/EHP1756.
Baccini, M.; Biggeri, A.; Accetta, G.; Kosatsky, T.; Katsouyanni, K.; Analitis, A.; Anderson, H. R.; Bisanti,
L.; D’Iippoliti, D.; Danova, J.; Forsberg, B.; Medina, S.; Paldy, A.; Rabczenko, D.; Schindler, C.;
Michelozzi, P. Heat effects on mortality in 15 European cities. Epidemiology 2008, 19, 711–719,
doi:10.1097/EDE.0b013e318176bfcd.
Ballester, J.; Robine, J.-M.; Herrmann, F. R.; Rodó, X. Long-term projections and acclimatization scenarios
of temperature-related mortality in Europe. Nat. Commun. 2011, 2, 358, doi:10.1038/ncomms1360.
Bao, J.; Li, X.; Yu, C. The construction and validation of the heat vulnerability index, a review. Int. J.
Environ. Res. Public Health 2015, 12, 7220–7234, doi:10.3390/ijerph120707220.
Barriopedro, D.; Fischer, E. M.; Luterbacher, J.; Trigo, R. M.; García-Herrera, R. The hot summer of 2010:
redrawing the temperature record map of Europe. Science (80-. ). 2011, 332, 220–224,
doi:10.1126/science.1201224.
Basu, R.; Samet, J. M. Relation between elevated ambient temperature and mortality: A review of the
epidemiologic evidence. Epidemiol. Rev. 2002, 24, 190–202, doi:10.1093/epirev/mxf007.
53
Benmarhnia, T.; Deguen, S.; Kaufman, J. S.; Smargiassi, A. Vulnerability to heat-related mortality: A
systematic review, meta-analysis, and meta-regression analysis. Epidemiology 2015, 26, 781–793,
doi:10.1097/EDE.0000000000000375.
Berry, H. L.; Bowen, K.; Kjellstrom, T. Climate change and mental health: A causal pathways framework.
Int. J. Public Health 2010, 55, 123–132, doi:10.1007/s00038-009-0112-0.
Beskow, J.; Kerkhof, A.; Kokkola, A.; Uutela, A. Suicide prevention in Finland 1986–1996: external
evaluation by an international peer group. Ministry Soc Aff Health Copies 1999:2. Ministry of Social
Affairs and Health, Helsinki.
Bille-Brahe, U.; Schmidtke, A.; Kerkhof, A.J.; De Leo, D.; Lonnqvist, J.; Platt, S.; Sampaio Faria, J.
Background and introduction to the WHO/EURO Multicentre Study on Parasuicide. Crisis 1995, 16, 72–
8:84.
Blazejczyk, K.; Epstein, Y.; Jendritzky, G.; Staiger, H.; Tinz, B. Comparison of UTCI to selected thermal
indices. Int. J. Biometeorol. 2012, 56, 515–535, doi:10.1007/s00484-011-0453-2.
Błazejczyk, K.; Jendritzky, G.; Bröde, P.; Fiala, D.; Havenith, G.; Epstein, Y.; Psikuta, A.; Kampmann, B.
An introduction to the Universal thermal climate index (UTCI). Geogr. Pol. 2013, 86, 5–10,
doi:10.7163/GPol.2013.1.
Bostwick, J. M.; Pankratz, V. S. Affective Disorders and Suicide Risk: A Reexamination. Am. J. Psychiatry
2000, 157, 1925–1932, doi:10.1176/appi.ajp.157.12.1925.
Bunker, A.; Wildenhain, J.; Vandenbergh, A.; Henschke, N.; Rocklöv, J.; Hajat, S.; Sauerborn, R. Effects of
Air Temperature on Climate-Sensitive Mortality and Morbidity Outcomes in the Elderly; a Systematic
Review and Meta-analysis of Epidemiological Evidence. EBioMedicine 2016, 6, 258–268,
doi:10.1016/j.ebiom.2016.02.034.
Cao, Y.; Wang, X.; Zheng, D.; Robinson, T.; Hong, D.; Richtering, S.; Leong, T. H.; Salam, A.; Anderson,
C.; Hackett, M. L. Air Pressure, Humidity and Stroke Occurrence: A Systematic Review and Meta-Analysis.
Int. J. Environ. Res. Public Health 2016, 13, 1–10, doi:10.3390/ijerph13070675.
Carter, T. R.; Fronzek, S.; Inkinen, A.; Lahtinen, I.; Lahtinen, M.; Mela, H.; O’Brien, K. L.; Rosentrater, L.
D.; Ruuhela, R.; Simonsson, L.; Terama, E. Characterising vulnerability of the elderly to climate change in
the Nordic region. Reg. Environ. Chang. 2016, 16, 43–58, doi:10.1007/s10113-014-0688-7.
Christodoulou, C.; Douzenis, A.; Papadopoulos, F. C.; Papadopoulou, A.; Bouras, G.; Gournellis, R.;
Lykouras, L. Suicide and seasonality. Acta Psychiatr. Scand. 2012, 125, 127–146, doi:10.1111/j.1600-
0447.2011.01750.x.
Coimbra, D. G.; Pereira e Silva, A. C.; De Sousa-Rodrigues, C. F.; Barbosa, F. T.; De Siqueira Figueredo,
D.; Araújo Santos, J. L.; Barbosa, M. R.; De Medeiros Alves, V.; Nardi, A. E.; De Andrade, T. G. Do
54
suicide attempts occur more frequently in the spring too? A systematic review and rhythmic analysis. J.
Affect. Disord. 2016, 196, 125–137, doi:10.1016/j.jad.2016.02.036.
Coomans, C. P.; Ramkisoensing, A.; Meijer, J. H. The suprachiasmatic nuclei as a seasonal clock. Front.
Neuroendocrinol. 2015, 37, 29–42, doi:10.1016/j.yfrne.2014.11.002.
Curriero, F. C.; Heiner, K. S.; Samet, J. M.; Zeger, S. L.; Strug, L.; Patz, J. A. Temperature and mortality in
11cities of the eastern United States. Am J Epidemiol 2002, 155, 80–87.
De Freitas, C. R.; Grigorieva, E. A. Role of acclimatization in weather-related human mortality during the
transition seasons of autumn and spring in a thermally extreme mid-latitude continental climate. Int. J.
Environ. Res. Public Health 2015, 12, 14974–14987, doi:10.3390/ijerph121214962.
Deisenhammer, E. A. Weather and suicide: the present state of knowledge on the association of
meteorological factors with suicidal behaviour. Acta Psychiatr Scand 2003, 108, 402–409.
Dixon, P. G.; McDonald, A. N.; Scheitlin, K. N.; Stapleton, J. E.; Allen, J. S.; Carter, W. M.; Holley, M. R.;
Inman, D. D.; Roberts, J. B. Effects of temperature variation on suicide in five U.S. counties, 1991-2001.
Int. J. Biometeorol. 2007, 51, 395–403, doi:10.1007/s00484-006-0081-4.
D’Ippoliti, D.; Michelozzi, P.; Marino, C.; Menne, B.; Cabre, M. G.; Katsouyanni, K.; Medina, S.; Paldy,
A.; Anderson, H. R.; Ballester, F. The Impact of Heat Waves on Mortality in 9 European Cities, 1990–2004.
Epidemiology 2008, 19, S286–S287.
Doherty, T. J.; Clayton, S. The Psychological Impacts of Global Climate Change. Am. Psychol. 2011, 66,
265–276, doi:10.1037/a0023141.
Donaldson, G. C.; Keatinge, W. R.; Näyhä, S. Changes in summer temperature and heat-related mortality
since 1971 in North Carolina, South Finland, and Southeast England. Environ. Res. 2003, 91, 1–7,
doi:10.1016/S0013-9351(02)00002-6.
Ebi, K. L.; Mills, D. Winter mortality in a warming climate: A reassessment. Wiley Interdiscip. Rev. Clim.
Chang. 2013, 4, 203–212, doi:10.1002/wcc.211.
El-Migdadi, F.; Nusier, M.; Bashir, N. Seasonal pattern of leutinizing, follicle-stimulating hormone,
testosterone and progesterone in adult population of both sexes in the Jordan Valley. Endocr Res 2000,
26:41–48.
Fanger, P.O. Thermal Comfort. 1970. Copenhagen, Danish Technical Press.
Gasparrini, A.; Armstrong, B.; Kenward, M. G. Distributed lag non-linear models. Stat. Med. 2010, 29,
2224–2234, doi:10.1002/sim.3940.
55
Gasparrini, A. Distributed Lag Linear and Non-Linear Models in R : The Package dlnm. J. Stat. Softw. 2011,
43, 1–20, doi:http://dx.doi.org/10.18637/jss.v043.i08.
Gasparrini, A.; Armstrong, B.; Kenward, M. G. Multivariate meta-analysis for non-linear and other multi-
parameter associations. Stat. Med. 2012, 31, 3821–3839, doi:10.1002/sim.5471.
Gasparrini, A.; Guo, Y.; Hashizume, M.; Lavigne, E.; Zanobetti, A.; Schwartz, J.; Tobias, A.; Tong, S.;
Rocklöv, J.; Forsberg, B.; Leone, M.; De Sario, M.; Bell, M. L.; Guo, Y. L. L.; Wu, C. F.; Kan, H.; Yi, S.
M.; De Sousa Zanotti Stagliorio Coelho, M.; Saldiva, P. H. N.; Honda, Y.; Kim, H.; Armstrong, B.
Mortality risk attributable to high and low ambient temperature: A multicountry observational study. Lancet
2015a, 386, 369–375, doi:10.1016/S0140-6736(14)62114-0.
Gasparrini, A.; Leone, M. Attributable risk from distributed lag models. BMC Med. Res. Methodol. 2014,
14, 55, doi:10.1186/1471-2288-14-55.
Gasparrini, A.; Guo, Y.; Sera, F.; Vicedo-Cabrera, A. M.; Huber, V.; Tong, S.; de Sousa Zanotti Stagliorio
Coelho, M.; Nascimento Saldiva, P. H.; Lavigne, E.; Matus Correa, P.; Valdes Ortega, N.; Kan, H.; Osorio,
S.; Kyselý, J.; Urban, A.; Jaakkola, J. J. K.; Ryti, N. R. I.; Pascal, M.; Goodman, P. G.; Zeka, A.;
Michelozzi, P.; Scortichini, M.; Hashizume, M.; Honda, Y.; Hurtado-Diaz, M.; Cesar Cruz, J.; Seposo, X.;
Kim, H.; Tobias, A.; Iñiguez, C.; Forsberg, B.; Åström, D. O.; Ragettli, M. S.; Guo, Y. L.; Wu, C.;
Zanobetti, A.; Schwartz, J.; Bell, M. L.; Dang, T. N.; Van, D. Do; Heaviside, C.; Vardoulakis, S.; Hajat, S.;
Haines, A.; Armstrong, B. Projections of temperature-related excess mortality under climate change
scenarios. Lancet Planet. Heal. 2017, 360–367, doi:10.1016/S2542-5196(17)30156-0.
Gosling, S. N.; Lowe, J. A.; McGregor, G. R.; Pelling, M.; Malamud, B. D. Associations between elevated
atmospheric temperature and human mortality: A critical review of the literature. Climatic Change 2009,
92, 299-341; doi:10.1007/s10584-008-9441-x.
Gosling, S.; Hondula, D.; Bunker, A.; Ibarreta, D.; Liu, J.; Zhang, X.; Sauerborn, R. Adaptation to climate
change: a comparative analysis of modelling methods for heat-related mortality. Environ. Health Perspect.
2017, 125, doi:10.1289/EHP634.
Grimaldi, S.; Partonen, T.; Saarni, S. I.; Aromaa, A.; Lönnqvist, J. Indoors illumination and seasonal
changes in mood and behavior are associated with the health-related quality of life. Heal. Qual Life
Outcomes 2008, 6, 56, doi:10.1186/1477-7525-6-56.
Grimaldi, S.; Englund, A.; Partonen, T.; Haukka, J.; Pirkola, S.; Reunanen, A.; Aromaa, A.; Lnnqvist, J.
Experienced poor lighting contributes to the seasonal fluctuations in weight and appetite that relate to the
metabolic syndrome. J. Environ. Public Health 2009a, 2009, doi:10.1155/2009/165013.
Grimaldi, S.; Partonen, T.; Haukka, J.; Aromaa, A.; Lönnqvist, J. Seasonal vegetative and affective
symptoms in the Finnish general population: Testing the dual vulnerability and latitude effect hypotheses.
Nord. J. Psychiatry 2009b, 63, 397–404, doi:10.1080/08039480902878729.
56
Guo, Y.; Gasparrini, A.; Armstrong, B. G.; Tawatsupa, B.; Tobias, A.; Lavigne, E.; Coelho, M. de S. Z. S.;
Pan, X.; Kim, H.; Hashizume, M.; Honda, Y.; Guo, Y.-L. L.; Wu, C.-F.; Zanobetti, A.; Schwartz, J. D.; Bell,
M. L.; Scortichini, M.; Michelozzi, P.; Punnasiri, K.; Li, S.; Tian, L.; Garcia, S. D. O.; Seposo, X.;
Overcenco, A.; Zeka, A.; Goodman, P.; Dang, T. N.; Dung, D. Van; Mayvaneh, F.; Saldiva, P. H. N.;
Williams, G.; Tong, S. Heat Wave and Mortality: A Multicountry, Multicommunity Study. Environ. Health
Perspect. 2017, 125, 87006–1, doi:10.1289/EHP1026.
Guo, Y.; Gasparrini, A.; Armstrong, B.; Li, S.; Tawatsupa, B.; Tobias, A.; Lavigne, E.; Zanotti, D. S.;
Coelho, S.; Leone, M.; Guo, Y. L.; Wu, C.; Punnasiri, K.; Yi, S.; Michelozzi, P.; Hilario, P.; Saldiva, N.;
Williams, G. Global Variation in the Effects of Ambient Temperature on Mortality. A Systematic
Evaluation. Epidemiology 2014, 25, 781–789, doi:10.1097/EDE.0000000000000165.
Hajat, S.; Armstrong, B. G.; Gouveia, N.; Wilkinson, P. Mortality Displacement of Heat-Related Deaths. A
Comparison of Delhi, São Paulo, and London. Epidemiology 2005, 16, 613–620,
doi:10.1097/01.ede.0000164559.41092.2a.
Hakko, H.; Rasanen, P.; Tiihonen, J. Seasonal variation in suicide occurrence in Finland. Acta Psychiatr.
Scand. 1998, 98, 92–97, doi:10.1111/j.1600-0447.1998.tb10048.x.
Hancock, A. M.; Witonsky, D. B.; Gorka, A.-A.; Beall, C. M.; Gebremedhin, A.; Sukernik, R.; Utermann,
G.; Pritchard, J.; Coop, G.; Di Rienzo, A. Adaptations to climate-mediated selective pressures in Humans.
PLOS Gentics 2011, 7, e1001375, doi:10.1371/journal.pgen.1001375.
Hastings, M. H.; Reddy, A. B.; Maywood, E. S. A clockwork web: Circadian timing in brain and periphery,
in health and disease. Nat. Rev. Neurosci. 2003, 4, 649–661, doi:10.1038/nrn1177.
Haukka, J.; Suominen, K.; Partonen, T.; Lönnqvist, J. Determinants and outcomes of serious attempted
suicide: A nationwide study in Finland, 1996-2003. Am. J. Epidemiol. 2008, 167, 1155–1163,
doi:10.1093/aje/kwn017.
Hiltunen, L.; Haukka, J.; Ruuhela, R.; Suominen, K.; Partonen, T. Local daily temperatures, thermal
seasons, and suicide rates in Finland from 1974 to 2010. Environ. Health Prev. Med. 2014, 19,
doi:10.1007/s12199-014-0391-9.
Hippocrates ca. 400 B.C. Airs, Waters, Places. Lloyd GER, ed 1978. Hippocratic writings. Harmondsworth,
England: Penguin Classics, 1983.
Hondula, D. M.; Davis, R. E.; Saha, M. V.; Wegner, C. R.; Veazey, L. M. Geographic dimensions of heat-
related mortality in seven U.S. cities. Environ. Res. 2015, 138, doi:10.1016/j.envres.2015.02.033.
Honma, S. The mammalian circadian system: a hierarchical multi-oscillator structure for generating
circadian rhythm. J. Physiol. Sci. 2018, 68, 207–219, doi:10.1007/s12576-018-0597-5.
Höppe, P. R. Heat balance modelling. Experientia 1993, 49, 741–746, doi:10.1007/BF01923542.
57
Höppe, P. The physiological equivalent temperature - a universal index for the biometeorological
assessment of the thermal environment. Int. J. Biometeorol. 1999, 43, 71–75, doi:10.1007/s004840050118.
Huang, C.; Barnett, A. G.; Wang, X.; Vaneckova, P.; FitzGerald, G.; Tong, S. Projecting Future Heat-
Related Mortality under Climate Change Scenarios: A Systematic Review. Environ. Health Perspect. 2011,
119, 1681–1690, doi:10.1289/ehp.1103456.
Jablonski, N. G.; Chaplin, G. Human skin pigmentation as an adaptation to UV radiation. Proc. Natl. Acad.
Sci. 2010, 107, 8962–8968, doi:10.1073/pnas.0914628107.
Jendritzky, G.; de Dear, R. Adaptation and Thermal Environment. In Biometeorology for Adaptation to
Climate Variability and Change; Ebi, K.L., Burton, I., McGregor, G.R., Eds.; Springer: The Netherlands,
2009; pp. 9–32. doi:10.1007/978-1-4020-8921-3.
Jia, H. M.; Lubetkin, E. I. Time Trends and Seasonal Patterns of Health-Related Quality of Life among US
Adults. Public Health Rep. 2009, 124, 692–701.
Karatsoreos, I. N. Links between Circadian Rhythms and Psychiatric Disease. Front. Behav. Neurosci.
2014, 8, 1–5, doi:10.3389/fnbeh.2014.00162.
Keatinge, W. R.; Donaldson, G. C.; Cordioli, E.; Martinelli, M.; Kunst, a E.; Mackenbach, J. P.; Nayha, S.;
Vuori, I. Heat related mortality in warm and cold regions of Europe: observational study. BMJ 2000, 321,
670–673, doi:10.1136/bmj.321.7262.670.
Kendrovski, V.; Baccini, M.; Martinez, G. S.; Paunovic, E.; Menne, B. Quantifying Projected Heat
Mortality Impacts under 21st-Century Warming Conditions for Selected European Countries. 2017, 14,
doi:10.3390/ijerph14070729.
Ketterer, C.; Matzarakis, A. Comparison of different methods for the assessment of the urban heat island in
Stuttgart, Germany. Int. J. Biometeorol. 2015, 59, 1299–1309, doi:10.1007/s00484-014-0940-3.
Kiefer, F. W. The significance of beige and brown fat in humans. Endocr. Connect. 2017, 6, R70–R79,
doi:10.1530/EC-17-0037.
Kim, S.; Sinclair, V. A.; Räisänen, J.; Ruuhela, R. Heat waves in Finland: Present and projected
summertime extreme temperatures and their associated circulation patterns. Int. J. Climatol. 2017,
doi:10.1002/joc.5253.
Kim, Y.; Kim, H.; Honda, Y.; Guo, Y. L.; Chen, B.; Woo, J.; Ebi, K. L. Suicide and ambient temperature in
East Asian countries : A time-stratified. Environ. Health Perspect. 2016, 124, 75–80,
doi:10.1289/ehp.1409392.
58
Kollanus, V.; Lanki, T. 2000-luvun pitkittyneiden helleaaltojen kuolleisuusvaikutukset Suomessa.
(Mortality effects of prolonged heat waves in the 2000s in Finland. Summary in English). Duodecim 2014
130, 983-90. duo11638 (011.638).
Koppe, C.; Jendritzky, G. Inclusion of short-term adaptation to thermal stresses in a heat load warning
procedure. Meteorol. Zeitschrift 2005, 14, 271–278, doi:10.1127/0941-2948/2005/0030.
Koppe, C.; Zacharias, S.; Bernhard, D. Repräsentativbefragung zur Wetterfühligkeit in Deutschland. Dtsch.
Wetterd. 2013. Report No UBA-FB 00. (Representative survey on weather sensitivity in Germany. German
Meteorological Service)
Kovats, R. S.; Hajat, S. Heat Stress and Public Health : A Critical Review. Annu. Rev. Public Heal. 2008,
29, 41–55, doi:10.1146/annurev.publhealth.29.020907.090843.
Lambert, G. W.; Reid, C.; Kaye, D. M.; Jennings, G. L.; Esler, M. D. Effect of sunlight and season on
serotonin turnover in the brain. Lancet 2002, 360, 1840–1842, doi:10.1016/S0140-6736(02)11737-5.
Leonard, W. R.; Sorensen, M. V.; Galloway, V. A.; Spencer, G. J.; Mosher, M. J.; Osipova, L.; Spitsyn, V.
A. Climatic influences on basal metabolic rates among circumpolar populations. Am. J. Hum. Biol. 2002, 14,
609–620, doi:10.1002/ajhb.10072.
Lian, H.; Ruan, Y.; Liang, R.; Liu, X.; Fan, Z. Short-term effect of ambient temperature and the risk of
stroke: A systematic review and meta-analysis. Int. J. Environ. Res. Public Health 2015, 12, 9068–9088,
doi:10.3390/ijerph120809068.
Liberman, A. R.; Halitjaha, L.; Ay, A.; Ingram, K. K. Modeling Strengthens Molecular Link between
Circadian Polymorphisms and Major Mood Disorders. J. Biol. Rhythms 2018, 33, 318–336,
doi:10.1177/0748730418764540.
Magnusson, A. An overview of epidemiological studies on seasonal affective disorder. Acta Psychiatr.
Scand. 2000, 101, 176–184, doi:10.1034/j.1600-0447.2000.101003176.x.
Matzarakis, A.; Mayer, H.; Iziomon, M. G. Application of a universal thermal index: physiological
equivalent temperature. Int. J. Biometeorol. 1999, 43, 76–84, doi:10.1007/s004840050119.
Matzarakis, A.; Rutz, F.; Mayer, H. Modelling radiation fluxes in simple and complex environments -
applications of the RayMan model. Int. J. Biometeorol. 2007, 51, 323–334, doi:10.1007/s00484-006-0061-8.
Matzarakis, A.; Rutz, F.; Mayer, H. Modelling radiation fluxes in simple and complex environments: basics
of the RayMan model. Int. J. Biometeorol. 2010, 54, 131–139, doi:10.1007/s00484-009-0261-0.
Medina-Ramón, M.; Schwartz, J. Temperature, temperature extremes, and mortality: A study of
acclimatisation and effect modification in 50 US cities. Occup. Environ. Med. 2007, 64, 827–833,
doi:10.1136/oem.2007.033175.
59
Mora, C.; Dousset, B.; Caldwell, I. R.; Powell, F. E.; Geronimo, R. C.; Bielecki, C. R.; Counsell, C. W. W.;
Dietrich, B. S.; Johnston, E. T.; Louis, L. V; Lucas, M. P.; McKenzie, M. M.; Shea, A. G.; Tseng, H.;
Giambelluca, T. W.; Leon, L. R.; Hawkins, E.; Trauernicht, C. Global risk of deadly heat. Nat. Clim. Chang.
2017, 7, 501–506, doi:10.1038/nclimate3322.
Morabito, M.; Crisci, A.; Messeri, A.; Capecchi, V.; Modesti, P. A.; Gensini, G. F.; Orlandini, S.
Environmental temperature and thermal indices: What is the most effective predictor of heat-related
mortality in different geographical contexts? Sci. World J. 2014, 2014, ID 961750,
doi:10.1155/2014/961750.
Muthers, S.; Matzarakis, A.; Koch, E. Climate change and mortality in vienna-A human biometeorological
analysis based on regional climate modeling. Int. J. Environ. Res. Public Health 2010, 7, 2965–2977,
doi:10.3390/ijerph7072965.
Mäkinen, T. M.; Raatikka, V. P.; Rytkönen, M.; Jokelainen, J.; Rintamäki, H.; Ruuhela, R.; Näyhä, S.;
Hassi, J. Factors affecting outdoor exposure in winter: Population-based study. Int. J. Biometeorol. 2006, 51,
27–36, doi:10.1007/s00484-006-0040-0.
Näyhä, S. Environmental temperature and mortality. Int. J. Circumpolar Health 2005, 64, 451–458,
doi:10.3402/ijch.v64i5.18026.
Näyhä, S. Heat mortality in Finland in the 2000s. Int. J. Circumpolar Health 2007, 66, 418–424,
doi:10.3402/ijch.v66i5.18313.
Näyhä, S.; Hassi, J.; Jousilahti, P.; Laatikainen, T.; Ikäheimo, T. M. Cold-related symptoms among the
healthy and sick of the general population: National FINRISK Study data, 2002. Public Health 2011, 125,
380–388, doi:10.1016/j.puhe.2011.02.014.
Näyhä, S.; Rintamäki, H.; Donaldson, G.; Hassi, J.; Jousilahti, P.; Laatikainen, T.; Jaakkola, J. J. K.;
Ikäheimo, T. M. Heat-related thermal sensation, comfort and symptoms in a northern population: The
National FINRISK 2007 study. Eur. J. Public Health 2014, 24, 620–626, doi:10.1093/eurpub/ckt159.
Näyhä, S.; Rintamäki, H.; Donaldson, G.; Hassi, J.; Jousilahti, P.; Laatikainen, T.; Jaakkola, J. J. K.;
Ikäheimo, T. M. The prevalence of heat-related cardiorespiratory symptoms: the vulnerable groups
identified from the National FINRISK 2007 Study. Int. J. Biometeorol. 2017, 61, 657–668,
doi:10.1007/s00484-016-1243-7.
OSF, 2015a, Official Statistics of Finland (SVT, Suomen virallinen tilasto): Causes of death [e-publication].
ISSN=1799-5078. 2015, 1. Causes of death in 2015. Helsinki: Statistics Finland [referred: 17.3.2018].
http://www.stat.fi/til/ksyyt/2015/ksyyt_2015_2016-12-30_kat_001_en.html
60
OSF 2015b, Official Statistics of Finland (SVT, Suomen virallinen tilasto): Deaths [e-publication].
ISSN=1798-2545. Annual Review 2015. Helsinki: Statistics Finland [referred: 18.3.2018].
http://www.stat.fi/til/kuol/2015/01/kuol_2015_01_2016-10-28_tie_001_en.html
O’Hare, C.; O’Sullivan, V.; Flood, S.; Kenny, R. A. Seasonal and meteorological associations with
depressive symptoms in older adults: A geo-epidemiological study. J. Affect. Disord. 2016, 191, 172–179,
doi:10.1016/j.jad.2015.11.029.
Oudin Åström, D.; Forsberg, B.; Rocklöv, J. Heat wave impact on morbidity and mortality in the elderly
population : A review of recent studies. Maturitas 2011, 69, 99–105, doi:10.1016/j.maturitas.2011.03.008.
Oudin Åström, D.; Forsberg, B.; Edvinsson, S.; Rocklöv, J. Acute fatal effects of short-lasting extreme
temperatures in Stockholm, Sweden: Evidence across a century of change. Epidemiology 2013, 24, 820–
829, doi:10.1097/01.ede.0000434530.62353.0b.
Oudin Åström, D.; Tornevi, A.; Ebi, K. L.; Rocklöv, J.; Forsberg, B. Evolution of minimum mortality
temperature in Stockholm, Sweden, 1901-2009. Environ. Health Perspect. 2016, 124, 740–744,
doi:10.1289/ehp.1509692.
Parsons, K. C. Human Thermal Environments; Second edit.; Taylor & Francis: London, 2003; ISBN 0-203-
34618-1.
Partonen, T. Hypothesis: Cryptochromes and brown fat are essential for adaptation and affect mood and
mood-related behaviors. Front. Neurol. 2012, NOV, 1–9, doi:10.3389/fneur.2012.00157.
Partonen, T.; Haukka, J.; Nevanlinna, H.; Lönnqvist, J. Analysis of the seasonal pattern in suicide. J. Affect.
Disord. 2004, 81, 133–139, doi:10.1016/S0165-0327(03)00137-X.
Petridou, E.; Papadopoulos, F. C.; Frangakis, C. E.; Skalkidou, A.; Trichopoulos, D. A role of sunshine in
the triggering of suicide. Epidemiology 2002, 13, 106–109, doi:10.1097/00001648-200201000-00017.
Preußner, M.; Heyd, F. Post-transcriptional control of the mammalian circadian clock: implications for
health and disease. Pflugers Arch. Eur. J. Physiol. 2016, 468, 983–991, doi:10.1007/s00424-016-1820-y.
Rakers, F.; Walther, M.; Schiffner, R.; Rupprecht, S.; Rasche, M.; Kockler, M.; Witte, O. W.; Schlattmann,
P.; Schwab, M. Weather as a risk factor for epileptic seizures: A case-crossover study. Epilepsia 2017, 58,
1287–1295, doi:10.1111/epi.13776.
Robine, J. M.; Cheung, S. L. K.; Le Roy, S.; Van Oyen, H.; Griffiths, C.; Michel, J. P.; Herrmann, F. R.
Death toll exceeded 70,000 in Europe during the summer of 2003. Comptes Rendus - Biol. 2008, 331, 171–
178, doi:10.1016/j.crvi.2007.12.001.
Rocklöv, J.; Forsberg, B. The effect of temperature on mortality in Stockholm 1998--2003: a study of lag
structures and heatwave effects. Scand J Public Heal. 2008, 36, 516–523, doi:10.1177/1403494807088458.
61
Rupp, R. F.; Vásquez, N. G.; Lamberts, R. A review of human thermal comfort in the built environment.
Energy Build. 2015, 105, 178–205, doi:10.1016/j.enbuild.2015.07.047.
Ruosteenoja, K.; Jylhä, K.; Kämäräinen, M. Climate projections for Finland under the RCP forcing
scenarios. Geophysica 2016, 51, 17–50.
Rush, L.; McCartney, G.; Walsh, D.; Mackay, D. Vitamin D and subsequent all-age and premature
mortality: a systematic review. BMC Public Health 2013, 13, 679, doi:10.1186/1471-2458-13-679.
Räsänen, P.; Hakko, H.; Jokelainen, J.; Tiihonen, J. Seasonal variation in specific methods of suicide: A
national register study of 20 234 Finnish people. J. Affect. Disord. 2002, 71, 51–59, doi:10.1016/S0165-
0327(01)00411-6.
Sarran, C.; Albers, C.; Sachon, P.; Meesters, Y. Meteorological analysis of symptom data for people with
seasonal affective disorder. Psychiatry Res. 2017, 257, 501–505, doi:10.1016/j.psychres.2017.08.019.
Shaposhnikov, D.; Revich, B.; Bellander, T.; Bedada, G. B.; Bottai, M.; Kharkova, T.; Kvasha, E.; Lezina,
E.; Lind, T.; Semutnikova, E.; Pershagen, G. Mortality Related to Air Pollution with the Moscow Heat
Wave and Wildfire of 2010. Epidemiology 2014, 25, 359–364, doi:10.1097/EDE.0000000000000090.
Smith, K. R.; Woodward, A.; Campbell-Lendrum, D.; Chadee, D.; Honda, Y.; Liu, Q.; Olwoch, J.; Revich,
B.; Sauerborn, R.; Dokken, D.; Mach, K.; Bilir, T.; Chatterjee, M.; Ebi, K.; Estrada, Y.; Genova, R. Human
Health: Impacts, Adaptation, and Co-Benefits. Clim. Chang. 2014 Impacts, Adapt. Vulnerability. Part A
Glob. Sect. Asp. Contrib. Work. Gr. II to Fifth Assess. Rep. Intergov. Panel Clim. Chang. [f. CB, Barros VR,
Dokken DJ, Mach KJ, Ma 2014, 709–754, doi:10.1017/CBO9781107415379.016.
Staddon, P. L.; Montgomery, H. E.; Depledge, M. H. Climate warming will not decrease winter mortality.
Nat. Clim. Chang. 2014, 4, 190–194, doi:10.1038/nclimate2121.
Taleghani, M.; Tenpierik, M.; Kurvers, S.; Van Den Dobbelsteen, A. A review into thermal comfort in
buildings. Renew. Sustain. Energy Rev. 2013, 26, 201–215, doi:10.1016/j.rser.2013.05.050.
Taylor, J.; Wilkinson, P.; Davies, M.; Armstrong, B.; Chalabi, Z.; Mavrogianni, A.; Symonds, P.;
Oikonomou, E.; Bohnenstengel, S. I. Mapping the effects of urban heat island, housing, and age on excess
heat-related mortality in London. Urban Clim. 2015, 14, 517–528, doi:10.1016/j.uclim.2015.08.001.
THL’s morbidity index 2012-2014. Statistical review 12/2016. http://urn.fi/URN:NBN:fi-fe2016111829194.
THL’s Morbidity Index, 2016. Available online: https://www.sotkanet.fi/sotkanet/en (accessed on 9 January
2018).
Thompson, R.; Hornigold, R.; Page, L.; Waite, T. Associations between high ambient temperatures and heat
waves with mental health outcomes: a systematic review. Public Health 2018, 161, 171–191,
doi:10.1016/j.puhe.2018.06.008.
62
Tobias, A.; Armstrong, B.; Gasparrini, A. Investigating Uncertainty in the Minimum Mortality Temperature.
Methods and Application to 52 Spanish Cities. Epidemiology 2017, 28, 72–76,
doi:10.1097/EDE.0000000000000567.
Trombley, J.; Chalupka, S.; Anderko, L. Climate Change and Mental Health. Am. J. Nurs. 2017, 117, 44–52,
doi:10.1097/01.NAJ.0000515232.51795.fa.
Von Mackensen, S.; Hoeppe, P.; Maarouf, A.; Tourigny, P.; Nowak, D. Prevalence of weather sensitivity in
Germany and Canada. Int. J. Biometeorol. 2005, 49, 156–166, doi:10.1007/s00484-004-0226-2.
Virtanen, J. K.; Nurmi, T.; Voutilainen, S.; Mursu, J.; Tuomainen, T. P. Association of serum 25-
hydroxyvitamin D with the risk of death in a general older population in Finland. Eur. J. Nutr. 2011, 50,
305–312, doi:10.1007/s00394-010-0138-3.
WHO, 2017a. Depression and Other Common Mental Disorders: Global Health Estimates. Geneva: World
Health Organization; 2017. Licence: CC BY-NC-SA 3.0 IGO
WHO, 2017b http://www.who.int/mental_health/prevention/suicide/suicideprevent/en/ (accessed
17.10.2017)
Williams, M. N.; Hill, S. R.; Spicer, J. Will climate change increase or decrease suicide rates? The differing
effects of geographical, seasonal, and irregular variation in temperature on suicide incidence. Clim. Change
2015, 130, 519–528, doi:10.10/s 0710584-015-1371-9.
Woo, J. M.; Okusaga, O.; Postolache, T. T. Seasonality of suicidal behavior. Int. J. Environ. Res. Public
Health 2012, 9, 531–547, doi:10.3390/ijerph9020531.
Wood, S. Package ‘mgcv’. 2016. Available online: https://cran.r-project.org/web/packages/mgcv/mgcv.pdf.
(accessed on 3 March 2016).
Vyssoki, B.; Kapusta, N. D.; Praschak-Rieder, N.; Dorffner, G.; Willeit, M. Direct Effect of Sunshine on
Suicide. JAMA Psychiatry 2014, 71, 1231, doi:10.1001/jamapsychiatry.2014.1198.
Ye, X.; Wolff, R.; Yu, W.; Vaneckova, P.; Pan, X.; Tong, S. Ambient Temperature and Morbidity: A
Review of Epidemiological Evidence. Environ. Health Perspect. 2011, 120, 19–28,
doi:10.1289/ehp.1003198.
Yoneshiro, T.; Matsushita, M.; Nakae, S.; Kameya, T.; Sugie, H.; Tanaka, S.; Saito, M. Brown adipose
tissue is involved in the seasonal variation of cold-induced thermogenesis in humans. Am. J. Physiol. -
Regul. Integr. Comp. Physiol. 2016, ajpregu.00057.2015, doi:10.1152/ajpregu.00057.2015.
Yu, W.; Mengersen, K.; Wang, X.; Ye, X.; Guo, Y.; Pan, X.; Tong, S. Daily average temperature and
mortality among the elderly: A meta-analysis and systematic review of epidemiological evidence. Int. J.
Biometeorol. 2012, 56, 569–581, doi:10.1007/s00484-011-0497-3.
Reprinted under the Creative Commons Attribution License
I
International Journal of
Environmental Research
and Public Health
Article
Biometeorological Assessment of Mortality Related toExtreme Temperatures in Helsinki Region,Finland, 1972–2014
Reija Ruuhela 1,*, Kirsti Jylhä 1, Timo Lanki 2, Pekka Tiittanen 2 and Andreas Matzarakis 3 ID
1 Finnish Meteorological Institute, P.O. Box 503, FI-00101 Helsinki, Finland; [email protected] National Institute for Health and Welfare, P.O. Box 95, FI-70701 Kuopio, Finland; [email protected] (T.L.);
[email protected] (P.T.)3 Research Center Human Biometeorology, German Meteorological Service, Stefan-Meier-Str. 4,
D-79104 Freiburg, Germany; [email protected]* Correspondence: [email protected]; Tel.: +358-500-424533
Received: 30 June 2017; Accepted: 18 August 2017; Published: 22 August 2017
Abstract: Climate change is expected to increase heat-related and decrease cold-related mortality.The extent of acclimatization of the population to gradually-changing thermal conditions is not wellunderstood. We aimed to define the relationship between mortality and temperature extremes indifferent age groups in the Helsinki-Uusimaa hospital district in Southern Finland, and changes insensitivity of the population to temperature extremes over the period of 1972–2014. Time series ofmortality were made stationary with a method that utilizes 365-day Gaussian smoothing, removestrends and seasonality, and gives relative mortality as the result. We used generalized additivemodels to examine the association of relative mortality to physiologically equivalent temperature(PET) and to air temperature in the 43-year study period and in two 21-year long sub-periods(1972–1992 and 1994–2014). We calculated the mean values of relative mortality in percentile-basedcategories of thermal indices. Relative mortality increases more in the hot than in the cold tail of thethermal distribution. The increase is strongest among those aged 75 years and older, but is somewhatelevated even among those younger than 65 years. Above the 99th percentile of the PET distribution,the all-aged relative mortality decreased in time from 18.3 to 8.6%. Among those ≥75 years old,the decrease in relative mortality between the sub-periods were found to be above the 90th percentile.The dependence of relative mortality on cold extremes was negligible, except among those ≥75 yearsold, in the latter period. Thus, heat-related mortality is also remarkable in Finland, but the sensitivityto heat stress has decreased over the decades.
Keywords: mortality; thermal comfort index; heat stress; cold stress; climate; acclimatization
1. Introduction
Temperature dependence of mortality is often described as U-, V-, or J-shaped with increasingmortality towards both hot and cold temperature extremes, e.g., [1–3]. The shape of themortality-temperature relationship and the temperature range of minimum mortality vary by latitudeand climatic zone [4,5]. The optimal, minimum mortality temperature is lower in cold than in warmclimates due to acclimatization of the population to their typical climatic conditions. In Finland,Northern Europe, the mortality is lowest when the daily mean temperature is in the range of 12–17 ◦C,while in Mediterranean countries the same is true at 22–25 ◦C [6–8].
Well-known past extreme events include the 2003 heat wave in Central and Western Europe,causing about 70,000 additional deaths [9], and the 2010 heat wave in Russia, leading to about 55,000additional deaths [10]. These heat waves also extended to Finland, which has about 5.5 million
Int. J. Environ. Res. Public Health 2017, 14, 944; doi:10.3390/ijerph14080944 www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2017, 14, 944 2 of 19
inhabitants. According to Kollanus and Lanki [11], the number of non-accidental extra deaths wasover 200 in 2003 and more than 300 in 2010. The impact of these two heat waves in Finland was mostsevere among the elderly, aged 75 years and older, as the increase in daily mortality was, on average,21%. The most severe heat wave that has been studied in Finland took place in 1972 and caused about800 extra deaths. After that, remarkable heat waves with increased mortality also occurred e.g., in 1973,1978, 1988, 1995, and 1997 [7].
Most of the temperature-related mortality burden is contributable to cold in different climaticzones—even in tropical and sub-tropical areas [12]. The impacts of cold thermal conditions on mortalityare more complex than the impacts of hot conditions due to different causal pathways. When theimpacts of hot weather tend to appear with short lags, on the same day or within a couple of days, theincrease in cold-related mortality can be found with some delay varying from days up to weeks [13–15].On the other hand, Ebi and Mills [16] questioned the assumption that temperature is the reason forthe strong seasonality with higher mortality in winter, particularly in case of cardiovascular, but alsorespiratory disease mortality.
Sensitivity to heat or cold stress varies according to many health and societal aspects, such as age,ill health, personal lifestyles, poverty, infrastructure and buildings, access to health care, and socialstructures [17,18]. Especially elderly and people with pre-existing medical conditions, such ascardio-vascular or respiratory diseases, diabetes, or chronic mental illnesses, are found to be vulnerableto temperature extremes [2,11,19,20]. Changes in sensitivity and adaptive capacity through, e.g.,improved population health and the provision of health services may also change temperature-relatedmortality [2,18].
Short-term acclimatization to seasonal temperature variations takes place within a couple ofweeks and should also be considered [21]. In the United States of America (USA), Lee et al. [22] foundthat heat effects were larger in the spring and early summer, and cold effects were larger in late fall.In addition, heat effects were larger in regions where high temperatures were less common, and viceversa for cold effects.
Long-term adaptation of the populations to their normal local climatic conditions takes placeover decades through physiological acclimatization processes or through behavioural adaptation, suchas housing or clothing suitable for the climate. According to Oudin Åström et al. [23] the minimummortality temperature increased in Stockholm over the course of the 20th century, suggesting thatautonomous adaptation took place in the Swedish population—in climatic conditions rather similar tothose in Southern Finland. In another study, Oudin Åström et al. [24] found that the relative risk ofmortality from cold and heat extremes remained stable over the period of 1980–2009.
Climate change is projected to increase heat-related mortality, especially due to more intense heatwaves and a decrease in cold-related mortality due to fewer cold extremes [25]. Thus, global warmingis expected to gradually alter the seasonal mortality cycle as well. Ballester et al. [3] assessed that inEurope, the rise in heat-related mortality would start to compensate for the reduction of deaths fromcold during the second half of the century.
This long-term adaptation causes substantial uncertainty in projected changes in mortality andshould be included in the assessments of climate change impacts on mortality. According to areview by Huang et al. [26], only half of the studies on projecting future heat-related mortalityincluded acclimatization. For instance, Muthers et al. [27] concluded that heat-related mortalitywould increase significantly in Vienna until the end of the 21st century, also in an approach wherelong-term adaptation was included. On the other hand, Ballester et al. [3] suggested that if societieseffectively adapt to a warmer climate, at the end of the century the total mortality due to thermalstress might decrease in Europe, when increases in heat-related mortality, decreases in cold-relatedmortality, and acclimatization are taken into account. Zacharias et al. [28] concluded that by theend of the 21st century, excess deaths due to ischemic heart diseases in Germany attributable toheat waves is expected to rise by factor 2.4 and 5.1 in the acclimatization and non-acclimatizationapproach, respectively.
Int. J. Environ. Res. Public Health 2017, 14, 944 3 of 19
Heat and cold stress depend on the energy exchange experienced by humans and are closelyrelated to human thermoregulatory mechanisms and the human circulatory system [29–32]. Thermalindices based on the human energy balance require several meteorological input parameters:air temperature, air humidity, wind speed, and the short- and long-wave radiation fluxes, the totalimpact of the latter two being summarized as the mean radiant temperature [30,32].
A widely-used human biometeorological thermal index, the physiologically equivalenttemperature (PET), is based on the human energy balance. It describes the temperature at a givenplace (outdoors or indoors) equivalent to the air temperature in a typical indoor setting with core andskin temperatures equal to those under the conditions being assessed. Units of PET, degrees Celsius,make the results comprehensible [32]. The standard PET values are valid for the assumed values ofactivity (80 W) and thermal resistance of the clothing (0.9) [31,33]. Moreover, the meteorological inputparameters are to be measured at, or transferred to, the average height of a standing person’s gravitycentre, 1.1 m above the ground [28,34].
Thermal indices such as PET describe the thermal environment of a person more realisticallythan a single parameter, such as outdoor air temperature [35]. However, it should be kept in mindthat the meteorological input parameters for thermal indices often show remarkable temporal andspatial variability, and are therefore sources of large uncertainties in the indices. Wind speed andmean radiant temperature show the highest variability and are modified strongly by surroundings andobstacles, especially in complex urban areas [36]. Therefore, it is not self-evident that the use of energybalance-based rational thermal indices would always be the best choice for studies if the health impactdata does not include precise spatial information. Quite often in studies on heat- and cold-relatedmortality, only outdoor air temperature is used as an explanatory variable.
The main aims of this study were (1) to quantify the increase in mortality related to both hot andcold temperature extremes according to different age groups in the Helsinki-Uusimaa hospital districtin Southern Finland; (2) to study potential changes in time of the temperature-mortality relationshipsover the period of 1972–2014; and (3) to examine whether outdoor air temperature and PET givesimilar temperature-mortality relationships. Finally, we discussed the results from the viewpoint ofpotential acclimatization and changes in sensitivity of the Finnish population to temperature extremes.
2. Materials and Methods
2.1. Mortality Data in the Study Area
Daily numbers of all-cause deaths and annual population in the Helsinki-Uusimaa hospitaldistrict in Finland from 1971 to 2015 were obtained from Statistics Finland (Helsinki, Finland). The dataincluded the number of all-aged deaths and the number of deaths in two age groups: 65–74 yearsand 75 years and older. From these data, we calculated the daily number of deaths in two additionalage groups: younger than 65 years, and 65 years and older. Total number of deaths in the studyperiod was 465,553, out of which 235,963 were 75 years and older, 98,967 were 65–74 years, and 130,623were younger than 65 years. We interpolated daily population values from the annual population,and used them to calculate the daily mortality (1/100,000). Linear time trends for mortality with a 95%confidence level were determined in each age group.
The population in this hospital district (area 9.097 km2) increased from about 1.0 million to about1.6 million during the study period. The age distribution during the study period changed remarkablyas the share of elderly (≥65 years) increased from 9.1 to 16.5%. The life-expectancy in Finland over thestudy period has increased markedly. In 1980, the median age at death was 68.2 years for men and75.4 years for women. In 2015, the corresponding values were 76.8 and 85.3 years, respectively [37].
2.2. Expected and Relative Mortality
In order to study the impacts of temperature extremes on mortality and changes in theseimpacts over the decades, we applied the following method developed by Koppe and Jendritzky [21].
Int. J. Environ. Res. Public Health 2017, 14, 944 4 of 19
As a baseline mortality in each age group, we used expected mortality that was calculated for eachday of the study period from the original mortality data using Gaussian smoothing with a filter of365 days. The time series of expected mortality includes the seasonal cycle, but day-to-day variabilityis smoothed out (Figure 1, middle). Relative mortality is then the deviation of mortality from theexpected mortality as a percentage. The time series of relative mortality exhibit day-to-day variability,whereas the seasonal cycle and long-term trends are removed (Figure 1, bottom).
The use of these stationary time series of relative mortality makes it possible to analyse impactsof temperature extremes over the decades regardless of the changes in population. The smoothingmethod shortens the time series of the mortality data from both ends. In the following, in comparisonsof relative mortality to the meteorological data, we use full calendar years, thus 1972–2014.
Int. J. Environ. Res. Public Health 2017, 14, 944 4 of 18
smoothed out (Figure 1, middle). Relative mortality is then the deviation of mortality from the expected mortality as a percentage. The time series of relative mortality exhibit day-to-day variability, whereas the seasonal cycle and long-term trends are removed (Figure 1, bottom).
The use of these stationary time series of relative mortality makes it possible to analyse impacts of temperature extremes over the decades regardless of the changes in population. The smoothing method shortens the time series of the mortality data from both ends. In the following, in comparisons of relative mortality to the meteorological data, we use full calendar years, thus 1972–2014.
Figure 1. Time series of mortality (1/100,000) (upper panel), expected mortality (1/100,000) (middle), and relative mortality (%) (lower panel) in Helsinki-Uusimaa hospital district, 1972–2014. All-cause deaths and all age groups.
2.3. Meteorological Data
Synoptic (three-hourly) temperature, relative humidity, wind speed, and global radiation data from the Helsinki-Vantaa weather station were used as input to calculate PET indices for the study period. We made the calculations with the RayMan model [38,39]. The daily mean values of PET were calculated from the three-hourly data to describe the daily average thermal conditions. While the emphasis of this paper is given to daily means of PET, we also calculated PET at 12 UTC (Universal Time Coordinated, at 15/14 local time during summer/winter) to describe the maximum daytime heat load and PET at 06 UTC (at 9/8 local time during summer/winter) to describe thermal conditions in the active morning hours.
In addition to the station-wise PET indices, we used spatial averages of daily mean air temperatures (Tavg) over the Helsinki-Uusimaa hospital district as an explanatory variable for relative mortality. Values of Tavg were derived from the gridded 10 × 10 km dataset of the Finnish Meteorological Institute [40]. Spatial averages of daily mean maximum and minimum temperatures were calculated likewise.
The use of two indices, PET, based on station data, and Tavg, based on gridded data, was motivated by the fact that weather conditions may vary remarkably within the Helsinki and Uusimaa province, especially depending on the distance from the Gulf of Finland and the wind direction, and the station-wise PET value may not describe thermal conditions in the hospital district. On the other hand, if both indices produce similar assessments of temperature-related mortality in our study area,
Figure 1. Time series of mortality (1/100,000) (upper panel), expected mortality (1/100,000) (middle),and relative mortality (%) (lower panel) in Helsinki-Uusimaa hospital district, 1972–2014. All-causedeaths and all age groups.
2.3. Meteorological Data
Synoptic (three-hourly) temperature, relative humidity, wind speed, and global radiation datafrom the Helsinki-Vantaa weather station were used as input to calculate PET indices for the studyperiod. We made the calculations with the RayMan model [38,39]. The daily mean values of PET werecalculated from the three-hourly data to describe the daily average thermal conditions. While theemphasis of this paper is given to daily means of PET, we also calculated PET at 12 UTC (UniversalTime Coordinated, at 15/14 local time during summer/winter) to describe the maximum daytime heatload and PET at 06 UTC (at 9/8 local time during summer/winter) to describe thermal conditions inthe active morning hours.
In addition to the station-wise PET indices, we used spatial averages of daily mean airtemperatures (Tavg) over the Helsinki-Uusimaa hospital district as an explanatory variable forrelative mortality. Values of Tavg were derived from the gridded 10 × 10 km dataset of the FinnishMeteorological Institute [40]. Spatial averages of daily mean maximum and minimum temperatureswere calculated likewise.
Int. J. Environ. Res. Public Health 2017, 14, 944 5 of 19
The use of two indices, PET, based on station data, and Tavg, based on gridded data, was motivatedby the fact that weather conditions may vary remarkably within the Helsinki and Uusimaa province,especially depending on the distance from the Gulf of Finland and the wind direction, and thestation-wise PET value may not describe thermal conditions in the hospital district. On the otherhand, if both indices produce similar assessments of temperature-related mortality in our study area,confidence in the results is increased. Linear time trends with 95% confidence level were determinedfrom the time series of the PET and Tavg.
2.4. Assessing the Relationship between Relative Mortality and Thermal Indices
The impacts of the thermal conditions on the relative mortality in different age groups werestudied by comparing the daily mean values of PET and Tavg with the mean value of relativemortality in the same and following day, thus applying a one-day lag. We used a generalized additivemodel (GAM) to describe the relationships between the relative mortality and the thermal indices.The exposure-response curves with 95% confidence intervals were fitted to the data by using penalizedthin plate regression splines in the R package “mgcv” [41]. In addition to the whole study period,we conducted the analysis for two 21 year-long sub-periods, 1972–1992 and 1994–2014, in order tovisualize potential changes in these relationships.
Another approach was used for quantitative analysis to assess relationships between relativemortality and the thermal indices: From the frequency distributions of daily values of PET and Tavg,we determined the following percentiles: 1%, 2.5%, 5%, 10%, 25%, 50%, 75%, 90%, 95%, 97.5%, and99% for the whole study period. These percentiles were used to classify the thermal indices into12 categories. Then the mean values of relative mortality in these percentile categories were calculatedwith 95% confidence intervals for various age groups: all, <65, 65–74, ≥75, and ≥65 years.
In order to assess whether the relationships between relative mortality and the thermal indiceshave changed over the study period, we applied linear regression to explore linear time trends of themean relative mortality in each percentile category and each age group for the whole study period.Furthermore, we concretized the changes in relative mortality by calculating the mean values ofrelative mortality in the two 21-year long sub-periods: 1972–1992 and 1994–2014 using the percentilecategories that were defined from the whole study period. We tested the statistical significance ofdifferences in the relative mortality between the sub-period percentile categories by Welch Two Samplet-test. The Shapiro-Wilk test was used to check the sub-period specific normality of the distributions ofrelative mortality.
The dependence between mortality and thermal conditions varies depending on the time window,thus, exposure time and lag, chosen for the study. The impacts of heat stress become apparent typicallyon a shorter time window than cold stress. In order to study impacts of prolonged heat waves orcold spells on mortality, we also calculated the relationships for longer time windows using 7- and14-day averages of relative mortality and thermal indices without lag. These time windows also partlyinclude delayed impacts and potential harvesting effects.
The statistical calculations were done with R version 3.2.3.
3. Results
3.1. Trends in the Thermal Indices and Mortality Data
Air temperature and PET have statistically significant increasing time trends (p-values < 0.001)indicating ongoing climate change. Regression coefficients for the linear time trends (with a 95%confidence level) in the period of 1972–2014 for PET daily mean value and daily mean temperaturein Helsinki-Vantaa were 0.37 ◦C/decade (0.22, 0.52) and 0.44 ◦C/decade (0.33, 0.55), respectively.Based on the gridded data, the trend for the daily mean temperature as an average over the hospitaldistrict was 0.36 ◦C/decade (0.24, 0.47).
Int. J. Environ. Res. Public Health 2017, 14, 944 6 of 19
Lengthening lifetime can be seen as statistically significant (p-values < 0.001) decreasing mortalitytrends in all age groups in the Helsinki-Uusimaa hospital district. The all-aged mortality (1/100,000)was 2.2 (±0.5) on average during the study period, with a decreasing trend of −0.116/decade (−0.122,−0.110). The mortality among 75 years and older during the study period was on average 24 ± 8and the decreasing trend in mortality was clear, −2.1/decade (−2.2, −2.0). In the relative mortality,there are no time trends, because the methodology to calculate relative mortality removes trends fromthe time series, as explained in the Methods section.
3.2. Relationships between Relative Mortality and Thermal Indices
In the following, we first visually inspect how daily values of relative mortality, i.e., the deviationsfrom expected mortality values, are related to two thermal indices, PET and the spatially-averagedtemperature (Tavg). Means, ranges, and temporal trends of relative mortality values are then studiedas a function of percentiles of the indices.
The dependencies of the two-day mean of relative mortality on the daily mean value of PET inthe whole study period (1972–2014) and in the two 21-year sub-periods (1972–1992 and 1994–2014)according to different age groups are presented in Figure 2. The modelled relative mortality-PETassociation curves show a clear increase from the expected mortality in all age groups at the PETmean value exceeding 20 ◦C. The slope is steepest in the age group ≥75 years. The increase in relativemortality is less in the cold than in the hot extreme conditions. In the cold thermal range, a slightincrease in relative mortality with decreasing PET is seen only in the age group ≥75 years at the PETvalues below −30 ◦C. Some changes in time in the modelled relative mortality-PET association curvescan be seen in both cold and hot extremes. In the latter sub-period, the increase in relative mortalitywith high values of PET is still apparent in all age groups, but smaller than previously. Moreover, in thelatter sub-period, the dependence of relative mortality on PET in the cold thermal range is negligible.
The relationship between relative mortality and spatially-averaged daily mean temperature in thehospital district shows similar characteristics to the dependence of relative mortality on PET, especiallyin the hot extreme temperature range (Figure 3). However, in the cold temperature range, the Tavg
dependencies differ somewhat from the PET dependencies: especially in the age group 65–74 years,the relative mortality-Tavg relationship in the cold range shows increases and decreases that are notseen in the relative mortality-PET relationship.
Relative mortality mean values in different percentile categories of daily mean PET and Tavg
distributions are presented in Tables 1 and 2, respectively, for different age groups. In the data coveringthe whole study period of 1972–2014, in the hot extreme end of the PET and Tavg distributions,the relative mortality values are elevated, by more than about 4% in the percentiles above 95%.The highest increase in mortality, 18.6%, was found above the 99th percentile of the Tavg distributionamong the elderly, 75 years and older, but even among the younger population (<65 years) the increasein mortality was almost 10% (Table 2). The increase in relative mortality in the highest percentilecategory appears to be larger for Tavg than for PET.
The increase of the relative mortality at the cold extreme end of the distributions is smaller thanin the hot extreme, typically only a few percent in the time period of 1972–2014. The largest deviationfrom the expected mortality, 5.2%, was found in the first percentile category of the PET distribution(below −25.1 ◦C) among the elderly, 75 years and older (Table 1).
Although the original time series of relative mortality by definition do not have long-term trendsin time, this is not necessarily true when relative mortality values in different percentile categories ofthermal indices are considered. Statistically significant decreasing linear trends in the relative mortalitywere indeed found in the highest percentile categories, above the 99th percentile of PET and Tavg in allage groups, except among those 65–74 years old. In the age group ≥75 years, the decreasing trend isseen also in more common thermal conditions, in PET and Tavg percentile categories above 90%. In thecold thermal range, only two statistically significant decreasing trends were found.
Int. J. Environ. Res. Public Health 2017, 14, 944 7 of 19
The considerations of the relative mortality in the various percentile categories of PET and Tavg
show clear differences between the two 21-year sub-periods (Tables 1 and 2, last columns). Above the99th percentile of the PET distribution, a decrease of about 10%, from 18.3% in the period of 1972–1992to 8.6% in 1994–2014, was found in all-aged relative mortality (Table 1). In the temperature distribution,the decrease was even greater, about 12%, from 22.5 to 10.7%, respectively (Table 2). In the age group≥75 years, the decreases between the two sub-periods in the percentiles above the 90th were clear,although they did not reach the level of statistical significance in either case. As an example, the meanrelative mortality was equal in the PET percentile class of 90 to 95th of the former sub-period and theclass 97.5 to 99th of the latter sub-period. It is also noteworthy that the increase in relative mortalityalmost disappeared between the 90th to 97.5th percentile of both PET and Tavg distributions in thelatter sub-period (Tables 1 and 2). Among the group aged <65 years, the decrease in time in relativemortality between the sub-periods above the 99th percentile was 11–12%, from 15.7 to 4.0% for PET(Table 1), and from 16.3 to 5.6% for Tavg (Table 2).
In the cold thermal range, the changes in relative mortality between the two sub-periods weresmall and inconsistent (Tables 1 and 2). However, in the latter sub-period, the dependence of relativemortality on cold extremes almost disappeared, except among those aged 75 years and older. In general,the changes between the two sub-periods were more consistent in the PET percentiles than in the Tavg
percentiles, and more consistent in the warm thermal range than in the cold thermal range.In addition to the narrow time window of two days for mortality, considered above, we studied
the effect of the length of the time window on the outcomes by also making calculations using 7- and14-day averages of mortality and explaining the thermal indices, without lag. The dependencies ofrelative mortality on PET (Supplementary Material, Figures S1 and S2 and Tables S1 and S2) and Tavg
are also J-shaped in these time windows, indicating that the impact of heat stress on relative mortalityis more pronounced than the impact of cold stress. In the longer time windows, the differences betweenthe sub-periods become statistically significant in a larger number of percentile categories than in thetwo-day time window. Hence, the calculations in these time windows confirm our previous findingthat the tendency of hot extremes to increase mortality has somewhat weakened over the decades.Furthermore, the outcomes in the longer time windows also suggest a decrease in the impact of thecoldest extreme conditions on mortality but, interestingly, also a small increase in the more commonwintertime thermal range with PET values around −10 ◦C among the group aged 65–74 years.
In this paper, we report the dependencies of relative mortality on daily mean values of PET andTavg. However, we made similar calculations also using PET at 12 and 06 UTC and daily maximumand minimum temperatures. Numerically, the relationships are somewhat different (data not shown)compared to the calculations with daily mean values, but the generic outcomes—increases in mortalityin the highest percentiles of the thermal range, and decreasing impacts of heat and cold stress from thefirst to the later sub-period—are the same as when using daily mean values of thermal indices.
Int. J. Environ. Res. Public Health 2017, 14, 944 8 of 19Int. J. Environ. Res. Public Health 2017, 14, 944 8 of 18
1972–2014 1972–1992 1994–2014 Age Group
All
≥75 years
65–74 years
<65 years
Figure 2. Scatterplots of relative mortality in different age groups and the daily mean value of physiologically equivalent temperature (PET) in the Helsinki-Vantaa weather station in the period of 1972–2014 and in two 21-year time periods, 1972–1992 and 1994–2014. Generalized additive models (GAM) fitted to the data (solid curves) with 95% confidence intervals (dashed curves).
Figure 2. Scatterplots of relative mortality in different age groups and the daily mean value ofphysiologically equivalent temperature (PET) in the Helsinki-Vantaa weather station in the period of1972–2014 and in two 21-year time periods, 1972–1992 and 1994–2014. Generalized additive models(GAM) fitted to the data (solid curves) with 95% confidence intervals (dashed curves).
Int. J. Environ. Res. Public Health 2017, 14, 944 9 of 19Int. J. Environ. Res. Public Health 2017, 14, 944 9 of 18
1972–2014 1972–1992 1994–2014 Age Group
All
≥75 years
65–74 years
<65 years
Figure 3. Scatterplots of relative mortality in different age groups and the daily mean temperature (Tavg) in the Helsinki-Uusimaa hospital district in the period of 1972–2014 and in two 21-year time periods, 1972–1992 and 1994–2014. GAM fitted to the data (solid curves) with 95% confidence intervals (dashed curves).
Figure 3. Scatterplots of relative mortality in different age groups and the daily mean temperature(Tavg) in the Helsinki-Uusimaa hospital district in the period of 1972–2014 and in two 21-year timeperiods, 1972–1992 and 1994–2014. GAM fitted to the data (solid curves) with 95% confidence intervals(dashed curves).
Int. J. Environ. Res. Public Health 2017, 14, 944 10 of 19
Table 1. Mean relative mortality (95% CI) of different age groups in percentiles of daily mean values of physiologically equivalent temperature (PET)at theHelsinki-Vantaa airport, and their linear time trends in the period of 1972–2014 with statistical significance. Mean relative mortality in two 21-year sub-periods,1972–1992 and 1994–2014, and the statistical significance of the differences in relative mortality between the sub-periods.
Percentiles PET Range Relative Mortality 1972–2014 (%) Trend (%/10 years) t-Test Relative Mortality 1972–1992 (%) Relative Mortality 1994–2014 (%) t-Test
All0–1 −42.0, −25.1 2.1 (−0.2, 4.4) 0.3 (−1.6, 2.2) 2.5 (−0.6, 5.6) 1.7 (−1.7, 5.1)
1–2.5 −25.1, −20.9 1.9 (0.2, 3.6) −0.6 (−2.0, 0.9) 2.7 (0.3, 5.0) 0.7 (−2.0, 3.3)2.5–5 −20.9, −17.1 0.9 (−0.5, 2.4) −0.9 (−2.1, 0.2) 2.4 (0.4, 4.3) −0.4 (−2.5, 1.7)5–10 −17.1, −12.8 0.2 (−0.8, 1.2) 0.5 (−0.3, 1.3) −0.4 (−1.8, 1.0) 0.9 (−0.7, 2.4)10–25 −12.8, −6.9 0.4 (−0.2, 0.9) 0.1 (−0.3, 0.6) 0.3 (−0.5, 1.2) 0.4 (−0.4, 1.2)25–50 −6.9, 0.4 −1.5 (−1.9, −1.1) 0.1 (−0.2, 0.5) −1.5 (−2.2, −0.9) −1.6 (−2.2, −1.0)50–75 0.4, 11.3 −0.9 (−1.4, −0.5) 0.3 (−0.1, 0.6) −1.3 (−1.9, −0.7) −0.6 (−1.2, −0.1)75–90 11.3, 17.3 0.0 (−0.6, 0.6) −0.3 (−0.8, 0.2) 0.3 (−0.6, 1.2) −0.2 (−1.0, 0.6)90–95 17.3, 20.4 2.5 (1.5, 3.6) −0.4 (−1.2, 0.4) 3.1 (1.6, 4.7) 2.1 (0.8, 3.4)
95–97.5 20.4, 22.6 4.4 (2.9, 5.9) −1.0 (−2.2, 0.2) 5.2 (3.1, 7.4) 3.5 (1.4, 5.6)97.5–99 22.6, 24.5 7.2 (5.2, 9.3) −1.5 (−3.2, 0.1) 9.5 (6.3, 12.8) 5.9 (3.2, 8.5)99–100 24.5, 29.4 11.7 (8.9, 14.5) −3.9 (−6.0, −1.8) *** 18.3 (12.4, 24.3) 8.6 (5.6, 11.5) **
Aged ≥75 years0–1 −42.0, −25.1 5.2 (2.5, 7.9) −0.4 (−2.7, 1.8) 5.7 (2.0, 9.3) 4.7 (0.6, 8.7)
1–2.5 −25.1, −20.9 2.1 (−0.3, 4.5) 0.3 (−1.7, 2.2) 2.1 (−1.3, 5.4) 1.8 (−1.7, 5.2)2.5–5 −20.9, −17.1 2.6 (0.7, 4.5) −2.0 (−3.5, −0.5) * 4.9 (2.1, 7.7) 0.4 (−2.2, 3.0) *5–10 −17.1, −12.8 0.0 (−1.5, 1.4) 0.6 (−0.6, 1.7) −0.9 (−2.9, 1.2) 0.9 (−1.1, 3.0)10–25 −12.8, −6.9 0.1 (−0.7, 0.9) 0.2 (−0.5, 0.9) 0.3 (−1.0, 1.6) 0.0 (−1.0, 1.1)25–50 −6.9, 0.4 −2.3 (−2.9, −1.7) 0.3 (−0.2, 0.8) −2.5 (−3.5, −1.6) −2.2 (−3.0, −1.4)50–75 0.4, 11.3 −0.8 (−1.4, −0.2) 0.6 (0.1, 1.1) * −1.6 (−2.5, −0.6) −0.1 (−0.9, 0.79 *75–90 11.3, 17.3 0.8 (0.0, 1.6) −0.4 (−1.0, 0.3) 1.4 (0.1, 2.7) 0.3 (−0.8, 1.4)90–95 17.3, 20.4 3.1 (1.6, 4.6) −1.3 (−2.5, −0.2) * 4.7 (2.2, 7.2) 1.9 (0.1, 3.7)
95–97.5 20.4, 22.6 4.0 (1.9, 6.1) −2.3 (−4.0, −0.6) ** 6.0 (2.7, 9.3) 1.8 (−1.0, 4.5) *97.5–99 22.6, 24.5 7.9 (4.7, 11.1) −3.6 (−6.1, −1.1) ** 13.4 (7.6, 19.3) 4.6 (0.9, 8.3) *99–100 24.5, 29.4 14.3 (10.4, 18.3) −4.8 (−7.8, −1.9) ** 21.0 (12.3, 29.7) 11.1 (7.1, 15.2) *
Int. J. Environ. Res. Public Health 2017, 14, 944 11 of 19
Table 1. Cont.
Percentiles PET Range Relative Mortality 1972–2014 (%) Trend (%/10 years) t-Test Relative Mortality 1972–1992 (%) Relative Mortality 1994–2014 (%) t-Test
Aged 65–74 years0–1 −42.0, −25.1 −0.7 (−5.3, 3.9) −1.6 (−5.5, 2.3) 1.2 (−4.8, 7.2) −2.9 (−10.2, 4.4)
1–2.5 −25.1, −20.9 2.4 (−1.5, 6.3) −2.4 (−5.6, 0.8) 4.7 (−0.2, 9.7) −0.2 (−6.4, 6.1)2.5–5 −20.9, −17.1 −1.6 (−4.4, 1.1) −0.4 (−2.6, 1.8) 0.0 (−4.0, 4.0) −3.2 (−7.0, 0.7)5–10 −17.1, −12.8 0.1 (−1.9, 2.1) 0.3 (−1.3, 1.9) 0.0 (−2.7, 2.7) −0.2 (−3.1, 2.7)10–25 −12.8, −6.9 1.8 (0.6, 2.9) 0.8 (−0.2, 1.7) 0.9 (−0.7, 2.6) 2.7 (1.0, 4.5)25–50 −6.9, 0.4 −1.6 (−2.5, −0.7) 0.3 (−0.5, 1.0) −1.6 (−2.9, −0.4) −1.5 (−2.9, −0.1)50–75 0.4, 11.3 −1.0 (−1.9, 0.0) −0.5 (−1.3, 0.2) −0.6 (−1.9, 0.7) −1.5 (−2.9, −0.1)75–90 11.3, 17.3 −0.6 (−1.8, 0.7) −0.2 (−1.2, 0.8) −0.4 (−2.2, 1.3) −0.7 (−2.5, 1.1)90–95 17.3, 20.4 3.1 (0.9, 5.2) 0.7 (−1.0, 2.4) 2.5 (−0.6, 5.5) 3.9 (0.8, 7.0)
95–97.5 20.4, 22.6 4.7 (1.6, 7.8) −0.2 (−2.7, 2.3) 5.3 (1.4, 9.1) 4.5 (−0.2, 9.3)97.5–99 22.6, 24.5 6.7 (2.5, 10.8) 0.3 (−3.0, 3.6) 7.2 (0.9, 13.5) 6.0 (0.5, 11.6)99–100 24.5, 29.4 10.9 (5.9, 15.9) −3.7 (−7.5, 0.1) 17.7 (7.9, 27.4) 7.7 (2.0, 13.4)
Aged <65 years0–1 −42.0, −25.1 −1.6 (−6.2, 3.0) 2.4 (−1.4, 6.3) −1.7 (−8.0, 4.7) −1.5 (−8.2, 5.3)
1–2.5 −25.1, −20.9 0.6 (−2.8, 4.0) −1.0 (−3.9, 1.8) 1.9 (−2.5, 6.2) −1.6 (−7.0, 3.8)2.5–5 −20.9, −17.1 0.6 (−1.8, 3.0) −0.1 (−2.1, 1.8) 1.1 (−2.1, 4.3) 0.1 (−3.7, 3.8)5–10 −17.1, −12.8 0.8 (−1.1, 2.6) 0.7 (−0.7, 2.2) 0.2 (−2.2, 2.6) 1.6 (−1.2, 4.5)10–25 −12.8, −6.9 −0.1 (−1.1, 0.9) −0.5 (−1.3, 0.3) 0.0 (−1.5, 1.4) −0.7 (−2.1, 0.7)25–50 −6.9, 0.4 0.0 (−0.8, 0.8) 0.1 (−0.5, 0.8) 0.0 (−1.2, 1.1) 0.1 (−1.2, 1.3)50–75 0.4, 11.3 −1.5 (−2.3, −0.7) 0.4 (−0.3, 1.0) −1.5 (−2.7, −0.4) −1.2 (−2.3, 0.0)75–90 11.3, 17.3 −0.8 (−1.8, 0.3) −0.5 (−1.4, 0.3) −0.7 (−2.2, 0.8) −1.1 (−2.6, 0.4)90–95 17.3, 20.4 1.7 (−0.1, 3.5) −0.1 (−1.5, 1.4) 1.7 (−1.0, 4.4) 1.3 (−1.2, 3.8)
95–97.5 20.4, 22.6 5.6 (2.7, 8.4) 0.0 (−2.3, 2.3) 5.0 (1.0, 9.0) 6.0 (2.1, 10.0)97.5–99 22.6, 24.5 7.0 (3.7, 10.4) 0.3 (−2.4, 2.9) 5.8 (1.5, 10.1) 7.9 (3.1, 12.6)99–100 24.5, 29.4 7.7 (3.7, 11.7) −3.1 (−6.2, −0.1) * 15.7 (8.4, 22.9) 4.0 (−0.7, 8.7) **
* p < 0.05, ** p < 0.01, *** p < 0.001.
Int. J. Environ. Res. Public Health 2017, 14, 944 12 of 19
Table 2. Mean relative mortality of different age groups (95% CI) in percentiles of daily mean temperature on average in the Helsinki-Uusimaa hospital district (Tavg),and time trends in the period of 1972–2014, and two 21-year sub-periods: 1972–1992 and 1994–2014.
Percentiles Tavg Range Relative Mortality 1972–2014 (%) Trend (%/10 years) t-Test Relative Mortality 1972–1992 (%) Relative Mortality 1994–2014 (%) t-Test
All0–1 −33.4, −18.5 0.5 (−1.7, 2.8) −0.3 (−2.2, 1.6) 1.5 (−1.5, 4.5) −0.9 (−4.3, 2.5)
1–2.5 −18.5, −14.1 3.2 (1.3, 5.0) −0.7 (−2.2, 0.8) 3.3 (0.8, 5.8) 2.6 (−0.1, 5.3)2.5–5 −14.1, −10.7 1.7 (0.2, 3.1) −0.8 (−1.9, 0.4) 2.8 (1.0, 4.7) 0.2 (−2.2, 2.5)5–10 −10.7, −6.6 −0.2 (−1.2, 0.8) 0.3 (−0.5, 1.2) −0.2 (−1.6, 1.1) 0.0 (−1.5, 1.6)10–25 −6.6, −0.8 0.3 (−0.1, 1.0) 0.2 (−0.2, 0.7) 0.1 (−0.8, 0.9) 0.5 (−0.3, 1.3)25–50 −0.8, 4.8 −1.4 (−1.7, −0.9) 0.2 (−0.2, 0.5) −1.4 (−2.0, −0.8) −1.4 (−2.0, −0.8)50–75 4.8, 12.7 −1.1 (−1.6, −0.7) 0.2 (−0.2, 0.5) −1.5 (−2.2, −0.9) −0.8 (−1.4, −0.2)75–90 12.7, 16.7 0.3 (−0.3, 0.9) −0.1 (−0.6, 0.4) 0.5 (−0.4, 1.4) 0.2 (−0.6, 1.0)90–95 16.7, 18.6 1.7 (0.8, 2.7) −0.5 (−1.2, 0.3) 2.4 (1.0, 3.9) 1.1 (−0.2, 2.4)
95–97.5 18.6, 19.9 4.3 (2.8, 5.8) −1.5 (−2.7, −0.3) * 6.0 (3.7, 8.3) 2.6 (0.6, 4.7) *97.5–99 19.9, 21.4 5.6 (3.5, 7.6) 0.4 (−1.2, 2.0) 4.6 (1.3, 7.9) 6.1 (3.5, 8.6)99–100 21.4, 25.4 15.2 (12.5, 18.0) −4.1 (−6.0, −2.3) *** 22.5 (17.7, 27.3) 10.7 (7.7, 13.8) ***
Aged ≥75 years0–1 −33.4, −18.5 3.9 (1.1, 6.8) −2.0 (−4.4, 0.4) 5.9 (2.0, 9.8) 0.9 (−3.1, 5.0)
1–2.5 −18.5, −14.1 4.1 (1.5, 6.7) 0.1 (−2.0, 2.2) 3.5 (−0.2, 7.2) 4.5 (0.9, 8.1)2.5–5 −14.1, −10.7 3.3 (1.4, 5.3) −0.8 (−2.4, 0.7) 4.4 (1.8, 7.0) 2.0 (−0.9, 4.9)5–10 −10.7, −6.6 −0.8 (−2.2, 0.6) 0.3 (−0.9, 1.4) −0.9 (−2.9, 1.1) −0.3 (−2.3, 1.7)10–25 −6.6, −0.8 0.0 (−0.8, 0.9) 0.2 (−0.5, 0.8) 0.0 (−1.3, 1.3) 0.1 (−1.0, 1.1)25–50 −0.8, 4.8 −2.0 (−2.6, −1.4) 0.3 (−0.2, 0.8) −2.2 (−3.1, −1.2) −1.9 (−2.7, −1.1)50–75 4.8, 12.7 −1.3 (−1.9, −0.7) 0.7 (0.2, 1.2) ** −2.3 (−3.3, −1.4) −0.3 (−1.1, 0.5)75–90 12.7, 16.7 1.0 (0.2, 1.8) −0.2 (−0.9, 0.4) 1.7 (0.4, 3.0) 0.5 (−0.6, 1.6)90–95 16.7, 18.6 2.2 (0.8, 3.6) −1.2 (−2.2, −0.1) * 3.7 (1.5, 6.0) 0.8 (−0.9, 2.5) *
95–97.5 18.6, 19.9 4.7 (2.5, 6.9) −3.3 (−5.0, −1.6) *** 8.2 (4.4, 11.9) 1.6 (−1.0, 4.2) **97.5–99 19.9, 21.4 5.7 (2.8, 8.7) −2.5 (−4.8, −0.1) * 8.2 (2.8, 13.7) 4.3 (0.8, 7.9)99–100 21.4, 25.4 18.6 (14.5, 22.7) −4.6 (−7.4, −1.8) ** 26.6 (18.7, 34.4) 13.7 (9.2, 18.1) **
Int. J. Environ. Res. Public Health 2017, 14, 944 13 of 19
Table 2. Cont.
Percentiles PET Range Relative Mortality 1972–2014 (%) Trend (%/10 years) t-Test Relative Mortality 1972–1992 (%) Relative Mortality 1994–2014 (%) t-Test
Aged 65–75 years0–1 −33.4, −18.5 −1.8 (−6.4, 2.7) 0.9 (−3.0, 4.8) −2.3 (−7.4, 2.8) −1.2 (−9.8, 7.4)
1–2.5 −18.5, −14.1 2.8 (−0.9, 6.4) −3.6 (−6.6, −0.6) * 5.9 (1.2, 10.6) −1.2 (−7.1, 4.7)2.5–5 −14.1, −10.7 −0.4 (−3.2, 2.5) −2.3 (−4.6, 0.0) 2.6 (−1.4, 6.5) −4.1 (−8.3, 0.1) *5–10 −10.7, −6.6 −0.5 (−2.4, 1.5) 0.6 (−1.0, 2.1) −0.8 (−3.4, 1.9) −0.2 (−3.1, 2.8)10–25 −6.6, −0.8 1.7 (0.5, 2.8) 0.9 (0.0, 1.9) 0.6 (−1.1, 2.2) 2.7 (1.0, 4.4)25–50 −0.8, 4.8 −1.4 (−2.3, −0.5) 0.3 (−0.4, 1.0) −1.3 (−2.5, −0.1) −1.3 (−2.7, 0.1)50–75 4.8, 12.7 −1.4 (−2.3, −0.4) −0.7 (−1.4, 0.1) −0.9 (−2.2, 0.4) −2.0 (−3.4, −0.6)75–90 12.7, 16.7 0.3 (−1.0, 1.5) 0.0 (−1.0, 0.9) 0.0 (−1.7, 1.7) 0.5 (−1.3, 2.3)90–95 16.7, 18.6 2.0 (−0.1, 4.29 −0.1 (−1.8, 1.6) 2.6 (−0.4, 5.6) 1.9 (−1.1, 5.0)
95–97.5 18.6, 19.9 3.7 (0.6, 6.8) 1.1 (−1.3, 3.5) 3.1 (−0.8, 7.0) 4.2 (−0.5, 8.9)97.5–99 19.9, 21.4 4.3 (0.1, 8.4) 3.2 (−0.1, 6.59) 0.2 (−5.8, 6.1) 6.5 (0.9, 12.2)99–100 21.4, 25.4 14.4 (9.6, 19.3) −5.9 (−9.2, −2.7) *** 24.7 (16.6, 32.8) 8.1 (2.3, 13.9) **
Aged <65 years0–1 −33.4, −18.5 −3.0 (−7.3, 1.3) 0.7 (−2.9, 4.4) −1.8 (−7.4, 3.8) −4.9 (−11.7, 2.0)
1–2.5 −18.5, −14.1 1.2 (−2.1, 4.6) −0.1 (−2.8, 2.6) 0.6 (−3.7, 4.9) 1.2 (−4.1, 6.5)2.5–5 −14.1, −10.7 0.0 (−2.5, 2.5) −0.5 (−2.5, 1.5) 0.8 (−2.2, 3.8) −1.4 (−5.6, 2.8)5–10 −10.7, −6.6 1.3 (−0.5, 3.1) 0.3 (−1.1, 1.8) 1.6 (−0.8, 4.0) 1.1 (−1.7, 3.8)10–25 −6.6, −0.8 −0.1 (−1.1, 1.0) −0.2 (−1.0, 0.6) 0.0 (−1.5, 1.4) −0.4 (−1.9, 1.0)25–50 −0.8, 4.8 −0.3 (−1.1, 0.5) 0.1 (−0.6, 0.7) −0.3 (−1.4, 0.8) −0.2 (−1.4, 1.0)50–75 4.8, 12.7 −1.0 (−1.9, −0.2) 0.1 (−0.6, 0.7) −1.0 (−2.2, 0.2) −0.9 (−2.0, 0.3)75–90 12.7, 16.7 −0.9 (−1.9, 0.2) −0.1 (−0.9, 0.8) −1.1 (−2.6, 0.4) −0.8 (−2.3, 0.7)90–95 16.7, 18.6 1.2 (−0.5, 3.0) 0.1 (−1.3, 1.4) 1.0 (−1.5, 3.5) 1.2 (−1.4, 3.7)
95–97.5 18.6, 19.9 5.0 (2.1, 7.8) −1.1 (−3.4, 1.1) 6.1 (2.2, 10.0) 3.5 (−0.6, 7.6)97.5–99 19.9, 21.4 7.1 (3.6, 10.5) 2.9 (0.2, 5.6) * 3.3 (−2.1, 8.7) 9.2 (4.7, 13.6)99–100 21.4, 25.4 9.7 (5.7, 13.7) −3.7 (−6.4, −1.0) ** 16.3 (10.2, 22.3) 5.6 (0.4, 10.8) **
* p < 0.05, ** p < 0.01, *** p < 0.001.
Int. J. Environ. Res. Public Health 2017, 14, 944 14 of 19
4. Discussion
Our study shows that the impact of heat stress on mortality is remarkable in the climate ofFinland, in Northern Europe. The increase in relative mortality appears above the 95th percentileof the thermal distribution and is highest above the 99th percentile among those 75 years and older.The increase in relative mortality is greater in hot than in cold extreme conditions. This outcomemay seem to contradict the common understanding that cold-related mortality is more remarkablethan heat-related mortality (e.g., [7]). However, this difference is due to the method used in ourstudy. Instead of absolute mortality we used relative mortality, defined as the deviation from theexpected mortality, which includes seasonal variation. Thus, we studied the weather dependence of themortality deviation from its expected seasonal value. According to our study, this method, developedby Koppe and Jendritzky [21], to calculate relative mortality is feasible especially when assessingimpacts of temperature extremes. The method has also been applied, e.g., by Muthers et al. [27],in mortality studies in Vienna, and by Zacharias et al. [28] for ischemic heart diseases in Germany.
The temperature dependence of mortality in the cold side of the thermal distribution is morecomplex than in the warm side, and might require further studies with more sophisticated methods,such as a distributed lag model between mortality and thermal conditions, e.g., [12,42]. The cold-relatedmortality may also be related to the behaviour of people and exposure to the cold stress. Based ona population study in Finland, people spent only 4% of their total time under cold exposure, most of it(71%) during leisure time [43]. In extremely cold conditions, people avoid outdoor activities and arenot exposed to cold stress for a long period of time, while in more typical cold conditions they mayspend more time outdoors and are therefore more exposed to the cold stress. Interestingly, we foundalso a small increase in sensitivity in the thermal range with PET values around −10 ◦C among thegroup aged 65–74 years that might be worth further studies. In Finland, the indoor conditions are alsotypically thermally comfortable in the cold season due to the heating of houses. On the other hand,the cooling of houses in summer is currently rare in Finland, especially in older buildings. The heatstress inside buildings during heat waves may be severe depending on the building characteristicsand intensity of the urban heat island (e.g., [44,45]).
The time series of relative mortality in our study are stationary, i.e., without a seasonal cycleor a trend in time. The use of the concept of relative mortality enables us to study changes in thetemperature dependence of mortality during the study period. Our study shows a decrease in theimpact of hot extremes on mortality in all age groups over four decades. This suggests that Finns,in general, have become less sensitive to very warm weather than previously. Thus, our studyin the Finnish population supports the findings of some other studies on decreasing sensitivity totemperature extremes, e.g., [42,46]. Our study also confirms the earlier studies of Donaldson et al. [47]and Näyhä [7], which conclude that heat-related mortality has decreased in Finland over decades.On the other hand, de’Donato et al. [48] reported a reduction in heat-related mortality in someEuropean cities, but suggested an increasing heat effect in Helsinki and Stockholm. However, in thatresearch two quite short time periods were compared, namely 1996–2002 and 2004–2010. Additionally,the latter period included one of the most significant heat waves in Finland in summer 2010.
In studies related to acclimatization or changes in sensitivity to temperature extremes, sufficientlylong time series, e.g., in our study over 40 years, should be used to capture inter-annual climatevariability and any potential long-term climatic and non-climatic trends. On the other hand, a timeperiod of about 20 years may be adequate, if the goal is to estimate impacts of extreme temperatures onmortality in the current climate and population structure. In our study, the increase in relative mortalitywith increasing temperature appeared to be larger when assessed in the whole study period than whenassessed in only the latter sub-period. As we can see in Figure 1, the level of expected mortality inthe Helsinki-Uusimaa hospital district has been on a fairly stable level since 2000. Accordingly, a timeperiod starting around 2000 could be sufficiently long for health impact assessments in the currentclimate of Finland.
Int. J. Environ. Res. Public Health 2017, 14, 944 15 of 19
Based on our results, the dependence of mortality on extreme temperatures has weakened intime, even among those 75 years and older. Nonetheless, it is beyond the scope of the presentpaper to firmly attribute the decreased temperature-related mortality burden to any physiologicalor behavioural changes of humans or infrastructural changes in society, such as improved publichealth or lengthening of life time. While our study strongly suggests that the sensitivity of theFinnish population to temperature extremes has decreased in the past decades, it does not providenew information about long-term acclimatization. Our study did not capture changes in minimummortality temperature, which would have been an important indicator for acclimatization as, e.g.,in studies of Oudin Åström et al. [23] and Ballester et al. [3]. However, the decreasing impact of hotextremes on mortality that we found in our study might partly be due to long-term acclimatizationas well.
In northern countries like Finland, the possibilities for future acclimatization to graduallyincreasing mean temperatures are good, since the limits of physiological acclimatization are nota threat, as in many parts of the world in hot climates. However, heat waves are expected to alsobecome more frequent and intensive in Finland [49], and therefore planned adaptation measures tominimize heat-related health risks should not be neglected.
A limitation in our research was that the dataset did not include information on causes of death.Hence, we were not able to investigate whether heat- and cold-related mortality decreased amongtemperature-sensitive risk groups, such as people with heart and respiratory illnesses, or whether thedecrease can be explained mainly by a healthier population in general. However, the recent researchof Ryti et al. [50] on cold spells and sudden cardiac deaths suggests that the use of cardioprotectivemedication decreases the risk of sudden death during cold spells.
An important motivation for our study was to provide information for assessing impacts ofclimate change on mortality. Since potential acclimatization to gradually-warming climate and changesin sensitivity of the population to temperature extremes are important sources of uncertainties instudies related to climate change impacts on temperature-related mortality, there is a clear need formulti-disciplinary research.
One of our aims was to study the feasibility of different biometeorological indices to describethe dependence of mortality on thermal conditions. Meteorological input variables to calculatephysiologically equivalent temperature, PET, are air temperature, humidity, wind speed, and solarradiation. Therefore, PET describes the thermal environment of a person more realistically than airtemperature alone. In our study, PET seemed to provide a more consistent mortality-thermal conditionrelationship than air temperature throughout the thermal distribution, especially in the cold thermalrange. Morabito et al. [51] also concluded that indices, which account for the extra effect of wind speedand solar radiation, fit best with all-cause elderly mortality.
The calculation of PET values requires sub-daily meteorological data and its spatial distribution isaffected remarkably by topography and the built environment. Here, we also reported the mortalitydependence on spatially-averaged gridded temperature data. Based on our results, temperature maygive results that are in many cases good enough for studies related to weather and climate changeimpacts on mortality, at least in the warm thermal range. The use of gridded temperature data insteadof station-wise meteorological data may, e.g., extend the possibilities to study weather impacts onmortality in sparsely-populated larger areas, which is the case in most parts of Finland and in areashaving a sparse weather station network. Furthermore, in climate change impact studies, it may beadequate to use temperature scenarios alone, without projections for all the input variables neededfor calculations of PET. Compared to inaccuracies arising from the use of temperature instead of PET,the other, non-climatic sources of uncertainty, discussed above, may be more influential.
The Finnish Meteorological Institute started to issue heat wave and cold spell warnings in 2011but, so far, they have not been fully utilized in the health sector. Our findings on decreasing sensitivityto temperature extremes over the decades in Finland show the importance of the work to improve
Int. J. Environ. Res. Public Health 2017, 14, 944 16 of 19
public health, and of well-functioning healthcare in the prevention of heat-related mortality andadaptation to climate change.
5. Conclusions
Heat-related mortality is also remarkable in northern countries such as Finland. The increase ofthe relative mortality in the hot extreme of the thermal distributions is more than in the cold extreme.On the other hand, from the 43-year long time series of mortality and meteorological data, we founda statistically significant decrease in relative mortality in upper percentiles of the thermal distribution,indicating a decreasing sensitivity to heat stress over the decades.
Both PET and spatially-averaged temperature are feasible indicators in modelling the dependenceof mortality on thermal environment, depending on the scope of the study and availability of the data.
Further studies to elaborate the dependence of mortality on thermal conditions would requiremodelling of delayed effects by applying, e.g., distributed lag non-linear models. In studies onacclimatization, long time series of mortality should be used; according to our study, a period of atleast several decades is needed to find time trends in sensitivity.
Supplementary Materials: The following are available online at www.mdpi.com/1660-4601/14/8/944/s1,Figure S1: Scatterplots of 7-day mean of relative mortality in different age groups and 7-day mean valuesof PET at Helsinki-Vantaa airport district in 1972–2014, and two 21-year time periods, 1972–1992 and 1994–2014and their relationships fitted with the generalized additive model, GAM (95% CI). Figure S2: Scatterplots of14-day means of relative mortality in different age groups and 14-day mean values of PET at Helsinki-Vantaaairport district in 1972–2014 and two 21-year time periods, 1972–1992 and 1994–014, and their relationships fittedwith the generalized additive model, GAM (95% CI). Table S1: 7-day mean values of relative mortality (95% CI) ofdifferent age groups in the percentiles of 7-day mean values of PET at Helsinki-Vantaa airport in 1972–2014, and intwo 21-year sub-periods, 1972–1992 and 1994–2014, and the statistical significance of the differences in relativemortality between the sub-periods. Table S2: 14-day mean values of relative mortality (95% CI) of different agegroups in the percentiles of 14-day mean values of PET at Helsinki-Vantaa airport in 1972–2014, and in two 21-yearsub-periods, 1972–1992 and 1994–2014, and the statistical significance of the differences in relative mortalitybetween the sub-periods.
Acknowledgments: The research presented in this paper is part of the PLUMES (Pathways Linking Uncertaintiesin Model Projections of Climate and its Effects) project funded by the Academy of Finland (decision number 278067)and it also contributes to the Nordic Centre of Excellence for Resilience and Societal Security—NORDRESS—whichis funded by the Nordic Societal Security Programme. We thank Dominik Fröhlich and Pentti Pirinen for assistingwith data and methods used in this research. We also thank the two anonymous reviewers whose commentshelped improve and clarify the manuscript.
Author Contributions: Reija Ruuhela made all the analysis and wrote the paper. Kirsti Jylhä contributed tothe manuscript as a climate expert, Timo Lanki as a health expert, and Pekka Tiittanen in statistical methods.Andreas Matzarakis also contributed to the biometeorological methods and the analysis tools used in the research.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Armstrong, B. Models for the relationship between ambient temperature and daily mortality. Epidemiology2006, 17, 624–631. [CrossRef] [PubMed]
2. Kovats, R.S.; Hajat, S. Heat stress and public health: A critical review. Annu. Rev. Public Health 2008, 29, 41–55.[CrossRef] [PubMed]
3. Ballester, J.; Robine, J.-M.; Herrmann, F.R.; Rodó, X. Long-term projections and acclimatization scenarios oftemperature-related mortality in Europe. Nat. Commun. 2011, 2, 358. [CrossRef] [PubMed]
4. Basu, R.; Samet, J.M. Relation between elevated ambient temperature and mortality: A review of theepidemiologic evidence. Epidemiol. Rev. 2002, 24, 190–202. [CrossRef] [PubMed]
5. McMichael, A.J.; Woodruff, R.E.; Hales, S. Climate change and human health: Present and future risks.Lancet 2006, 367, 859–869. [CrossRef]
6. Keatinge, W.R.; Donaldson, G.C.; Cordioli, E.; Martinelli, M.; Kunst, A.E.; Mackenbach, J.P.; Näyhä, S.;Vuori, I. Heat related mortality in warm and cold regions of Europe: Observational study. BMJ 2000,321, 670–673. [CrossRef] [PubMed]
Int. J. Environ. Res. Public Health 2017, 14, 944 17 of 19
7. Näyhä, S. Environmental temperature and mortality. Int. J. Circumpolar Health 2005, 64, 451–458. [CrossRef][PubMed]
8. Näyhä, S. Heat mortality in Finland in the 2000s. Int. J. Circumpolar Health 2007, 66, 418–424. [CrossRef][PubMed]
9. Robine, J.M.; Cheung, S.L.K.; Le Roy, S.; Van Oyen, H.; Griffiths, C.; Michel, J.P.; Herrmann, F.R. Death tollexceeded 70,000 in Europe during the summer of 2003. C. R. Biol. 2008, 331, 171–178. [CrossRef] [PubMed]
10. Barriopedro, D.; Fischer, E.M.; Luterbacher, J.; Trigo, R.M.; García-Herrera, R. The hot summer of 2010:Redrawing the temperature record map of Europe. Science 2011, 332, 220–224. [CrossRef] [PubMed]
11. Kollanus, V.; Lanki, T. 2000-luvun pitkittyneiden helleaaltojen kuolleisuusvaikutukset Suomessa. (Mortalityeffects of prolonged heat waves in the 2000s in Finland. Summary in English). Duodecim 2014, 130, 983–990.[PubMed]
12. Gasparrini, A.; Guo, Y.; Hashizume, M.; Lavigne, E.; Zanobetti, A.; Schwartz, J.; Tobias, A.; Tong, S.;Rocklöv, J.; Forsberg, B.; et al. Mortality risk attributable to high and low ambient temperature:A multicountry observational study. Lancet 2015, 386, 369–375. [CrossRef]
13. Anderson, B.G.; Bell, M.L. Weather-Related Mortality. Epidemiology 2009, 20, 205–213. [CrossRef] [PubMed]14. Rocklöv, J.; Forsberg, B. The effect of temperature on mortality in Stockholm 1998–2003: A study of lag
structures and heatwave effects. Scand. J. Public Health 2008, 36, 516–523. [CrossRef] [PubMed]15. Yu, W.; Mengersen, K.; Wang, X.; Ye, X.; Guo, Y.; Pan, X.; Tong, S. Daily average temperature and mortality
among the elderly: A meta-analysis and systematic review of epidemiological evidence. Int. J. Biometeorol.2012, 56, 569–581. [CrossRef] [PubMed]
16. Ebi, K.L.; Mills, D. Winter mortality in a warming climate: A reassessment. Wiley Interdiscip. Rev. Clim. Chang.2013, 4, 203–212. [CrossRef]
17. Bao, J.; Li, X.; Yu, C. The construction and validation of the heat vulnerability index, a review. Int. J. Environ.Res. Public Health 2015, 12, 7220–7234. [CrossRef] [PubMed]
18. Carter, T.R.; Fronzek, S.; Inkinen, A.; Lahtinen, I.; Lahtinen, M.; Mela, H.; O’Brien, K.L.; Rosentrater, L.D.;Ruuhela, R.; Simonsson, L.; et al. Characterising vulnerability of the elderly to climate change in the Nordicregion. Reg. Environ. Chang. 2016, 16. [CrossRef]
19. Oudin Åström, D.; Forsberg, B.; Rocklöv, J. Heat wave impact on morbidity and mortality in the elderlypopulation: A review of recent studies. Maturitas 2011, 69, 99–105. [CrossRef] [PubMed]
20. Bunker, A.; Wildenhain, J.; Vandenbergh, A.; Henschke, N.; Rocklöv, J.; Hajat, S.; Sauerborn, R. Effects of AirTemperature on Climate-Sensitive Mortality and Morbidity Outcomes in the Elderly; a Systematic Reviewand Meta-analysis of Epidemiological Evidence. EBioMedicine 2016, 6, 258–268. [CrossRef] [PubMed]
21. Koppe, C.; Jendritzky, G. Inclusion of short-term adaptation to thermal stresses in a heat load warningprocedure. Meteorol. Z. 2005, 14, 271–278. [CrossRef]
22. Lee, M.; Nordio, F.; Zanobetti, A.; Kinney, P.; Vautard, R.; Schwartz, J. Acclimatization across space and timein the effects of temperature on mortality: A time-series analysis. Environ. Health 2014, 13, 89. [CrossRef][PubMed]
23. Oudin Åström, D.; Tornevi, A.; Ebi, K.L.; Rocklöv, J.; Forsberg, B. Evolution of minimum mortalitytemperature in Stockholm, Sweden, 1901–2009. Environ. Health Perspect. 2016, 124, 740–744. [CrossRef]
24. Oudin Åström, D.; Forsberg, B.; Ebi, K.L.; Rocklöv, J. Attributing mortality from extreme temperatures toclimate change in Stockholm, Sweden. Nat. Clim. Chang. 2013, 3, 1050–1054. [CrossRef]
25. Smith, K.R.; Woodward, A.; Campbell-Lendrum, D.; Chadee, D.; Honda, Y.; Liu, Q.; Olwoch, J.; Revich, B.;Sauerborn, R.; Dokken, D.; et al. Human Health: Impacts, Adaptation, and Co-Benefits. In Climate Change2014: Impacts, Adaptation, and Vulnerability; Cambridge University Press: Cambridge, UK, 2014; pp. 709–754.
26. Huang, C.; Barnett, A.G.; Wang, X.; Vaneckova, P.; FitzGerald, G.; Tong, S. Projecting Future Heat-RelatedMortality under Climate Change Scenarios: A Systematic Review. Environ. Health Perspect. 2011,119, 1681–1690. [CrossRef] [PubMed]
27. Muthers, S.; Matzarakis, A.; Koch, E. Climate change and mortality in Vienna—A human biometeorologicalanalysis based on regional climate modeling. Int. J. Environ. Res. Public Health 2010, 7, 2965–2977. [CrossRef][PubMed]
28. Zacharias, S.; Koppe, C.; Mücke, H.-G. Climate Change Effects on Heat Waves and Future HeatWave-Associated IHD Mortality in Germany. Climate 2015, 3, 100–117. [CrossRef]
29. Höppe, P.R. Heat balance modelling. Experientia 1993, 49, 741–746. [CrossRef] [PubMed]
Int. J. Environ. Res. Public Health 2017, 14, 944 18 of 19
30. Verein Deutscher Ingenieure (VDI). Handbuch Reinhaltung der Luft Band 1: Umweltmeteorologie.In Methoden zur Human-Biometeorologischen Bewertung von Klima und Lufthygiene für die Stadt-undRegionalplanung; Beuth Verlag: Düsseldorf, Germany, 1998; pp. 25–27.
31. Höppe, P. The physiological equivalent temperature—A universal index for the biometeorological assessmentof the thermal environment. Int. J. Biometeorol. 1999, 43, 71–75. [CrossRef] [PubMed]
32. Matzarakis, A.; Mayer, H.; Iziomon, M.G. Application of a universal thermal index: Physiological equivalenttemperature. Int. J. Biometeorol. 1999, 43, 76–84. [CrossRef] [PubMed]
33. Matzarakis, A.; Mayer, H. Heat stress in Greece. Int. J. Biometeorol. 1997, 41, 34–39. [CrossRef] [PubMed]34. Matzarakis, A.; Rocco, M.; Najjar, G. Thermal bioclimate in Strasbourg—The 2003 heat wave.
Theor. Appl. Climatol. 2009, 98, 209–220. [CrossRef]35. Jendritzky, G.; de Dear, R. Adaptation and Thermal Environment. In Biometeorology for Adaptation to Climate
Variability and Change; Ebi, K.L., Burton, I., McGregor, G.R., Eds.; Springer: Dordrecht, The Netherlands,2009; pp. 9–32. [CrossRef]
36. Matzarakis, A.; Endler, C. Climate change and thermal bioclimate in cities: Impacts and options foradaptation in Freiburg, Germany. Int. J. Biometeorol. 2010, 54, 479–483. [CrossRef] [PubMed]
37. Official Statistics of Finland (OSF). Deaths; Official Statistics of Finland: Helsinki, Finland, 2016;ISSN 1798-2545. Available online: http://tilastokeskus.fi/til/kuol/2015/01/kuol_2015_01_2016-10-28_tie_001_en.html (accessed on 7 March 2017).
38. Matzarakis, A.; Rutz, F.; Mayer, H. Modelling radiation fluxes in simple and complexenvironments—Applications of the RayMan model. Int. J. Biometeorol. 2007, 51, 323–334. [CrossRef][PubMed]
39. Matzarakis, A.; Rutz, F.; Mayer, H. Modelling radiation fluxes in simple and complex environments: Basicsof the RayMan model. Int. J. Biometeorol. 2010, 54, 131–139. [CrossRef] [PubMed]
40. Aalto, J.; Pirinen, P.; Jylhä, K. New gridded daily climatology of Finland: Permutation-based uncertaintyestimates and temporal trends in climate. J. Geophys. Res. Atmos. 2016, 121, 3807–3823. [CrossRef]
41. Wood, S. Package ‘mgcv’. 2016. Available online: https://cran.r-project.org/web/packages/mgcv/mgcv.pdf. (accessed on 3 March 2016).
42. Gasparrini, A.; Guo, Y.; Hashizume, M.; Kinney, P.L.; Petkova, E.P.; Lavigne, E.; Zanobetti, A.; Schwartz, J.D.;Tobias, A.; Leone, M.; et al. Temporal variation in heat–mortality associations: A multicountry study.Environ. Health Perspect. 2015, 123. [CrossRef] [PubMed]
43. Mäkinen, T.M.; Raatikka, V.P.; Rytkönen, M.; Jokelainen, J.; Rintamäki, H.; Ruuhela, R.; Näyhä, S.; Hassi, J.Factors affecting outdoor exposure in winter: Population-based study. Int. J. Biometeorol. 2006, 51, 27–36.[CrossRef] [PubMed]
44. Mavrogianni, A.; Wilkinson, P.; Davies, M.; Biddulph, P.; Oikonomou, E. Building characteristics asdeterminants of propensity to high indoor summer temperatures in London dwellings. Build. Environ. 2012,55, 117–130. [CrossRef]
45. Taylor, J.; Wilkinson, P.; Davies, M.; Armstrong, B.; Chalabi, Z.; Mavrogianni, A.; Symonds, P.; Oikonomou, E.;Bohnenstengel, S.I. Mapping the effects of urban heat island, housing, and age on excess heat-relatedmortality in London. Urban Clim. 2015, 14, 517–528. [CrossRef]
46. Arbuthnott, K.; Hajat, S.; Heaviside, C.; Vardoulakis, S. Changes in population susceptibility to heat andcold over time: Assessing adaptation to climate change. Environ. Health 2016, 15. [CrossRef] [PubMed]
47. Donaldson, G.C.; Keatinge, W.R.; Näyhä, S. Changes in summer temperature and heat-related mortalitysince 1971 in North Carolina, South Finland, and Southeast England. Environ. Res. 2003, 91, 1–7. [CrossRef]
48. De’Donato, F.K.; Leone, M.; Scortichini, M.; De Sario, M.; Katsouyanni, K.; Lanki, T.; Basagaña, X.; Ballester, F.;Åström, C.; Paldy, A.; et al. Changes in the effect of heat on mortality in the last 20 years in nine Europeancities. Results from the PHASE project. Int. J. Environ. Res. Public Health 2015, 12, 15567–15583. [CrossRef][PubMed]
49. Kim, S.; Sinclair, V.A.; Räisänen, J.; Ruuhela, R. Heat waves in Finland: Present and projected summertimeextreme temperatures and their associated circulation patterns. Int. J. Climatol. 2017. [CrossRef]
Int. J. Environ. Res. Public Health 2017, 14, 944 19 of 19
50. Ryti, N.R.I.; Mäkikyrö, E.M.S.; Antikainen, H.; Junttila, M.J. Cold spells and ischaemic sudden cardiac death:Effect modification by prior diagnosis of ischaemic heart disease and cardioprotective medication. Sci. Rep.2017, 7. [CrossRef] [PubMed]
51. Morabito, M.; Crisci, A.; Messeri, A.; Capecchi, V.; Modesti, P.A.; Gensini, G.F.; Orlandini, S. Environmentaltemperature and thermal indices: What is the most effective predictor of heat-related mortality in differentgeographical contexts? Sci. World J. 2014, 2014, 961750. [CrossRef] [PubMed]
© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open accessarticle distributed under the terms and conditions of the Creative Commons Attribution(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Supplementary Materials: Biometeorological Assessment of Mortality from Extreme
Temperatures in Helsinki Region, Finland, in 1972-2014
Reija Ruuhela, Kirsti Jylhä, Timo Lanki, Pekka Tiittanen and Andreas Matzarakis
Figure S1. Scatterplots of 7-day mean of relative mortality in different age groups and 7-day mean values of PET at
Helsinki-Vantaa airport district in 1972–2014, and two 21-year time periods, 1972–1992 and 1994–2014 and their relationships
fitted with the generalized additive model, GAM (95% CI).
Figure S2. Scatterplots of 14-day means of relative mortality in different age groups and 14-day mean values of PET at
Helsinki-Vantaa airport district in 1972–2014 and two 21-year time periods, 1972–1992 and 1994–014, and their relationships
fitted with the generalized additive model, GAM (95% CI).
Table S1. 7-day mean values of relative mortality (95% CI) of different age groups in the percentiles of 7-day mean values
of PET at Helsinki-Vantaa airport in 1972–2014, and in two 21-year sub-periods, 1972–1992 and 1994–2014, and the statistical
significance of the differences in relative mortality between the sub-periods.
Table S2. 14-day mean values of relative mortality (95% CI) of different age groups in the percentiles of 14-day mean
values of PET at Helsinki-Vantaa airport in 1972–2014, and in two 21-year sub-periods, 1972–1992 and 1994–2014, and the
statistical significance of the differences in relative mortality between the sub-periods.
1972–2014 1972–1992 1994–2014
Age
group
All
>75 y
65–74 y
<65 y
Figure S1. Scatterplots of 7-day mean of relative mortality in different age groups and 7-day mean values of PET at
Helsinki-Vantaa airport district in 1972–2014, and two 21-year time periods, 1972–1992 and 1994–2014 and their
relationships fitted with the generalized additive model, GAM (95% CI).
1972–2014 1972–1992 1994–2014
Age
group
All
>75 y
65–74 y
<65 y
Figure S2. Scatterplots of 14-day means of relative mortality in different age groups and 14-day mean values of PET
at Helsinki-Vantaa airport district in 1972–2014 and two 21-year time periods, 1972–1992 and 1994–014, and their
relationships fitted with the generalized additive model, GAM (95% CI).
Table S1. 7-day mean values of relative mortality (95% CI) of different age groups in the percentiles of 7-day mean
values of PET at Helsinki-Vantaa airport in 1972–2014, and in two 21-year sub-periods, 1972–1992 and 1994–2014,
and the statistical significance of the differences in relative mortality between the sub-periods.* p < 0.05, ** p < 0.01,
*** p < 0.001.
a) All
Percentiles PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -38.7, -22.2 1.7 (0.5, 2.9) 1.6 (0.2, 3.1) 1.8 (-0.3, 3.9)
1–2.5 -22.2, -18.9 2.9 (1.9, 3.9) 4.7 (3.4, 6.0) 1.1 (-0.3, 2.6) ***
2.5–5 -18.9, -15.9 1.1 (0.2, 1.9) 1.6 (0.4, 2.8) 0.6 (-0.6, 1.7)
5–10 -15.9, -12.5 1.1 (0.4, 1.7) 0.8 (0.0, 1.6) 1.4 (0.4, 2.4)
10–25 -12.5, -7.4 0.5 (0.2, 0.8) 0.3 (-0.2, 0.8) 0.5 (0.1, 1.0)
25–50 -7.4, 0.4 -1.9 (-2.1, -1.6) -1.7 (-2.1, -1.4) -2.0 (-2.3, -1.6)
50–75 0.4, 11.6 -0.9 (-1.1, -0.6) -1.3 (-1.7, -1.0) -0.4 (-0.8, -0.1) ***
75–90 11.6, 16.8 0.3 (0.0, 0.7) 0.7 (0.3, 1.2) -0.2 (-0.7, 0.3) **
90–95 16.8, 19.4 1.7 (1.1, 2.2) 2.5 (1.7, 3.3) 1.0 (0.2, 1.7) **
95–97.5 19.4, 21.3 4.1 (3.3, 4.9) 5.1 (3.7, 6.5) 3.5 (2.5, 4.5)
97.5–99 21.3, 23.0 6.1 (4.8, 7.3) 7.9 (5.8, 10.0) 4.8 (3.3, 6.3) *
99–100 23.0, 27.3 13.7 (11.8, 15.6) 21.5 (17.9, 25.2) 10.5 (8.5, 12.4) ***
b) Age >75 years
Percentiles PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -38.7, -22.2 4.3 (2.8, 5.9) 5.3 (3.5, 7.1) 2.9 (0.2, 5.7)
1–2.5 -22.2, -18.9 4.6 (3.3, 5.9) 6.7 (5.0, 8.4) 2.5 (0.7, 4.4) **
2.5–5 -18.9, -15.9 2.0 (1.0, 3.1) 2.4 (0.8, 4.0) 1.7 (0.3, 3.1)
5–10 -15.9, -12.5 1.5 (0.6, 2.3) 0.6 (-0.5, 1.8) 2.5 (1.2, 3.7) *
10–25 -12.5, -7.4 0.3 (-0.1, 0.8) 0.4 (-0.3, 1.1) 0.2 (-0.4, 0.9)
25–50 -7.4, 0.4 -2.8 (-3.1, -2.5) -2.9 (-3.4, -2.4) -2.8 (-3.2, -2.3)
50–75 0.4, 11.6 -0.5 (-0.9, -0.2) -1.5 (-2.0, -0.9) 0.2 (-0.2, 0.7) ***
75–90 11.6, 16.8 0.6 (0.2, 1.1) 1.7 (1.0, 2.3) -0.3 (-0.9, 0.3) ***
90–95 16.8, 19.4 1.6 (0.8, 2.4) 3.4 (2.2, 4.5) 0.1 (-0.8, 1.1) ***
95–97.5 19.4, 21.3 4.7 (3.5, 6.0) 7.3 (4.9, 9.7) 3.2 (1.7, 4.6) **
97.5–99 21.3, 23.0 6.7 (4.7, 8.7) 10.5 (6.6, 14.3) 4.2 (2.2, 6.2) **
99–100 23.0, 27.3 16.7 (14.2, 19.1) 23.5 (19.0, 28.1) 13.8 (11.1, 16.6) ***
c) Age 65–74 years
Percentiles
PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -38.7, -22.2 -0.6 (-2.9, 1.7) -2.0 (-4.9, 0.8) 1.4 (-2.5, 5.2)
1–2.5 -22.2, -18.9 2.3 (0.2, 4.3) 5.3 (2.7, 7.8) -0.7 (-3.8, 2.5) **
2.5–5 -18.9, -15.9 -0.3 (-1.8, 1.3) 2.3 (0.1, 4.6) -2.9 (-5.1, -0.7) **
5–10 -15.9, -12.5 1.5 (0.5, 2.6) 2.4 (0.8, 3.9) 0.7 (-0.9, 2.3)
10–25 -12.5, -7.4 1.1 (0.5, 1.7) -1.0 (-1.8, -0.1) 3.1 (2.2, 4.0) ***
25–50 -7.4, 0.4 -1.9 (-2.4, -1.4) -1.2 (-1.8, -0.6) -2.5 (-3.2, -1.7) *
50–75 0.4, 11.6 -1.2 (-1.7, -0.7) -0.9 (-1.6, -0.2) -1.6 (-2.3, -0.9)
75–90 11.6, 16.8 0.2 (-0.5, 0.9) -0.2 (-1.1, 0.7) 0.5 (-0.5, 1.5)
90–95 16.8, 19.4 2.7 (1.6, 3.8) 2.8 (1.2, 4.4) 2.7 (1.1, 4.3)
95–97.5 19.4, 21.3 4.6 (3.0, 6.2) 5.4 (2.6, 8.1) 4.2 (2.1, 6.2)
97.5–99 21.3, 23.0 6.3 (4.1, 8.5) 8.1 (5.2, 11.0) 5.0 (1.9, 8.2)
99–100 23.0, 27.3 11.8 (9.0, 14.5) 20.0 (14.6, 25.5) 8.4 (5.4, 11.4) ***
d) Age <65 years
Percentiles
PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -38.7, -22.2 -1.0 (-2.9, 0.9) -1.5 (-4.2, 1.3) -0.4 (-3.0, 2.1)
1–2.5 -22.2, -18.9 0.7 (-0.9, 2.4) 2.0 (-0.2, 4.2) -0.5 (-2.9, 2.0)
2.5–5 -18.9, -15.9 0.3 (-1.0, 1.6) 0.2 (-1.6, 2.0) 0.4 (-1.5, 2.4)
5–10 -15.9, -12.5 -0.1 (-1.1, 0.9) 0.2 (-1.1, 1.4) -0.7 (-2.3, 0.9)
10–25 -12.5, -7.4 0.3 (-0.2, 0.9) 0.9 (0.2, 1.7) -0.6 (-1.4, 0.2) **
25–50 -7.4, 0.4 -0.2 (-0.6, 0.2) -0.5 (-1.1, 0.1) 0.3 (-0.4, 0.9)
50–75 0.4, 11.6 -1.5 (-1.9, -1.0) -1.6 (-2.2, -1.0) -1.0 (-1.6, -0.5)
75–90 11.6, 16.8 0.1 (-0.5, 0.7) 0.1 (-0.7, 1.0) -0.5 (-1.3, 0.4)
90–95 16.8, 19.4 1.2 (0.2, 2.3) 1.2 (-0.2, 2.6) 1.3 (-0.3, 2.8)
95–97.5 19.4, 21.3 3.4 (2.1, 4.8) 2.3 (0.2, 4.4) 4.1 (2.3, 5.8)
97.5–99 21.3, 23.0 5.6 (3.8, 7.4) 5.6 (2.9, 8.2) 5.6 (3.2, 8.0)
99–100 23.0, 27.3 8.6 (6.2, 10.9) 19.9 (15.4, 24.4) 4.0 (1.7, 6.3) ***
Table S2. 14-day mean values of relative mortality (95% CI) of different age groups in the percentiles of 14-day
mean values of PET at Helsinki-Vantaa airport in 1972–2014, and in two 21-year sub-periods, 1972–1992 and 1994–
2014, and the statistical significance of the differences in relative mortality between the sub-periods.
* p < 0.05, ** p < 0.01, *** p < 0.001.
a) All
Percentiles PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -33.7, -20.8 2.7 (1.7, 3.6) 2.8 (1.9, 3.8) 2.3 (-0.2, 4.7)
1–2.5 -20.8, -17.7 2.5 (1.7, 3.3) 3.6 (2.4, 4.8) 1.4 (0.3, 2.6) *
2.5–5 -17.7, -15.5 1.9 (1.2, 2.5) 1.7 (0.8, 2.7) 2.1 (1.2, 2.9)
5–10 -15.5, -12.4 0.7 (0.2, 1.2) 1.0 (0.3, 1.6) 0.4 (-0.3, 1.0)
10–25 -12.4, -7.6 0.6 (0.3, 0.8) 0.4 (0.0, 0.8) 0.5 80.1, 0.9)
25–50 -7.6, 0.4 -1.9 (-2.1, -1.8) -2.0 (-2.2, -1.8) -1.8 (-2.1, -1.6)
50–75 0.4, 11.6 -0.7 (-0.9, -0.5) -1.1 (-1.3, -0.8) -0.4 (-0.6, -0.1) ***
75–90 11.6, 16.6 0.5 (0.3, 0.8) 1.1 (0.7, 1.4) -0.1 (-0.4, 0.3) ***
90–95 16.6, 18.8 1.7 (1.3, 2.1) 2.1 (1.5, 2.7) 1.4 (0.8, 1.9)
95–97.5 18.8, 20.5 3.4 (2.8, 4.1) 5.6 (4.7, 6.6) 2.0 (1.2, 2.8) ***
97.5–99 20.5, 22.0 4.2 (3.3, 5.1) 4.1 (2.2, 6.1) 4.3 (3.2, 5.3)
99–100 22.0, 25.4 12.0 (10.8, 13.1) 16.8 (14.8, 18.8) 9.4 (8.2, 10.5) ***
b) Age >75 years
Percentiles PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -33.7, -20.8 5.3 (4.0, 6.6) 6.4 (5.3, 7.6) 2.8 (-0.4, 6.1) *
1–2.5 -20.8, -17.7 3.4 (2.3, 4.4) 5.0 (3.5, 6.4) 1.8 (0.3, 3.2) **
2.5–5 -17.7, -15.5 3.1 (2.3, 3.8) 2.9 (1.8, 4.0) 3.2 (2.1, 4.3)
5–10 -15.5, -12.4 1.2 (0.6, 1.8) 1.4 (0.4, 2.3) 1.0 (0.2, 1.9)
10–25 -12.4, -7.6 0.5 (0.1, 0.9) 0.3 (-0.3, 0.8) 0.6 (0.1, 1.2)
25–50 -7.6, 0.4 -2.8 (-3.0, -2.6) -2.9 (-3.3, -2.6) -2.7 (-3.0, -2.4) 50–75 0.4, 11.6 -0.4 (-0.6, -0.1) -1.3 (-1.6, -0.9) 0.4 (0.1, 0.7) ***
75–90 11.6, 16.6 0.9 (0.6, 1.2) 2.2 (1.7, 2.6) -0.4 (-0.8, 0.0) ***
90–95 16.6, 18.8 1.5 (1.0, 2.1) 2.6 (1.7, 3.5) 0.7 (0.0, 1.5) **
95–97.5 18.8, 20.5 3.8 (2.9, 4.7) 6.6 (4.8, 8.4) 2.0 (1.0, 3.0) ***
97.5–99 20.5, 22.0 4.4 (3.2, 5.6) 4.9 (2.2, 7.6) 4.2 (2.9, 5.6)
99–100 22.0, 25.4 14.5 (12.6, 16.4) 19.9 (16.3, 23.5) 11.5 (9.5, 13.5) ***
c) Age 65–74 years
Percentiles PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -33.7, -20.8 0.6 (-0.6, 1.8) 0.4 (-1.1, 1.9) 1.0 (-0.9, 3.0)
1–2.5 -20.8, -17.7 1.9 (0.5, 3.3) 3.4 (1.5, 5.3) 0.4 (-1.6, 2.4) *
2.5–5 -17.7, -15.5 2.3 (1.1, 3.5) 2.6 (1.0, 4.2) 2.0 (0.3, 3.7)
5–10 -15.5, -12.4 0.2 (-0.6, 1.0) 1.0 (-0.1, 2.2) -0.8 (-1.8, 0.3) *
10–25 -12.4, -7.6 0.8 (0.4, 1.2) -0.2 (-0.8, 0.4) 1.8 (1.1, 2.5) ***
25–50 -7.6, 0.4 -1.8 (-2.1, -1.5) -1.7 (-2.1, -1.3) -1.9 (-2.4, -1.3)
50–75 0.4, 11.6 -1.0 (-1.4, -0.7) -0.5 (-0.9, 0.0) -1.7 (-2.2, -1.2) ***
75–90 11.6, 16.6 0.5 (0.0, 0.9) 0.1 (-0.5, 0.7) 1.0 (0.3, 1.6) *
90–95 16.6, 18.8 2.6 (1.8, 3.3) 2.4 (1.3, 3.6) 2.7 (1.7, 3.8)
95–97.5 18.8, 20.5 4.1 (3.0, 5.2) 7.1 (5.7, 8.6) 2.1 (0.6, 3.5) ***
97.5–99 20.5, 22.0 4.8 (3.2, 6.4) 3.9 (1.5, 6.3) 5.1 (3.1, 7.2)
99–100 22.0, 25.4 10.8 (9.1, 12.5) 14.8 (12.2, 17.4) 8.6 (6.5, 10.7) ***
d) Age <65 years
Percentiles PETmean
range
Relative Mortality
1972–2014 [%]
Relative Mortality
1972–1992 [%]
Relative Mortality
1994–2014 [%]
t-
test
0–1 -33.7, -20.8 -0.3 (-1.7, 1.1) -1.2 (-2.9, 0.5) 1.8 (1.8, 2.5) *
1–2.5 -20.8, -17.7 2.0 (0.8, 3.2) 2.3 (0.5, 4.2) 1.7 (1.7, 1.5)
2.5–5 -17.7, -15.5 -0.2 (-1.2, 0.7) -0.1 (-1.6, 1.3) -0.4 -0.4, 1.2)
5–10 -15.5, -12.4 0.3 (-0.4, 0.9) 0.5 (-0.4, 1.4) -0.2 (-0.2, 0.9)
10–25 -12.4, -7.6 0.3 (-0.1, 0.7) 1.0 (0.5, 1.5) -0.7 (-0.7, 0.6) ***
25–50 -7.6, 0.4 -0.4 (-0.6, -0.1) -0.8 (-1.2, -0.4) 0.3 (0.3, 0.4) ***
50–75 0.4, 11.6 -1.4 (-1.7, -1.1) -1.5 (-1.9, -1.1) -1.1 (-1.1, 0.4)
75–90 11.6, 16.6 0.4 (-0.1, 0.8) 0.4 (-0.2, 1.0) -0.1 (-0.1, 0.6)
90–95 16.6, 18.8 1.3 (0.6, 2.0) 1.0 (0.1, 1.9) 1.5 (1.5, 1.0)
95–97.5 18.8, 20.5 2.7 (1.6, 3.7) 3.5 (2.1, 5.0) 2.1 (2.1, 1.4)
97.5–99 20.5, 22.0 3.5 (2.2, 4.8) 3.3 (0.7, 5.9) 3.6 (3.6, 1.5)
99–100 22.0, 25.4 8.2 (6.9, 9.6) 15.5 (14.0, 17.1) 4.3 (4.3, 1.4) ***
Reprinted under the Creative Commons Attribution License
II
International Journal of
Environmental Research
and Public Health
Article
Regional Assessment of Temperature-RelatedMortality in Finland
Reija Ruuhela *, Otto Hyvärinen and Kirsti Jylhä
Finnish Meteorological Institute, P.O. Box 503, FI-00101 Helsinki, Finland; [email protected] (O.H.);[email protected] (K.J.)* Correspondence: [email protected]; Tel.: +358-500-424533
Received: 1 February 2018; Accepted: 23 February 2018; Published: 27 February 2018
Abstract: The aim of this study was to assess regional differences in temperature–mortality relationshipsacross 21 hospital districts in Finland. The temperature dependence of the daily number of all-cause,all-aged deaths during 2000–2014 was studied in each hospital district by using daily mean temperatures,spatially averaged across each hospital district, to describe exposure to heat stress and cold stress.The relationships were modelled using distributed lag non-linear models (DLNM). In a simple modelversion, no delayed impacts of heat and cold on mortality were taken into account, whereas a morecomplex version included delayed impacts up to 25 days. A meta-analysis with selected climaticand sociodemographic covariates was conducted to study differences in the relationships betweenhospital districts. A pooled mortality-temperature relationship was produced to describe the averagerelationship in Finland. The simple DLNM model version without lag gave U-shaped dependenciesof mortality on temperature almost without exception. The outputs of the model version with a 25-daylag were also U-shaped in most hospital districts. According to the meta-analysis, the differences inthe temperature-mortality relationships between hospital districts were not statistically significant onthe absolute temperature scale, meaning that the pooled mortality–temperature relationship can beapplied to the whole country. However, on a relative temperature scale, heterogeneity was found,and the meta-regression suggested that morbidity index and population in the hospital districtsmight explain some of this heterogeneity. The pooled estimate for the relative risk (RR) of mortalityat a daily mean temperature of 24 ◦C was 1.16 (95% CI 1.12–1.20) with reference at 14 ◦C, which isthe minimum mortality temperature (MMT) of the pooled relationship. On the cold side, the RR ata daily mean temperature of −20 ◦C was 1.14 (95% CI 1.12–1.16). On a relative scale of daily meantemperature, the MMT was found at the 79th percentile.
Keywords: temperature-related mortality; distributed lag models; regional differences; meta-analysis
1. Introduction
Mortality-temperature association is often schematically described as a U-shaped (or V-, J-shaped)curve with a trough at a so-called minimum mortality temperature (MMT) and increasing mortalitytowards both hot and cold tails of the temperature distribution. However, the exposure–responserelationship is quite complex because of the non-linear and delayed impacts of thermal stress onhuman health. The shape of this relationship also varies regionally and in relation to sociodemographicfactors [1–4].
The MMT is lower in cooler compared to warmer climatic zones [5,6], meaning that differencesin MMT across climatic zones can be considered as an indicator of acclimatization of a population totheir typical climatic environment. According to Guo et al. [6], on the basis of data from 12 countries indifferent climatic zones, the MMT is found at approximately the 75th percentile of the temperaturedistribution, varying between the 66th and 80th percentiles. However, Tobias et al. [7] detected a much
Int. J. Environ. Res. Public Health 2018, 15, 406; doi:10.3390/ijerph15030406 www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2018, 15, 406 2 of 13
wider range for the MMT among cities in Spain. Earlier studies indicated that in Finland the MMT isabout 12–17 ◦C, while, in Mediterranean countries, the lowest mortality is found at 22–25 ◦C [5,8,9].
No systematic climate zone dependence has been found otherwise in the shape of thetemperature-mortality relationship, thus in the steepness of the slopes around the minimum mortalitytemperature [5,6]. However, for example Curriero et al. [10] found a strong association of thetemperature–mortality relationship with latitude in U.S. cities. People in colder climates are moresensitive to high temperatures than people in warm climates. Meta-analyses across European andU.S. cities [11–13] have also indicated heterogeneity in temperature–mortality relationships. The riskvaries by community and country [6], and differences in vulnerability and sensitivity of the populationto temperature extremes depend also on socioeconomic and non-climatic environmental factors.For instance, in a study comparing U.S. cities, Hondula et al. [14] concluded that places with a greaterrisk had a developed urban environment, higher percentage of children, elderly, and minority residents,and lower income and educational attainment; however, the key explanatory variables varied fromone city to another.
The effects of hot temperatures on mortality appear immediately on the same day and usuallylast a few days, while the effects of cold appear typically after a couple of days and last about10 days [6,15,16] or even weeks [4]. The impacts of extreme temperatures vary seasonally, and forinstance the impacts of heat waves in early summer may be more severe than later in the season.Short-term acclimatization to seasonal variation takes place typically within a couple of weeks [17].Furthermore, the heat and cold stress may lead to a displacement of mortality, called “harvesting”,with death taking place earlier than it would have happened otherwise [2].
The complexity of the mortality–temperature relationship creates challenges for modellingand assessing the overall impacts of hot and cold stress on mortality. In simple statistical models,mortality and temperature on the same day are compared. However, in recent years, more complexdistributed lag non-linear models (DLNM) have been developed and applied in studies on themortality–temperature relationship [2,18–21]. DLNM has been applied in several multi-country, (e.g., [6,22]),and multi-city studies including Helsinki, the capital of Finland [23], and at the country level, e.g., inEstonia, a neighbouring country of Finland [24].
In vulnerability assessment of the elderly to climate change in Finland [25], one mortality-temperaturerelationship was applied to the whole country. However, there are substantial regional differences inmortality (Figure 1, right) and morbidity in Finland. According to a morbidity index of the NationalInstitute of Health and Welfare in Finland, people in western and southern Finland are healthier thanin eastern and northern Finland, and there are significant differences between municipalities in theprevalence of chronic disease [26]. The main motivation for modelling the mortality–temperaturerelationship regionally is to further improve the vulnerability assessment.
Thermal stress experienced by people depends not only on temperature but also on humidity,wind, and radiation balance. Thermal indices based on human energy balance describe the thermalenvironment more realistically than air temperature alone [27]. However, so far there is no consensuson the best thermal index for predicting temperature-related mortality. Based on a comparison of twoindices, Physiologically Equivalent Temperature (PET) and outdoor air temperature, Ruuhela et al. [28]concluded that although PET appeared to better describe impacts of cold stress, the latter was applicableas well. Therefore, in this study, temperature is used as the indicator for heat and cold stress.
Our initial hypotheses were that: (1) there are regional differences in the mortality–temperaturerelationships between hospital districts in Finland; (2) these differences can be partly explained byclimatic and sociodemographic factors. In this paper, we show that these hypotheses could not be fullyconfirmed. Furthermore, we study applicability of spline modelling methods in sparsely populatedareas using models with and without delayed temperature impacts.
Int. J. Environ. Res. Public Health 2018, 15, 406 3 of 13Int. J. Environ. Res. Public Health 2018, 15, x FOR PEER REVIEW 3 of 13
Figure 1. Hospital districts (left; 2011 regionalization), annual mean temperature in Finland in the
climate normal period 1981–2010 (middle, [29]) and mortality (1/100,000) by hospital district in 2014
(right, [30]).
2. Materials and Methods
2.1. Data
The regional administrative level of health services in Finland consists of 21 hospital districts
(HD) (Figure 1, left). All‐cause daily number of deaths and annual population in the HDs for the
study period of 2000–2014 were obtained from Statistics Finland. Finland is a sparsely populated
country with a somewhat higher population density in the southern part of the country. The
population is highest, about 1.5 million, in the Helsinki‐Uusimaa hospital district (HD1)
(Supplementary Materials Table S1). In five hospital districts, the population is less than 0.1 million
and, in the remaining hospital districts, it varies between 0.1 and 0.5 million. The share of elderly (75
years of age and older) in the hospital district population varies from 5% to 10%, and the highest
values are found in eastern Finland HDs. The daily number of deaths in HD1 varied between 11 and
57, with a median of 30 deaths per day during our study period (Table S1). In the smallest hospital
districts, the median of daily deaths was less than five, while in most of the hospital districts the
median of daily deaths was about 10 with a maximum of around 20 deaths per day. In meta‐
regression analysis we also used hospital district‐level morbidity indices from the National Institute
for Health and Welfare (THL). In these indices, selected disease groups are weighted on the basis of
their significance for mortality, disability, quality of life, and health‐care costs in the population
[26,31]. The morbidity index varied between 66 and 147, with the highest values in eastern Finland
hospital districts.
Daily numbers of deaths were compared to daily mean temperatures that were calculated as a
spatial average over each hospital district on the basis of a FMI (Finnish Meteorological Institute)
daily gridded (10 km × 10 km) temperature data. The gridded data were produced by kriging
interpolation, in which topography and water bodies are taken into account besides the measured
temperature observations [32]. As the distance from the southernmost to the northernmost point in
Finland is more than 1000 km, there are substantial temperature differences between southern and
northern parts of the country (Figure 1, middle). In southwestern hospital districts, the spatially
averaged daily temperature varied between –25 °C and +25 °C during our study period. More
continental climatic conditions, with a wider temperature range, are found in eastern and northern
hospital districts, e.g., in Northern Carelia (HD12 = Pohjois−Karjala) the range was from –35 °C to +27
°C. The spatial temperature variation on the daily level within each hospital district greatly depends
on the prevailing weather conditions. The spatial temperature differences may be negligible, less than
Figure 1. Hospital districts (left; 2011 regionalization), annual mean temperature in Finland in theclimate normal period 1981–2010 (middle, [29]) and mortality (1/100,000) by hospital district in 2014(right, [30]).
2. Materials and Methods
2.1. Data
The regional administrative level of health services in Finland consists of 21 hospital districts (HD)(Figure 1, left). All-cause daily number of deaths and annual population in the HDs for the studyperiod of 2000–2014 were obtained from Statistics Finland. Finland is a sparsely populated countrywith a somewhat higher population density in the southern part of the country. The population ishighest, about 1.5 million, in the Helsinki-Uusimaa hospital district (HD1) (Supplementary MaterialsTable S1). In five hospital districts, the population is less than 0.1 million and, in the remaining hospitaldistricts, it varies between 0.1 and 0.5 million. The share of elderly (75 years of age and older) in thehospital district population varies from 5 to 10%, and the highest values are found in eastern FinlandHDs. The daily number of deaths in HD1 varied between 11 and 57, with a median of 30 deaths perday during our study period (Table S1). In the smallest hospital districts, the median of daily deathswas less than five, while in most of the hospital districts the median of daily deaths was about 10with a maximum of around 20 deaths per day. In meta-regression analysis we also used hospitaldistrict-level morbidity indices from the National Institute for Health and Welfare (THL). In theseindices, selected disease groups are weighted on the basis of their significance for mortality, disability,quality of life, and health-care costs in the population [26,31]. The morbidity index varied between 66and 147, with the highest values in eastern Finland hospital districts.
Daily numbers of deaths were compared to daily mean temperatures that were calculated as aspatial average over each hospital district on the basis of a FMI (Finnish Meteorological Institute) dailygridded (10 km × 10 km) temperature data. The gridded data were produced by kriging interpolation,in which topography and water bodies are taken into account besides the measured temperatureobservations [32]. As the distance from the southernmost to the northernmost point in Finland is morethan 1000 km, there are substantial temperature differences between southern and northern parts of thecountry (Figure 1, middle). In southwestern hospital districts, the spatially averaged daily temperaturevaried between –25 ◦C and +25 ◦C during our study period. More continental climatic conditions,with a wider temperature range, are found in eastern and northern hospital districts, e.g., in NorthernCarelia (HD12 = Pohjois−Karjala) the range was from –35 ◦C to +27 ◦C. The spatial temperaturevariation on the daily level within each hospital district greatly depends on the prevailing weather
Int. J. Environ. Res. Public Health 2018, 15, 406 4 of 13
conditions. The spatial temperature differences may be negligible, less than 0.5 ◦C, during cloudyweather, while the differences may be more than 20 ◦C in clear sky conditions during winter in Lapland,largely because of topographic variations in the region.
2.2. Methods
In modelling the relationship between the number of daily deaths and daily mean temperaturesin the hospital districts, we applied the distributed lag non-linear model (DLNM) and quasi-Poissondistribution for the number of deaths [19,20,33]. The general model definition is:
g(µt) = α+ s(xt;β) +J
∑j=1
hj(cti;γi)
where g is a log link function of the expectation µt ≡ E(Yt), with Yt being the time series daily mortalitycounts in hospital district, α is an intercept, s(xt;β) is an exposure-response function to temperature(xt) defined by β and it is chosen as quadratic B-spline defined by internal knots. In addition,the cross-basis matrix of coefficients also describes lag effects of temperature, defined by knots forlag on a logarithmic scale. Here, we studied delayed temperature effects on mortality with a lag ofup to 25 days. Confounding factors (cti) in this study were day of the week and elapsed time fromthe beginning of the time series. Day of the week is modelled as a categorical variable and elapsedtime as a natural cubic spline with 7 degrees of freedom (df) per year to control seasonal variation andlong-term trend.
We applied different versions of DLNM, with varying lags and internal knots and with orwithout confounding factors. A simple version of the model, with no delayed temperature impacts(lag = 0), no confounding factors, and 4 internal knots for temperature distribution, was first appliedfor each hospital district (HD) separately. After this first-stage modelling, we studied potentialregional differences in mortality–temperature relationship between the hospital districts with the aidof meta-regression analysis, following the method developed by Gasparrini et al. [34]. In orderto reduce the effects of random variation in the relationships, especially in the less populatedHDs, best linear unbiased predictions (BLUP) were applied. These BLUP estimates convergethe HD-specific relationships towards a pooled, averaged exposure–response relationship. In themeta-analysis, Cochran Q test and I2 were used to study heterogeneity across the BLUP estimatesof the relationships in hospital districts. Because temperature ranges deviate between the hospitaldistricts, the meta-regression was done on both absolute and relative temperature scales.
We used selected characteristics of the hospital districts as covariates in meta-regression in orderto assess their effect in explaining potential heterogeneity in temperature-mortality relationshipsbetween hospital districts. These covariates were climatological mean temperature, ranges of dailymean temperatures, morbidity indices, population, and share of elderly (75 years of age and older)in the hospital districts. LR test and Wald test were also applied to study the significance of thesecovariates to explain heterogeneity.
For sensitivity analysis of the modelled temperature-mortality relationship, we used three slightlydifferent versions of the simple model: (i) three knots for temperature range; (ii) a model with theconfounding effects of weekday and seasonal variation and long-term trend; (iii) a model with anexposure term of a moving average of daily mean temperature on the same and five previous days.The last analysis covers part the lagged effects of the heat and cold exposure.
For studying the impacts of temperature on mortality with long delay, we applied a more complexDLNM with lag up to 25 days. Here, the first modelling was performed using the lowest AkaikeInformation Criteria (AIC) value as a criterion for selecting the number of knots for the model in eachhospital district. Thereby, the lag structure and optimal knots for the temperature distribution couldbe determined. The AIC-based number of knots varied from only one knot for both temperature andlag, to five knots for temperature and three knots for lag. On average, the best models had three knots
Int. J. Environ. Res. Public Health 2018, 15, 406 5 of 13
for temperature and two knots for lag, and in the following step the modelling was done using thesefixed numbers of knots for each hospital district.
The temperature–mortality relationships were concretized by reporting the relative risks ofmortality (RR), with 95% confidence intervals (CI) at selected values (+24, +20, –15, –20, and –25 ◦C)of spatially averaged temperature (Tavg). The baseline for RR is the mortality at minimum mortalitytemperature. MMT and its 95% confidence intervals (CI) were determined with the help of a simulationmethod developed in Tobias et al. [7]. Furthermore, the lag distributions were drawn from the DLNMoutput. The R packages dlnm [20] and mvmeta [34] were used to conduct the studies.
3. Results
3.1. Model without Lag and Meta-Analysis
The simple model version without lag produced U-shaped relationships in all hospital districts,except for the two least-populated ones, which have populations of less than 50,000. Another district,HD22, located in an archipelago in the Baltic Sea, was left out from the subsequent analysis. In Figure 2b,hospital district-specific mortality–temperature relationships, based on the first-stage modelling,are presented together with the pooled, average relationship. The HD-specific, first-stage modellingresults show large differences and variation, especially on the cold temperature scale, but the BLUPestimation converges the relationships towards the average, and the differences between hospitaldistricts almost disappear (Figure 2c). Examples of how the BLUP estimations differ from the first-stagemodels and from the pooled average in the hospital districts are presented in Figure 2d,e. In theHelsinki–Uusimaa hospital district (HD1), all the three relationships are fairly similar for typicaltemperatures, but in the cold extreme of the temperature distribution (below −20 ◦C), the first stagemodel outcome is remarkable higher than the BLUB estimate. In the less populated HD12 in easternFinland, the BLUP estimate is drawn towards the average throughout the temperature scale morestrongly than in HD1.
Exposure to high or low temperatures increased the mortality risk: the pooled estimate for therelative risk (RR) of mortality at a daily mean temperature of 24 ◦C was 1.16 (95% CI 1.12–1.20) whencompared to mortality at 14 ◦C, which is the MMT of the pooled relationship. On the cold side, at adaily mean temperature of −20 ◦C, RR was 1.14 (95% CI 1.12–1.16). On the relative scale of daily meantemperature, the MMT was at the 79th percentile.
The pooled relationship had relatively narrow confidence intervals (Figure 2a), and the meta-analysisconfirmed that there was no statistically significant heterogeneity in mortality–temperature relationshipsamong the hospital districts when the analysis was made on an absolute temperature scale (Table 1).However, on the relative temperature scale, heterogeneity was found (Q test p = 0.029), and 21% ofvariation in the relationships would be explained by heterogeneity (I2 = 21.1%). According to the Waldtest, morbidity index and population in the hospital districts explain heterogeneity on a statisticallysignificant level, but the LR tests do not support these findings. The climatological mean temperatureand temperature range do not explain a considerable amount of heterogeneity. Figure 3 demonstratesthe small impacts of selected meta-regression covariates and the differences in mortality–temperaturerelationship at the 25th and 75th percentiles of the covariate. On the basis of this meta-analysis,we can conclude that, since there are no statistically significant differences in temperature–mortalityrelationships between hospital districts on the absolute temperature scale, the same relationship canbe applied in all parts of the country. On the other hand, the meta-analysis on the relative temperaturescale suggests that morbidity index and population in the hospital districts might explain some of thesmall regional heterogeneity of the temperature–mortality association.
The sensitivity analysis of the modelled temperature–mortality relationships supports the findingabove that there is no statistically significant heterogeneity in the temperature–mortality relationshipacross the hospital districts on the absolute temperature scale, while some heterogeneity is found onthe relative temperature scale. Sensitivity analysis was performed using the simple model relationship
Int. J. Environ. Res. Public Health 2018, 15, 406 6 of 13
with different parameterizations, and pooled outcomes of the results are presented in SupplementaryMaterials (Figure S1).Int. J. Environ. Res. Public Health 2018, 15, x FOR PEER REVIEW 6 of 13
(a)
(b) (c)
(d) (e)
Figure 2. Temperature‐mortality relationships based on the model without lag. (a) Pooled, average
relationship (solid curve) with 95% confidence interval (shaded), dots indicating location of knots
(up), relative risk (RR) reference at T = 14 °C, (b) hospital district (HD)‐specific models (dash curves)
with the pooled relationship (solid curve), (c) best linear unbiased predictions (BLUP) estimations for
hospital districts. Examples on how BLUP estimation pulls the relationships towards the average in
two hospital districts: (d) HD1, in southern Finland and (e) HD12, in eastern Finland.
Figure 2. Temperature-mortality relationships based on the model without lag. (a) Pooled, averagerelationship (solid curve) with 95% confidence interval (shaded), dots indicating location of knots (up),relative risk (RR) reference at T = 14 ◦C; (b) hospital district (HD)-specific models (dash curves) with thepooled relationship (solid curve); (c) best linear unbiased predictions (BLUP) estimations for hospitaldistricts. Examples on how BLUP estimation pulls the relationships towards the average in two hospitaldistricts: (d) HD1, in southern Finland and (e) HD12, in eastern Finland.
Int. J. Environ. Res. Public Health 2018, 15, 406 7 of 13
Table 1. Heterogeneity tests on absolute and relative temperature scales. The hospital district(HD)-level meta-analysis covariates are the following: long-term climatological mean temperature;range of daily mean temperatures, THL’s morbidity index, population, and share of elderly (75+ years).Statistically significant p-values with bold.
CovariateCochran Q Test I2 Information
Criteria LR Test Wald Test
Q df p (%) AIC BIC Stat df p Stat df p
Absolute scaleIntercept-only 129.2 114 0.156 11.8 −304.8 −229.6Relative scaleIntercept-only 144.4 114 0.029 21.1 −301.0 −225.7
Climate-Tmean 134.8 108 0.041 19.9 −294.0 −202.0 5.0 6 0.544 5.6 6 0.465Temperature range 134.1 108 0.045 19.5 −297.6 −205.6 8.6 6 0.196 11.0 6 0.089
Morbidity index 129.1 108 0.081 16.4 −298.5 −206.5 9.5 6 0.146 13.5 6 0.036Population 127.5 108 0.097 15.3 −298.5 −206.5 9.5 6 0.146 14.9 6 0.021
Share of elderly 131.3 108 0.063 17.7 −295.8 −203.9 6.9 6 0.335 11.4 6 0.077
Int. J. Environ. Res. Public Health 2018, 15, x FOR PEER REVIEW 7 of 13
Table 1. Heterogeneity tests on absolute and relative temperature scales. The hospital district (HD)‐
level meta‐analysis covariates are the following: long‐term climatological mean temperature; range
of daily mean temperatures, THL’s morbidity index, population, and share of elderly (75+ years).
Statistically significant p‐values with bold.
Covariate Cochran Q Test I2 Information Criteria LR Test Wald Test
Q df p (%) AIC BIC Stat df Stat df
Absolute scale
Intercept‐only 129.2 114 0.156 11.8 −304.8 −229.6
Relative scale
Intercept‐only 144.4 114 0.029 21.1 −301.0 −225.7
Climate‐Tmean 134.8 108 0.041 19.9 −294.0 −202.0 5.0 6 0.544 5.6 6 0.465
Temperature range 134.1 108 0.045 19.5 −297.6 −205.6 8.6 6 0.196 11.0 6 0.089
Morbidity index 129.1 108 0.081 16.4 −298.5 −206.5 9.5 6 0.146 13.5 6 0.036
Population 127.5 108 0.097 15.3 −298.5 −206.5 9.5 6 0.146 14.9 6 0.021
Share of elderly 131.3 108 0.063 17.7 −295.8 −203.9 6.9 6 0.335 11.4 6 0.077
Figure 3. Meta‐analysis of the temperature–mortality relationship on the relative temperature scale.
Modelled relationships with the 25th and 75th percentiles of hospital district‐level covariates:
climatological mean temperature (°C), temperature range (°C), share of elderly (%), and morbidity
index.
3.2. Model with 25‐Day Lag
The modelling with the more complex DLNM version with 25‐day lag, day of the week, seasonal
variation, and long‐term trend as confounding factors, gave realistic U‐shaped relationships in most
of the hospital districts, even in those with a low population. A downside is that increased complexity
in the DLNM version led to increased uncertainty in the model outcomes. In a few hospital districts,
the modelled relationships appeared unrealistic, e.g., with continuously decreasing mortality with
increasing temperature. Examples of 3D visualizations of the DLNM outputs and overall effects of
temperature on the relative risk, aggregated from the 25‐day delayed effects, are presented from two
hospital districts in Figure 4. These hospital districts represent different climatic conditions. Helsinki–
Uusimaa, HD1, is located in southern Finland, and its climate is affected by the Baltic Sea. During
Figure 3. Meta-analysis of the temperature–mortality relationship on the relative temperature scale.Modelled relationships with the 25th and 75th percentiles of hospital district-level covariates: climatologicalmean temperature (◦C), temperature range (◦C), share of elderly (%), and morbidity index.
3.2. Model with 25-Day Lag
The modelling with the more complex DLNM version with 25-day lag, day of the week,seasonal variation, and long-term trend as confounding factors, gave realistic U-shaped relationshipsin most of the hospital districts, even in those with a low population. A downside is that increasedcomplexity in the DLNM version led to increased uncertainty in the model outcomes. In a few hospitaldistricts, the modelled relationships appeared unrealistic, e.g., with continuously decreasing mortalitywith increasing temperature. Examples of 3D visualizations of the DLNM outputs and overall effectsof temperature on the relative risk, aggregated from the 25-day delayed effects, are presented fromtwo hospital districts in Figure 4. These hospital districts represent different climatic conditions.Helsinki–Uusimaa, HD1, is located in southern Finland, and its climate is affected by the Baltic Sea.
Int. J. Environ. Res. Public Health 2018, 15, 406 8 of 13
During our study period, the daily mean temperature as a spatial average in the HD1 varied between−24.5 ◦C and +25.6 ◦C, with a median of 5.7 ◦C. Pohjois-Karjala, HD12, in eastern central Finland,has a more continental type of climate with a wider temperature variation range: the daily meantemperature varied between −34.1 ◦C and +26.9 ◦C, with a median of 3.2 ◦C.
Int. J. Environ. Res. Public Health 2018, 15, x FOR PEER REVIEW 8 of 13
our study period, the daily mean temperature as a spatial average in the HD1 varied between −24.5
°C and +25.6 °C, with a median of 5.7 °C. Pohjois‐Karjala, HD12, in eastern central Finland, has a
more continental type of climate with a wider temperature variation range: the daily mean
temperature varied between −34.1 °C and +26.9 °C, with a median of 3.2 °C.
(a) (b)
(c) (d)
Figure 4. Outcomes of DLNM for hospital districts HD1 and HD12 with a model version with 25‐day
delayed temperature effects: (a,c) 3D visualizations of the relative risk of mortality as a function of
temperature and lag; (b,d) overall relative risk (curves) with 95% CI (shaded). The point of minimum
mortality temperature is indicated with a solid vertical line (95% CI, dotted vertical lines). HD1 is in
southern and HD12 in eastern‐central Finland. Modelling with fixed number of knots: 3 for
temperature and 2 for lag.
Delayed impacts had common characteristics in most of DLNM outcomes: increase in the heat‐
related RR was apparent immediately on the same day and it disappeared within a few days. The
cold‐related RR appeared after a few days but lasted for several days or even weeks. Aggregating
delayed impacts over 25 days increased the cold impact more than the heat impact. Examples of the
structure of the lagged impacts of heat stress (at T = +24 °C) and cold stress (at T = −20 °C) on the all‐
aged mortality in HD1 are presented in Figure 5.
Figure 4. Outcomes of DLNM for hospital districts HD1 and HD12 with a model version with 25-daydelayed temperature effects: (a,c) 3D visualizations of the relative risk of mortality as a function oftemperature and lag; (b,d) overall relative risk (curves) with 95% CI (shaded). The point of minimummortality temperature is indicated with a solid vertical line (95% CI, dotted vertical lines). HD1 is insouthern and HD12 in eastern-central Finland. Modelling with fixed number of knots: 3 for temperatureand 2 for lag.
Delayed impacts had common characteristics in most of DLNM outcomes: increase in theheat-related RR was apparent immediately on the same day and it disappeared within a fewdays. The cold-related RR appeared after a few days but lasted for several days or even weeks.Aggregating delayed impacts over 25 days increased the cold impact more than the heat impact.Examples of the structure of the lagged impacts of heat stress (at T = +24 ◦C) and cold stress (at T = −20 ◦C)on the all-aged mortality in HD1 are presented in Figure 5.
Examples of how inclusions of the delayed impacts may affect the estimated relative risks ofall-aged mortality at various temperature levels are in Table 2 for two hospital districts. The outcomesof three model versions are presented: (i) the pooled model results without lag and without seasonalvariation; (ii) hospital district-specific simple models without lag and without seasonal variation;(iii) a complex DLNM model with 25-day lag, seasonal variation, and a fixed number of knots fortemperature and for lag. Our preliminary results suggest that including delayed impacts with long lag
Int. J. Environ. Res. Public Health 2018, 15, 406 9 of 13
may double or triple the mortality risks, but since the meta-analysis of the complex model version wasbeyond this study, we cannot make firm conclusions from this model version.Int. J. Environ. Res. Public Health 2018, 15, x FOR PEER REVIEW 9 of 13
(a) (b)
Figure 5. Examples of the effects of lag on relative mortality risk (RR) at two different temperatures,
(a) Tavg = 24 °C, representing heat stress and (b) Tavg = −20 °C, cold stress. Examples of all‐aged
mortality in HD1, Helsinki–Uusimaa hospital district. DLNM with 3 knots for temperature and 2 for
lag. Tavg: spatially averaged daily mean temperature.
Examples of how inclusions of the delayed impacts may affect the estimated relative risks of all‐
aged mortality at various temperature levels are in Table 2 for two hospital districts. The outcomes
of three model versions are presented: (i) the pooled model results without lag and without seasonal
variation; (ii) hospital district‐specific simple models without lag and without seasonal variation; (iii)
a complex DLNM model with 25‐day lag, seasonal variation, and a fixed number of knots for
temperature and for lag. Our preliminary results suggest that including delayed impacts with long
lag may double or triple the mortality risks, but since the meta‐analysis of the complex model version
was beyond this study, we cannot make firm conclusions from this model version.
Table 2. Comparison of the relative mortality risk, RR (95% CI), in HD1 and HD12 at selected
temperatures in three model versions: pooled model without lag and seasonal variation, a simple
model without lag and seasonal variation, a DLNM with 25‐day lag and seasonal variation.
Tavg RR (95% CI)
Pooled Estimate 24 1.16 (1.12, 1.20)
‐Without Lag 20 1.04 (1.03, 1.05)
−15 1.13 (1.12, 1.15)
−20 1.14 (1.12, 1.16)
−25 1.16 (1.13, 1.19)
HD1 HD12
Simple Model 24 1.15 (1.10, 1.20) 1.25 (1.14, 1.38)
‐Without Lag 20 1.06 (1.04, 1.08) 1.10 (1.05, 1.15)
−15 1.14 (1.11, 1.16) 1.21 (1.17, 1.26)
−20 1.14 (1.09, 1.18) 1.25 (1.20, 1.31)
−25 1.13 (1.06, 1.20) 1.29 (1.21, 1.39)
DLNM 24 1.34 (1.22, 1.48) 1.55 (1.24, 1.94)
‐With 25‐Day Lag 20 1.10 (1.06, 1.13) 1.18 (1.07, 1.31)
−15 1.20 (1.09, 1.31) 1.26 (1.05, 1.52)
−20 1.25 (1.11, 1.39) 1.34 (1.07, 1.68)
−25 1.32 (1.00, 1.74) 1.37 (1.09, 1.72)
4. Discussion
We conducted regional assessments of the temperature–mortality relationship in hospital
districts in Finland by using different versions of distributed lag non‐linear models (DLNM) and
meta‐analysis. Finland is a sparsely populated country, which generates challenges and large
uncertainties in modelling. Earlier studies have focused on the capital city of Helsinki [11,23] or on
larger areas such as southern Finland or northern Finland [5,8,9].
Figure 5. Examples of the effects of lag on relative mortality risk (RR) at two different temperatures,(a) Tavg = 24 ◦C, representing heat stress and (b) Tavg = −20 ◦C, cold stress. Examples of all-agedmortality in HD1, Helsinki–Uusimaa hospital district. DLNM with 3 knots for temperature and 2 forlag. Tavg: spatially averaged daily mean temperature.
Table 2. Comparison of the relative mortality risk, RR (95% CI), in HD1 and HD12 at selectedtemperatures in three model versions: pooled model without lag and seasonal variation, a simplemodel without lag and seasonal variation, a DLNM with 25-day lag and seasonal variation.
Tavg RR (95% CI)
Pooled Estimate 24 1.16 (1.12, 1.20)-Without Lag 20 1.04 (1.03, 1.05)
−15 1.13 (1.12, 1.15)−20 1.14 (1.12, 1.16)−25 1.16 (1.13, 1.19)
HD1 HD12
Simple Model 24 1.15 (1.10, 1.20) 1.25 (1.14, 1.38)-Without Lag 20 1.06 (1.04, 1.08) 1.10 (1.05, 1.15)
−15 1.14 (1.11, 1.16) 1.21 (1.17, 1.26)−20 1.14 (1.09, 1.18) 1.25 (1.20, 1.31)−25 1.13 (1.06, 1.20) 1.29 (1.21, 1.39)
DLNM 24 1.34 (1.22, 1.48) 1.55 (1.24, 1.94)-With 25-Day Lag 20 1.10 (1.06, 1.13) 1.18 (1.07, 1.31)
−15 1.20 (1.09, 1.31) 1.26 (1.05, 1.52)−20 1.25 (1.11, 1.39) 1.34 (1.07, 1.68)−25 1.32 (1.00, 1.74) 1.37 (1.09, 1.72)
4. Discussion
We conducted regional assessments of the temperature–mortality relationship in hospital districtsin Finland by using different versions of distributed lag non-linear models (DLNM) and meta-analysis.Finland is a sparsely populated country, which generates challenges and large uncertainties inmodelling. Earlier studies have focused on the capital city of Helsinki [11,23] or on larger areassuch as southern Finland or northern Finland [5,8,9].
To our knowledge, this is the first attempt to model the temperature–mortality relationshipsat the hospital district level, i.e., at the regional administrative level of health services in Finland.The characteristics—e.g., area, population, public health, and climate—of hospital districts vary
Int. J. Environ. Res. Public Health 2018, 15, 406 10 of 13
substantially, and our aim was to assess how heat- and cold-related health risks vary betweenhospital districts.
Following the conclusions in Ruuhela et al. [28], we used the spatially averaged daily meantemperature in each hospital district as an indicator for thermal environment. Spatially averagedtemperatures describe the thermal environment in a larger area better than single station-wisetemperatures, since they include the topographic effects on the temperature variation in the area.We used 10 km × 10 km gridded climate data from the Finnish Meteorological Institute [30].
We found that modelling at the hospital district level gives realistic, U-shaped relationships evenin sparsely populated areas, but with some limitations: the outcomes of the model version without lagwere unrealistic in two hospital districts, possibly because of a small population, of less than 50,000,in those hospital districts. When complexity in modelling was increased by taking into account 25-daylag impacts of heat and cold, the uncertainties in the temperature-mortality relationships increased,and their shapes became unrealistic in a few more hospital districts in the northern and western partof the country. However, in regions where the modelling with a 25-day lag is possible, these modelscan provide a better understanding of the overall health impacts of heat and cold than the modelwithout lag.
Studies that have found heterogeneity in temperature–mortality relationships were made in largerareas such as Europe or the USA [11–13]. Our study concentrated only on one country, and we could notconfirm heterogeneity across Finland on the absolute temperature scale. According to the meta-analysisand BLUP estimates of temperature–mortality relationships in the hospital districts, the smalldifferences in relationships across hospital districts were not statistically significant and cannot beexplained by climatic or sociodemographic factors. Therefore, a pooled, average mortality–temperaturerelationship can be applied throughout the country, e.g., in climate change impact studies.
However, on the relative temperature scale, some heterogeneity was detected in thetemperature-mortality association between the hospital districts. Sociodemographic factors suchas morbidity index and population might partly explain these differences. Yet, we cannot firmlyconclude these outcomes because of the contradicting outcomes of the statistical tests.
The minimum mortality temperature (MMT) in Finland is clearly lower than among populationsin warmer climates. Based on the pooled, average temperature–mortality relationship, the MMT inFinland is 14 ◦C. In a relative temperature scale, the MMT is found at the 79th percentile of the daily meantemperature distribution, thus somewhat higher than in studies from other countries, (e.g., [6]). We didnot find regional differences in MMT, thus we cannot conclude that there would be regional differencesin acclimatization in the Finnish population either. One explanation for this conclusion could beinternal migration, which is substantial in Finland, mainly towards the south. However, in their studyabout internal migration and mortality, Saarela and Finnäs [35] found that the birth region is a verydecisive mortality determinant. One limitation in our study was that we were not able to conductresearch at the individual level and track the place of birth for each individual.
The relative risk (RR) of mortality in the hot tail of the temperature distribution appeared quitesimilar regardless of the DLNM version. In the cold tail of the temperature distribution, the variationand uncertainty in the modelled relationship was larger than in the hot tail, at least partly because of thelong-lasting delayed impacts, as can be seen in Figure 5b. In this study, we focused on studying regionaldifferences in the temperature–mortality relationship and finding the location of MMT, but modellingwith various versions of DLNM model provided interesting outcomes for further studies anddiscussions on cold-related mortality. The simple model without lag or controlling seasonality producesclassic U-shaped association between temperature and mortality. When seasonality was included intothe model as a confounding factor, the cold tail of the association became flat, thus mortality risk didnot appear to depend on temperature on the cold side. When delayed impacts were included into themodel, the cold-related mortality risk was again elevated (Figure S1). When extending the lag up to25 days, the cold impacts become more pronounced (Figure 4).
Int. J. Environ. Res. Public Health 2018, 15, 406 11 of 13
These various model versions indicate that more research is needed on the winter mortality andimpacts of cold. The study of Ebi and Mills [36] questioned the assumption that seasonality withhigher mortality in winter than in summer would be attributable to temperature. Higher wintermortality may also be related to other factors that vary seasonally, such as influenza epidemics [37].A climatologically interesting candidate to partly explain excess winter mortality, especially in highlatitude countries, might be vitamin D concentration, which typically varies seasonally depending onsolar radiation. In the meta-analysis of Rush et al. [38], vitamin D status was inversely associated withall-cause mortality. Similar conclusion was reported also in Finland [39].
According to the morbidity index of the National Institute of Health and Welfare, there aresignificant regional differences in public health in Finland [26]. We expected that these regional healthdifferences would reflect into the temperature–mortality relationships as well, but our study did notfully support that. In the future, the regional differences in health risks due to heat and cold couldbe studied using, as additional explanatory variables, also other relevant socioeconomic and healthindicators, such as the incidences of weather-sensitive chronic diseases. However, the small number ofcases would limit the statistical power of such studies.
In areas of northern Finland with larger differences in elevation, the spatially aggregatedtemperature data may not describe well the exposure to heat or cold stress, because valleys aremore populated than higher-altitude areas. One way forward in modelling the mortality–temperaturerelationship might be using population-weighted gridded temperature data.
The outcomes of this study can be used, e.g., in increasing the awareness of temperature-relatedhealth risks in high-latitude countries and, specifically, in discussions on the potential benefits of earlywarning systems in the health sector in Finland.
5. Conclusions
Modelling temperature dependence of mortality by DLNM without delayed impacts producedU-shaped relationships even in sparsely populated areas if the population is over 50,000. The DLNMwith a 25-day lag also produced U-shaped relationships in most of the hospital districts and couldthus provide a better understanding of the delayed health impacts of heat and cold than the modelwithout delayed effects.
We did not find statistically significant differences in the temperature–mortality relationships betweenhospital districts on the absolute temperature scale, thus no differences in acclimatization betweenhospital district populations were found. Therefore, a pooled national average temperature–mortalityrelationship can be applied throughout the country in studies about future climate change impacts ontemperature-related mortality in Finland. However, on the relative temperature scale, some heterogeneitywas found, and the meta-regression suggested that morbidity index and population in the hospitaldistricts might explain part of the heterogeneity.
Supplementary Materials: The following are available online at http://www.mdpi.com/1660-4601/15/3/406/s1,Table S1: Characteristics of the hospital districts in 2000–2014: daily number of deaths, population, shareof elderly, morbidity index, and spatially averaged daily temperature in the hospital districts, Figure S1:Pooled temperature–mortality relationships with different model parameters.
Acknowledgments: The research presented in this paper is part of the PLUMES project (Pathways LinkingUncertainties in Model Projections of Climate and its Effects, the Academy of Finland decision number 278067),iSCAPE project (Improving the Smart Control of Air Pollution in Europe, H2020-SC5-2015, grant agreementnumber 689954) and URCLIM project (Urban Climate Services), which has received funding from the EuropeanUnion’s Horizon 2020 research and innovation programme under grant agreement number 690462. It alsocontributes to the Nordic Centre of Excellence for Resilience and Societal Security-NORDRESS-which is fundedby the Nordic Societal Security Programme. We thank Pentti Pirinen for assisting with FMI climate data andCurtis Wood for providing language help. Furthermore, we thank the anonymous reviewers, whose commentsand suggestions remarkably helped us improve and clarify the manuscript.
Author Contributions: Reija Ruuhela made all the analysis and wrote the paper. Kirsti Jylhä contributed to themanuscript as a climate expert and Otto Hyvärinen as an expert of statistical methods in climate impacts studies.
Conflicts of Interest: The authors declare no conflict of interest.
Int. J. Environ. Res. Public Health 2018, 15, 406 12 of 13
References
1. Basu, R.; Samet, J.M. Relation between elevated ambient temperature and mortality: A review of theepidemiologic evidence. Epidemiol. Rev. 2002, 24, 190–202. [CrossRef] [PubMed]
2. Armstrong, B. Models for the relationship between ambient temperature and daily mortality. Epidemiology2006, 17, 624–631. [CrossRef] [PubMed]
3. Kovats, R.S.; Hajat, S. Heat stress and public health: A critical review. Annu. Rev. Public Health 2008, 29, 41–55.[CrossRef] [PubMed]
4. Anderson, B.G.; Bell, M.L. Weather-Related Mortality. Epidemiology 2009, 20, 205–213. [CrossRef] [PubMed]5. Keatinge, W.R.; Donaldson, G.C.; Cordioli, E.; Martinelli, M.; Kunst, A.E.; Mackenbach, J.P.; Näyhä, S.; Vuori, I.
Heat related mortality in warm and cold regions of Europe: Observational study. BMJ 2000, 321, 670–673.[CrossRef] [PubMed]
6. Guo, Y.; Gasparrini, A.; Armstrong, B.; Li, S.; Tawatsupa, B.; Tobias, A.; Lavigne, E.; Zanotti, D.S.; Coelho, S.;Leone, M.; et al. Global Variation in the Effects of Ambient Temperature on Mortality A Systematic Evaluation.Epidemiology 2014, 25, 781–789. [CrossRef] [PubMed]
7. Tobias, A.; Armstrong, B.; Gasparrini, A. Investigating Uncertainty in the Minimum Mortality Temperature.Methods and Application to 52 Spanish Cities. Epidemiology 2017, 28, 72–76. [CrossRef] [PubMed]
8. Näyhä, S. Environmental temperature and mortality. Int. J. Circumpolar Health 2005, 64, 451–458. [CrossRef][PubMed]
9. Näyhä, S. Heat mortality in Finland in the 2000s. Int. J. Circumpolar Health 2007, 66, 418–424. [CrossRef][PubMed]
10. Curriero, F.C.; Heiner, K.S.; Samet, J.M.; Zeger, S.L.; Strug, L.; Patz, J.A. Temperature and mortality in 11 cities ofthe eastern United States. Am. J. Epidemiol. 2002, 155, 80–87. [CrossRef] [PubMed]
11. Baccini, M.; Biggeri, A.; Accetta, G.; Kosatsky, T.; Katsouyanni, K.; Analitis, A.; Anderson, H.R.; Bisanti, L.;D’Iippoliti, D.; Danova, J.; et al. Heat effects on mortality in 15 European cities. Epidemiology 2008, 19, 711–719.[CrossRef] [PubMed]
12. D’Ippoliti, D.; Michelozzi, P.; Marino, C.; Menne, B.; Cabre, M.G.; Katsouyanni, K.; Medina, S.; Paldy, A.;Anderson, H.R.; Ballester, F. The Impact of Heat Waves on Mortality in 9 European Cities, 1990–2004.Epidemiology 2008, 19, S286–S287.
13. Medina-Ramón, M.; Schwartz, J. Temperature, temperature extremes, and mortality: A study of acclimatisationand effect modification in 50 US cities. Occup. Environ. Med. 2007, 64, 827–833. [CrossRef] [PubMed]
14. Hondula, D.M.; Davis, R.E.; Saha, M.V.; Wegner, C.R.; Veazey, L.M. Geographic dimensions of heat-relatedmortality in seven U.S. cities. Environ. Res. 2015, 138, 439–452. [CrossRef] [PubMed]
15. Yu, W.; Mengersen, K.; Wang, X.; Ye, X.; Guo, Y.; Pan, X.; Tong, S. Daily average temperature and mortalityamong the elderly: A meta-analysis and systematic review of epidemiological evidence. Int. J. Biometeorol.2012, 56, 569–581. [CrossRef] [PubMed]
16. Rocklov, J.; Forsberg, B. The effect of temperature on mortality in Stockholm 1998–2003: A study of lagstructures and heatwave effects. Scand. J. Public Health 2008, 36, 516–523. [CrossRef] [PubMed]
17. Koppe, C.; Jendritzky, G. Inclusion of short-term adaptation to thermal stresses in a heat load warningprocedure. Meteorol. Z. 2005, 14, 271–278. [CrossRef]
18. Muggeo, V.M.; Hajat, S. Modelling the non-linear multiple-lag effects of ambient temperature on mortalityin Santiago and Palermo: A constrained segmented distributed lag approach. Occup. Environ. Med. 2009, 66,584–591. [CrossRef] [PubMed]
19. Gasparrini, A.; Armstrong, B.; Kenward, M.G. Distributed lag non-linear models. Stat. Med. 2010, 29, 2224–2234.[CrossRef] [PubMed]
20. Gasparrini, A. Distributed Lag Linear and Non-Linear Models in R: The Package dlnm. J. Stat. Softw. 2011, 43,1–20. [CrossRef] [PubMed]
21. Gasparrini, A.; Scheipl, F.; Armstrong, B.; Kenward, M.G. A penalized framework for distributed lag non-linearmodels. Biometrics 2017, 938–948. [CrossRef] [PubMed]
22. Gasparrini, A.; Guo, Y.; Hashizume, M.; Lavigne, E.; Zanobetti, A.; Schwartz, J.; Tobias, A.; Tong, S.; Rocklöv, J.;Forsberg, B.; et al. Mortality risk attributable to high and low ambient temperature: A multicountry observationalstudy. Lancet 2015, 386, 369–375. [CrossRef]
Int. J. Environ. Res. Public Health 2018, 15, 406 13 of 13
23. De’Donato, F.K.; Leone, M.; Scortichini, M.; De Sario, M.; Katsouyanni, K.; Lanki, T.; Basagaña, X.; Ballester, F.;Åström, C.; Paldy, A.; et al. Changes in the effect of heat on mortality in the last 20 years in nine Europeancities. Results from the PHASE project. Int. J. Environ. Res. Public Health 2015, 12, 15567–15583. [CrossRef][PubMed]
24. Oudin Åström, D.; Åström, C.; Rekker, K.; Indermitte, E.; Orru, H. High summer temperatures and mortalityin Estonia. PLoS ONE 2016, 11, e0155045. [CrossRef] [PubMed]
25. Carter, T.R.; Fronzek, S.; Inkinen, A.; Lahtinen, I.; Lahtinen, M.; Mela, H.; O’Brien, K.L.; Rosentrater, L.D.;Ruuhela, R.; Simonsson, L.; et al. Characterising vulnerability of the elderly to climate change in the Nordicregion. Reg. Environ. Chang. 2016, 16, 43–58. [CrossRef]
26. THL’s Morbidity Index 2012–2014. Statistical Review 12/2016. Available online: http://urn.fi/URN:NBN:fi-fe2016111829194 (accessed on 25 February 2018).
27. Jendritzky, G.; de Dear, R. Adaptation and Thermal Environment. In Biometeorology for Adaptation to ClimateVariability and Change; Ebi, K.L., Burton, I., McGregor, G.R., Eds.; Springer: Dordrecht, The Netherlands, 2009;pp. 9–32.
28. Ruuhela, R.; Jylhä, K.; Lanki, T.; Tiittanen, P.; Matzarakis, A. Biometeorological assessment of mortalityrelated to extreme temperatures in Helsinki region, Finland, 1972–2014. Int. J. Environ. Res. Public Health2017, 14, 944. [CrossRef] [PubMed]
29. Pirinen, P.; Simola, H.; Aalto, J.; Kaukoranta, J.-P.; Karlsson, P.; Ruuhela, R. Tilastoja Suomen ilmastosta1981–2010—Climatological statistics of Finland 1981–2010. In Ilmatieteen Laitos Raportteja—Finnish MeteorologicalInstitute Reports; Finnish Meteorological Institute: Helsinki, Finland, 2012; Volume 1, ISBN 9789516977655.
30. Statistical Information on Welfare and Health in Finland. Available online: https://www.sotkanet.fi/sotkanet/en/kartta?indicator=szYKBwA=®ion=s7YssM7SM4y3BAA=&year=sy4rBQA=&gender=t(accessed on 22 November 2017).
31. THL’s Morbidity Index. Available online: https://www.sotkanet.fi/sotkanet/en/taulukko/?indicator=szYOAgA=®ion=s7YssM7SM4y3BAA=&year=sy4rtc7X0zUEAA==&gender=t&abs=f&color=f&buildVersion=3.0-SNAPSHOT&buildTimestamp=201709141202 (accessed on 9 January 2018).
32. Aalto, J.; Pirinen, P.; Heikkinen, J.; Venäläinen, A. Spatial interpolation of monthly climate data for Finland:Comparing the performance of kriging and generalized additive models. Theor. Appl. Climatol. 2013, 112,99–111. [CrossRef]
33. Gasparrini, A.; Leone, M. Attributable risk from distributed lag models. BMC Med. Res. Methodol. 2014, 14, 55.[CrossRef] [PubMed]
34. Gasparrini, A.; Armstrong, B.; Kenward, M.G. Multivariate meta-analysis for non-linear and othermulti-parameter associations. Stat. Med. 2012, 31, 3821–3839. [CrossRef] [PubMed]
35. Saarela, J.; Finnäs, F. Internal migration and mortality: The case of Finland. Environ. Health Insights 2008, 2, 1–12.[CrossRef] [PubMed]
36. Ebi, K.L.; Mills, D. Winter mortality in a warming climate: A reassessment. Wiley Interdiscip. Rev. Clim. Chang.2013, 4, 203–212. [CrossRef]
37. Staddon, P.L.; Montgomery, H.E.; Depledge, M.H. Climate warming will not decrease winter mortality.Nat. Clim. Chang. 2014, 4, 190–194. [CrossRef]
38. Rush, L.; McCartney, G.; Walsh, D.; Mackay, D. Vitamin D and subsequent all-age and premature mortality:A systematic review. BMC Public Health 2013, 13, 679. [CrossRef] [PubMed]
39. Virtanen, J.K.; Nurmi, T.; Voutilainen, S.; Mursu, J.; Tuomainen, T.P. Association of serum 25-hydroxyvitaminD with the risk of death in a general older population in Finland. Eur. J. Nutr. 2011, 50, 305–312. [CrossRef][PubMed]
© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open accessarticle distributed under the terms and conditions of the Creative Commons Attribution(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Supplementary Materials:
Regional assessment of temperature‒related mortality in Finland
Table S1. Characteristics of the hospital districts in 2000–2014: daily number of deaths, population, share of
elderly (75 years and older), morbidity index (2001–2014) and spatially averaged daily temperature in the
hospital districts
Reija Ruuhela, Otto Hyvärinen and Kirsti Jylhä
Table S1. Characteristics of the hospital districts in 2000–2014: daily number of deaths, population, share
of elderly, morbidity index and spatially averaged daily temperature in the hospital districts.
Figure S1. Pooled temperature‒mortality relationships with different model parameters.
Daily deaths, all Population, all
Share of elderly,
75+ years [%] Morbidity index Daily mean T [°C]
Hospital district Median Min Max Median Min Max Median Min Max Mean Min Max Median Min Max
HD1, Helsinki-Uusimaa 30 11 57 1471550 1375772 1599390 5.6 5.0 6.4 83.0 76.7 88.0 5.7 –24.5 25.4
HD3, Varsinais-Suomi 12 2 27 461928 449316 449316 8.5 7.3 9.9 108.6 94.9 116.1 5.7 –25.2 25.7
HD4, Satakunta 7 0 20 227590 223983 234228 9.4 7.7 10.4 108.2 103.4 112.3 5.3 –26.6 25.5
HD5, Kanta-Häme 5 0 16 170703 165190 175481 8.9 7.7 9.7 102.4 100.2 105.0 5.1 –27.3 25.7
HD6, Pirkanmaa 12 1 27 472015 443763 500265 7.9 6.8 8.7 101.8 98.2 104.6 4.7 –28.9 25.6
HD7, Päijät-Häme 6 0 17 210934 208837 213542 8.4 6.8 9.7 116.2 107.6 123.1 4.9 –28.0 25.5
HD8, Kymenlaakso 5 0 16 176675 172908 181119 9.4 7.8 10.4 119.1 111.8 125.7 5.2 –28.1 26.2
HD9, Etelä-Karjala 4 0 13 134169 131764 136524 9.5 7.7 10.8 113.1 105.2 117.4 4.6 –30.9 26.9
HD10, Etelä-Savo 3 0 13 108212 103873 113157 10.1 8.4 11.0 128.3 123.3 130.7 4.4 –28.8 26.4
HD11, Itä-Savo 1 0 8 47083 44051 50110 10.9 9.1 11.4 126.3 122.6 130.0 4.2 –30.7 26.2
HD12, Pohjois-Karjala 5 0 16 171326 168896 177235 9.0 7.3 9.9 139.2 129.5 147.4 3.2 –34.9 26.9
HD13, Pohjois-Savo 7 0 18 249184 247943 255451 9.0 7.1 9.9 141.1 137.4 145.5 3.6 –30.8 27.8
HD14, Keski-Suomi 7 0 18 270457 265131 275360 8.1 6.6 9.2 111.7 106.6 121.9 4.0 –29.7 26.6
HD15, Etelä-Pohjanmaa 8 0 21 198832 198242 201999 9.6 8.0 10.3 115.4 111.2 118.2 4.2 –28.8 24.7
HD16, Vaasa 4 0 14 162864 161112 169652 9.2 8.2 9.7 93.9 87.2 97.4 4.7 –27.7 24.0
HD17, Keski-Pohjanmaa 2 0 9 74703 74180 75494 8.3 6.8 9.2 112.7 105.1 116.9 3.7 –32.4 24.8
HD18, Pohjois-Pohjanmaa 8 1 25 388683 369835 408536 6.4 5.2 7.4 121.7 110.9 127.6 2.4 –33.2 24.5
HD19, Kainuu 2 0 11 80477 76119 86940 9.2 7.1 9.8 134.2 131.4 135.9 2.2 –32.2 25.9
HD20, Länsi-Pohja 2 0 9 66042 63603 69240 8.5 6.6 9.3 131.2 116.9 141.6 2.1 –33.4 24.6
HD21, Lappi 3 0 13 118620 118145 125105 7.7 5.6 9.3 119.3 115.2 125.4 0.5 –31.6 22.5
HD22, Ahvenanmaa 1 0 6 27038 25706 25706 8.3 8.3 9.7 75.6 65.7 83.1 6.2 –22.3 24.1
Figure S1. Pooled mortality-temperature relationships with different model parameters.
Exposure: Daily mean temperature (left), 5 day moving temperature average (right)
No confounding factors. Shaded areas represent 95% confidence intervals.
Confounding factors: day of the week and seasonal variation and long-term trend included;
Exposure: Daily mean temperature (left), 5 day moving temperature average (right)
Exposure: Daily mean temperature, model with 3 internal knots for temperature. No confounding
factors.
© 2009 Springer Nature Reprinted, with permission, from
International Journal of Biometeorology 53:167-175.
doi:10.1007/s00484-008-0200-5.
III
Erratum
In Table 2, the last time frame in the first column should be 1991–2003.
ORIGINAL PAPER
Climate impact on suicide rates in Finland from 1971 to 2003
Reija Ruuhela & Laura Hiltunen & Ari Venäläinen &
Pentti Pirinen & Timo Partonen
Received: 10 April 2008 /Revised: 16 November 2008 /Accepted: 24 November 2008 / Published online: 20 December 2008# ISB 2008
Abstract Seasonal patterns of death from suicide are well-documented and have been attributed to climatic factorssuch as solar radiation and ambient temperature. However,studies on the impact of weather and climate on suicide arenot consistent, and conflicting data have been reported. Inthis study, we performed a correlation analysis betweennationwide suicide rates and weather variables in Finlandduring the period 1971–2003. The weather parametersstudied were global solar radiation, temperature andprecipitation, and a range of time spans from 1 month to1 year were used in order to elucidate the dose-responserelationship, if any, between weather variables and suicide.Single and multiple linear regression models show weakassociations using 1-month and 3-month time spans, butrobust associations using a 12-month time span. Cumula-tive global solar radiation had the best explanatory power,while average temperature and cumulative precipitation hadonly a minor impact on suicide rates. Our results demon-strate that winters with low global radiation may increasethe risk of suicide. The best correlation found was for the5-month period from November to March; the inter-annualvariability in the cumulative global radiation for that periodexplained 40 % of the variation in the male suicide rate and14 % of the variation in the female suicide rate, both at astatistically significant level. Long-term variations in globalradiation may also explain, in part, the observed increasing
trend in the suicide rate until 1990 and the decreasing trendsince then in Finland.
Keywords Climate impact . Global solar radiation . Season .
Suicide .Weather impact
Introduction
Suicide is a public health problem in many Europeancountries, with an average suicide rate of 17.5 per 100,000in the WHO European Region (WHO 2005). In Finland,the average suicide rate was 20.3 per 100,000 in 2004 ascompared, for example, with 13.2 per 100,000 in Swedenin 2002, 11.5 per 100,000 in Norway in 2004, and 34.3 per100,000 in Russia in 2004. To visualise the impact, suicidewas ranked as the fourth leading cause of death among bothmen and women of working age in Finland in 2006. Inother words, in 2006, there were 663 and 206 deaths fromsuicide, representing 8.9% and 6.5% of all deaths ofindividuals aged 15 to 64 years. Suicide mortality variesbetween men and women in different parts of Finland,being highest for men in northern and north-eastern areas,and highest for women in the south of Finland, since the1970s (Partonen et al. 2003).
Suicide is a complex phenomenon, usually having morethan one risk factor such as mental illness, unemployment,low income and a family history of suicide (Qin et al. 2003;Mann et al. 2005). In addition, suicidal behaviour follows aseasonal pattern, with a peak in the spring and earlysummer in both the northern and southern hemispheres(Näyhä 1981; Hakko et al. 1998a; Rock et al. 2003;Partonen et al. 2004a; Rocchi et al. 2004; Bridges et al.2005; Zonda et al. 2005). For women, there seems to be asecond peak in the autumn (Meares et al. 1981; Näyhä
Int J Biometeorol (2009) 53:167–175DOI 10.1007/s00484-008-0200-5
R. Ruuhela (*) :A. Venäläinen : P. PirinenClimate Change Research, Finnish Meteorological Institute,P.O. Box 503, 00101 Helsinki, Finlande-mail: [email protected]
L. Hiltunen : T. PartonenNational Public Health Institute,Helsinki, Finland
1982; Micciolo et al. 1989). In addition, violent methods ofdeath from suicide appear to have a more seasonal patternthan other forms (Hakko et al. 1998b; Räsänen et al. 2002).
Weather varies from season to season, and influencespopulations living under these conditions. Weather has animpact on individuals in their habitat at the social,psychological and physiological levels, and this may beseen in a range of behaviours (Voracek et al. 2007). So far,however, no specific meteorological condition can belabelled as “suicide weather” (Dixon and Shulman 1983;Yan 2000; Deisenhammer 2003; Lee et al. 2006). Highertemperatures appear to increase the risk of suicide (Lee et al.2006; Page et al. 2007; Preti et al. 2007), but negativecorrelations have also been found (Souêtre et al. 1987,1989), as well as no correlation at all (Partonen et al. 2004b;Dixon et al. 2007). In Finland, drops in temperature onsuccessive days during spring have been found to be relatedto suicide rates in northern Finland (Partonen et al. 2004c).
Daily sunshine duration has a positive correlation withsuicide rates (Souêtre et al. 1987; Petridou et al. 2002;Lambert et al. 2003), contributing in part to the seasonaleffect on suicide mortality, which increases with increasinglatitudes (Parker et al. 2001; Heerlein et al. 2006). Sunlightmight possibly act as a trigger for suicide (Papadopouloset al. 2005), because an increase in daylight is specificallyable to stimulate the production of serotonin in the brain(Lambert et al. 2002), and among suicide victims there areindeed some abnormalities in serotonin metabolism in thebrain (Dwivedi et al. 2001; Pandey et al. 2002). However,negative correlations between daily sunshine duration andsuicide rates have also been reported (Stoupel et al. 1995;Preti 1998).
Most studies in this field analyse daily sunshineduration, whereas solar radiance has been used for analysisin only one earlier study, reporting data from Greece(Papadopoulos et al. 2005). The conflicting results fromearlier studies, as summarised above, thus gave a rationalefor the present study, and data on global solar radiation isan added feature of this analysis.
One reason for the contradictions in earlier studies on theclimate dependence of suicides may be that they have beenperformed in different climatic conditions. Finland, situatedin northern Europe, has a strong seasonal variationdetermined mainly by temporal variations in the solarradiation energy received in the area. Most of the solarradiation is received in the summer, when the monthly sumof global radiation is about 600 MJ/m2, while in the wintermonths the monthly global radiation is typically only one-tenth of summer-time values. The annual global solarradiation varies from 2,800 MJ/m2 in the north to3,400 MJ/m2 in southern Finland. The monthly durationof sunshine in Lapland varies on average from 270 h inJune to a period of darkness in the winter. In southern
Finland, the sun shines on average 300 h in June and only30 h in December. The annual mean temperature variesbetween −2°C in northern Lapland to +5°C in southernFinland, and the annual precipitation from 500 mm innorthern Lapland to 650 mm in southern Finland. Thewinter (December–February) is usually cold, and the meantemperature varies from −4°C in southern Finland to −16°Cin northern Lapland. The summer (June–August) meantemperature varies from 12 to 16°C; precipitation duringsummer months exceeds that in winter (Drebs et al. 2002).
Our aims were to test whether global solar radiation(hereafter, global radiation) explains the inter-annual andintra-seasonal variation in suicide rates better than dailysunshine duration (hereafter, sunshine hours) and, if so, isglobal radiation the most informative of the explanatoryweather parameters. In addition, we hypothesised that climatemay have a cumulative impact on suicide rate. Longerperiods, i.e. season(s) to year(s), of climatic anomalies mighthave an effect on the number of potential suicide candidates,whereas shorter periods of days to week(s) might act as atrigger for the actual decision to commit suicide.
Methods
Our material contained data on all the 43,393 deaths fromsuicide (33,993 men and 9,400 women) in Finland (whichhas a population of approximately 5 million), over the35-year period from 1969 to 2003, both years inclusive.The data, derived from official death certificates, wereobtained from Statistics Finland (www.stat.fi) andcontained the following information for each case: the dateof death, date of birth, sex, and residence. The codes of deathfrom suicide according to the World Health Organization’sInternational Classifications of Diseases include X60-X84and Y87.0 (ICD-10), E950A-E959X (ICD-9) and E950.9-E959.9 (ICD-8). The detailed content of the Finnish deathcertificate provides excellent accuracy and reliability ofinformation on the cause of death. Finnish law requires amedico-legal (forensic) investigation of the cause and themanner of death whenever it has been unnatural, or issuspected to be so, or whenever it has been sudden orunexpected. The proportion of investigations for suicide hasbeen stable for many decades (Näyhä 1981), and the overallautopsy rate has remained relatively high, both of whichallow for reliable conclusions on suicide trends over time.
Daily climate data from the Finnish MeteorologicalInstitute (www.fmi.fi) were used in this study. Fortemperature and precipitation, we used data that have beeninterpolated from weather observations onto a grid of10 km×10 km. For global radiation, grid-based data arenot available, so we used data that were measured at aweather station located at Jokioinen, in south-western
168 Int J Biometeorol (2009) 53:167–175
Finland, this being well enough representative of thevariability in global radiation in the most denselypopulated (southern and central) areas of Finland, where80% of suicides take place. Global radiation is measuredwith a pyranometer at 1-min intervals. Daily cumulativeobservations of radiation were available in the databasefrom 1971 onwards, which thus defined the study periodfor analysis.
Suicide rates and weather variables were analysed usingtime frames ranging from the monthly to the annual level.For each time frame, the cumulative values of suicide rate,global radiation, sunshine hours and precipitation werecalculated as the sum of daily values for the period inquestion, whereas temperature was averaged over the wholeperiod. The definition of the four seasons used in ouranalysis was as follows: spring is from March to May,summer is from June to August, autumn is from Septemberto November, and winter is from December to February.Suicide rates were analysed for the total population as wellas for men and women separately.
Statistics
Both simple and multivariate linear regression analyseswere performed. Linear univariate regression modelswere calculated using the suicide rate as the dependentvariable and measured global radiation, sunshine hours,average temperature or precipitation as explanatoryvariables. For the multivariate models the measuredglobal radiation, average temperature and precipitationwere included stepwise as explanatory variables. Sun-shine hours were not used, since this parameter can nolonger be considered to be an independent variablewhen global radiation is included in the model. Themodel follows Eq. 1:
SR ¼ aþ b Gð Þ þ c Tavgð Þ þ d RRð Þ ð1Þwhere SR is suicide rate (per 100,000 persons); a, b, c, andd are regression coefficients; G is cumulative globalradiation (MJ/m2); Tavg is average temperature (°C); andRR is accumulated precipitation (mm). The regressionanalysis was executed using a range of time spans fromthe monthly to the annual level. In order to study the effectof the long-term variation in climate factors, a regressionanalysis was first performed with the data including trendsand then with the residual data after removing the effect oftrends. The trends were filtered by fitting linear trendsboth to the suicide and weather data for two sub-periods,i.e. the periods of increasing suicide rates from 1971 to1990 and of decreasing suicide rates from 1991 to 2003;the regression was analysed using the residual of suiciderate as the dependent variable and the residuals of theweather parameters as the explanatory variables.
Results
Annual suicide rates and weather parameters
There was a marked inter-annual variability in suicide andweather variables (Fig. 1). Concerning global radiation,there is both inter-annual and long-term variability with noclear trend, but a decreasing trend can be interpreted for theperiod of 1971–1990 and an increasing trend thereafter. Thecurrent climate change can be discerned in the annual meantemperature, but the inter-annual variability is even stronger.As a sign of ongoing climate change, we noted that since 1989no years have been as cold as they still were in the 1970s and1980s, and that there is an increasing trend (+0.2°C in10 years) in the annual mean temperature in the 1990s. At theannual level there is no precipitation trend.
At the annual level, global radiation correlated best withsuicide rates, with the correlation being stronger than thatfor sunshine hours (Table 1). The correlation coefficient (R)between sunshine hours and global radiation was only0.839. Global radiation is a parameter that measures theamount of energy received at ground level from both thedirect and diffuse solar radiation. It gives a more validestimate of luminance than sunshine hours, becausesunshine hours accumulate only when there are no cloudsor only thin high-level clouds in front of the sun. Globalradiation is often strong in situations of changing cumulusor scattered middle-level altocumulus clouds. In addition,an altostratus cloud cover may let a large proportion ofradiation through to the ground.
Regression analyses show that statistically significant(P<0.01) models for both the total suicide rate and the malesuicide rate were achieved using only global radiation asthe independent variable, and in these models R2 was asgreat as 0.23 (Table 2). Temperature or precipitation alonemade no significant contribution to the suicide rates (datanot shown). However, using temperature and precipitationtogether with global radiation as the independent variables,the model explained a greater proportion of the variation(R2=0.32) in the suicide rates (Table 2). Because the femalesuicide rate was lower than the male suicide rate, thestochastic variations in suicide incidence may be larger.
Next, the regression analysis was computed separatelyfor the periods of increasing (1971–1990) and decreasing(1991–2003) suicide rates. The dependence on weather wasmore robust for the former period (Table 2). For the latterperiod, any dependence on the weather disappeared (Table 2).It is important to note that the results from these differentperiods are not totally comparable because the smallernumber of observations in the latter period makes thestatistical significance weaker.
In view of these trends, the suicide rates and weathervariables were analysed further using the residuals after
Int J Biometeorol (2009) 53:167–175 169
fitting the trends to the data, even though this was ratherhypothetical in the case of the weather data. In this way wesought to discover whether the variation in the weathervariables correlated with the variation in the suicide rates(Table 3).
After trend filtering, the coefficients for global radiationbecame smaller and less significant (Tables 2, 3). On theother hand, the changes in the coefficients for temperatureand precipitation were small, but they became moresignificant. Thus, the impact of temperature and precipita-
Table 1 Pearson correlation matrix for annual suicide rates and weather parameters in Finland in the period 1971–2003
Suicide rate Male suicide rate Female suicide rate Global radiation Sunshine hours Average temperature Precipitation
Suicide rate 1.000
Male suicide rate 0.991 1.000
Female suicide rate 0.814 0.729 1.000
Global radiation −0.475 −0.475 −0.349 1.000
Sunshine hours −0.276 −0.317 −0.047 0.839 1.000
Average temperature 0.071 0.016 0.278 0.096 0.138 1.000
Precipitation 0.108 0.094 0.121 −0.572 −0.527 0.312 1.000
-1
0
1
2
3
4
5
6
7
8
1971 1975 1979 1983 1987 1991 1995 1999 2003
Tav
g (
oC
)
100
200
300
400
500
600
700
800
Pre
cip
itat
ion
(m
m/y
ear)
Tavg / year
Precipitation
0
10
20
30
40
50
1971 1975 1979 1983 1987 1991 1995 1999 2003
Su
icid
es /
100
000
Total suicide rate
Male suicide rate
Female suicide rate
1800
2000
2200
2400
2600
2800
3000
3200
3400
3600
3800
1971 1975 1979 1983 1987 1991 1995 1999 2003
Glo
bal
rad
iati
on
(M
J/m
2)
1400
1600
1800
2000
2200
2400
Su
nsh
ine
ho
urs
Global radiation
Sunshine hours
Fig. 1 Total, male and femalesuicide rates (top panel), annualglobal solar radiation and sun-shine hours at Jokioinen (middlepanel), and annual mean tem-perature and precipitation (bot-tom panel) in Finland from 1971to 2003
170 Int J Biometeorol (2009) 53:167–175
Table 3 Regression coefficients for the residuals of weather parameters explaining suicide rates after fitting linear trends for the years of 1971–1990 and 1991–2003
Suicide rate total Suicide rate male Suicide rate female
Coeffient P Coeffient P Coeffient P
Suicide rates explained by radiation, temperature and precipitation in 1971–2003G −0.004* 0.019* −0.006* 0.014* −0.001 0.336Tavg 0.627* 0.002* 0.960* 0.005* 0.307* 0.006*RR −0.010* 0.004* −0.017* 0.005* −0.004 0.052Multiple R2 0.35 0.32 0.26P 0.005* 0.010* 0.033*Suicide rates explained by radiation, temperature and precipitation in 1971–1990G −0.004 0.061 −0.008* 0.041* −0.001 0.400Tavg 0.692* 0.009* 1.102* 0.001* 0.299* 0.034*RR −0.012* 0.013* −0.020* 0.011* −0.004 0.103Multiple R2 0.42 0.43 0.28P 0.028* 0.027* 0.149Suicide rates explained by radiation, temperature and precipitation in 1991–2003G −0.001 0.497 −0.003 0.526 0.000 0.870Tavg 0.365 0.373 0.307 0.710 0.428 0.121RR −0.003 0.563 −0.004 0.696 −0.002 0.557Multiple R2 0.05 0.05 0.26P 0.914 0.914 0.410
*P<0.05
Table 2 Regression coefficients for annual global radiation (G), average temperature (Tavg), and precipitation (RR) explaining annual suicide rates
Total suicide rate Male suicide rate Female suicide rate
Coefficient P Coefficient P Coefficient P
Suicide rates explained by global radiation1971–2003 G −0.007* 0.005* −0.012* 0.005* −0.002* 0.047*
R2 0.23 0.23 0.121971–1990 G −0.006* 0.027* −0.011* 0.020* −0.002 0.115
R2 0.24 0.27 0.131991–2003 G −0.008 0.093 −0.015 0.083 −0.001 0.302
R2 0.24 0.245 0.10Suicide rates explained by global radiation, temperature and precipitation1971–2003 G −0.010* 0.001* −0.018* 0.001* −0.003* 0.006*
Tavg 0.609 0.140 0.833 0.259 0.364* 0.016*RR −0.014 0.075 −0.023 0.087 −0.004 0.102Multiple R2 0.32 0.31 0.29P 0.010* 0.013* 0.018*
1971–1990 G −0.010* 0.003* −0.017* 0.002* −0.003* 0.026*Tavg 0.829* 0.032* 1.320* 0.035* 0.355 0.051RR −0.013 0.066 −0.022 0.056 −0.004 0.171Multiple R2 0.47 0.48 0.33P 0.016* 0.013* 0.084
1971–1990 G −0.011 0.065 −0.019 0.156 −0.003 0.233Tavg −0.051 0.154 −0.448 0.882 0.328 0.487RR −0.014 0.976 −0.024 0.572 −0.005 0.471Multiple R2 0.27 0.29 0.17P 0.389 0.364 0.619
*P<0.05
Int J Biometeorol (2009) 53:167–175 171
tion on the suicide rate appeared similar, whereas theimpact of global radiation raised new questions.
Monthly and seasonal suicide rates and weather parameters
The significant associations found at the annual level raisedthe question of whether there is any particular period of theyear during which these correlations are at their strongest.In other words, is the dependence on weather parametersconstant throughout the year, and does the relative impactof weather parameters change during the course of theyear? To that end, we wanted to analyse these relationshipsat the monthly and seasonal (3-month) levels. The monthlyand seasonal suicide rates, global radiation, mean temper-ature and precipitation were modelled in a similar way tothose at the annual level.
Analysis of the monthly weather data yielded fewsignificant associations. The male suicide rate was relatedto global radiation in March (b=−0.004, P=0.050) andSeptember (b=−0.006, P=0.015). The female suicide ratewas related to global radiation (b=−0.011, P=0.026) andmean temperature (c=0.046, P=0.005) in November.
Next, seasonal analysis using 3-month cumulative valuesyielded no significant correlation between the suicide ratesand the weather parameters for spring or summer. For autumn,there was a negative correlation between global radiation andthe suicide rate for both men (R2=0.17; b=−0.012, P=0.017)and women (R2=0.20; b=−0.004, P=0.009). In the winter,there was a positive correlation between mean temperatureand the suicide rate for men (R2=0.16; c=0.16, P=0.022)only. Multiple regression with global radiation, meantemperature and precipitation as the independent variablesimproved the explanatory power of the findings in autumnbut not those in winter.
Cumulative suicide rates and global radiation
Finally, we wanted to test whether a lack of global radiationresults in a gradual increase in suicide rates. To address thisquestion we analysed the cumulative effect using a range oftime periods. For this purpose we computed correlationsbetween suicide rates and global radiation starting fromNovember, which is the beginning of the darkest, andtherefore potentially the most depressing period of the year.From November onwards, daylight time is short (less than9 h, even in southern Finland) and the weather is quite oftencloudy and rainy, which reduces solar radiation on theground. It is noteworthy that the selection of a “suicideyear” as the 12-month period of November to October hadan influence on the results, with those based on the calendaryear presented in Table 2 being slightly different.
The analysis of cumulative global radiation showed thatthe longer the cumulative period, the stronger the associationof suicide rates with global radiation (Table 4). The bestcorrelation was found for the period of November to March,especially for the male suicide rate; 40% of the variation inthis rate was explained by cumulative global radiation.
From April onwards, the impact of the cumulative globalsolar radiation on the suicide rates decreased, being at itslowest for the period of November to July. After summer,these negative correlations between suicide rates andcumulative global solar radiation again reached significance.
Discussion
Our key finding was that global solar radiation fromNovember to March had an influence on the subsequentnumber of deaths from suicide in Finland. The less global
Table 4 Dependence of suicide rates on cumulative global radiation (Glob) by period, starting from November onwards. Data for this analysis:time period November 1971–October 2003, nationwide suicide rates, global solar radiation from the Jokioinen weather station
Suicide rate total Suicide rate male Suicide rate female
Time period Coefficient Glob R2 P Coefficient Glob R2 P Coefficient Glob R2 P
November −0.012 0.08 0.123 −0.014 0.04 0.306 −0.011* 0.14* 0.036*November-December −0.013 0.07 0.157 −0.018 0.04 0.274 −0.008 0.08 0.108November-January −0.011 0.03 0.312 −0.011 0.01 0.556 −0.011 0.12 0.058November-February −0.012 0.06 0.176 −0.017 0.04 0.260 −0.006 0.07 0.071November-March −0.015* 0.39* 0.0001* −0.026* 0.40* 0.0001* −0.004* 0.14* 0.035*November-April −0.007* 0.23* 0.005* −0.013* 0.23* 0.006* −0.002 0.08 0.112November-May −0.005 0.12 0.051 −0.008 0.11 0.064 −0.001 0.06 0.188November-June −0.004 0.10 0.075 −0.007 0.09 0.106 −0.001 0.07 0.138November-July −0.003 0.03 0.333 −0.005 0.03 0.326 −0.000 0.004 0.728November-August −0.004 0.07 0.137 −0.007 0.07 0.138 −0.001 0.02 0.425November-September −0.006* 0.17* 0.018* −0.011* 0.18* 0.017* −0.006* 0.17* 0.018*November-October −0.011* 0.17* 0.019* −0.011* 0.17* 0.019* −0.006* 0.17* 0.021*
*P<0.05
172 Int J Biometeorol (2009) 53:167–175
radiation there is during the winter, the higher the risk ofsuicide will be, from November to March in particular.These results do not contradict, but rather extend a series ofearlier findings regarding suicidal behaviour from nation-wide study samples in Finland. In Finland, the incidence ofattempted suicide is lowest in December and highest inApril (Haukka et al. 2008), whereas the risk of death fromsuicide is greatest in May. The latter is not only associatedwith high levels of solar radiation but is also mostpronounced in springs of longer-term periods, which haverelatively low suicide rates (Partonen et al. 2004b). Forviolent suicides there is a clear spring peak, whereas non-violent suicides are more likely to occur in the autumn(Hakko et al. 1998a). However, by analysing time spansfrom 1 month to a year, we found no positive correlationbetween the suicide rate and global radiation, and no furthersupport for the view that the spring peak in suicidemortality is due to increases in the amount of sunlight.Concerning suicide mortality from April to June in particular,analysis of time spans shorter than a month may be of addedvalue but is beyond the scope of our current report.
Global solar radiation is unique among meteorologicalvariables as an independent risk factor, since it explainedthe suicide rates in a significant way. However, after trendfiltering, its impact was lessened and was statistically lesssignificant in our analysis. Longer-term trends in suicidemortality have been considered to be due to socio-economicor psychological factors. As health authorities becameconcerned at the increasing mortality due to suicide in the1970s, a nationwide suicide-prevention program wasstarted in Finland in 1986, leading to a decreasing trendafter 1990 (Beskow et al. 1999). This change specificallyparallels the decrease seen in violent suicides (Hakko et al.1998b). Here, on the basis of our results, the long-termtrends seen in suicide mortality may not just be an outcomeof the sum of the aforementioned factors; longer-termvariability in global solar radiation may also be relevant. Alimitation of our analysis, however, is that individual dataconcerning the age, education, civil status, socio-economicstatus or mental illness were not available and therefore nottaken into account as risk factors. It also needs to be notedthat variations in physical factors such as sunshine hoursmay result in changes in people's social activities and, ifthese changes end in difficulties in social relationships andassociated negative outcomes, contribute to suicide rates.
Papadopoulos et al. (2005) reported that solar radianceduring the preceding day increased the risk of suicide by2% per each MW/m2. The finding seems to contradict ours,as we found that the less solar radiance there was during thewinter, the higher the risk of suicide, but the two studiesapplied different time spans in that Papadopoulos et al.focussed on time spans shorter than a month whereas weanalysed those longer than a month. In the data from
Greece, however, the past solar radiance was related onaverage to risk of suicide on a daily level, indicating thatthere might be a mechanism with a duration effect. Ourresults from Finland support the finding that there was aduration effect. They also found that longer sunshineexposure was needed in men to trigger suicide. In thisstudy, we demonstrate the magnitude of the weather impacton suicide occurrence by applying coefficients from Table 2(suicide rates explained by global radiation) for the Finnishpopulation. A variability of twice the standard deviation (of180MJ/m2) in global radiation would mean a variation of 130suicide cases per year. Even the smaller coefficients of globalradiation (from Table 3) would mean a variation of 65 suicidecases per year. To conclude, variability in global radiation hasa marked influence on suicide mortality and is therefore ofinterest to public health and suicide prevention programs.Precipitation and mean temperature also have an effect,though smaller (71 and 66 suicide cases, respectively).
Another of our key findings was that the dose-responseperiod of time may be shorter for women. For women, thesignificant correlation between suicide rate and globalradiation disappears early in March, whereas for men thecorrelation remains stronger for a longer period. Interestingly,these negative correlations were stronger for men thanwomen, indicating that men might be more sensitive tovariability in global radiation than women. On the other hand,the mean temperature was more strongly related to the suiciderate for women than for men, thereby suggesting that bothmen and women react to climate, but in a different way.
The impact of climate on the suicide rates, and therelative value of each weather parameter to explainvariation, fluctuates over the year. Using the 1-month timespan there is no relationship between suicide rates andweather for most months. One reason may be that thestochastic variability is greater at the monthly than at theannual level. Therefore, a calendar month may not be avalid frame for estimation of the impact of weather onsuicide. In addition, the dose-response effect may lead to anonsensical correlation if the suicides rates of a givenmonth are compared with the weather parameters of thesame month, and any possible lag is ignored.
Our current results indicate that the period of autumnand winter months having a lack of global solar radiation,in other words being relatively dark, increases the numberof people at risk of committing suicide. Our monthly andseasonal time span analysis showed that the meantemperature had a positive correlation with the femalesuicide rate in November and with the male suicide rate inwinter (December to February). These findings support theresults from two recent studies (Ajdacic-Gross et al. 2007;Preti et al. 2007) suggesting that it may not be the heat assuch but the lack of cold that is related to suicideoccurrence.
Int J Biometeorol (2009) 53:167–175 173
The suicide peak in spring has often been attributed tosudden increases in luminance and sunshine. However, inthis study, using 1-month or 3-month time spans we foundno positive correlation between suicide and global radiationduring spring. On the contrary, there were negativecorrelations only at the annual, seasonal (autumn) andmonthly (March, September, November) levels.
There are some limitations to our analysis. First, thespatial resolution of our temperature and precipitation dataare high, but for global radiation we used measurementsfrom only one weather station at Jokioinen. However, thisparticular station is located in the middle of the mostdensely populated (south-western) part of Finland andtherefore the weather impact is very similar if the numberof suicides completed near Jokioinen is used as thedependent variable. Therefore, in order to generalise theresults, we decided to use nationwide suicide rates for ouranalysis. Second, the characteristics of weather and climatethat impact on suicide might have been elucidated in moredetail by using longer cumulative periods to analysewhether weather affects the number of people at risk ofcommitting suicide, and by using shorter study periods tostudy whether weather is a short-acting trigger for suicide.We also need to keep in mind the fact that weatherparameters are not truly independent. However, our resultssuggest that weather and climate factors may be significantand worthy of further studies in suicide research.
Our finding regarding the negative association of annualglobal radiation levels with suicide rates is of continuinginterest, and may be a key to suicide mortality statistics atthe global level. Suicide rates tend to be higher at higherlatitudes than at lower latitudes (Lester 1970; Terao et al.2002; Lawrynowicz and Baker 2005). Our results nowshow that there is a significant association between suiciderates and global radiation, and thus suggest that a lack ofshort-wave radiation, especially during the winter time inhigher latitudes, may make a difference to suicide mortality.The relative importance of global radiation and otherclimate factors additionally raises the question of whetherthe current climate change is having an effect on suicidemortality. In Finland, there has been a decreasing trend insuicide mortality since 1990, with male and female suiciderates being reduced by 39% and 24%, respectively, to 2006.Due to projected climate change, however, global radiationis likely to be reduced in winter because of increasingprecipitation and cloudiness, and shorter periods with snowcover. For northern Europe, winter (December–February)precipitation is projected to increase by 9–25% by the endof this century (Christensen et al. 2007), and the number ofdays with snow cover to decrease by 30–40% on average ascompared with the current climate (Jylhä et al. 2008). Theworst scenario is that the positive development in suicideprevention and mental health promotion will come to a halt
to be replaced by an increasing trend in suicide in the nextdecades.
In conclusion, our results show that global solarradiation gives a more accurate representation of the impactof climate on suicide rates than sunshine hours. Cumulativeperiods ranging across seasons rather than the commonlyused calendar months give a more robust association ofweather parameters with suicide rates. Variability in annualglobal radiation explains a large proportion of the suicideoccurrence at the annual level and may also closely reflectlong-term trends in suicide rates.
Acknowledgement This study was supported in part by the grants#201097 and #210262 (to T.P.) from the Academy of Finland.
References
Ajdacic-Gross V, Lauber C, Sansossio R, Bopp M, Eich D, Gostynski M,Gutzwiller F, Rössler W (2007) Seasonal associations betweenweather conditions and suicide—evidence against a classic hypothesis.Am J Epidemiol 165:561–569. doi:10.1093/aje/kwk034
Beskow J, Kerkhof A, Kokkola A, Uutela A (1999) Suicideprevention in Finland 1986–1996: external evaluation by aninternational peer group. Ministry Soc Aff Health Copies 1999:2.Ministry of Social Affairs and Health, Helsinki
Bridges FS, Yip PS, Yang KC (2005) Seasonal changes in suicide inthe United States, 1971 to 2000. Percept Mot Skills 100:920–924
Christensen JH, Hewitson B, Busuioc A, Chen A, Gao X, Held I,Jones R, Kolli RK, Kwon W-T, Laprise R, Magaña Rueda V,Mearns L, Menéndez CG, Räisänen J, Rinke A, Sarr A, WhettonP (2007) Regional climate projections. In: Solomon S, Qin D,Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, MillerHL (eds) Climate Change 2007: the physical science basis.contribution of working group I to the fourth assessment report ofthe intergovernmental panel on climate change. CambridgeUniversity Press, Cambridge, UK
Deisenhammer EA (2003) Weather and suicide: the present state ofknowledge on the association of meteorological factors withsuicidal behaviour. Acta Psychiatr Scand 108:402–409.doi:10.1046/j.0001-690X.2003.00209.x
Dixon KW, Shulman MD (1983) A statistical investigation into therelationship between meteorological parameters and suicide. Int JBiometeorol 27:93–105. doi:10.1007/BF02185738
Dixon PG, McDonald AN, Scheitlin KN, Stapleton JE, Allen JS,Carter WM, Holley MR, Inman DD, Roberts JB (2007) Effects oftemperature variation on suicide in five U.S. counties, 1991–2001.Int J Biometeorol 51:395–403. doi:10.1007/s00484-006-0081-4
Drebs A, Nordlund A, Karlsson P, Helminen J, Rissanen P (2002)Climatological statistics of Finland 1971–2000 (in Finnish).Finnish Meteorological Institute, Helsinki
Dwivedi Y, Rizavi HS, Roberts RC, Conley RC, Tamminga CA,Pandey GN (2001) Reduced activation and expression of ERK1/2MAP kinase in the post-mortem brain of depressed suicide subjects.J Neurochem 77:916–928. doi:10.1046/j.1471-4159.2001.00300.x
Hakko H, Räsänen P, Tiihonen J (1998a) Seasonal variation in suicideoccurrence in Finland. Acta Psychiatr Scand 98:92–97.doi:10.1111/j.1600-0447.1998.tb10048.x
Hakko H, Räsänen P, Tiihonen J (1998b) Secular trends in the ratesand seasonality of violent and nonviolent suicide occurrence inFinland during 1980–95. J Affect Disord 50:49–54. doi:10.1016/S0165-0327(98)00031-7
174 Int J Biometeorol (2009) 53:167–175
Haukka J, Suominen K, Partonen T, Lönnqvist J (2008) Determinantsand outcomes of serious attempted suicide: a nationwide study inFinland, 1996–2003. Am J Epidemiol 167:1155–1163. doi:10.1093/aje/kwn017
Heerlein A, Valeria C, Medina B (2006) Seasonal variation in suicidaldeaths in Chile: its relationship to latitude. Psychopathology39:75–79. doi:10.1159/000090596
Jylhä K, Fronzek S, Tuomenvirta H, Carter TR, Ruosteenoja K (2008)Changes in frost, snow and Baltic sea ice by the end of the twenty-first century based on climate model projections for Europe. ClimChange 86:441–462. doi:10.1007/s10584-007-9310-z
Lambert GW, Reid C, Kaye DM, Jennings GL, Esler MD (2002)Effect of sunlight and season on serotonin turnover in the brain.Lancet 360:1840–1842. doi:10.1016/S0140-6736(02)11737-5
Lambert G, Reid C, Kaye D, Jennings G, Esler M (2003) Increasedsuicide rate in the middle-aged and its association with hours ofsunlight. Am J Psychiatry 160:793–795. doi:10.1176/appi.ajp.160.4.793
Lawrynowicz AEB, Baker TD (2005) Suicide and latitude inArgentina: Durkheim upside-down. Am J Psychiatry 162:1022.doi:10.1176/appi.ajp.162.5.1022
Lee HC, Lin HC, Tsai SY, Li CY, Chen CC, Huang CC (2006) Suiciderates and the association with climate: a population based study. JAffect Disord 92:221–226. doi:10.1016/j.jad.2006.01.026
Lester D (1970) Completed suicide and latitude. Psychol Rep 27:818Mann JJ, Apter A, Bertolote J, Beautrais A, Currier D, Haas A, Hegerl
U, Lonnqvist J, Malone K, Marusic A, Mehlum L, Patton G,Phillips M, Rutz W, Rihmer Z, Schmidtke A, Shaffer D,Silverman M, Takahashi Y, Varnik A, Wasserman D, Yip P,Hendin H (2005) Suicide prevention strategies: a systematicreview. JAMA 294:2064–2074. doi:10.1001/jama.294.16.2064
Meares R, Mendelsohn FAO, Milgrom-Friedman J (1981) A sexdifference in the seasonal variation of suicide rate, a single cyclefor men, two cycles for women. Br J Psychiatry 138:321–325.doi:10.1192/bjp.138.4.321
Micciolo R, Zimmermann-Tansella C, Williams P, Tansella M (1989)Seasonal variation in suicide: is there a sex difference? PsycholMed 19:199–203
Näyhä S (1981) Short and medium-term variations in mortality inFinland: a study on cyclic variations, annual and weekly periodsand certain irregular changes in mortality in Finland duringperiod 1868–1972. Scand J Soc Med 8(Suppl 21):1–101
Näyhä S (1982) Autumn incidence of suicides pre-examined: datafrom Finland by sex, age and occupation. Br J Psychiatry141:512–517. doi:10.1192/bjp.141.5.512
Page LA, Hajat S, Kovats RS (2007) Relationship between dailysuicide counts and temperature in England and Wales. Br JPsychiatry 191:106–112. doi:10.1192/bjp.bp.106.031948
Pandey GN, Dwivedi Y, Rizavi HS, Ren X, Pandey SC, Pesold C,Roberts RC, Conley RR, Tamminga CA (2002) Higher expressionof serotonin 5-HT 2A receptors in the postmortem brains ofteenage suicide victims. Am J Psychiatry 159:419–429.doi:10.1176/appi.ajp.159.3.419
Papadopoulos FC, Frangakis CE, Skalkidou A, Petridou E, StevensRG, Trichopoulos D (2005) Exploring lag and duration effect ofsunshine in triggering suicide. J Affect Disord 88:287–297.doi:10.1016/j.jad.2005.08.010
Parker G, Gao F, Machin D (2001) Seasonality of suicide inSingapore: data from the equator. Psychol Med 31:549–553
Partonen T, Haukka J, Lönnqvist J (2003) Suicide mortality in Finland1979–2001 (in Finnish). Duodecim 119:1827–1834
Partonen T, Haukka J, Pirkola S, Isometsä E, Lönnqvist J (2004a)Time patterns and seasonal mismatch in suicide. Acta PsychiatrScand 109:110–115. doi:10.1046/j.0001-690X.2003.00226.x
Partonen T, Haukka J, Nevanlinna H, Lönnqvist J (2004b) Analysis ofthe seasonal pattern in suicide. J Affect Disord 81:133–139.doi:10.1016/S0165-0327(03)00137-X
Partonen T, Haukka J, Viilo K, Hakko H, Pirkola S, Isometsä E,Lönnqvist J, Särkioja T, Väisänen E, Räsänen P (2004c) Cyclictime patterns of death from suicide in northern Finland. J AffectDisord 78:11–19. doi:10.1016/S0165-0327(02)00236-7
Petridou E, Papadopoulos FC, Frangakis CE, SkalkidouA, Trichopoulos D(2002) A role of sunshine in the triggering of suicide. Epidemiology13:106–108. doi:10.1097/00001648-200201000-00017
Preti A (1998) The influence of climate on suicidal behaviour in Italy.Psychiatry Res 78:9–19. doi:10.1016/S0165-1781(97)00154-6
Preti A, Lentini G, Maugeri M (2007) Global warming possibly linkedto an enhanced risk of suicide: data from Italy, 1974–2003. JAffect Disord 102:19–25. doi:10.1016/j.jad.2006.12.003
Qin P, Agerbo E, Mortensen PB (2003) Suicide risk in relation tosocioeconomic, demographic, psychiatric, and familial factors: anational register-based study of all suicides in Denmark, 1981–1997.Am J Psychiatry 160:765–772. doi:10.1176/appi.ajp.160.4.765
Räsänen P, Hakko H, Jokelainen J, Tiihonen J (2002) Seasonal variationin specific methods of suicide: a national register study of 20,234Finnish people. J Affect Disord 71:51–59. doi:10.1016/S0165-0327(01)00411-6
Rocchi MB, Miotto P, Preti A (2004) Seasonal variation in suicidesand in deaths by unintentional illicit acute drug intoxications.Addict Biol 9:255–263. doi:10.1080/13556210412331292587
Rock D, Greenberg DM, Hallmayer JF (2003) Increasing seasonalityof suicide in Australia 1970–1999. Psychiatry Res 120:43–51.doi:10.1016/S0165-1781(03)00165-3
Souêtre E, Salvati E, Belugou JL, Douillet P, Braccini T, Darcourt G(1987) Seasonality of suicides: environmental, sociological andbiological covariations. J Affect Disord 13:215–225. doi:10.1016/0165-0327(87)90040-1
Souêtre E, Wehr TA, Douillet P, Darcourt G (1989) Influence ofenvironmental factors on suicidal behavior. Psychiatry Res32:253–263. doi:10.1016/0165-1781(90)90030-9
Stoupel E, Petrauskiene J, Kalediene R, Abramson E, Sulkes J (1995)Clinical cosmobiology: the Lithuanian study 1990–1992. Int JBiometeorol 38:204–208. doi:10.1007/BF01245390
Terao T, Soeda S, Yoshimura R, Nakamura J, Iwata N (2002) Effect oflatitude on suicide rates in Japan. Lancet 360:1892. doi:10.1016/S0140-6736(02)11761-2
VoracekM, Tran US, Sonneck G (2007) Facts and myths about seasonalvariation in suicide. Psychol Rep 100:810–814. doi:10.2466/PR0.100.3.810-814
WHO (2005) Mental Health: facing the challenges, building solutions,suicide prevention. Report from the WHO European MinisterialConference. World Health Organization, Geneva, pp 75–82
Yan YY (2000) Geophysical variables and behavior: LXXXXIX. Theinfluence of weather on suicide in Hong Kong. Percept MotSkills 91:571–577. doi:10.2466/PMS.91.6.571-577
Zonda T, Bozsonyi K, Veres E (2005) Seasonal fluctuation of suicidein Hungary between 1970-2000. Arch Suicide Res 9:77–85.doi:10.1080/13811110590512967
Int J Biometeorol (2009) 53:167–175 175
© 2012 Springer Nature Reprinted, with permission, from
International Journal of Biometeorology 56:6, 1045-1053
doi: 10.1007/s00484-011-0518-2
IV
ORIGINAL PAPER
Atmospheric pressure and suicide attemptsin Helsinki, Finland
Laura Hiltunen & Reija Ruuhela & Aini Ostamo &
Jouko Lönnqvist & Kirsi Suominen & Timo Partonen
Received: 4 October 2010 /Revised: 15 December 2011 /Accepted: 23 December 2011# ISB 2012
Abstract The influence of weather on mood and mentalhealth is commonly debated. Furthermore, studies concerningweather and suicidal behavior have given inconsistent results.Our aim was to see if daily weather changes associate with thenumber of suicide attempts in Finland. All suicide attemptstreated in the hospitals in Helsinki, Finland, during two sep-arate periods, 8 years apart, were included. Altogether, 3,945suicide attempts were compared with daily weather parame-ters and analyzed with a Poisson regression. We found thatdaily atmospheric pressure correlated statistically significantlywith the number of suicide attempts, and for men the correla-tion was negative. Taking into account the seasonal normalvalue during the period 1971–2000, daily temperature, globalsolar radiation and precipitation did not associate with thenumber of suicide attempts on a statistically significant levelin our study. We concluded that daily atmospheric pressure
may have an impact on suicidal behavior, especially on sui-cide attempts of men by violent methods (P<0.001), and mayexplain the clustering of suicide attempts. Men seem to bemore vulnerable to attempt suicide under low atmosphericpressure and women under high atmospheric pressure. Weshow only statistical correlations, which leaves the exactmechanisms of interaction between weather and suicidal be-havior open. However, suicidal behavior should be assessedfrom the point of view of weather in addition to psychiatricand social aspects.
Keywords Global solar radiation . Precipitation .
Temperature . Season .Weather
Introduction
Despite the common belief that weather has an effect onmental health, the scientific results do not agree. In theNetherlands, no correlation between weather and mood ingeneral population was found (Huibers et al. 2010), but inItaly a positive correlation between behavior and affectivestates and weather was found among day-care children(Ciucci et al. 2011), and it has also been suggested that thereis an association between low barometric pressure and actsof violence and emergency psychiatric visits (Schory et al.2003). Furthermore, sunlight exposure seems to have aninfluence on cognitive function among depressed individu-als (Kent et al. 2009), and bright-light therapy is effective inthe treatment of seasonal affective disorder (Gooley 2008).Although, some criticism against the seasonal pattern hasemerged (Ajdacic-Gross et al. 2010), the seasonality ofaffective disorders (Wehr and Rosenthal 1989) and suicidalbehavior (Barker et al. 1994; Partonen et al. 2004; Valtonenet al. 2006; Yip and Yang 2004) have been explained partly
Electronic supplementary material The online version of this article(doi:10.1007/s00484-011-0518-2) contains supplementary material,which is available to authorized users.
L. Hiltunen :A. Ostamo : J. Lönnqvist :K. Suominen :T. PartonenNational Institute for Health and Welfare,Helsinki, Finland
L. Hiltunen : J. Lönnqvist :K. SuominenDepartment of Psychiatry, University of Helsinki,Helsinki, Finland
R. RuuhelaFinnish Meteorological Institute,Helsinki, Finland
L. Hiltunen (*)Department of Mental Health and Substance Abuse Services,National Institute for Health and Welfare,PO Box 30, FI-00271 Helsinki, Finlande-mail: [email protected]
Int J BiometeorolDOI 10.1007/s00484-011-0518-2
by meteorological variations. It may therefore be thatthe depressed are more sensitive to weather signals thanothers.
In seasonal as well as in non-seasonal depression (Germainand Kupfer 2008; Monteleone and Maj 2008), which is ofteninvolved in suicidal behavior (Henriksson et al. 1993;Suominen et al. 1996), disturbances in circadian rhythms areoften seen, but it is not known whether the circadian clock ofdepressed individuals reacts differently to environmentalsignals. Normally, changes in the rhythm of daylight(Roenneberg et al. 2007) and in ambient temperature(Majercak et al. 1999) are zeitgebergs, that is, the primarysignals that help the circadian clock to synchronize the endog-enous circadian rhythms with environmental variations suchas seasonal changes.
Sunshine hours have been demonstrated to have a posi-tive correlation with suicides (Lambert et al. 2003, 2002;Petridou et al. 2002; Salib and Gray 1997; Souetre et al.1987, 1990, 1991), but a negative correlation has also beenreported (Preti 1998; Stoupel et al. 1995). In our earlierstudy, we showed that low levels of cumulative global solarradiation during winter in Finland are associated with ahigher number of suicides (Ruuhela et al. 2009), althoughthe highest number of suicides occurred around the summersolstice (Hiltunen et al. 2011). A number of studies on thecorrelation between ambient temperature and suicide mor-tality (Ajdacic-Gross et al. 2007; Lee et al. 2006; Lester1986; Page et al. 2007; Salib and Gray 1997; Souetre et al.1987, 1990) have given inconsistent results, as have studieson the interaction between atmospheric pressure and suicidalbehavior (Barker et al. 1994; Breuer et al. 1986; Dexter 1904;Schory et al. 2003; Wang et al. 1997; Yan 2000).
Attempted suicides have received much less attentioncompared with completed suicides. However, the incidenceof attempted suicides is more than ten-fold that of completedsuicides, and attempted suicides increase markedly the riskof completed suicides (Haukka et al. 2008; Owens et al.2002; Sinclair et al. 2010; Suokas et al. 2001; Suominen etal. 2004), and also the risk of suicidal behavior in followinggenerations (Brent et al. 2002).
The interaction between weather, mental health and sui-cidal behavior seems to be a complex one, requiring furtherstudy (Deisenhammer 2003; Dexter 1904; Mills 1934). Sev-eral explanations for the inconsistent results observed havebeen proposed (Dixon and Kalkstein 2009).
Our aim was to focus on possible interactions betweenweather and suicide attempts. We aimed to study which dailyweather variables or which specific weather types, if any, areassociated with suicide attempts. In order to take into accountthe weather as a whole, we chose to use synoptic weatheranalyses in addition to analyzing single weather variables. Wealso attempted to assess if a weather impact appears particu-larly on days or weeks when the number of suicide attempts is
clearly above the average (hereafter called “cluster days” and“cluster weeks”, see also the exact definition in Materials andmethods) and, for comparison, if a certain weather type can beconsidered to exert a protective effect, with no or only a lownumber of suicide attempts.
Materials and methods
Subjects
Our data consist of two periods of time; the first from 1January 1989 to 31 July 1990 (I), and the second from 15January 1997 to 14 January 1998 (II): altogether 942 days.During these periods of time, all suicide attempts treated inhospitals in Helsinki (N03,945) were recorded as part of theWorld Health Organization Multicenter Study on Parasui-cide (Bille-Brahe et al. 1995). The date, sex, and the method(violent /non-violent) were recorded for each case (for exactnumbers, see Table 1). Statistics Finland provided the num-ber of completed suicides in Helsinki for the same periods(Table 1). Information concerning the cause of death wasobtained from the official death certificates. Since 1987,death certificate forms have complied with the recommen-dations of the International Classification of Diseases of theWorld Health Organization. However, because of the lowdaily number of completed suicides, we decided to concen-trate on analyzing only suicide attempts.
Meteorological data
The Finnish Meteorological Institute provided the dailymeteorological data from Helsinki Kaisaniemi weather sta-tion for the study periods: temperature (mean, maximumand minimum), precipitation, global solar radiation, sun-shine hours, and atmospheric pressure, which was takenfrom synoptic weather observations made eight times perday. For temperature and global solar radiation, the devia-tion from the climatic normal value (period 1971–2000) wasalso used in order to take into account the potential impactof acclimatization on seasons. The absolute daily values ofglobal solar radiation varied between 100 and 30,500 KJ/m2
and the temperature varied from −21°C to +28°C in ourdata.
Synoptic and statistical weather analyses
Descriptive synoptic weather analyses were conducted forthe days with high number of suicide attempts, that is thecluster days, and for the days with no self-harm, in order tostudy the complex nature of association between weatherand suicidal behavior, and to explore a hypothesis for fur-ther statistical analyses. The criterion for cluster days was
Int J Biometeorol
defined as the probability for the daily number of suicideattempts being less than 0.01 according to a Poisson distribu-tion, ending up with seven or more suicide attempts per dayfor men and women separately, or at least ten suicide attemptsfor both sexes counted together (Fig. 1, Table 2). Chi-squaredtest was used to compare the frequency distribution of atmo-spheric pressure during cluster days versus all days during thestudy periods. For testing purposes, the pressure data wasclassified as followed: low (<1,005 hPa), normal (1,005–1,020 hPa) and high (>1,020 hPa) (Table 3). Since socialactivities may also have an impact on the risk of self-harm,day of the week effect (Chiu 1988; Maldonado and Kraus1991; Pirkola et al. 1997) was considered. Furthermore, fordays with no self-harm, short-term mortality displacement orharvesting (Huynen et al. 2001) was considered in order toexclude the no-self-harm-days that follow (immediately after)a cluster day, or a period of a few days with elevated numbersof self-harm cases.
In order to take into account any possible lag effect andexposure time of weather impact on incidents of self-harm,we also made descriptive synoptic analyses on cluster weeksin each season, that is, weeks with the highest incidence of
self-harm during any 1-week period in the season. Theresults and discussion of these analyses are provided assupplementary material.
For the whole data set of both study periods we analyzedthe association between daily weather variables and dailysuicide attempts with a Poisson regression because the dailynumber of suicide attempts follows an approximate Poissondistribution with an average of two attempts per day for bothmen and women (Fig. 1). Poisson analysis was performedseparately also on violent suicide attempts for both genders.Rate ratios (RR) and confidence intervals per change ofweather unit were calculated (Table 5). Furthermore, RRs
0
50
100
150
200
250
300
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Daily suicide attempts
Number of days
Women Men
Fig. 1 Frequency distribution of daily number of suicide attempts forwomen and men in Helsinki
Table 2 Cluster days of suicide attempts for women (W) and men (M)separately and together(W+M)
Year Date of thecluster day
Day of the week Number ofsuicide attempts
W M W+M
1989 14 March Tue 6 5 11a
18 March Sat 3 8a 11
18 May Thu 6 7a 13
5 July Wed 8a 2 10
1990 2 Januaryb Tue 4 7 11
11 February Sun 7a 1 8
7 March Wed - 7a 7
18 April Wed 3 9a 12
1997 28 March Fri - 8a 8
18 April Fri 8a 2 10
23 Junec Mon 2 7a 9
24 June Tue 2 7a 9
21 October Thu 4 8a 12
23 October Thu 7a 2 9
7 Decemberd Sun 8a 1 9
1989 1 Januarye Thu 5 6 11a
11 January Sun 5 6 11a
a Cluster day selection criteriab After New Yearc After Midsummerd After Independence Daye New year’s day
Table 1 Number of attemptedand completed suicides formen and women in Helsinki,Finland, during periods I and II
aI0 1 January 1989–31 July1990, II 0 15 January 1997–14January 1998
Perioda Men Women Total
Suicide attempts (of which violent attempts) I 1,198 (202) 1,111 (104) 2,309
II 796 (88) 840 (57) 1,636
I+II 1,994 (290) 1,951 (161) 3,945
Completed suicides I 193 86 279
II 96 46 142
I+II 289 132 421
Int J Biometeorol
for a change of weather parameter by 10 units were calcu-lated (Table 5).
Results
Cluster days of suicide attempts
We found 17 days in total that fulfilled the cluster daycriteria (Table 2), of which 9 comprised men’s suicideattempts, 5 women’s attempts, and 3 men’s and women’sattempts counted together. The cluster days were differentfor women and men.
According to the synoptic weather analyses, cluster daysof men’s suicide attempts occurred more often in the contextof low pressure and cloudy weather with rain or snow. Rapidand marked pressure changes, for example a decrease of36 hPa per day and an increase of 20 hPa per day alsooccurred during those days. However, in April and May,men’s cluster days seemed more likely to occur in a contextof high pressure with sunny weather. Cluster days of wom-en’s suicide attempts seemed to be linked more often withhigh pressure weather conditions. The 3 days that fulfilledthe cluster day criteria according to the total number ofsuicide attempts occurred in winter, on days with highpressure and cloudy weather. Out of all of the weathervariables studied, atmospheric pressure seemed to have thestrongest association with the cluster days, which was alsostatistically significant according to the chi-squared test (P<0.05) (Table 3). Global solar radiation, temperature andprecipitation did not associate with the cluster days in thesynoptic analyses. The day of the week effect did not appearin the cluster days (Table 2).
The synoptic analyses of cluster weeks of suicideattempts revealed some interesting characteristics that couldbe used as potential hypotheses for further studies. Howev-er, due to the short study periods, the amount of data in ourstudy was insufficient for statistical tests for cluster weeksand therefore these descriptive analyses are provided assupplementary material.
Days with no suicide attempts or completed suicides
In order to study the potential “protective” weather type inrespect with self-harm, synoptic analyses were made for thedays with no-self-harm, i.e., days with no suicide attemptsor completed suicides by either sex. Synoptic analyses wereperformed also for women and men separately, but withperiods of 2 or more consecutive days without self-harm.
These no-self-harm-days can be explained in part byharvesting; the number of self-harm cases is low immedi-ately after a cluster day or after a period of a few days withthe elevated numbers of self-harm cases. Therefore, becausewe were interested in potential protective weather type, thedays explained by harvesting were excluded from the anal-ysis. Thereafter a synoptic analysis was performed for 7 dayswith no self-harm of either sex and for 5 periods of men andfor 14 periods of women (Table 4).
Firm conclusions cannot be made from these 7 dayswithout self-harm of either sex, but it seems that ‘protective’weather, i.e., weather conditions that carry the lowest self-harm risk, could be described as warmer than usual for theseason and partly sunny. The results are more interestingwhen analyzing the periods of no-self-harm for women andmen separately. For men, ‘protective’ weather, carrying alow self-harm risk could be described as days with normalatmospheric pressure, but in other respects the weatherconditions such as temperature, precipitation and cloudinessseemed to vary considerably during these days. A day of theweek effect was observed, and the most probable protectiveday was Monday (Table 4). For women, the protectiveweather comprised low pressure conditions, with or withoutprecipitation, and warmer than usual for the season. Over-cast or mainly cloudy weather often seemed to have pre-vailed on those days as well. The day of the week effect wasclear and the most probable protective day was Monday(Table 4).
Day of the week effect can be considered to be a conse-quence of variation in social activities. In cluster days theday of the week effect was not seen, and we concluded thatthe weather impact dominates cluster days of suicide
Table 3 Frequencies of days by different categories of atmospheric pressure during cluster days and no-self-harm days
Type of days Total no.of days
Atmospheric pressure Chi-squared test
Low:<1,005 hPa
Normal:1,005-1,020 hPa
High:>1,020 hPa
Value Significance
All days in study periods 942 246 494 202
Cluster daysa 17 9 4 4 7.24 0.003
No self harm days, men 5 1 4 0 1.86 0.39
No self harm days, women 14 9 4 1 10.37 0.006
a A minimum of seven suicide attempts per day for men and women separately, or a minimum of ten suicide attempts per day, sexes countedtogether
Int J Biometeorol
attempts. However, for no-self-harm days both weather de-pendence and day of the week effect, and thus impact ofsocial activities, can be seen.
Poisson regression of weather–suicide attempts
Analyses of the weather dependence of suicide attempts oncluster days and on no-self-harm days can be considered asspecial cases. In order to achieve more generic results on theweather dependence, we analyzed the daily data for bothperiods using Poisson regression with various weatherparameters as independent variables to explain the rates ofsuicide attempts.
For women, the number of daily suicide attempts showed asmall, but statistically significant positive correlation withatmospheric pressure (RR01.005, P0.0.026, Table 5). Thiscorrelation remained significant also when the daily values ofatmospheric pressure during the 1 and 2 days before thesuicide attempt (RR01.005, P0.0.027, and RR01.006,
P 0 0.011, Table 5) were taken into account as a mean of thepressure over these days. For men, the number of daily suicideattempts showed a small, but statistically significant negativecorrelation with atmospheric pressure (RR0 .0.995, P0.0.016,Table 5), and the correlation remained statistically significantalso when the value of atmospheric pressure the day before theattempt was taken into consideration (RR0 .0.996, P0.0.040).Furthermore, this correlation was even stronger for violentsuicide attempts by men (RR0 .0.983, P0<0.000), and lastedalso when the values of atmospheric pressure 1 and 2 daysbefore the attempt was included (RR0 .0.984, P0.0.002,RR0 .0.985, P 0 0.006, respectively, Table 5). Accordingly,rate ratios for a change of 10 units per each weather parameterwere calculated (Table 5), which was considered clinicallymeaningful especially for atmospheric pressure for which achange over 10 hPa per day is quite common. The maximumchange per day in our data was 36 hPa.
Global solar radiation (proportion to the normal), tem-perature (deviation from the normal) and precipitation did
Table 4 Days and periodswithout self-harm (attempted/completed suicides) by eithersex (W/M), by women (W) or bymen (M) in separate. Days withpotential impact of harvestingare excluded
aMonday is the most probableday without self-harm
Year Date/period without self-harm Day of the week No self- harm by
W M W/M
1989 9–11 January Mona-Wed x
22–23 January Sun-Mona x
6–7 February Mona-Tue x
17–19 February Fri-Sun x
17–19 April Mona-Wed x
30 June–1 July Fri-Sat x
27 July Thu x
11August Fri x
19–20 September Mona-Wed x
15–16 October Sun-Mona x
6 November Mona x
5–6 November Sun-Mona x
18–19 November Sat-Sun x
19 November Sun x
1990 18–19 January Thu-Fri x
12–14 March Mona-Wed x
23 May Wed x
23–24 May Wed-Thu x
16 July Mona x
1997 8–9 February Sat-Sun x
27–28 March Thu-Fri x
26–27 April Sat-Sun x
1–2 May Thu-Fri x
15 June Sun x
15–16 June Sun-Mona x
1998 12–13 January Mona-Tue x
Int J Biometeorol
not show statistically significant daily correlation with sui-cide attempts for either sex by Poisson regression.
Discussion
We found that atmospheric pressure shows a small, butstatistically significant correlation with attempted suicideson a daily time span, but that the correlation is oppositebetween genders (Fig. 2). Men seem to be more prone toattempt suicide under low pressure conditions, while womenseem to be more prone to do so under high atmosphericpressure. The result was even stronger for the violent suicideattempts of men, which is in line with the prior notion thatsuicides committed by violent methods show a strongercorrelation with weather than suicides committed by non-
violent methods (Lester and Frank 1988; Maes et al. 1994;Preti and Miotto 2000), although the division into violentand non-violent suicidal behavior may be somewhat arbitrary(Ajdacic-Gross et al. 2010). High-pressure situations often lastsome time, whereas troughs pass by much more quickly, whichprovides a logical explanation to our findings that the associa-tion between women’s suicide attempts and high pressure wasseen for up to 3 days, whereas the association between men’ssuicide attempts and low pressure was seen only for 2 days inPoisson regression.
A limitation of our study is that the study periods wererather short. Furthermore, due to the relatively short periodsand the low numbers studied, we were unable to take mentaldisorders or psychosocial factors into account in the analy-sis. Psychiatric disorders and psychosocial factors (Hendin1965) are much stronger risk factors for suicidal behavior
Table 5 Rate ratios for daily global solar radiation (G), temperature(T), precipitation (P) and mean atmospheric pressure (AP) and dailynumber of suicide attempts, analyzed by a Poisson regression for men
and women separately. Daily violent suicide attempts and mean atmo-spheric pressures are analyzed separately. Confidence intervals and Pvalues are calculated according to changes per one unit
Daily Rate ratio for men Rate ratio for women
Suicideattempts
Weatherparameters
Per unit /per 10 units
95% CIper unit
P value Per unit /per 10 units
95% CIper unit
P value
All G* (MJ/m2)a 1.003/1.030 0.915-1.099 n.s. 0.995/0.951 0.907-1.092 n.s.
T** (°C)b 1.009/1.094 0.997-1.021 n.s. 1.001/1.010 0.989-1.013 n.s.
P (mm) 1.001/1.010 0.989-1.013 n.s. 0.995/0.951 0.983-1.008 n.s.
AP (hPa) D−0c 0.995/0.951 0.991-0.999 0.016 1.005/1.051 1.001-1.009 0.026
D−1 0.996/0.961 0.991-1.000 0.040 1.005/1.062 1.001-1.009 0.027
D−2 0.997/0.970 0.993-1.001 n.s. 1.006/0.990 1.001-1.011 .011
Violent AP (hPa) D−0 0.983/0.844 0.974-0.993 <.001 0.999/0.990 0.985-1.013 n.s.
D−1 0.984/0.852 0.974-0.994 0.002 0.999/0.990 0.985-1.014 n.s.
D−2 0.985/0.861 0.974-0.996 0.006 1.002/1.020 0.987-1.018 n.s.
a Proposition to seasonal normal value, period 1971–2000b Anomaly from the seasonal normal value, period 1971–2000a D−00 day of suicide attempt, D−102-day mean; the day of suicide attempt and 1 day prior to it, D−203-day mean; the day of suicide attempt and2 days prior to it
1008
1009
1010
1011
1012
1013
0 1 2 3 4
Number of suicide attempts per day
Women
Men
Average atmospheric pressure (hPa) (hPa)
Fig. 2 Daily atmosphericpressure on average as afunction of daily number ofsuicide attempts, men andwomen shown separately
Int J Biometeorol
compared to weather impact, but may be inadequate toexplain the timing of suicidal behavior. However, sincestatistics on suicide attempts are not compiled on a regularbasis in Finland, our data, collected as a part of a WHOproject, by specially trained nursing staff, represents a rareand an extensive community-based sample from the capitalregion in Finland—a well-defined urban catchment area,with more or less homogeneous social surroundings duringthe study periods. It is also noteworthy that the effect ofatmospheric pressure on an individual can be estimated tobe equal indoors and outdoors, and much less subject totechnical alterations than is the case with lighting and heat-ing for example. Consequently it is a valid variable inrelation to all the individuals who attempted suicide in thestudy area.
Gender
We found that men and women seem to react in oppositeways to changes in atmospheric pressure. This tendency ofgenders to react in opposite ways towards weather variableshas been found also in several prior studies (Barker et al.1994; Breuer et al. 1986; Tsai 2010; Wang et al. 1997; Yan2000). This is also in line with the common knowledge thatthe profile of suicidal behavior in general differs betweenmen and women in many ways. Men and women havedistinct patterns of seasonality in attempted (Yip and Yang2004) and completed (Lester and Frank 1988; Meares et al.1981; Micciolo et al. 1989) suicides, men prefer violentmethods whereas women prefer non-violent methods (Pretiand Miotto 2000). Women have a greater number of non-fatal suicide attempts, whereas men complete three to fourtimes as many suicides compared to women. Furthermore,men also have a different profile in social risk factors(Nordentoft and Branner 2008; Qin et al. 2003; Rocchi etal. 2007). It is beyond the scope our study to find explan-ations for this sex difference, but some involvement of sex-specific hormones has been found for the varying responseto atmospheric pressure (El-Migdadi et al. 2000). Overall, inour opinion, the distinct pattern of suicidal behavior amonggenders should be accorded more attention and taken intoaccount also in future studies.
Interaction between weather and seasons
Our study found an association between atmospheric pres-sure and number of suicide attempts. However, possibleother alternative associations between weather and suicideattempts were also seen depending on the season (see Sup-plementary material on cluster weeks). The impact of asingle weather parameter, such as temperature, or the devi-ation of temperature from the seasonal normal value, candiffer depending on the season. For example, in temperate
climate during summer, weather that is unusually warm forthe season may cause a strong heat stress, while duringautumn a similar anomaly may be thermally comfortable.As far as atmospheric pressure is considered, its impactshould be more equal throughout the year, although obvi-ously things may be not that simple. Accordingly, eventhough the men’s cluster days seemed more likely to occurin a context of low pressure, the cluster days in April andMay were quite the opposite and many suicide attemptswere made on high pressure days with sunny weather. Forwomen, fewer cluster days emerged, but they seemed tocorrelate more often with high than low atmospheric pres-sure. The 3 cluster days defined by the total number ofsuicide attempts per day, without sex specific definition,occurred in winter in high pressure conditions with cloudyweather.
It is a characteristic of the Finnish climate that, under highatmospheric pressure, during summer the weather is typicallywarm and sunny, but during winter the weather can be eithervery cold with a clear sky, or overcast with relatively mildweather. Under low pressure, rain or snowfall and cloudyweather are typical characteristics, and during summer theweather is usually cool, but during winter mild for the season.Unfortunately the various combinations of weather parame-ters on suicide attempts could not be assessed from our datawith the short study periods available.
Large seasonal variability in weather variables mightpartly explain some of the inconclusive associations be-tween weather and suicidal behavior found in differentcountries (Dixon and Kalkstein 2009; Tsai 2010). Accord-ingly, climate zones with marked climate variability fromseason to season could be considered more demanding toacclimate to, especially for individuals with other risk fac-tors such as psychiatric illnesses and, for example, withpossible underlying malfunctions of the circadian clock.
Effects of atmospheric pressure on the central nervoussystem
The biological mechanisms by which suicidal behavior andweather variables might interact with each other are stillunsolved. However, some interesting biological theorieshave emerged. Variation in the amount of sunshine hours,but also light deprivation has been shown to influence brainmetabolism connected to depression in humans (Lambert etal. 2002) and rats (Gonzalez and Aston-Jones 2008). Tem-perature might have an influence through an over-activatedbrown adipose tissue metabolism found among the de-pressed (Huttunen and Kortelainen 1990). An associationbetween atmospheric pressure and levels of monoamines ortheir metabolites, which are involved in the brain metabo-lism of mood disorders, has been shown (Eklundh et al.1994, 2000; Heilig et al. 1996; Nordin et al. 1992).
Int J Biometeorol
Furthermore, in rats, the rate and magnitude of a pressurechange have been shown to increase pain-related behavior(Funakubo et al. 2011; Messlinger et al. 2010; Sato et al.2004) through the inner ear (Funakubo et al. 2010), andneuronal activity in rat spinal trigeminal nucleus (Messlingeret al. 2010). For humans, the biological mechanisms in-volved in the aggravation of neuropathic pain, as a con-sequence of changes in atmospheric pressure, remainunsolved. However, antidepressants are often used in thetreatment of chronic pain (Dharmshaktu et al. 2012).
As for differences between genders, the direction of theassociation between atmospheric pressure and suicideattempts has varied also in previous studies (Maldonadoand Kraus 1991; Nordstrom et al. 1994), although our resultsfor men were opposite. Serum levels of sex-hormones, that is,luteinizing hormone and testosterone in men, and levels offollicle-stimulating hormone and progesterone in women havebeen shown to correlate positively with higher atmosphericpressures (El-Migdadi et al. 2000).
To sum up, our findings and the existing data on neuro-biological mechanisms in the central nervous system sug-gest that several neurotransmitters and also sex-specifichormones may be involved in the interaction of weatherwith suicidal behavior. The clinical importance of the pos-sible association of the atmospheric pressure with suicideattempts as reported here remains to be elucidated.
Acknowledgments The Finnish Cultural Foundation, Finnish Na-tional Graduate School for Clinical Investigation, and Finnish GraduateSchool of Psychiatry allocated scholarships (to L.H.) for this project buthad no further role in study design. We thank Dr. Jari Haukka from theNational Institute of Health and Welfare for the statistical advice.
References
Ajdacic-Gross V, Lauber C, Sansossio R, Bopp M, Eich D, GostynskiM, Gutzwiller F, Rossler W (2007) Seasonal associations betweenweather conditions and suicide—evidence against a classic hy-pothesis. Am J Epidemiol 165:561–569
Ajdacic-Gross V, Bopp M, Ring M, Gutzwiller F, Rossler W (2010)Seasonality in suicide—a review and search of new concepts forexplaining the heterogeneous phenomena. Soc Sci Med 71:657–666
Barker A, Hawton K, Fagg J, Jennison C (1994) Seasonal and weatherfactors in parasuicide. Br J Psychiatry 165:375–380
Bille-Brahe U, Schmidtke A, Kerkhof AJ, De Leo D, Lonnqvist J, PlattS, Sampaio Faria J (1995) Background and introduction to theWHO/EURO Multicentre Study on Parasuicide. Crisis 16(72–8):84
Brent DA, Oquendo M, Birmaher B, Greenhill L, Kolko D, Stanley B,Zelazny J, Brodsky B, Bridge J, Ellis S, Salazar JO, Mann JJ(2002) Familial pathways to early-onset suicide attempt: risk forsuicidal behavior in offspring of mood-disordered suicide attemp-ters. Arch Gen Psychiatry 59:801–807
Breuer HW, Breuer J, Fischbach-Breuer BR (1986) Social, toxicolog-ical and meteorological data on suicide attempts. Eur Arch Psy-chiatry Neurol Sci 235:367–370
Chiu LP (1988) Do weather, day of the week, and address affect therate of attempted suicide in Hong Kong? Soc Psychiatry PsychiatrEpidemiol 23:229–235
Ciucci E, Calussi P, Menesini E, Mattei A, Petralli M, Orlandini S(2011) Weather daily variation in winter and its effect on behaviorand affective states in day-care children. Int J Biometeorol55:327–337. doi:10.1007/s00484-010-0340-2
Deisenhammer EA (2003) Weather and suicide: the present state ofknowledge on the association of meteorological factors withsuicidal behaviour. Acta Psychiatr Scand 108:402–409
Dexter EG (1904) Chapter XI: Suicide and the weather. In: Weatherinfluences: an empirical study of the mental and physiologicaleffects of definite meteorological conditions. MacMillan, NewYork, pp 198–218
Dharmshaktu P, Tayal V, Kalra BS (2012) Efficacy of antidepressantsas analgesics: a review. J Clin Pharmacol 52:6–17. doi:10.1177/0091270010394852
Dixon PG, Kalkstein AJ (2009) Climate–suicide relationships: a re-search problem in need of geographic methods and cross-disciplinary perspectives. Geogr Compass 3:1–14
Eklundh T, Fernstrom V, Nordin C (1994) Influence of tapping-timeand atmospheric pressure on concentrations of monoaminemetabolites in the cerebrospinal fluid: a prospective study infemale volunteers. J Psychiatr Res 28:511–517
Eklundh T, Gunnarsson T, Ornhagen H, Nordin C (2000) Cerebrospi-nal fluid levels of monoamine compounds and cholecystokininpeptides after exposure to standardized barometric pressure. AviatSpace Environ Med 71:1131–1136
El-Migdadi F, Nusier M, Bashir N (2000) Seasonal pattern of leutinizing,follicle-stimulating hormone, testosterone and progesterone in adultpopulation of both sexes in the Jordan Valley. Endocr Res 26:41–48
Funakubo M, Sato J, Honda T, Mizumura K (2010) The inner ear isinvolved in the aggravation of nociceptive behavior induced bylowering barometric pressure of nerve injured rats. Eur J Pain14:32–39
Funakubo M, Sato J, Obata K, Mizumura K (2011) The rate andmagnitude of atmospheric pressure change that aggravate pain-related behavior of nerve injured rats. Int J Biometeorol 55:319–326. doi:10.1007/s00484-010-0339-8
Germain A, Kupfer DJ (2008) Circadian rhythm disturbances in de-pression. Hum Psychopharmacol 23:571–585
Gonzalez MM, Aston-Jones G (2008) Light deprivation damagesmonoamine neurons and produces a depressive behavioral phe-notype in rats. Proc Natl Acad Sci USA 105:4898–4903
Gooley JJ (2008) Treatment of circadian rhythm sleep disorders withlight. Ann Acad Med Singapore 37:669–676
Haukka J, Suominen K, Partonen T, Lonnqvist J (2008) Determinantsand outcomes of serious attempted suicide: a nationwide study inFinland, 1996-2003. Am J Epidemiol 167:1155–1163
Heilig M, Mansson JE, Blennow K (1996) Cerebrospinal fluid mono-amine metabolites and atmospheric pressure. Biol Psychiatry39:299–301
HENDIN H (1965) Cross-cultural studies. Psychiatry Dig Jan 25–34Henriksson MM, Aro HM, Marttunen MJ, Heikkinen ME, Isometsa
ET, Kuoppasalmi KI, Lonnqvist JK (1993) Mental disorders andcomorbidity in suicide. Am J Psychiatry 150:935–940
Hiltunen L, Suominen K, Lonnqvist J, Partonen T (2011) Relationshipbetween daylength and suicide in Finland. J Circadian Rhythms9:10. doi:10.1186/1740-3391-9-10
Huibers MJ, de Graaf LE, Peeters FP, Arntz A (2010) Does the weathermake us sad? Meteorological determinants of mood and depres-sion in the general population. Psychiatry Res 180:143–146.doi:10.1016/j.psychres.2009.09.016
Huttunen P, Kortelainen ML (1990) Long-term alcohol consumptionand brown adipose tissue in man. Eur J Appl Physiol OccupPhysiol 60:418–24
Int J Biometeorol
Huynen MM, Martens P, Schram D, Weijenberg MP, Kunst AE (2001)The impact of heat waves and cold spells on mortality rates in theDutch population. Environ Health Perspect 109(5):463–70
Kent ST,McClure LA, CrossonWL, Arnett DK,WadleyVG, SathiakumarN (2009) Effect of sunlight exposure on cognitive function amongdepressed and non-depressed participants: a REGARDS cross-sectional study. Environ Health 8:34
Lambert GW, Reid C, Kaye DM, Jennings GL, Esler MD (2002) Effectof sunlight and season on serotonin turnover in the brain. Lancet360:1840–1842
Lambert G, Reid C, Kaye D, Jennings G, Esler M (2003) Increasedsuicide rate in the middle-aged and its association with hours ofsunlight. Am J Psychiatry 160:793–795
Lee HC, Lin HC, Tsai SY, Li CY, Chen CC, Huang CC (2006) Suiciderates and the association with climate: a population-based study. JAffect Disord 92:221–226
Lester D (1986) Suicide and homicide rates: their relationship tolatitude and longitude and to the weather. Suicide Life ThreatBehav 16(3):356–9
Lester D, Frank ML (1988) Sex differences in the seasonal distributionof suicides. Br J Psychiatry 153:115–117
Maes M, De Meyer F, Thompson P, Peeters D, Cosyns P (1994) Syn-chronized annual rhythms in violent suicide rate, ambient tempera-ture and the light-dark span. Acta Psychiatr Scand 90:391–396
Majercak J, Sidote D, Hardin PE, Edery I (1999) How a circadianclock adapts to seasonal decreases in temperature and day length.Neuron 24:219–230
Maldonado G, Kraus JF (1991) Variation in suicide occurrence by timeof day, day of the week, month, and lunar phase. Suicide LifeThreat Behav 21(2):174–87
Meares R, Mendelsohn FA, Milgrom-Friedman J (1981) A sex differ-ence in the seasonal variation of suicide rate: a single cycle formen, two cycles for women. Br J Psychiatry 138:321–325
Messlinger K, Funakubo M, Sato J, Mizumura K (2010) Increases inneuronal activity in rat spinal trigeminal nucleus following changesin barometric pressure–relevance for weather-associated headaches?Headache 50:1449–1463. doi:10.1111/j.1526-4610.2010.01716.x
Micciolo R, Zimmermann-Tansella C, Williams P, Tansella M (1989)Seasonal variation in suicide: is there a sex difference? PsycholMed 19:199–203
Mills CA (1934) Suicides and homicides in their relation to weatherchanges. Am J Psychiatry 91:669–677
Monteleone P, Maj M (2008) The circadian basis of mood disorders:recent developments and treatment implications. Eur Neuropsy-chopharmacol 18:701–711
Nordentoft M, Branner J (2008) Gender differences in suicidal intent andchoice of method among suicide attempters. Crisis 29:209–212
Nordin C, Swedin A, Zachau A (1992) CSF 5-HIAA and atmosphericpressure. Biol Psychiatry 31:644–645
Nordstrom P, Samuelsson M, Asberg M, Traskman-Bendz L, Aberg-Wistedt A, Nordin C, Bertilsson L (1994) CSF 5-HIAA predictssuicide risk after attempted suicide. Suicide Life Threat Behav 24:1–9
Owens D, Horrocks J, House A (2002) Fatal and non-fatal repetition ofself-harm. Systematic review. Br J Psychiatry 181:193–199
Page LA, Hajat S, Kovats RS (2007) Relationship between dailysuicide counts and temperature in England and Wales. Br J Psy-chiatry 191:106–112
Partonen T, Haukka J, Pirkola S, Isometsa E, Lonnqvist J (2004) Timepatterns and seasonal mismatch in suicide. Acta Psychiatr Scand109:110–115
Petridou E, Papadopoulos FC, Frangakis CE, Skalkidou A, TrichopoulosD (2002) A role of sunshine in the triggering of suicide. Epidemi-ology 13:106–109
Pirkola S, Isometsa E, Heikkinen M, Lonnqvist J (1997) Employmentstatus influences the weekly patterns of suicide among alcoholmisusers. Alcohol Clin Exp Res 21(9):1704–6
Preti A (1998) The influence of climate on suicidal behaviour in Italy.Psychiatry Res 78:9–19
Preti A, Miotto P (2000) Influence of method on seasonal distribu-tion of attempted suicides in italy. Neuropsychobiology 41:62–72
Qin P, Agerbo E, Mortensen PB (2003) Suicide risk in relation tosocioeconomic, demographic, psychiatric, and familial factors: anational register-based study of all suicides in Denmark, 1981–1997. Am J Psychiatry 160:765–772
Rocchi MB, Sisti D, Miotto P, Preti A (2007) Seasonality of suicide:relationship with the reason for suicide. Neuropsychobiology56:86–92
Roenneberg T, Kumar CJ, Merrow M (2007) The human circadianclock entrains to sun time. Curr Biol 17:R44–5
Ruuhela R, Hiltunen L, Venalainen A, Pirinen P, Partonen T (2009)Climate impact on suicide rates in Finland from 1971 to 2003. IntJ Biometeorol 53:167–175
Salib E, Gray N (1997) Weather conditions and fatal self-harm in NorthCheshire 1989–1993. Br J Psychiatry 171:473–477
Sato J, Aoyama M, Yamazaki M, Okumura S, Takahashi K, FunakuboM, Mizumura K (2004) Artificially produced meteorologicalchanges aggravate pain in adjuvant-induced arthritic rats. Neuro-sci Lett 354:46–49
Schory TJ, Piecznski N, Nair S, el-Mallakh RS (2003) Barometricpressure, emergency psychiatric visits, and violent acts. Can JPsychiatry 48(9):624–6
Sinclair JM, Hawton K, Gray A (2010) Six year follow-up of a clinicalsample of self-harm patients. J Affect Disord 121:247–252
Souetre E, Salvati E, Belugou JL, Douillet P, Braccini T, Darcourt G(1987) Seasonality of suicides: environmental, sociological andbiological covariations. J Affect Disord 13:215–225
Souetre E, Wehr TA, Douillet P, Darcourt G (1990) Influence ofenvironmental factors on suicidal behavior. Psychiatry Res32:253–263
Souetre E, Salvati E, Candito M, Darcourt G (1991) Biological clocksin depression: phase-shift experiments revisited. Eur Psychiatry6:21–29
Stoupel E, Petrauskiene J, Kalediene R, Abramson E, Sulkes J (1995)Clinical cosmobiology: the Lithuanian study 1990–1992. Int JBiometeorol 38:204–208
Suokas J, Suominen K, Isometsa E, Ostamo A, Lonnqvist J (2001)Long-term risk factors for suicide mortality after attempted sui-cide—findings of a 14-year follow-up study. Acta PsychiatrScand 104:117–121
Suominen K, Henriksson M, Suokas J, Isometsa E, Ostamo A,Lonnqvist J (1996) Mental disorders and comorbidity in attemp-ted suicide. Acta Psychiatr Scand 94:234–240
Suominen K, Isometsa E, Suokas J, Haukka J, Achte K, Lonnqvist J(2004) Completed suicide after a suicide attempt: a 37-yearfollow-up study. Am J Psychiatry 161:562–563
Tsai JF (2010) Socioeconomic factors outweigh climate in the regionaldifference of suicide death rate in Taiwan. Psychiatry Res179:212–216. doi:10.1016/j.psychres.2008.06.044
Valtonen H, Suominen K, Partonen T, Ostamo A, Lonnqvist J(2006) Time patterns of attempted suicide. J Affect Disord90:201–207
Wang YT, Wang D, Wang XY (1997) Suicide and meteorologicalfactors in Huhhot, Inner Mongolia. Crisis 18:115–117
Wehr TA, Rosenthal NE (1989) Seasonality and affective illness. Am JPsychiatry 146:829–839
Yan YY (2000) Geophysical variables and behavior: LXXXXIX. Theinfluence of weather on suicide in Hong Kong. Percept Mot Skills91:571–577
Yip PS, Yang KC (2004) A comparison of seasonal variation betweensuicide deaths and attempts in Hong Kong SAR. J Affect Disord81:251–257
Int J Biometeorol
1
Supplementary material
Synoptic analyses of cluster weeks in each season
The main focus of our article was on daily associations between weather and suicide attempts. However, it may
be beneficial to study also longer periods such as weeks in order to understand possible effects of a longer
exposure time or lag. Furthermore, weather impact may vary also according to the season. In this
supplementary material we describe the weather type during the week with highest self-harm incidents in each
season. The average expected number of self-harm per week calculated from our data was 32 with 29 suicide
attempts. About half of the attempts were by women. Because of the shortness of the study periods statistical
methods were not applied to this weekly approach and only descriptive analyses are provided.
Cluster week in spring
The highest number of self-harm cases during one week period occurred from 12 to 18 March 1989 (Sunday to
Saturday). The total number of attempted and completed suicides during this period was 56 (51 attempts) of
which 25 (22 attempts) by women. Within this period there were two cluster days of suicide attempts (Tuesday
14 and Saturday 18 March, Table 2).
The week prior to the cluster week was very dark with overcast skies and high atmospheric pressure. Weather
was warm for the season and snow cover had melted. On Sunday 12 March high pressure started to weaken,
gradually at first, and then on the cluster day on Tuesday14 March a rapid and large pressure decrease, more
than 10 hPa per 12 h occurred, preceding the passage of a low pressure system over Finland towards the
northeast. It also rained and air temperature was above zero degrees of Celsius. The low pressure situation was
followed by another low centre passing over southern Finland on Friday 17 March. On the male cluster day of
suicide attempts (Saturday 18 March 1989) the centre of the low was already over northern Finland and
atmospheric pressure was increasing. There was no precipitation and the skies were clear, with more than 8
hours of sunshine.
Besides the high numbers of men’s suicide attempts during the two cluster days, men dominated the overall
number of suicide attempts during this cluster week. This is well in line with our results of the daily association
between men’s suicide attempts and low atmospheric pressure since during this week low atmospheric pressure
and rapid pressure changes had a clear role determining the weather. Another interesting characteristic was
2
cloudiness. Mainly overcast weather prevailed until the cluster day on 18 March when suddenly the sky became
clear and amount of solar radiation increased.
Cluster week in summer
In summer the highest number of self-harm cases occurred during the 7-day period from 5 to 11 July 1989
(Wednesday to Tuesday). During this week there were 48 incidents of self-harm (42 attempts); of which by
women 23 (22 attempts). One cluster day occurred on Wednesday 5 July with 8 attempts by women (Table 2).
A period of high pressure over Finland with clear skies and hot weather prevailed from Monday 3 July until 9
July. The hottest days were Tuesday 4 and Wednesday 5 with maximum temperatures exceeding +28 oC. For the
Finnish climate the weather is considered hot when temperature exceeds +25 oC. Besides the cluster day of
suicide attempts by women, the number of self-harm cases continued to be somewhat elevated until the end of
the hot period. The weather began to cool, and by the 11 July the maximum temperature was only 18 oC and the
daily precipitation about 10 mm. The interesting weather phenomena during this period were high pressure
together with a heat-wave that culminated into a thunderstorm and heavy precipitation. This week also supports
our findings from the daily association between high atmospheric pressure and women’s suicide attempts, since
women’s attempts dominate during this week, although only slightly. However, the heat stress may have played
a part during this week as well, since association between temperature and increased suicide mortality have been
suspected in many other studies.
Cluster weeks in autumn
Considering incidents of self-harm, September 1997 was black all in all and we chose two periods for the
analyses. A week with 55 incidents of self-harm (48 attempts) of which 30 (28 attempts) by women occurred
from 10 to 16 September (Wednesday to Tuesday). The number of female suicide attempts was elevated during
the whole one-week period, but there were no single days filling our criteria for a cluster day. The weather
pattern was dominated by a low pressure activity, and it was partly cloudy together with rain showers, yet it was
warm for the season. It rained on six days out of seven.
Towards the end of the month there was an additional one-week period of note. During the week from 20 to 26
September (Saturday to Friday) there were 51 incidents of self-harm of which 19 (18 attempts) by women. In
this period, it was the number of male suicide attempts that was elevated during the period, but there were no
single cluster days. The atmospheric pressure would be classified mainly as normal during this period. The sky
3
was partly cloudy and it rained on 3 days during the week. The weather was some degrees cooler than normally
for the season (compared to normal 1971-2000).The results of these two cluster weeks are not in line with our
findings concerning the daily association between atmospheric pressure and suicide attempts. Risk factors other
than weather may be more important. The possible weather association could come across through relatively
high number of rainy days.
Cluster week in winter
The week with the highest incidents of self-harm in winter occurred from 2 to 8 January 1990 (Tuesday to
Monday). There were 54 cases of self-harm (46 attempts) of which 23 (20 attempts) by women. One cluster day
of suicide attempts occurred on Tuesday 2 January (Table 2.), in which New Year festivities may have had some
impact. However, after that, the number of suicide attempts remained high with 9 attempts on two days.
The main characteristics of the weather would be summarised as persistent high pressure with overcast sky,
temperature mainly below 0 oC. In the beginning of the week it was cold for the season, but later on warm for
the season. This week supports our finding of association between high pressure and suicide attempts of women.
Lack of sunshine could be another weather parameter to explain high number of self-harm incidents. Again,
some other risk factors, such as the effect of social activities during holiday seasons may well be responsible for
the high numbers of suicide attempts during this week.
Supplementary table 1. Daily distribution of suicide attempts during cluster weeks in each season,
for women (W) and men (M) in separate.
Season and dates of cluster week
Gender No of suicide attempts per day Tot. No of sa. per week
Spring: 12.-18.3.1997 W 3 2 6 2 1 5 3 22
M 6 4 5 2 2 2 8 29
Summer:5.-11.7.1989 W 8 3 2 1 2 4 2 22
M 2 3 3 3 2 4 3 20
Autumn 10.-16.9.1997 W 6 5 3 2 3 5 4 28
M 3 3 4 6 2 1 1 20
20.-26.9.1997 W 4 1 2 3 2 3 3 18
M 5 5 4 4 4 4 5 31
Winter: 2.-8.1.1990 W 4 2 6 0 5 1 2 20
M 7 3 3 3 4 3 3 26
FINNISH METEOROLOGICAL INSTITUTEErik Palménin aukio 1
P.O. Box 503
FI-00101 HELSINKI
tel. +358 29 539 1000
WWW.FMI.FI
FINNISH METEOROLOGICAL INSTITUTE
CONTRIBUTIONS No. 147
ISBN 978-952-336-059-4 (paperback)
ISSN 0782-6117
Erweko
Helsinki 2018
ISBN 978-952-336-060-0 (pdf)
Helsinki 2018