Introduction

The thermal environment can affect health and well-being (World Health Organization (WHO) 2022): adverse outcomes have been associated with both hot and cold ambient conditions (Zhao et al. 2021; Gasparrini et al. 2015). More specifically, cardiovascular, respiratory, and cerebrovascular diseases are exacerbated in extreme thermal conditions (Turner et al. 2012; Liu et al. 2022; Ma et al. 2018); certain pregnancy outcomes, such as a preterm birth, low birth weight, and stillbirths, are more likely to occur when the air temperature increases (Chersich et al. 2020); adverse mental health outcomes have been associated with elevated ambient temperatures (Liu et al. 2021); the incidence of infectious diseases varies across different thermal environments (Fong and Smith 2022; Liang et al. 2021); and mild heat-related symptoms may even be experienced at elevated, though climatologically normal, ambient temperatures (Pantavou et al. 2016, 2020).

Globally, the exposure of human populations to unfavorable thermal conditions is increasing and the frequency of extreme weather events is rising due to climate change (World Health Organization (WHO) 2018). Vulnerable populations include people with underlying health conditions, the elderly, pregnant women, infants and children, workers outdoors, athletes, and poor people (World Health Organization (WHO) 2018). The vulnerability of a population can be affected by urbanization, population growth and aging, and several socio-economic factors (Ebi et al. 2021). Developing countries with weak health infrastructures are considered to be the most vulnerable areas (World Health Organization (WHO) 2018).

Recent research findings suggest a greater effect of cold than of warm thermal conditions on mortality (Gasparrini et al. 2015; Chigozie et al. 2022). Moreover, temperature-related mortality projection scenarios show that vulnerability could decrease over time in some regions, although they still suggest that premature mortality remains an important public health risk (Potchter et al. 2018). At the same time, integrated approaches assessing the thermal environment are increasingly used. Given that air temperature alone cannot sufficiently describe the thermal environment, several studies (Romaszko et al. 2022; Avalos et al. 2017; Pantavou et al. 2021) incorporated thermal indices into their methodologies as measures of the effect of the thermal environment on human health. The physiologically equivalent temperature (PET) and the universal thermal climate index (UTCI) are two of the most widely used indices in research (Potchter et al. 2018) and are derived from multi-node human thermal energy models (Mayer and Höppe 1987; Hoppe 1993; Höppe 1999; Fiala et al. 2012; Kruger et al. 2012; Bröde et al. 2012).

The harmful thermal conditions are well known and precautionary measures have been defined and communicated to people in order to prevent and mitigate health risks. Nevertheless, heat- or cold-related casualties are still reported around the world, although the number of casualties varies with the region and climate, mainly due to adaptation (Gasparrini et al. 2015; Liu et al. 2022; Chigozie et al. 2022; Kephart et al. 2022; Scortichini and De’Donato et al. 2018; Evangelopoulos et al. 2021). The minimum mortality temperature within the U-, V-, or J-shaped pattern of the temperature–mortality relationship has been found to vary by location (Gasparrini et al. 2015) and to be linked to the temperature most commonly experienced locally (Yin et al. 2019). This local average temperature is the one people are typically exposed to and adapt to physiologically. The Mediterranean region is considered a climate change hotspot that is at risk from significant extreme-weather-related social and economic impacts (Hochman et al. 2022). The aim of the present study was to examine the effect of the thermal environment on all-cause mortality in the Republic of Cyprus, an island in the eastern Mediterranean Sea.

Materials and methods

Mortality data

Daily mortality data from 1 January 2009 to 31 December 2018 were provided by the Health Monitoring Unit of the Ministry of Health of the Republic of Cyprus (2023). There was information on the date of death, age, sex, district of residence (i.e., Nicosia, Limassol, Larnaca, Pafos, and Ammochostos), and cause of death according to the International Statistical Classification of Diseases and Related Health Problems (ICD-10 version 2010). A map of the Republic of Cyprus is shown in Fig. 1.

Fig. 1
figure 1

Map of Cyprus and locations of the capitals of the five administrative districts (Nicosia, Limassol, Larnaca, Paphos, and Ammochostos) in the Republic of Cyprus. Each districts bears the same name as its capital

Full size image

The Cyprus National Bioethics Committee (EEBE/EP2018/48) approved the protocol of this study.

Themal envirionment and air quality

Hourly data on the air temperature (Tair, °C; at 2 m height), dew-point temperature (Td, °C; at 2 m height), eastward and northward components of the wind speed (WS, m/s; at 10 m height), and solar radiation (SR) reaching the Earth’s surface were retrieved from the ERA5-Land reanalysis database of the European Centre for Medium-Range Weather Forecasts (Muñoz Sabater 2012). The ERA5-Land datasets are available with hourly time intervals and a spatial horizontal resolution of 0.1° × 0.1°.

Hourly concentrations of nitrogen monoxide (NO, μg/m3), dioxide (NO2, μg/m3), nitrogen oxides (NOx, μg/m3), ozone (O3, μg/m3), carbon monoxide (CΟ, μg/m3), sulfur dioxide (SO2, μg/m3), and benzene (μg/m3) and the daily concentrations of particulate matter ≤ 10 μm in diameter (PM10, μg/m3) and ≤ 2.5 μm in diameter (PM2.5, μg/m3) were obtained by the Department of Labour Inspection of the Ministry of Labour, Welfare, and Social Insurance of the Republic of Cyprus (Ministry of Finance of the Republic of Cyprus 2021).

The meteorological data were retrieved for the closest available grid point of ERA5-Land and the air quality data were obtained for the traffic station in each district.

Data processing

The PET (°C) (Mayer and Höppe 1987; Höppe 1999) and UTCI (°C) (Fiala et al. 2012) were used to assess the thermal environment. The PET and UTCI are models, namely thermal indices, that were developed to assess thermal perception using Tair, relative humidity (Rh, %), WS, and SR in one value. The value of a thermal index can be expressed as the degree of thermal sensation or stress according to the relevant assessment scales, which include a neutral point (i.e., 0/“comfortable-neutral”/“no thermal stress”; 18 °C < PET < 23 °C, 9 °C < UTCI < 26 °C) and two opposite subscales ranging from a slightly cold/warm sensation/stress to an extreme cold/warm sensation/stress (Table 2 in the Appendix). The PET and UTCI provide a thermophysiological approach to the evaluation of human physiological responses to variation in the thermal environment and are widely used in outdoor thermal perception studies (Potchter et al. 2018). Both indices were estimated using the RayMan software (Matzarakis et al. 2010, 2007) and the hourly values of the meteorological data. The RH was calculated using Td. The WS was computed at a height of 2 m using the logarithmic wind profile for a roughness length of 1 m (Stull 2011). The personal data were configured as follows: a male individual with a height of 1.8 m, a weight of 75 kg, and an age of 35 years. The clothing and activity levels were set at 0.90 clo and 80 W/m2, respectively.

Daily mortality per 105 population was calculated by dividing the daily number of deaths by the population of each district. Daily mean values of Tair, PET, UTCI, and the concentrations of air pollutants were calculated using hourly values and were matched to the daily number of deaths.

Statistical analysis

The statistical analysis was conducted for groups A–R (non-accidental causes) of ICD-10 mortality. Records with an unspecified or unknown city of residence and non-permanent residents of Cyprus were excluded from the analysis.

The descriptive analysis included the calculation of the mean, median, standard deviation, maximum, and minimum values and the interquartile range. Time trends were examined using Spearman’s rho. Differences between means were examined using analysis of variance (ANOVA). Negative binomial regression was applied—as appropriate for count and overdispersed data—to examine the association between the response outcome (daily number of deaths) and the predictive variables (Tair, PET, and UTCI). The models were adjusted for year, district, and air pollutant concentrations. The negative binomial regression coefficient represents the expected change in the difference of the natural logs of expected counts (i.e., the natural log of the ratio of expected counts) for the response variable for one unit change in the predictor variable when the other predictor variables in the model are constant. The exponentiated regression coefficient is the incidence rate ratio (IRR). A 95% confidence interval (CI) that did not contain 1 (p value less than 0.05) was considered to be statistically significant.

Subgroup analysis was carried out for cold (December to March) and warm (April to November) periods, for winter (December, January, February) and summer (June, July, August), for people older than 64 years, and for cardiovascular and respiratory causes (ICD-10 groups I and J). The delayed effect of thermal conditions was examined using mean daily values averaged over 2 (0–1), 3 (0–2), 4 (0–3), 5 (0–4), and 6 (0–5) days.

Piecewise linear regression was used to check for potential break points (thresholds) in the associations of daily mean Tair, PET, and UTCI with mortality. Tair, PET, and UTCI values were divided into 1 °C intervals (independent variables) and regression lines were fitted to subsets, modeling the association between the independent variable and mortality. The break points were defined based on whether there was a statistically significant difference in the regression coefficients.

The analysis was performed using the statistical software STATA 17.0 (Stata Corp., College Station, TX, USA).

Results

Mortality data

The dataset included 49,866 records (mean age ± standard deviation: 77.8 ± 14 years) due to non-accidental causes of mortality (Table 1), of which 52% (n = 25,934) were for males (mean age ± standard deviation: 75.7 ± 14.2 years) and 48% (n = 23,932) were for females (mean age ± standard deviation: 80.1 ± 13.4 years). Most of the deaths were attributed to cardiovascular diseases (ICD-10: group I; 36.5%, n = 18,207) followed by neoplasms (ICD-10: group C00-D48; 25.2%, n = 12,562). Respiratory causes (ICD-10: group J) accounted for 8.8% (n = 4397) of the deaths. The mean daily number of deaths per 105 population ranged between 1.7 in Nicosia and 3.1 in Ammochostos (Table 1). Ammochostos also experienced the highest daily mortality for people over 64 years (2.9 ± 1.4) and for mortality due to cardiovascular (2.4 ± 0.8) and respiratory (2.2 ± 0.4) causes. This can be partially attributed to the fact that Ammochostos had the highest percentage of males in its population. The daily mean mortality rate (per 105 population) was positively correlated to the percentage of males in each district (r = 0.89, p = 0.044). The annual mean total mortality (per 105 population) increased (p value < 0.001) over the years, from 1.7 in 2010 to 2.3 in 2017.

Table 1 Number of deaths and mean daily mortality per 105 population in the five districts (Nicosia, Limassol, Larnaca, Pafos, and Ammochostos) of the Republic of Cyprus during 2009–2018
Full size table

Thermal environment and air quality

Mean daily Tair ranged between 5.1 °C (Larnaca) and 34.8 °C (Limassol), with the lowest average of the daily mean Tair recorded in Pafos (19.3 °C) and the highest in Larnaca (20.8 °C; Table S1 in the supplemental file). PET and UTCI predictions varied with respect to the degree of thermal perception. The 5th percentiles of the thermal indices were 8.8 °C (category “cool”) for PET and 10.3 °C (category “no thermal stress”) for UTCI. The 95th percentiles of the thermal indices were 33.7 °C (category “warm”) for PET and 33.4 °C (category “strong heat stress”) for UTCI. The mean daily Tair, PET, and UTCI differed among the years (p value < 0.001), and an increasing trend was seen for Tair (p value = 0.048) and UTCI (p value = 0.020). On the contrary, there was a decreasing trend (− 0.42 ≤ Spearman’s rho ≤ − 0.04; p value < 0.037) for the daily mean concentrations of all pollutants, which were below the thresholds in the European Commission guidelines (European Commission 2017).

Mortality and thermal environment

The highest percentage of deaths (28.7%) was observed in winter (December, January, and February; Fig. 2a) and the lowest (22.8%) was observed in summer (June, July, and August). The mean number of deaths per 1 °C ranged between 2.1 and 2.8 (per 105 population), while evidence of increased mortality was observed for a daily mean Tair of ≤ 17 °C (Fig. 2b).

Fig. 2
figure 2

a Monthly average of the air temperature and percentage of all deaths for each calendar month. b Mean number of deaths per 105 population in 1 °C intervals of air temperature and the frequency that the daily mean air temperature is within each of these 1 °C intervals in Cyprus during 2009–2018

Full size image

Figure 3 shows the IRR and the 95% CI for the negative binomial regression models for different population groups (Fig. 3a) and for the cold period (December to March; Fig. 3b) and warm period (April to November; Fig. 4c). The air quality parameters that remained in the overall models were PM2.5 and NOx. PM2.5 (warm period) and NOx (cold period) concentrations were associated with an increase in mortality.

Fig. 3
figure 3

Estimated incidence rate ratio (IRR) and 95% confidence interval (95% CI) obtained using negative binomial regression for the impacts of air temperature (Tair), physiologically equivalent temperature (PET), and universal thermal climate index (UTCI) on mortality (dependent variable) a overall, b in the cold period of the year (December to March), and c in the warm period of the year (April to November). All models were adjusted for year, district, and air pollutant concentrations. The IRR is statistically significant if the 95% CI does not contain 1

Full size image
Fig. 4
figure 4

Estimated incidence rate ratio (IRR) and 95% confidence interval obtained using negative binomial regression for the impact of a air temperature (Tair), b physiologically equivalent temperature (PET), and c universal thermal climate index (UTCI) on mortality (deaths per 105 population) for each district of the Republic of Cyprus during 2009–2018

Full size image

All associations in the overall analysis (Fig. 3a) were statistically significant, with IRR ranging between 0.981 and 0.997 (p values < 0.001; Fig. 3a and Table S2 in the supplemental file), indicating an increase of 0.3–1.9% in mortality for a 1 °C reduction in Tair, PET, or UTCI. In particular, the analysis of all data showed that a reduction of 1 °C in Tair, PET, or UTCI was associated with an increase in mortality of about 1% (IRR: 0.99; p values < 0.001) (Fig. 3a and Table S2). This negative association persisted in subgroup analyses by sex (0.9–1.2%; p values < 0.001), for people older than 64 years (1.7%, IRR: 0.98 for Tair; 1.3% and 1.4%, IRR: 0.99 for PET and UTCI, respectively; p values < 0.001), and for deaths attributed to cardiovascular causes (1.9%, IRR: 0.98 for Tair; 1.4% and 1.5%, IRR: 0.99 for PET and UTCI, respectively; p values < 0.001; Fig. 3a and Table S2).

Similar results were found in the subgroup analysis by district (Fig. 4, Table S3). A 1 °C decrease in Tair, PET, or UTCI was associated with an increase in all-cause mortality of about 1% (IRR: 0.99; p values ≤ 0.001), with the highest percent change estimated for Limassol (1.5%, IRR: 0.98 for Tair; 1.1%, IRR: 0.99 for PET, and 1.2%, IRR: 0.99 for UTCI; p values < 0.001) and the lowest for Ammochostos (0.8%, IRR: 0.99 for Tair; 0.6%, IRR: 0.99 for PET, and 0.5%, IRR: 0.99 for UTCI; p values ≤ 0.001).

The effects of Tair, PET, and UTCI on mortality (increase in mortality per 1 °C decrease) were largest in the cold period (Fig. 3b, Table S2) for all deaths (3.3%, IRR: 0.97 for Tair; 1.8%, IRR: 0.98 for PET and UTCI; p values < 0.001), for males (3.3%, IRR: 0.97 for Tair; 2.3%, IRR: 0.98 for PET; p values < 0.001), and for people over 64 years (4%, IRR: 0.96 for Tair; 2.2%, IRR: 0.98 for PET and UTCI; p values < 0.001). In the warm period, the association reversed, meaning that a 1 °C increase in Tair was associated with an increase in mortality of about 1.4% (IRR: 1.01; p value < 0.001) (Fig. 3c and Table S2). Similarly, a 1 °C increase in PET or UTCI was associated with an increase in mortality of about 1.1% (IRR: 1.01; p values < 0.001) (Fig. 3c and Table S2). Nevertheless, the results of subgroup analyses were statistically significant only for males (0.9% for Tair; 0.7% for PET and UTCI, IRR: 1.01; p values < 0.01), for females (1.1% for Tair; 0.9% for PET and UTCI, IRR: 1.01; p values < 0.01), and for the group of people older than 64 years (1.3%, for Tair; 1%, for PET, IRR: 1.01; p values < 0.001).

Several subgroup analyses, such as those for females and for people with cardiovascular and respiratory causes of death, became statistically significant when the analysis focused on the winter and summer seasons (Fig. 5, Table S4). The increase in mortality for a 1 °C change in temperature was higher in winter (December to February) and summer (June to August) compared to the cold and warm periods, respectively (Fig. 5). The percent increase in mortality (all data) varied between 2% (UTCI, IRR: 0.98; p value < 0.001) and 3.6% (Tair, IRR: 0.96; p value < 0.001) in winter and between 1.2% (UTCI, IRR: 1.01; p = 0.002) and 4.1% (Tair, IRR: 1.04; p value < 0.001) in summer.

Fig. 5
figure 5

Estimated incidence rate ratio (IRR) and 95% confidence interval (95% CI) obtained using negative binomial regression for the impacts of air temperature (Tair), physiologically equivalent temperature (PET), and universal thermal climate index (UTCI) on mortality (dependent variable) in a winter (December, January, February) and b summer (June, July, August). All models were adjusted for year, district, and air pollutant concentrations. The IRR is statistically significant if the 95% CI does not contain the value 1

Full size image

Figure 6 shows the lag effect of the outdoor thermal environment on mortality. The effect of cool temperatures increases as the lag in days increases (p value ≤ 0.001), reaching up to 5.1% (IRR: 0.95, p value < 0.001) for Tair, 3.7% (IRR: 0.96, p value < 0.001) for PET, and 3.5% (IRR: 0.97, p value < 0.001) for UTCI at a lag of 5 days (Table S5). In summer, the effects of PET and UTCI increase as the lag in days increases, reaching an increase in mortality of up to 2.6% (IRR: 1.026, p value < 0.001) and 1.8% (IRR: 1.018, p value < 0.001), respectively, for a 1 °C increase in the past 6 days (0–5) on average.

Fig. 6
figure 6

Estimated incidence rate ratio (IRR) and 95% confidence interval (95% CI) obtained using negative binomial regression for the lag effects of air temperature (Tair), physiologically equivalent temperature (PET), and universal thermal climate index (UTCI) on mortality in a winter (December, January, February) and b summer (June, July, August). All models were adjusted for year, district, and air pollutant concentrations. The IRR is statistically significant if the 95% CI does not contain the value 1

Full size image

The threshold temperatures above which there are changes in the associations of the daily mean Tair, PET, and UTCI with mortality were identified using piecewise linear regression. The slope of the association between Tair and mortality was significantly lower for Tair ≤ 17 °C than for Tair > 17 °C (p value = 0.019) and higher for Tair ≥ 29.1 °C than for Tair < 29.1 °C (p value = 0.011; Figure S1). A PET threshold of 18 °C (p value = 0.028) was identified for cold temperatures, while the UTCI thresholds were 18.5 °C (p value = 0.034) and 29 °C (p value = 0.007).

Discussion

The present study examined the association between thermal conditions and all-cause mortality from 2009 to 2018 in the Republic of Cyprus. The analysis was carried out throughout the year and considered integrated approaches for assessing the thermal environment (i.e., PET and UTCI) and air pollutant concentrations as confounding factors. Analysis of cold and warm periods showed that both heat and cold increase all-cause mortality, although a greater impact was found for cold conditions. Notably, PM2.5 and NOx appeared to act as confounding factors in this relationship.

The findings of this research are consistent with the results of previous studies. Non-optimal cold or warm thermal conditions have been linked with an increased risk of death (Liu et al. 2022; Chigozie et al. 2022; Kephart et al. 2022). In fact, it has been suggested that moderately cold and hot temperatures may contribute most to the total mortality burden, meaning that only a low proportion of the burden is due to extreme thermal conditions (Gasparrini et al. 2015). This has been also suggested to be particularly true for Mediterranean cities as compared to cities of Northern Europe (Scortichini et al. 2018). In line with our study, cold has been found to be associated with a higher mortality than heat (Gasparrini et al. 2015; Chigozie et al. 2022), and the mortality burden was found to be larger among older people and for people with cardiovascular causes of death (Kephart et al. 2022). A multi-country study (Gasparrini et al. 2015) showed that the majority of temperature-related deaths were primarily due to the influence of cold temperatures, especially on moderately cold days. The same trend was particularly evident among the Mediterranean countries included in the study (Gasparrini et al. 2015).

Previous studies in Cyprus have mainly focused on heat-related effects (Pantavou et al. 2020; Heaviside et al. 2016; Lubczyńska et al. 2015; Tsangari et al. 2016), probably due to Cyprus’s warm climate (type Csa and BSh; Kottek et al. 2006). These studies also support a positive association between high temperatures and all-cause and cardiovascular morbidity (Pantavou et al. 2021; Tsangari et al. 2016) and mortality (Heaviside et al. 2016; Lubczyńska et al. 2015; Tsangari et al. 2015). The effect of low temperature was absent for all-cause hospital admissions examined for the same period (2009–2018) in all general hospitals of the Republic of Cyprus (Pantavou et al. 2021). This could be due to the different characteristics of the datasets of mortality and hospital admissions. Specifically, the mean age of the hospitalized patients (50.3 years for males and 49 years for females) was lower compared to those in the mortality dataset. In terms of air quality, there was no evidence of any effect of particulate matter on heat-related mortality (Heaviside et al. 2016; Tsangari et al. 2015), while an effect of NO2 on heat-related morbidity was found (Pantavou et al. 2021).

A limitation of this study is that an adjustment for social and economic factors was not possible. Moreover, the mortality attributed to influenza could not be identified in our dataset. On the other hand, this study provides a comprehensive assessment of the association between the thermal environment and mortality risk during a prolonged period of time.

Conclusions

A significant burden of mortality is attributed to less-than-optimal outdoor thermal conditions worldwide. This study has focused on Cyprus, a country in the eastern Mediterranean Sea. The analysis showed a higher mortality rate in winter than in summer, with a larger proportion of deaths observed during the colder months. Mortality rates were elevated among older people, as were mortality rates due to cardiovascular causes. Interestingly, vulnerability to cold weather was more pronounced among males than among females, while females appeared to be more susceptible than males in warmer months.

The findings suggest that public awareness of the health risks of both cold and warm thermal conditions should be raised. People in warm settings, such as in Cyprus’s climate, are probably adapted, well informed (through awareness campaigns), and prepared (through effective weather forecasts and measures issued by public health authorities) to deal with heat. On the contrary, they may underestimate the effects of exposure to colder conditions. Thus, awareness campaigns during the cold period are also needed. The public should be aware that cold conditions can be as harmful as heat, especially for vulnerable groups like children, the elderly, and people experiencing circulation-related disorders and diseases. Additionally, other factors, such as receiving certain medications, may increase susceptibility to cold. Special attention should also be given to encouraging people to stay updated through weather forecasts, to follow warnings and advisories, and to be well prepared to effectively handle cold conditions.

Given that extreme weather events are expected to occur often in the coming decades, awareness should be kept high, and adaptation measures should be improved and updated constantly.