Temperature and hospital admissions in the Eastern Mediterranean: a case study in Cyprus

Exposure to extreme temperatures can trigger a cascade of adverse cardiovascular and respiratory events. However, in Cyprus, a hotspot of climate change in the Eastern Mediterranean region, little is known about the temperature-related cardiorespiratory morbidity risks. We analyzed daily counts of hospital admissions for cardiovascular and respiratory diseases from four general hospitals in three districts in Cyprus from 2000 through 2019. For each district, we fitted time-series quasi-Poisson regression with distributed lag non-linear models to analyze the associations between daily mean temperature (lag 0–21 d) and hospital admissions for cardiorespiratory, cardiovascular, and respiratory diseases. A random-effects meta-analytical model was then applied to pool the district-specific estimates and obtain the national average associations. We analyzed 20 years of cause-specific hospitalization data with a total of 179 988 cardiovascular and respiratory events. The relationships between cardiorespiratory morbidity and temperature were overall U-shaped. During extreme temperature days, 15.85% (95% empirical CI [eCI]: 8.24, 22.40%) excess cardiovascular hospitalizations and 9.59% (95% eCI: −0.66, 18.69%) excess respiratory hospitalizations were attributable to extreme cold days (below the 2.5th percentile). Extreme hot days (above the 97.5th percentile) accounted for 0.17% (95% eCI: 0.03, 0.29%) excess cardiovascular hospitalizations and 0.23% (95% eCI: 0.07, 0.35%) excess respiratory hospitalizations. We found evidence of increased cardiovascular morbidity risk associated with extreme temperatures in Cyprus. Our study highlights the necessity to implement public health interventions and adaptive measures to mitigate the related temperature effects in an understudied region.


Introduction
In the context of global warming, climate extremes that have resulted in adverse impacts on nature and humans are likely to be more frequent and intense in the future [1].High and low non-optimal temperatures have been recently included as a leading risk factor in the Global Burden of Disease Study 2019 [2].Extreme cold or hot conditions are well recognized to lead to a greater risk of morbidity and mortality from a wide range of causes [3][4][5][6][7].The related risk is projected to rise even more with higher degrees of global warming over time [8].
Cardiorespiratory diseases are among the main health outcomes attributable to extreme temperatures worldwide [9].A growing body of epidemiological evidence demonstrates the adverse impacts of extreme temperatures on cardiovascular and respiratory mortality [4,[9][10][11].In contrast, there is less research investigating non-fatal cardiovascular and respiratory outcomes in relation to temperature.In Phung et al [12]'s systematic review, researchers reported an elevated risk of cold-induced cardiovascular morbidity associated with short-term cold exposures in the general population; however, no significant effect was estimated for heat exposure.While there were null associations between heat exposure and cardiovascular hospitalizations in multiple European countries, such as Spain and Catalonia [13][14][15], several studies from Australia [16], Vietnam [17], and New York, United States of America [18] reported an elevated risk associated with extremely high temperature.On the other hand, the temperature effect on respiratory morbidity is less explored.Existing findings are mixed as some studies identified an increased risk of respiratory hospitalizations due to heat or cold exposures [13,14,19], whereas a non-significant effect of extreme temperatures was reported in a systematic review article summarizing the global evidence [20].Overall, the temperature-related health effects display a large geographical heterogeneity.These relationships can be possibly modified by climatic characteristics, demographic and socioeconomic factors, as well as population acclimatization [21,22].Therefore, evidence from local epidemiology is of utmost significance to enacting adaptation measures and reducing the public health burden caused by increasing unusual temperature events.
Over the recent decades, the Eastern Mediterranean region has been one of the featured climate change hotspots experiencing warming at almost two times faster than the global average [23].Cyprus is an island in this region where the temperature is projected to rise by up to 4.5 • C-5 • C by the end of the 21st century [24].The increased frequency and intensity of climate extremes are likely to take a greater toll on public health.However, the evidence on temperature-morbidity association remains scarce and inconclusive in Cyprus [25][26][27].To the best of our knowledge, only a recent study examined the effect of thermal conditions specifically on hospital admissions for cardiovascular and respiratory diseases in public hospitals in Cyprus using negative binomial regression.It presented a negative association of temperature with respiratory admissions and a null association with cardiovascular admissions between 2009 and 2018 [26].To date, the existing few studies focused on the health effects of exposure to thermal environments but underappreciated the cold-related effects [25,26].In addition, they did not fully capture the potential lagged non-linearity of health effects in relation to temperature, a common problem when imposing linear models [25,26] or descriptive approaches [27].We used a time-series analysis with distributed lag non-linear model (DLNM) to investigate the association between ambient temperatures and daily hospital admissions for cardiovascular and respiratory diseases collected from four general hospitals in the three largest districts in Cyprus from 2000 to 2019.

Study area and population
Cyprus is an island located in the Eastern Mediterranean region at 35 • North and 33 • East.Cyprus has a typical temperate Mediterranean climate characterized by mild humid winter and hot dry summer.The summers last from mid-May to mid-September, and the winters last from November to mid-March; autumn and spring seasons are short with rapid changes in weather conditions [28,29].
Nicosia, Limassol, and Larnaca are the largest three administrative districts in Cyprus.Nicosia district is the capital of the island with the largest resident population of 351 600 (38%) (Census 2021) [30].Limassol and Larnaca are the second and third largest urban areas, which account for about 28% and 17% of the total population of 918 100, respectively (Census 2021) [30].

Data source
We used data on daily number of inpatient discharges collected by the Division of Demographic, Social and Tourism Statistics of the Statistical Service of Cyprus (CYSTAT) from two general hospitals located in Nicosia, one general hospital in Limassol, and one general hospital in Larnaca over the study period, from 1 January 2000 to 31 December 2019.Causes of hospital admissions were classified by the primary discharge diagnosis code from the International Classification of Diseases, Tenth Revision (ICD-10).Hospital admissions for cardiovascular diseases were identified by ICD10 I00-I99, and hospital admissions for respiratory diseases were identified by ICD10 J00-J99.The data for seven months from June to December 2008 had an unknown misreporting issue in the discharge hospital records for Nicosia.We also identified a high percentage of sporadic missing data in 2011 (38%) and 2012 (43%), which may be indicative of a high degree of systematic discharge reporting errors.Therefore, these two years were excluded from the main analysis but were incorporated in the sensitivity analysis.
We obtained meteorological data from Cyprus Department of Meteorology of the Ministry of Agriculture, Rural Development and Environment for each urban district.Meteorological data included daily mean temperature ( • C) and relative humidity (%).The daily mean temperature was averaged over the corresponding maximum and minimum values for each day, and the daily mean relative humidity was averaged over hourly measurements for each day.Daily average concentrations of particulate matter less than 10 µm in aerodynamic diameter (PM 10 , µg m −3 ) and ozone (O 3 , µg m −3 ), were obtained from the Air Quality and Strategic Planning Section, Department of Labour Inspection, Ministry of Labour and Social Insurance.While the above two meteorological parameters cover the whole study period, the data on PM 10 concentration were not available until 2005 and the data on O 3 concentration were not available until 2010.

Statistical analysis
We conducted a time-series analysis to assess the association between mean ambient temperature and daily counts of hospital admissions for cardiorespiratory diseases.The study period covered all the seasons and spanned from January 2000 to December 2019.The analysis followed two stages: (1) estimation of the risks for each district using the local meteorological data and hospital discharge records, and (2) pooling the effect estimates in a meta-analytical model.
In the first stage, we estimated the district-specific exposure-response curves for cardiorespiratory, cardiovascular, and respiratory hospital admissions.DLNMs were applied to explain the potential nonlinearity of the exposure-response relationship and the lagged effects, where the temperature and lag terms were modeled in a cross-basis function simultaneously [31].The quasi-Poisson distribution was used to account for the overdispersion of daily hospitalization counts for cardiovascular or respiratory diseases.It is widely accepted that the health effects of cold temperatures could last for up to three or four weeks, while the effects of hot temperatures are more acute and could last for several days within a week [21,32].Therefore, we determined the maximum lag period as 21 d to explain the delayed effects of temperature.Our choice of the 21 day lag period could also account for the potential harvesting effect of temperature, which has been reported in prior mortality and morbidity studies [33,34].We relied on a quasi-Akaike's Information Criterion (q-AIC) to evaluate the goodness of fit and seek the optimal model parameters [6,35] including spline types, degrees of freedom for ambient temperature and lag effects, and degree of freedom for seasonality.Given that Nicosia is the most populated district with the highest cardiorespiratory admission counts in Cyprus, we made the modeling choices based on the best-fitting model applied to the Nicosia dataset.The q-AIC values of the DLNM models with different statistical modeling choices are reported in table S1.In the final model, we applied quadratic b-splines for the temperature term with three internal knots placed in the 10th, 75th, and 90th percentiles, and natural cubic splines for the lag effect with three knots equally placed on the log scale with a maximum lag period of 21 d [21].The models were adjusted for daily mean relative humidity in a linear term.We controlled for seasonality and long-time trend by applying a natural spline function for the day of study with four degrees of freedom per year and an additional one degree of freedom per decade.We included day of the week as a categorical term as well as an indicator variable for public holidays in the model.The model is presented as follows: where Y i is the observed number of hospitalizations on day i; β 0 is the intercept; β 1 to β 3 are matrix and vectors of regression coefficients for their respective term; cb (T i,l ) is the matrix for DLNM cross-basis for temperature for a lag from day l; ns (time i ) is natural cubic spline function of time with four degrees of freedom per year and one degree of freedom per decade; DOW i is the categorical variable for day of the week; and Holiday i is a dummy variable for public holidays.
In the second stage, for each hospital admission cause, we performed a random-effect meta-analysis to pool the district-specific estimates that described the overall reduced cumulative exposure-response relationships [36].The heterogeneity in the association across districts was examined by the Cochran Q test and I 2 statistic, which can represent the percentage of the total variability in effect sizes among districts [37].Then we derived the Best Linear Unbiased Prediction (BLUP) of the overall cumulative temperature-morbidity association for each district, which allows the district with the least data to borrow information from other districts [38].We identified the cause-specific minimum morbidity temperatures (MMTs) for each district that corresponded to the temperature that had the least morbidity risk.With the extreme cold temperature and extreme hot temperature defined as the 1st and 99th percentiles of the temperature distribution, respectively [39][40][41], we reported the relative risk (RR) comparing the extreme temperature point to the MMT.To present the disease burden in a more interpretable way, we further presented the excess morbidity risk by computing the corresponding number (attributable number, AN) and the fraction (attributable fraction, AF) attributable to ranges of cold days (below the 2.5th temperature percentile) and hot days (above the 97.5th temperature percentile) [3,21,42,43].These two temperature percentiles have been utilized previously to define extreme weather [21,42,44,45].The empirical confidence intervals (eCI) for attributable morbidity risk were calculated using Monte Carlo simulations, assuming a normal distribution of the BLUPs of the reduced coefficients.
In the main analyses, we removed the misreported seven months from June to December 2008.To further reduce the bias introduced by reporting errors, we ran analyses with and without 2011 and 2012, however, in the main analysis these two years were excluded.A number of sensitivity analyses were conducted.First, we explored the robustness of our results to the adjustment for two air pollutants, PM 10 and O 3 .Since the data were only available from 2005 for PM 10 and from 2010 for O 3 , we additionally adjusted for PM 10 using the data from 2005 to 2019 and adjusted for PM 10 and/or O 3 using the data from 2010 to 2019, respectively, in linear terms.To reveal solely the effects of adjustment for air pollutants on results, we repeated the analysis with the period of 2005-2019 and 2010-2019 as a comparison, respectively, which can also suggest the potential temporal variations in the associations.Second, to understand the potential influences of missing and underreported data, we included the hospitalization data for 2011 and 2012 and further adjusted for the suspected underreported days.We assumed that the months with a monthly number of days without any cause-specific cases beyond the 95th percentile of the numbers for all the months might have a higher underreporting probability.We therefore created a dummy variable (0,1) for the suspected underreported days and included it in the model.Third, we tested the robustness of the results by applying alternative modeling choices.Specifically, we fitted the models using a natural spline with five degrees of freedom per year to control for seasonality instead of four degrees of freedom per year.In addition, we extended the maximum lags from 21 d to 28 d to estimate the overall cumulative association.
All statistical analyses were performed using R statistical software version 4.1.2.The dlnm and mixmeta packages were used for non-linear delayed ambient temperature effects and pooling, respectively.A p-value < 0.05 was considered statistically significant.

Results
The summary of hospital admissions for cardiorespiratory and specific causes, meteorological conditions, and air pollutant concentrations in three districts are presented in table 1.In total, we included 98 339 hospital admissions for cardiovascular diseases and 81 649 hospital admissions for respiratory diseases.The largest daily number of hospital admissions for cardiovascular diseases was observed in Nicosia, the capital city with the largest population, with a mean of 8 cases per day, whereas the least was seen in Larnaca with a mean of 2 cases per day.The average daily number of hospital admissions for respiratory diseases seemed to be comparable between Nicosia and Limassol (5 cases per day), which was higher than that in Larnaca (3 cases per day).Over the study period, the average daily mean temperatures in Nicosia, Limassol, and Larnaca were 20.3 ± 7.3 • C, 21.1 ± 5.8 • C, and 20.3 ± 6.0 • C, respectively.The seasonal pattern of daily mean temperature across these cities appeared to be highly consistent (figure S1).
Figures 1 and 2 present the district-specific cold effects (1st percentile vs. MMT) and heat effects (99th percentile vs. MMT) of temperature on cardiorespiratory hospitalizations at varying lag days, respectively.The RRs at the lags of 0, 5, 7, 14, and 21 d are shown in table S2.The corresponding cumulative effects over varying lag periods (0-5, 0-7, 0-14, and 0-21 d) are summarized in table S3.The cold effects displayed a delayed tendency with a pronounced adverse effect appearing within the first week and subsequently persisting for two weeks or even more.In general, we found that the heat effects occurred acutely within 2-5 d and then diminished to a low level.
Figure 3 shows the district-specific and pooled overall cumulative exposure-response curves of the relationship between mean ambient temperature and cardiorespiratory hospital admissions over 21 day lags after the second stage meta-analysis.Overall, all the exposure-response curves were U-shaped reflecting the non-linear dependencies of cardiorespiratory morbidity and temperature with escalated risk, as the temperature approached the extremes.In most scenarios of different districts and causes, the morbidity risk tended to increase at a steeper slope for high temperatures above the MMT (29.4 • C-31.8 • C), while the risk increased rather modestly for cold exposures below the MMT.
The pooled cold effect over 21 day lag was quantified by a 26% increased risk of total cardiorespiratory hospitalizations (RR = 1.26; 95% CI: 1.02, 1.56) (table 2).However, the overall heat effect on cardiorespiratory hospitalizations showed about 6% increase (RR = 1.06; 95% CI: 0.99, 1.13).The three cities demonstrated a generally similar pattern in terms of the associations between daily mean temperature and hospital admissions for cardiorespiratory diseases.In detail, the strongest effects of extreme cold and heat were observed for Nicosia, which corresponded to the RRs of 1.44 (95% CI: 1.11, 1.87) and 1.11 (95% CI: 1.01, 1.22), respectively.Despite similar trends with Nicosia, the estimated effects of extreme temperatures for Limassol and Larnaca were not as pronounced as that for Nicosia.Overall, there were no discernable statistical differences in the effect estimates for both cardiovascular morbidity (Cochran Q p-value = 0.83, I 2 = 0.0%) and respiratory morbidity (Cochran Q p-value = 0.26, I 2 = 18.8%) across the three districts.With regards to the total cardiorespiratory morbidity burden, there was a substantial fraction of 11.82% (95% eCI: 3.67, 18.44%) attributable to cold days characterized by extreme range of low temperatures below the 2.5th percentile, but a smaller fraction of 0.20% (95% eCI: 0.08, 0.29%) attributable to hot days characterized by extreme range of high temperatures above the 97.5th percentile.Specifically, cold days and hot days were responsible for 15.85% (95% eCI: 8.24, 22.40%) and 0.17% (95% eCI: 0.03, 0.29%) excess cardiovascular hospital admissions.In parallel, cold days and hot days were responsible for 9.59% (95% eCI: −0.66, 18.69%) and 0.23% (95% eCI: 0.07, 0.35%) excess respiratory hospital admissions.
Generally, the adjustment for PM 10 or O 3 did not substantially change the results, except that adjusting for PM 10 seemed to increase the cold-related respiratory morbidity risk during the period of 2010-2019 (table S4).We also observed some temporal variations in the temperature effect by comparing the analysis of different time periods (table S4).Relative to the whole nearly 20 year study period, the cold-related respiratory morbidity risk became pronounced after 2005 and continuously increased between 2005 and 2010.Moreover, the risk of cardiovascular admissions from heat increased in the last decade (2010-2019 vs. 2000-2019).
In the sensitivity analysis for the statistical modeling choices, the adjustment for suspected underreported days did not substantially change our results (figure S2), which implied the potentially small magnitude of underreporting effect.The estimated cumulative cold effect on hospital admissions for cardiovascular diseases was slightly attenuated when applying five degrees of freedom per year (figure S3).Increasing the maximum lag period from 21 d to 28 d gave slightly higher but similar effects of extremely low temperatures (figure S4).The results were stable and the pooled effect of extreme cold on cardiovascular morbidity remained statistically significant after the above modifications to the modeling.

Discussion
In the present study, we analyzed nearly 180 000 hospital admissions over two decades (2000-2019) from the largest morbidity dataset in Cyprus.We found that extreme temperatures increased cardiovascular and respiratory morbidity risks across the most populous districts in the country.Cold appeared to be responsible for a higher fraction of excess hospital admissions for cardiorespiratory diseases than heat in Cyprus.Nevertheless, the steeply increasing pattern in RR at heat extremes is a cause for concern if the region continues to warm at such a fast pace.Our results are also suggestive of an upward maladaptive trend in cold-related respiratory morbidity risk and heat-related cardiovascular morbidity risk over the past two decades (2010-2019 vs. 2000-2019).The temperature-related cardiovascular and respiratory morbidity burdens in Cyprus may help us understand the effects in other countries that share climatic and demographic similarities in the Eastern Mediterranean region.
To the best of our knowledge, this is the first study to investigate the health effect of extreme cold in Cyprus.Over the past two decades, our results demonstrated that extremely low temperatures increased the risk of cardiovascular morbidity.This finding is in line with many previous studies in a global range [9,12,46,47].Several studies have documented increased incidence of cardiovascular hospitalizations due to cold exposure in neighboring regions in the Mediterranean basin, including Spain [13], Greece [48], and Catalonia [15].Moreover, in Europe, the cold effect tends to be exacerbated in warmer cities located in the South [49].A large multicounty mortality study showed that the risk of cardiovascular mortality from extreme cold in Cyprus was comparable with South European countries, but generally higher than other regions in the world [42].
Our findings also supplement the knowledge about the heat-related health effects in Cyprus.We observed that extremely high temperatures and increased risk of cardiovascular and respiratory hospitalizations were both marginally associated.Very few studies explored the health effects of temperature locally and suggested that the increased risk of morbidity is associated with elevated air temperature during warm periods or with extreme weather.However, these results used broad and nondefinite metrics for the exposure or health outcome.Only one previous study in Cyprus focused on the specific cardiovascular and respiratory causes of morbidity [26].The authors fitted negative binomial regression to investigate the associations between air temperature and daily number of hospital admissions from eight public hospitals in five districts.They found that despite a 0.6% increase in the risk of all-cause hospital admissions associated with each 1 • C increase in temperature, the temperature was negatively associated with respiratory admissions and not associated with cardiovascular admissions.In comparison, we fully accounted for the potential nonlinearity and lag structures of the temperature-morbidity association with evidence-based implications.In addition, no significant heat impact on cardiovascular or respiratory hospitalizations was reported in multiple European cities [13][14][15], which is comparable with the present study.However, our U-shaped exposure-response curves revealed an increasing tendency in hospitalization risk towards the heat extremes of the temperature distribution, in contrast to the flat and even declining trend in risk for higher temperatures in Spanish cities [13].
In addition, we presented evidence that the significant lag effects of extreme cold temperatures on cardiovascular and respiratory hospitalizations in Cyprus may persist for more than two weeks.However, the effects of extreme hot temperatures appeared immediate with shorter lags within a week before reaching marginal lag-response relationships.This discrepancy in cold-and heat-related lag structures of mortality and morbidity are commonly reported in previous studies [21,32].Gaining a comprehensive understanding of the lag patterns of extreme cold and heat weather can be significant in informing timely preventative measures and improving preparedness of healthcare service to reduce the temperature-attributable hospitalization risks.
Different effect estimates of extreme cold and heat across regions can be partially explained by the geographical differences in climate, age structure, socioeconomic characteristics, and acclimatization capacity of the local population [22].For example, out of concern towards increasingly warmer weather, a series of protective practices and actions such as the heat-health warning systems and heat-health action plans have been widely implemented in warmer regions close to the Mediterranean; other protective measures include increased access to air conditioning and longer stay during warmer months [26,50,51].In Cyprus, there is a substantial demand for cooling in households during the long summers with about 80% of households equipped with at least one air-conditioning unit [52,53].Moreover, potentially more adapted housing conditions in response to heat extremes, as compared to cold extremes, could provide an additional explanation for a more pronounced cold-attributable morbidity risk in Cyprus, which has been suggested in other warmer regions with Mediterranean climate [15,54].These factors may have contributed to increased adaptation to hotter environments and led to a reduction in heat-related disease burden.A relatively high MMT for cardiorespiratory causes reported in Cyprus by our analysis (30.5 • C) compared to other European areas (ranges from 14.7 • C to 29.5 • C) may indicate a larger adaptive capacity to warmer temperatures [14].
The exposure-response curves showed that RRs of cardiorespiratory hospitalization in Cyprus escalated sharply with very extreme hot temperatures.However, the excess hospitalizations from extreme heat were considerably less than those from cold because of the high MMT, which allowed only very few days with sharply increased RRs to contribute as extremely hot days.The temperature-related associations were generally consistent among three districts, although the magnitudes of risks corresponding to extreme temperatures slightly differed.The discrepancy could be partially due to the larger population and number of cases in Nicosia than Limassol and Larnaca.In addition, the extreme cold (or hot) temperature in Nicosia was obviously lower (or higher) compared to the other two districts, which might also explain the most pronounced risks associated with extreme temperatures in Nicosia.
Our results were generally robust to the adjustment for PM 10 and/or O 3 , which suggested little confounding bias was present in the relationship between temperature and cardiorespiratory morbidity.Similar findings have been reported in some research on the effect of on health [55][56][57].However, adjusting PM 10 appeared to the risk of respiratory morbidity associated with extreme cold temperature, which might be subject to measurement errors of PM 10 during dust periods and Note: a Cold effect was calculated by the RR comparing the temperature at the 1st percentile vs. the MMT.Heat effect was calculated by the RR comparing the temperature at the 99th percentile vs. the MMT.b AF and AN of hospital admissions represented the excess morbidity risk attributable to ranges of extremely cold days (below the 2.5th temperature percentile) and extremely hot days (below the 97.5th temperature percentile).AN was computed as the yearly number of cases.
c Bolded number represents statistical significance at the 0.05 level.
needs further research to clarify.Air pollution is widely believed to be associated with respiratory and cardiovascular disease [58,59], meanwhile, air pollutants are affected by meteorological variables and the correlation between air pollutants (e.g.tropospheric O 3 ) and temperature is present.This seems to provide a basis for the rationale of treating air pollution as potential confounders.However, of note, the roles of air pollutants in the relationship between temperature and morbidity or mortality have been heavily debated.While many studies included air pollution as confounding variables in the analysis, other studies investigated it as a co-exposure or effect modifier [60][61][62].A systematic review did not summarize a modification effect on cardiovascular and respiratory mortality by PM 10 or O 3 but the results may vary by region [61].Some research also asserted them as causal intermediates that can be affected by temperature and meanwhile influence the risk of disease [63][64][65].
Multiple physiological mechanisms and pathways are proposed to explain the cold-related health effect on the cardiovascular system in humans.Lower temperatures can activate the sympathetic nervous system, which will increase blood viscosity, raise the cardiac workload, and lead to vasoconstriction in veins and arteries [66][67][68][69].These physiological alternations can subsequently cause rises in blood pressure, which is an important risk factor for major cardiovascular disease events [70].The cold environments may also stimulate the increase in blood lipids and uric acid, further contributing to the development of atherosclerosis [71].Additionally, acute respiratory infection is considered a risk factor for cardiovascular events, given that it could cause the release of pro-inflammatory cytokines and induce pro-coagulant and hemodynamic effects, increasing the risk of atherosclerosis [72,73].During heat stress, hyperthermia triggers fluid depletion, hemoconcentration and electrolyte disturbances through peripheral vasodilatation and excessive sweating, which also activate sympathetic response and induce tachycardia [74].Consequently, the resulting hypercoagulability could increase the risk of ischemia or atherosclerotic plaque rupture.In addition, impaired cellular endothelial function and conformational changes in proteins might occur in response to extreme heat, causing systemic inflammation [74][75][76].
Extreme cold is well known to impact the respiratory tract.When passing through airways, cold air can weaken the physiological defense function of the upper respiratory tract, thus increasing the risk of infection [19].Moreover, some studies suggest that cold exposure can lead to a rise in fibrinogen levels that may further contribute to the occurrence of seasonal respiratory infections [77,78].Similarly, albeit with limited knowledge about the biological mechanism, extreme heat conditions have been hypothesized to trigger acute lung inflammation and damage in the form of respiratory distress syndrome via a series of physiological alterations [79,80].Sustained extreme heat may also alter the spatiotemporal distribution of environmental allergens and some infectious disease vectors, thus increasing the risk of respiratory etiologies [81].
Our study has several strengths.First, we applied state-of-the-art analytical models to provide the most comprehensive evaluation of the temperature-related morbidity risk in Cyprus.Specifically, we considered the non-linearity and lagged effects of ambient temperature simultaneously and presented the varying effect estimate within the whole temperature range via the exposure-response curves.Second, we used the largest available dataset over two decades from the most populous three districts in Cyprus, which improved the statistical power and increased the reliability of our estimates.In addition, we evaluated the potential geographical variability in the association by fitting separate models for each district and summarized the homogenous temperature effects on cardiorespiratory morbidity risk in a novel meta-analytic approach [82].Furthermore, by quantifying the excess hospitalizations attributable to extreme temperature days, we provided more interpretable and policy-informative evidence than the RR measures to promote the local climatic adaptation strategies and mitigate the temperature-attributable public health burden.
We need to acknowledge some limitations present in our study as well.First, the study sample may not fully represent the general population in Cyprus, since our analysis only used the hospital discharge data from public hospitals but did not include those from rural and private hospitals.Systematic reporting issues in hospital discharge could also impair the data quality and lead to potential bias, including the inevitable underreporting issue in the hospitalization data.In the present study, we excluded data for quality control purposes to reduce the potential measurement error in the outcome.The results did not change after correcting for the suspected underreported months.Second, the ambient temperature data cannot reflect the precise individual exposures because they were obtained from the district-specific ground monitoring stations and had low spatial resolution.This kind of exposure misclassification is not related to the outcome and hence would non-differentially bias the estimate towards the null.Third, we used the daily mean air temperature as the only exposure metric and did not use other physiologically relevant temperature indicators.This choice, despite its physiological limitation, summarizes the population-level exposure for actionable policy-making (e.g.heat-warning systems) [83,84].Other research did not have definite preferences among different temperature metrics that generated similar results [85,86].Fourth, we could not differentiate the first hospitalizations from the repeated hospitalizations due to data restrictions.In addition, there is a lack of data on seasonal epidemics from flu or other respiratory viruses that may confound the association between cold days and cardiovascular and respiratory hospital admissions.Moreover, while potential confounding bias by unmeasured individual socioeconomic factors that are assumed to be time-invariant can be removed by time-series design, we did not explore whether the relationship can be modified by these factors, such as heating/cooling methods availability, nutrition, physical activity, and smoking.Finally, our findings of the temperature-cardiorespiratory morbidity associations have a regional scope and may not be generalized to other regions due to the discrepancies in climate characteristics, demographic and socioeconomic characteristics, lifestyle factors, etc.

Conclusions
In conclusion, we described the short-term adverse impacts of extremely low and high ambient temperatures on the risk of cardiovascular and respiratory hospital admissions in Cyprus.Extreme cold appears to show a prolonged effect and accounts for a higher fraction of excess cardiorespiratory morbidity risk, while the effect of extreme heat is more immediate and presents a public health concern in the context of climate warming.Our findings from a rapidly warming part of the world highlight the urgent need for preventative strategies to reduce the temperature-associated risks to population health and increase the adaptative capability to unfavorable temperatures, such as improving the thermal comfort in households and weather early warning system for climate change.

Figure 1 .
Figure 1.Relative risk of hospital admissions for (a) cardiorespiratory, (b) cardiovascular, and (c) respiratory causes associated with extreme cold temperatures (defined as the 1st percentile of the distribution of daily mean temperatures) at different lag times relative to the minimum morbidity temperature using the distributed lag non-linear models (DLNM).

Figure 2 .
Figure 2. Relative risk of hospital admissions for (a) cardiorespiratory, (b) cardiovascular, and (c) respiratory causes associated with extreme hot temperatures (defined as the 99th percentile of the distribution of daily mean temperatures) at different lag times relative to the minimum morbidity temperature using the distributed lag non-linear models (DLNM).

Figure 3 .
Figure 3. District-specific and pooled cumulative exposure-response curves of the relationship between mean ambient temperature and daily hospital admissions for (a) cardiorespiratory, (b) cardiovascular, and (c) respiratory causes over 21 day lags.Solid black line represents the first stage regression, whereas the dashed blue line represents the best linear unbiased prediction (BLUP) after the second stage meta-analysis.Dotted vertical line represents the minimum morbidity temperature (MMT), and the two dashed vertical lines represent the 1st and the 99th percentiles.
Note: a The summary of PM10 daily average concentrations was calculated from available data over the period 2005-2019.The summary of O3 daily average concentrations was calculated from available data over the period 2010-2019.

Table 2 .
District-specific and pooled cumulative extreme cold and hot effects of mean temperature on daily hospital admissions for cardiorespiratory, cardiovascular, and respiratory causes over 21 day lags.