Cardiovascular morbidity risk attributable to thermal stress: analysis of emergency ambulance dispatch data from Shenzhen, China

Abstract

Background

Climate change has raised scientific interest in examining the associations of weather conditions with adverse health effects, while most studies determined human thermal stress using ambient air temperature rather than the thermophysiological index.

Objectives

To evaluate the association between emergency ambulance dispatches (EADs) related to cardiovascular causes and heat/cold stress in Shenzhen, a city in southern China, with the aim of providing new insights for local policymakers.

Methods

A time series analysis using ambulance dispatch data of cardiovascular diseases in Shenzhen, China (2013–2019) was conducted. A quasi-Poisson nonlinear distributed lag model was applied to explore the relationship between emergency ambulance dispatches (EADs) due to cardiovascular causes and thermal stress (determined by Universal Thermal Climate Index, UTCI). Attributable fractions were estimated to identify which UTCI ranges have a greater health impact.

Results

The relationship between UTCI and EADs due to cardiovascular diseases exhibits a reverse J-shaped curve. The effects of cold stress were immediate and long-lasting, whereas the effects of heat stress were non-significant. Compared with the optimal equivalent temperature (71st percentile of UTCI, 29.22 °C), the relative risks for cumulative (0–21 days) exposures to cold stress (1st percentile, − 0.13 °C; 5th percentile, 7.68 °C) were 1.55 (95%CI:1.28,1.88) and 1.44 (95%CI:1.22,1.69), respectively. Thermal (cold and heat) stress was responsible for 10.81% (95%eCI: 5.67%,15.43%) of EADs for cardiovascular diseases, with 9.46% (95%eCI: 3.98%,14.40%) attributed to moderate cold stress (2.5th ~ 71st percentile). Greater susceptibility to cold stress was observed for males and the elderly. Heat stress showed harmful effects in the warm season.

Conclusions

Our results demonstrated that cold exposure elevates the risk of EADs for cardiovascular causes in Shenzhen, and moderate cold stress cause the highest burden of ambulance dispatches. Health authorities should consider effective adaptation strategies and interventions responding to cold stress to reduce the morbidity of cardiovascular diseases.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-024-20144-1.

Introduction

Anthropogenic climate change is expected to lead to more extreme climatic events across the globe [1], with abundant evidence linking cold or heat extremes to an increase in mortality and morbidity, particularly for cardio/cerebro-vascular diseases [2–5]. Analysis across 43 countries estimated that there were approximately 5 million temperature-related excess deaths each year, accounting for 9.43% of total deaths [6]. A systematic review has reported that the risk of cardiovascular hospitalization elevated by 2.8% for exposure to cold exposure and that the heatwave had a substantial influence on cardiovascular morbidity [7]. Bunker et al. [3] highlighted temperature fluctuations were related to increased mortality/morbidity rates of cerebrovascular, cardiovascular, respiratory, genitourinary, diabetes, and infectious diseases in the elderly. Given the intensified climate change events, a better understanding of the relationship between thermal conditions and human health is crucial to developing adaptation and mitigation strategies and improving medical service [8].

The existing literature has focused on mortality or hospital-based data [9]. Mortality data is the highest quality disease indicator, which would provide information on the fatal effects of climate-induced changes. However, mortality data is available only after a certain period due to the time required for data collection, verification, and reporting. This delay makes it less efficient as a health surveillance outcome [10, 11]. Likewise, hospital-based morbidity data analysis would help to improve health services to better protect the public’s health during cold and heat episodes, there were barriers to accessing the hospital-based morbidity data for the total population in some regions or countries [11, 12]. Emergency ambulance dispatch data, on the other hand, can offer temporal and geographical details on acute health occurrences and may help in detecting the early signs of health effects of climatic change [13–15].

Prior epidemiologic studies have shown that ambulance calls or dispatches are associated with ambient thermal conditions [10, 15–18]. However, most studies quantified these associations regarding relative risk (RR), with fewer estimated potential public health burden due to heat/cold stress [19, 20]. Attributable risk measures, such as attributable fraction (AF) or attributable number (AN), are extensions of relative risk that could better depict the impact of different exposure ranges, which is essential for health authorities to design more targeted health interventions [21, 22]. Additionally, prior studies commonly employed ambient temperature as an exposure parameter, despite it has been argued that using a single temperature to assess thermal stress is oversimplifying the net heat load of human exposure [23, 24]. So far, over 100 indices have been proposed to measure thermal stress [23, 25, 26]. Among others, the Universal Thermal Climate Index (UTCI) based on a multi-node model incorporates cutting-edge knowledge from a wide range of fields including climate science, epidemiology, mathematics, and physiology, making it more suitable for depicting the actual impacts of thermal conditions on humans [27–29]. Recent studies have revealed the applicability of UTCI in the epidemiology and medical science fields, suggesting that UTCI has great potential to serve as a predictor in the models of climate change impact on human health [30, 31].

Over the past three decades, the global burden of cardiovascular diseases has grown substantially [32]. In China, the incidence, prevalence, and mortality rates of cardiovascular diseases increased by 132.82%, 140.02%, and 89.12%, respectively [33]. As China confronts the dual challenges of accelerated climate change and an aging population, the burden of cardiovascular diseases is expected to continue increasing [34, 35]. A better understanding of how cardiovascular diseases are affected by climatic fluctuation is essential for achieving the goal of reducing cardiovascular mortality to 190.7 per 100 thousand population in China by 2030 [36]. This study determines thermal stress by a human thermophysiological index UTCI to analyze the relationship between emergency ambulance dispatches (EAD) due to cardiovascular causes and heat/cold stress in the southern city of Shenzhen, China. Specifically, we aimed to: 1) estimate the risks associated with heat/cold stress, 2) identify high-risk subpopulations, and 3) estimate the attributable fractions and numbers for thermal stress exposure, as well as relative contributions from moderate and extreme heat/cold stress.

Material and methods

Data and study area

As one of the core cities of the Guangdong-Hong Kong-Marco Greater Bay Area, Shenzhen has an approximate land area of 1997.47km², with 10 administrative districts under its jurisdiction. The city’s permanent population reached 17.63 million at the end of 2020 [37]. The climate of Shenzhen is typically subtropical monsoon, with mild but dry winters and hot, muggy, and rainy summers.

Daily records of emergency ambulance dispatch between 1st January 2013 to 31st December 2019 were obtained from the Shenzhen Emergency Medical Center. This center coordinates 75 emergency network hospitals and 33 independent emergency stations and collects ambulance dispatch information for the entire city [38]. Each record included information on the age and sex of the patient, the time and location of the ambulance call, symptoms of the patient, and both the initial and supplementary diagnoses. EADs due to cardiovascular diseases (International Classification of Diseases, 10th Revision; ICD-10: I00-I99) were extracted from the database based on strict diagnostic criteria, including only cases where both the initial and supplementary diagnoses indicated cardiovascular disease.

Exposure metric

Thermophysiological index

UTCI is derived from the UTCI-Fiala model, which combines an advanced physiological model of human thermoregulation with a temperature-varying clothing insulation model [28, 29, 39–41]. Conceptually, UTCI is defined as the equivalent temperature (ET)(°C) of the reference conditions (calm air, 50% relative humidity, and no extra thermal radiation), which would elicit the same dynamic physiological response to actual climatic conditions [28, 29]. For any combination of air temperature, radiation, humidity, and wind speed, UTCI can be approximated by a six-order polynomial regression equation [42]. We obtained the UTCI from the high-spatial-resolution database of human thermal stress indices over South and East Asia (HiTiSEA), a newly released gridded dataset derived from ERA5-Land and ERA5 reanalysis [43]. The HiTiSEA dataset contains mean radiant temperature (MRT), three types of UTCI (outdoor unshaded UTCI, outdoor shaded UTCI, and indoor UTCI), and eight other thermal stress indices at 0.1° × 0.1° spatial resolution from 1981 to 2019 for the area of South and East Asia [43]. We extracted the gridded daily mean UTCI (outdoor unshaded) over Shenzhen for the period 2013–2019 using Python software (version 3.9; packages ‘geopandas’ and ‘salem’), and further calculated the arithmetic mean of daily mean gridded UTCI over the study area to derive daily UTCI exposure.

Meteorological data

The daily meteorological data on mean temperature (°C) and relative humidity (%) were collected through ground-based meteorological monitors and obtained from the Shenzhen Meteorological Service Center.

Statistical analysis

We investigated the effect of thermal stress on EADs due to cardiovascular disease using a quasi-likelihood Poisson regression model coupled with a distributed lag non-linear model (DLNM) [44]. The Poisson model account for overdispersion in the ambulance data [45, 46]. The DLNM was utilized to assess the non-linear and lagged associations between the exposure metric and EADs due to cardiovascular causes via a cross-basis function [44]. The model was specified as follows:

$$\begin{array}{c}log\left(E,\left(Y_t\right)\right)=\alpha +cb{\left(U,T,C,I\right)}_{t,l}+s\left(t,i,m,e,,,7,*,y,e,a,r\right)+\phi Do{\omega }_t\\ +\gamma Holida{y}_t+offset\left({\mathit{pop}}_i\right)\end{array}$$ 1

where $E\left(Y_t\right)$ is the expected number of EADs due to cardiovascular diseases on day t; α is the intercept; cb refers to cross-basis function, a bi-dimensional matrix of functions that would simultaneously explore the delayed and nonlinear associations between EADs for cardiovascular causes and UTCI [44, 47]. l refers to maximum lag days. The ‘cb’ term was defined by natural cubic splines with 3 degrees of freedom (df) for the UTCI predictor, and 2-df for the lag dimension, with a lag period of 21 days. The df for the two-dimensional spaces were selected through Quasi Akaike information criteria (QAIC) (see Table S1) [44]. s() refers to natural cubic spline; In line with prior studies, 7 df per year is used to control the seasonality and long-term trend [18, 20, 45]. Doω and Holiday are categorical variables, representing the day of the week and the public holiday, respectively. pop_i represents the year-end permanent population in the year i [48]. φ and γ are coefficients.

We introduced a new metric: optimum equivalent temperature (OET), which corresponds to the UTCI of minimum ambulance dispatch risk [18]. The relative risks (RRs) of EADs for cardiovascular diseases were computed for exposure to cold stress (1st and 5th percentile of UTCI distribution) and heat stress (95th and 99th percentiles), using the OET as the reference. We reported the results on the same day (lag 0) and different lag periods (lag 0–3, lag 0–7, lag 0–14, and lag 0–21) to represent the immediate and delayed effects, respectively [49]. Stratified analyses were performed by sex, age (< 65 and ≥ 65 years), and season (warm, May to October; cold, November–April).

We also calculated the Attributable fraction and number through a backward approach within the DLNM framework, which assumes the current burden attributed to a series of exposures in the past [21]. The AF and AN of ambulance dispatches caused by exposure to cold and heat stress were calculated using the OET as the null exposure condition [50]. In addition, AFs and AN were computed for exposure to extreme cold stress (below the 2.5th percentile of UTCI), extreme heat stress (above the 97.5th percentile of UTCI), along with moderate cold stress (the range between OET and 2.5th percentile of UTCI) and moderate heat stress (the range between OET and 97.5th percentile of UTCI) (see Supplemental Material Fig S1). Empirical Confidence intervals (eCIs) of AFs were obtained by 10000 Monte Carlo simulations [21, 51].

Sensitivity analyses were performed by altering the df of time (6 and 8) and extending the maximum lag period (25, 28, and 31 days). All statistical analyses were conducted in R software version 4.1.3 (packages ‘dlnm’ [52]); function ‘attrdl’ [21].

Results

Characteristics of cardiovascular EADs and exposure metric

Within the study period, there were 64,890 EADs related to cardiovascular diseases, with a daily mean (± standard deviation, SD) dispatches of 25.39 (± 7.05). Males and those aged below 65 made up the majority (59.7% and 61.6%, respectively) of cases. Over half of cases occurred in the cold season (52.2%) (Table 1). We observed an overall increasing trend in the EADs due to cardiovascular causes (Fig. 1).

Table 1.

Descriptive statistics of EADs due to cardiovascular causes, thermophysiological index and weather conditions in Shenzhen, 2013–2019

Variables	Mean (SD)	Min	25th	50th	75th	Max
EADs related to cardiovascular diseases
(n = 64,890)	25.39(7.05)	9.00	20.00	25.00	30.00	51.00
Sexa
Male (n = 38,732)	15.15(5.03)	2.00	12.00	15.00	18.00	39.00
Female (n = 26,157)	10.23(3.66)	1.00	8.00	10.00	13.00	25.00
Age (years)b
< 65 (n = 39,986)	15.64(5.21)	3.00	12.00	15.00	19.00	43.00
≥ 65 (n = 24,900)	9.75(3.67)	1.00	7.00	9.00	12.00	28.00
Season
Cold (n = 33,891)	26.73(7.30)	10.00	21.00	26.00	32.00	51.00
Warm (n = 30,999)	24.07(6.54)	9.00	19.00	24.00	28.00	48.00
Thermophysiological index
UTCI (°C)	22.88(8.31)	− 13.89	17.30	24.53	29.95	35.62
Daily meteorological measure
Mean temperature (°C)	23.52(5.35)	3.50	19.50	24.80	28.10	33.00
Relative humidity (%)	75.64(13.07)	19.00	70.00	78.00	85.00	100.00

EADs Emergency ambulance dispatches, Min Minimum, Max Maximum, SD Standard deviation, UTCI Universal Thermal Climate Index

^a1 missing value in sex

^b4 missing values in age

Fig. 1.

Time series of emergency ambulance dispatches due to cardiovascular causes, UTCI exposure, mean temperature, and relative humidity in Shenzhen, 2013 to 2019. Note: EADs, emergency ambulance dispatches; CVD, cardiovascular disease; UTCI, Universal Thermal Climate Index; RH, relative humidity

The daily mean (± SD) UTCI and temperature were 22.88 °C (± 8.31 °C) and 23.52 °C (± 5.35 °C), respectively. The daily mean relative humidity (± SD) was 75.64% (± 13.07%) (Table 1). We also observed that both the daily mean UTCI and temperature showed clear seasonality, but UTCI showed larger amplitudes than temperature. The relative humidity reached high levels in the middle of each year (Fig. 1). As shown in Table S2, EADs due to cardiovascular causes were negatively correlated with UTCI (r = − 0.15, p < 0.01), and the correlation between UTCI and mean temperature was significant (r = 0.96, p < 0.01). Moreover, the UTCI values across the 26 grid points were similar, with consistent temporal trends observed in each grid point's time series (Table S3, Fig S2).

Thermal stress-emergency ambulance dispatches association

The exposure-lag-response surface revealed that the relationship between thermal stress and EADs due to cardiovascular causes was nonlinear, and the effects of cold stress were pronounced on the day of exposure and persisted for 2 to 3 weeks, whereas no significant effect of heat stress. The overall UTCI-dispatch curve was a reversed ‘J’ shape, with significantly higher risks for cold stress. The OET was identified as 29.22 °C, corresponding to the 71st percentile of UTCI distribution (Fig. 2). As shown in Fig. 3, the effects of heat and cold stress were distinctly different by lag periods. The risks of EADs due to cardiovascular causes associated with cold stress were strongest on the day of exposure (RR_1st = 1.04, 95%CI: 1.03 ~ 1.06; RR_5th = 1.03, 95%CI: 1.02 ~ 1.05), and decreased to lag day 15. Compared with the OET, the risks increased with the duration of cold stress, RRs reached maximum at lag 0–21 days (RR_1st = 1.55, 95%CI: 1.28 ~ 1.88; RR_5th = 1.44, 95%CI: 1.22 ~ 1.69) (Table 2). In contrast, the risks for exposure to heat stress were not statistically significant in all lag periods.

Fig. 2.

Distributed-lag-nonlinear association between UTCI and EADs due to cardiovascular causes (Left panel) and cumulative effects (lag 0–21 days) of UTCI on EADs related to cardiovascular disease with UTCI distribution (Right panel). The solid blue curve represents point estimates, and the grey area shows 95% confidence intervals; Red solid line indicates optimal equivalent temperature (29.22 °C, 71st percentile of UTCI), and black dotted vertical lines present the 2.5th (3.91 °C) and 97.5th (33.48 °C) percentiles of UTCI

Fig. 3.

Lag effects for cold stress at the 1st and 5th percentiles and heat stress at the 95th and 99th percentiles compared with the optimal equivalent temperature (29.22 °C, 71st percentile of UTCI) over 21 lag days for emergency ambulance dispatches due to cardiovascular disease

Table 2.

Relative risks of EADs due to cardiovascular causes for different UTCI levels in Shenzhen from 2013 to 2019, with OET (29.22 °C, 71st percentile of UTCI) as the reference

Different UTCI levels	RR (95%CI)
	All cases	Males	Females	< 65 years old	≥ 65 years old
1st percentile (− 0.13 °C)
Lag 0	1.04(1.03,1.06)	1.05(1.03,1.07)	1.03(1.00,1.05)	1.04(1.02,1.06)	1.04(1.02,1.07)
Lag 0–3	1.16(1.11,1.22)	1.21(1.13,1.29)	1.10(1.02,1.19)	1.15(1.08,1.22)	1.18(1.09,1.27)
Lag 0–7	1.31(1.20,1.43)	1.40(1.25,1.57)	1.19(1.04,1.36)	1.27(1.13,1.43)	1.35(1.18,1.54)
Lag 0–14	1.50(1.30,1.72)	1.63(1.36,1.95)	1.30(1.06,1.60)	1.39(1.16,1.67)	1.61(1.30,2.00)
Lag 0–21	1.55(1.28,1.88)	1.67(1.30,2.14)	1.39(1.03,1.87)	1.37(1.07,1.77)	1.79(1.34,2.40)
5th percentile (7.68 °C)
Lag 0	1.03(1.02,1.05)	1.04(1.03,1.06)	1.02(1.00,1.04)	1.03(1.02,1.05)	1.03(1.01,1.05)
Lag 0–3	1.13(1.08,1.18)	1.17(1.10,1.23)	1.08(1.01,1.15)	1.13(1.07,1.20)	1.12(1.04,1.20)
Lag 0–7	1.24(1.15,1.34)	1.32(1.19,1.45)	1.14(1.01,1.29)	1.24(1.13,1.37)	1.23(1.09,1.39)
Lag 0–14	1.39(1.23,1.57)	1.51(1.30,1.77)	1.21(1.02,1.45)	1.36(1.17,1.59)	1.41(1.17,1.70)
Lag 0–21	1.44(1.22,1.69)	1.58(1.28,1.95)	1.25(0.97,1.60)	1.36(1.11,1.68)	1.55(1.20,1.99)
95th percentile (32.87 °C)
Lag 0	1.00(1.00,1.01)	1.00(0.99,1.01)	1.00(0.99,1.01)	1.01(1.00,1.01)	1.00(0.99,1.01)
Lag 0–3	1.01(0.99,1.03)	1.01(0.98,1.04)	1.01(0.97,1.04)	1.02(0.99,1.05)	0.99(0.96,1.02)
Lag 0–7	1.02(0.98,1.06)	1.02(0.97,1.07)	1.01(0.95,1.08)	1.04(0.98,1.09)	0.99(0.93,1.05)
Lag 0–14	1.03(0.96,1.09)	1.02(0.94,1.11)	1.03(0.94,1.13)	1.06(0.98,1.15)	0.98(0.88,1.08)
Lag 0–21	1.03(0.94,1.12)	1.01(0.91,1.13)	1.05(0.91,1.19)	1.07(0.95,1.19)	0.97(0.85,1.12)
99th percentile (34.00 °C)
Lag 0	1.00(1.00,1.01)	1.00(0.99,1.02)	1.00(0.99,1.02)	1.01(1.00,1.02)	1.00(0.98,1.01)
Lag 0–3	1.02(0.98,1.05)	1.02(0.98,1.06)	1.01(0.96,1.06)	1.03(0.99,1.07)	0.99(0.94,1.04)
Lag 0–7	1.03(0.97,1.09)	1.03(0.96,1.11)	1.02(0.94,1.11)	1.06(0.98,1.13)	0.98(0.90,1.07)
Lag 0–14	1.04(0.95,1.14)	1.04(0.93,1.16)	1.04(0.91,1.18)	1.09(0.97,1.21)	0.97(0.85,1.12)
Lag 0–21	1.04(0.93,1.18)	1.03(0.88,1.20)	1.07(0.89,1.28)	1.10(0.94,1.28)	0.97(0.80,1.17)

EADs Emergency ambulance dispatches, UTCI Universal Thermal Climate Index, OET Optimal equivalent temperature, RRs Relative Risks

Statistically significant results (p < 0.05) are marked in bold

The UTCI-dispatch curves show similar reverse J-shaped in sex and age groups, referencing the OET (29.22 °C) (Fig. 4). In terms of sex, males tend to be more vulnerable to cold stress (RR_{1st, lag0-21} = 1.67, 95%CI: 1.30 ~ 2.14; RR_{5th, lag0-21} = 1.58, 95%CI: 1.28 ~ 1.95) compared to females (RR_{1st, lag0-21} = 1.39, 95%CI: 1.03 ~ 1.87; RR_{5th, lag0-21} = 1.25, 95%CI: 0.97 ~ 1.60). When accounting for age, the estimated risks of EADs due to cardiovascular disease were higher for elderly people (RR_1st = 1.79, 95%CI: 1.34 ~ 2.40; RR_5th = 1.55, 95%CI: 1.20 ~ 1.99), compared with young people (RR_1st = 1.37, 95%CI: 1.07 ~ 1.77; RR_5th = 1.36, 95%CI: 1.11 ~ 1.68) over the lag of 0–21 days.

Fig. 4.

Associations between UTCI and EADs due to cardiovascular causes stratified by sex and age. The optimal equivalent temperature (29.22 °C, 71st percentile of UTCI) as reference

In addition, we investigated the association between UTCI and EADs due to cardiovascular causes within cold season (November to April) and warm season (May to October) (Table S4). The results showed the effects of cold stress were consistently significant in the cold season, with RRs slightly decreased compared to the main results. Surprisingly, we observed a significant effect of heat stress on the day of exposure during the warm season (RR_{1st, lag0} = 1.01, 95%CI: 1.001 ~ 1.02), and the effect increased with cumulative exposure to heat stress for two weeks (RR_{95th, lag0-14} = 1.06, 95%CI: 1.001 ~ 1.02).

Attributable risk

Table 3 provides the estimates of the backward attributable fractions and numbers for EADs related to cardiovascular diseases attributed to cold and heat stresses. The results indicated that 10.81% (7019 cases) of cardiovascular ambulance dispatches were caused by exposure to cold and heat stresses in Shenzhen during the study period. A greater percentage of thermal stress-attributable cardiovascular ambulance dispatches were observed in cold stress (10.42%, 95%eCI: 4.99% ~ 15.51%) compared to heat stress (0.39%, 95%eCI: − 1.11% ~ 1.79%). The results from two components of cold stress clearly showed that moderate cold stress contributed to the largest fractions for cardiovascular ambulance dispatches (9.46%, 95%eCI: 3.98% ~ 14.40%).

Table 3.

Estimated attributable numbers and fractions of cardiovascular ambulance dispatches from exposure to cold and heat stress

	AN (95%eCI)		AF (%) (95%eCI)
Overall	7019	(3589,10,058)	10.81	(5.67,15.43)
Cold stressa	6764	(2733,9377)	10.42	(4.99,15.51)
Extreme cold stressb	767	(438,1082)	1.18	(0.67,1.67)
Moderate cold stressb	6141	(2656,9333)	9.46	(3.98,14.40)
Heat stressa	255	(− 742,1176)	0.39	(− 1.11,1.79)
Extreme heat stressb	68	(− 128,258)	0.10	(− 0.19,0.39)
Moderate heat stressb	187	(− 580,918)	0.29	(− 0.89,1.45)

AN Attributable number, AF Attributable fraction, eCI empirical confidential interval

^aCold and heat stresses were defined as UTCI below and above optimal equivalent temperature (OET) (29.22 °C, 71st percentile of UTCI)

^bExtreme cold stress was defined as UTCI at the 2.5th (3.91 °C) percentile or below. Moderate cold stress was defined as UTCI range between the 2.5th (3.91 °C) percentile and the OET. Extreme heat stress was defined as UTCI at the 97.5th (33.48 °C) percentile or above. Moderate cold stress was defined as UTCI range between the OET and the 97.5th (33.48 °C) percentile

Statistically significant results (p < 0.05) are marked in bold

We additionally estimated the AF and AN for different subgroups that attributed to extreme/moderate cold stress (Table 4). We found males and those aged 65 and above showed higher risks of EADs due to cardiovascular disease attributed to cold stress, with AF of 1.35% and 1.73% for extreme cold stress, 12.63% and 11.79% for moderate cold stress.

Table 4.

Estimated attributable fractions and numbers of cardiovascular ambulance dispatches from exposure to extreme and moderate cold stress stratified by sex and age

	Extreme cold stressa		Moderate cold stressb
	AF (%) (95% eCI)	AN (95%eCI)	AF (%) (95% eCI)	AN (95%eCI)
Males	1.35(0.71,1.97)	526(272,764)	12.63(5.86,18.51)	4893(2388,7156)
Females	0.90(0.10,1.67)	237(25,431)	4.44(− 5.10,12.53)	1161(− 1280,3245)
< 65 years old	0.79(0.17,1.39)	319(67,559)	8.21(1.30,14.47)	3285(532,5734)
≥ 65 years old	1.73(0.90,2.52)	432(224,626)	11.79(2.93,19.22)	2937(763,4822)

AN Attributable number, AF Attributable fraction, eCI empirical confidential interval

^aExtreme cold stress was defined as UTCI at the 2.5th (3.91 °C) percentile or below

^bModerate cold was defined as UTCI range between the 2.5th (3.91 °C) percentile and Optimal equivalent temperature (29.22 °C, 71st percentile)

Statistically significant results (p < 0.05) are marked in bold

Sensitivity analysis

When we altered the degree of freedom of time trend to 6 or 8, the results were comparable to the main findings (Table S5). When we extended the maximum lag days to 25, 28, or 31, the cumulative effects of cold and heat stresses were not substantially changed (Fig S3). The above sensitivity analyses proved that our results were stable.

Discussion

This study explored the relationships between EADs due to cardiovascular causes and thermal stress in Shenzhen using the advanced human thermophysiological index (UTCI). and assessed the cardiovascular ambulance dispatch risks attributed to cold or heat stresses. We observed a reversed J-shape exposure–response association, the risks of EADs due to cardiovascular causes significantly increased with the intensity of cold stress. The effects of cold stress were found to be strongest on the day of exposure, and detectable up through the lag 15 days, whereas the effects of heat stress were non-significant over the lag period. More thermal-attributable cardiovascular ambulance dispatches were caused by cold than heat stress, and moderate cold stress contributed to the largest attributable fraction and number of ambulance dispatches.

Our findings of the J-shaped association between thermal stress and cardiovascular morbidity were broadly consistent with those of studies conducted in England [18], Spain [20], Australia [8, 15], Japan [19], and one central Chinese city [16]. However, there are differences in the temporal pattern for the effects of cold stress between our findings and prior studies that utilized ambient temperature as an exposure metric. Previous studies observed the health impact of cold temperatures was delayed and long-lasting. For instance, a study in Australia reported that the effect of cold on ambulance attendance for cardiovascular initially observed two or three days (lag 2 or 3) following exposure, the cumulative effects over lag 2–15 showed that per 1 °C decrease in mean temperature above the threshold (22 °C) associated with a 1.63% (95%CI: 0.64% ~ 2.62%) increase in ambulance attendance for cardiovascular [15]. Similarly, a study conducted in one central Chinese city, Luoyang, reported that significant cold effects were observed at lags 2 to 6 days (lag 50-145 h), and a decrease in the hourly temperature from 26.2 °C to − 2.5 °C led to a cumulative 25% (95%CI: 18% ~ 33%) increase in emergency calls for cardiovascular disease over lag 0–170 h [16].

One possible explanation for the discrepancy between our findings and previous studies is that UTCI accounts for the effects of wind, which ambient temperature alone does not. Specifically, UTCI tends to amplify the wind chill effect, particularly at high wind speed [42], which potentially categorizes conditions with strong wind but moderate cold as extremely cold [47]. However, given the climatic characteristics of Shenzhen, where wind speeds are relatively mild throughout the year, using UTCI to assess cold effects is indeed more appropriate as it better represents the thermal environment and likely yields more realistic cold-related health outcomes.

Additionally, since UTCI integrates multiple meteorological factors, it may offer a more comprehensive assessment of cold-related risks compared to ambient temperature, which has been reported might underestimate the magnitude of cold-related health outcomes [53]. Future research could benefit from further comparing the effect estimates from thermal indices and ambient temperature to better understand the potential differences in health outcome estimations.

Unlike the significant effects of cold stress, we did not observe any significant associations between heat stress and EADs for cardiovascular disease. This is similar to findings from a recent study in Spain, which reported that heat effects on cardiovascular ambulance dispatches did not display a significant pattern [20]. There were, however, other studies that have come to a different conclusion [15, 18]. Studies in London, England found that heat effects on cardiovascular ambulance dispatches were generally immediate and short-lived [18]. One possible reason for the differences may rely on the fact that Shenzhen is at a lower latitude compared to London, which means the residents in Shenzhen are more acclimatized to hot and humid weather [54, 55]. Additionally, the widespread use of air conditioners in nearly all indoor spaces and public transport shields locals from long, hot summers. Moreover, another potential reason could be the relatively younger population structure of Shenzhen. Younger individuals tend to have better cardiovascular regulation, allowing them to support thermoregulation and maintain hemodynamics more effectively when dealing with heat, compared to older adults [56].

The results from the sex-specific analysis suggested that males in Shenzhen were more vulnerable to cold stress. Specifically, over fourfold cold-related cardiovascular ambulance dispatches and more elevated risks of cardiovascular ambulance dispatches were observed for males compared to females. This might relate to physiological differences in thermoregulation between males and females [57]. Previous studies reported that males experienced greater decreases in core temperatures than females during cold stress [58, 59]. Additionally, increased cold stress association in males could also be explained by behavioral factors, for instance, males are more likely to engage in outdoor work or participate in outdoor exercises [60, 61]. In terms of age, the effect estimates of the elderly under cold stress were found to be more pronounced than young people. This finding is consistent with prior studies of the association between cold and cardiovascular morbidity based on hospital admissions or emergency room visit records [61, 62]. Previous studies reported that thermoregulatory functions would deteriorate naturally with advancing age, resulting in a limited ability to maintain thermal homeostasis under cold stress in the elderly [63, 64]. Besides, other social factors such as reduced financial access to healthcare and poor living conditions may increase their susceptibility to extreme thermal environments [3, 65]. In the seasonal analyses, consistent significant effects of cold stress on cardiovascular ambulance dispatches were found in cold season, and interestingly, there was a short-term increased risk of ambulance dispatch for cardiovascular diseases driven by heat stress in warm season. A similar prior study conducted in Shenzhen to investigate temperature-related emergency ambulance dispatch found that the effects of high temperatures on ambulance dispatches for cardiovascular were not statistically significant during the warm season [11]. One possible reason for the inconsistencies may be the shorter time series in this prior study (2015 ~ 2016), resulting in insufficient statistical efficacy. Another possible explanation may be the high climate sensitivity of UTCI, which better captures thermal stimuli similar to the human body compared to the ambient temperature [25, 27].

We additionally assessed attributable fractions and numbers with a backward perspective measure based on DLNM to determine the burden of the cold/heat-related ambulance dispatch. A greater percentage of thermal-attributable ambulance dispatch for cardiovascular were found in cold stress compared to heat stress. Although the relative risks of EADs due to cardiovascular causes increased with the intensity of cold stress as described in the previous section, a higher fraction of cardiovascular ambulance dispatch was found attributable to moderate cold stress than extreme cold stress. The most probable explanation is that the extreme cold stress range included only a small proportion of days, and therefore the contribution to ambulance dispatch is relatively smaller [20, 49]. Our finding is consistent with a nationwide study in Japan, which reported that moderate low temperature was responsible for the majority of temperature-related emergency transport for cardiovascular [19]. Similarly, studies based on circulatory emergency hospitalizations or mortality also have demonstrated that most of the attributable risk happened on moderately cold days [49, 51, 66]. We further estimated the AF and AN for cold stress by demographic characteristics, which have rarely been reported in prior studies. Consistent with results from relative risks, we found the AFs for males and the elderly were higher on both moderately and extremely cold stress days compared with females and younger people. Of note, despite the higher AF for the elderly than that for younger ones in the moderate cold stress range (11.79% vs 8.21%), the AN for young people was greater than for the elderly. This is related to the disparity in the proportion of age groups in EADs due to cardiovascular disease in Shenzhen for the period 2013–2019, with more than 60% of cases being individuals younger than 65 years old (Table 1). Our results indicate that the current public health policies, which focused primarily on climate-induced heat events, should seriously take into account the effects associated with cold stress, and improve interventions for higher-risk people [49, 67].

There are several strengths in our study. First, instead of choosing the commonly used exposure indicator ambient temperature, we utilized the state-of-the-art thermophysiological index UTCI with a high spatial resolution (0.1° × 0.1°), which could better depict climate variability [25]. Second, EAD data is considered a more sensitive morbidity indicator because it can provide information about when and where acute health events occur that admission data might not capture [15, 18]. The dispatch data were obtained from Shenzhen Emergency Medical Center which oversees all emergency dispatches in the city. This comprehensive coverage allows for a more accurate assessment of the impact of heat/cold stress on cardiovascular-related EADs. Third, we estimated the attributable risk of thermal stress on EADs for cardiovascular diseases, which strengthens the knowledge of the health impacts of cold and heat stress and allows health authorities to better allocate medical resources and initiate public health interventions for higher-risk subpopulations.

This study also has some limitations. First, although the high-spatial-resolution dataset of the thermal index has been widely used in epidemiological studies, it still might introduce some measurement errors in the exposure assessment. Second, due to the limited facilities for detailed diagnoses, there might be some outcome misclassification in dispatch data. Third, missing or incomplete information in dispatch data such as patients' addresses and medical history, limits the precision and depth of data analysis. Fourth, our study findings are based on a single subtropical city (Shenzhen), caution is needed when generalizing the results to areas with different climatic conditions. Of note, we did not adjust any air pollutants in this study. As Reid et al. [68] reported the role of air pollutants in the temperature-mortality associations are more likely to be intermediates than confounders. Therefore, adjusting air pollutants in quantifying the total effects of thermal stress on cardiovascular morbidity might be inappropriate [69]. Moreover, prior studies that include the air pollutants in the main models have reported negligible change compared to unadjusted effect estimates [18, 70].

Conclusion

Our findings demonstrate that increased risks of emergency ambulance dispatch due to cardiovascular causes were related to cold stress exposure, especially among males and individuals older than 65 years. The highest risks of ambulance dispatch were observed at more extreme cold conditions, while moderate cold stress was responsible for the majority of thermal-related cardiovascular ambulance dispatch. Given the intensified climatic change, policymakers should inform more effective adaptation strategies and initiate targeted interventions, such as increasing affordable heating equipment, increasing public awareness, and operating cold-warning systems that could minimize the health impacts during cold episodes [71]. Future research may want to consider utilizing human thermo-physiological indices as exposure parameters in climate change modeling of health outcomes [31].

Authors’ contributions

MJ contributed to the formal analysis, wrote and revised the manuscript. ZY contributed to data curation and revised the manuscript. SH, NL, JJ, and ZL contributed to data collection. PW revised the manuscript. JP and JC supervised the research project. PY helped design the study and revised the manuscript.

Funding

The work was supported by the National Natural Science Foundation of China (Grant Nos. 82173628 and 81973004).

Data availability

The UTCI dataset is freely accessible on the Scientific Data (https://springernature.figshare.com/collections/A_High-spatial-resolution_Dataset_of_Human_Thermal_Stress_Indices_over_South_and_East_Asia/5196296). The ambulance data is not publicly available due to the access agreement, but it can be available with reasonable request and permission from the Shenzhen Center for Disease Control and Prevention.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.