The Impact of Ambient Temperature on the Cardiovascular and Cerebrovascular Disease Incidence during 2009-2023 in Zhanjiang, Guangdong, China

Abstract

Aim: To investigate the effects of temperature exposure on the incidence of cardiovascular and cerebrovascular diseases in Zhanjiang, a subtropical coastal city in China.

Methods: We analyzed hospitalization data from 2009 to 2023 using generalized additive models (GAMs) and distributed lag nonlinear models (DLNMs) to examine the exposure-response relationships and lagged effects of daily maximum temperature on stroke (CVA), acute coronary syndrome (ACS), and acute aortic syndrome (AAS).

Results: Distinct temperature-associated risk patterns were observed for each disease type. ACS showed the strongest association, with significant risk increases at 1-day (relative risk [RR] = 1.19) and 5-day lags (RR = 1.12). The risk of CVA was moderately elevated at lower temperatures (22.7℃; RR = 1.08), whereas AAS exhibited a delayed peak at 5 days (RR = 1.29).

Conclusions: These subtropical patterns differ from those reported in temperate regions. Our findings support climate-specific warnings: namely, multi-day monitoring for CVA at lower temperatures, immediate and sustained alerts for ACS, and focused surveillance approximately 5 days after exposure for AAS.

1. Introduction

Climate change has emerged as a major global public health challenge in the 21st century, with extreme temperature events (e.g., heatwaves and cold spells) receiving significant attention for their health impacts. Extensive epidemiological studies have shown that abnormal temperature fluctuations can substantially increase the mortality rates¹⁾. As global warming intensifies, heat-related health risks have become increasingly evident, and the effects of cold weather remain equally concerning, particularly in temperate and subtropical regions, where temperature-health relationships often exhibit complex patterns²⁾.

In China, the vast territory and diverse climate types create marked regional differences in the vulnerability to extreme temperatures. For instance, northern regions, which experience prolonged winter cold spells, face more persistent cardiovascular effects from low temperatures, whereas southern regions face greater health risks from summer heatwaves owing to high heat-humidity conditions³⁾. In addition, rapid urbanization has intensified the urban heat island effect, thereby further increasing the heat exposure risk for urban residents⁴⁾. Therefore, multi-city studies across different climate zones are essential to explore the regional heterogeneity in temperature-health effects and inform targeted public health interventions.

While early studies predominantly employed linear regression models to explore the relationships between environmental factors and health outcomes, temperature-health effects are typically nonlinear and may involve lagged effects. Therefore, sophisticated models are necessary to accurately characterize these exposure-response relationships. Recent advances in generalized additive models (GAMs) and distributed lag nonlinear models (DLNMs) have significantly advanced this field⁵⁾. GAMs flexibly capture nonlinear associations, whereas DLNMs further account for lag effects, thus enabling a more precise estimation of both short-term (e.g., immediate heatwave effects) and long-term (e.g., cumulative cold spell effects) health risks⁶⁾. This methodological framework has been widely applied to define the disease-specific risk of exposure windows. For example, studies of acute heat effects on respiratory and cerebrovascular outcomes often use lag periods of 0-7 days to capture immediate to subacute physiological stress, as shown by research in Nanjing on bronchial diseases⁷⁾ and in Chengdu on cardiovascular mortality⁸⁾. In contrast, to fully capture the protracted impacts of cold exposure, studies conducted in colder regions, such as Northern Greece and Northeast China, have employed longer maximum lag periods of 21-27 days, revealing that cold effects on circulatory mortality can persist and accumulate over several weeks^{9, 10)}. Similarly, research in Beijing on respiratory morbidity confirmed that the cumulative cold effect peaked at a lag of 21 days¹¹⁾. These consistent findings highlight the critical importance of selecting lag periods that align with both the expected pathophysiological mechanisms and regional climate characteristics. For instance, Yang et al. (2016) demonstrated that cold weather may have more prolonged impacts on stroke mortality than heat, particularly in northern China³⁾, further underscoring the need for a regionally calibrated analytical approach when assessing temperature-health effects.

Although temperature-health relationships in temperate zones have been well studied, adaptive mechanisms may differ in subtropical regions, such as southern China. For instance, a study in Shanghai found that the impact of heat on mental health hospitalizations was more pronounced than in temperate zones, potentially due to high summer humidity¹²⁾. Moreover, subtropical populations exhibit reduced cold adaptation, meaning that even modest winter temperature drops can elevate cardiovascular risks¹³⁾. However, research on temperature stratification in subtropical areas remains limited, particularly regarding regional heatwave threshold definitions and vulnerability variations across subpopulations (e.g., urban-rural disparities and age groups), warranting further investigation.

Building on this background and existing gaps, this study focuses on Zhanjiang, Guangdong Province. Using hospital data and gridded meteorological observations from 2009-2023, we employed GAM and DLNM to systematically evaluate the acute effects of extreme temperatures on cardiovascular and cerebrovascular mortality. Our innovation lies in optimizing temperature-health assessment methods for subtropical climates. The results will inform regional early warning systems for extreme weather and enhance public health preparedness. Ultimately, this research aims to provide scientifically robust evidence for climate-adaptive health policies, thus supporting the strategic goals of “Healthy China 2030”.

2. Data and Methods

2.1 Data Sources

2.1.1 Hospitalization Records Collection and Management

This study utilized hospitalization records for cardiovascular and cerebrovascular diseases from a major hospital in Zhanjiang City between November 13, 2009, and December 31, 2023. The extracted variables included admission date, age, sex, diagnostic type, and related indicators. Diagnoses were defined using ICD-10 codes (see Supplementary Table 1).

Supplementary Table 1.Inclusion criteria of diagnoses

Disease Category	Specific Diagnosis	ICD-10 Code
Cerebrovascular Accident (CVA)	Ischemic stroke (IS)	I63.xx
	Intracerebral hemorrhage (ICH)	I61.xx
	Subarachnoid hemorrhage (SAH)	I60.xx
	Transient ischemic attack (TIA)	G45.9
Acute Coronary Syndrome (ACS)	Acute myocardial infarction (AMI)	I21.xx
	Unstable angina (UA)	I20.xx
	Ventricular tachycardia (VT)	I47.2
	Acute heart failure exacerbation	I50.23
	Atrial fibrillation (AF)*	I48.xx
Acute Aortic Syndrome (AAS)	Intramural hematoma (IMH)	I71.2
	Aortic dissection (AD)	I71.0x
	Aortic rupture (AAR)	I71.1

^*Only acute cases were included

Data were retrieved from the hospital’s integrated Electronic Medical Record (EMR) system (Research Data Repository, RDR, version 4.3.7; manufactured by Guangzhou AID Cloud Technology Co., Ltd). For each inpatient discharge, a structured abstract was automatically populated with demographic variables, admission/discharge dates, and the full ICD-10 code string assigned by the attending physician. Within 24 h of discharge, trained nosologists (certified by the Guangdong Provincial Health Commission) reviewed the abstracts for completeness, correct principal diagnosis, and ICD-10 coding accuracy, modifying codes when necessary. The finalized abstract is locked in the EMR and simultaneously pushed to an encrypted SQL server database located on the hospital’s internal network.

To ensure data quality, the following steps were implemented: (i) real-time validation rules (e.g., logic checks for impossible dates, mandatory fields); (ii) monthly random audits of 5 % of records by an independent quality-control team; and (iii) annual external audits by the municipal Health Information Center.

Patient privacy was protected by replacing personal identifiers with a one-way hashed 18-digit master patient key. All analysts accessed the data through a virtual private desktop located inside the hospital firewall; login required two-factor authentication, and all query logs were retained for 5 years. The study protocol was approved by the Ethics Committee of our hospital (approval No. 2022-047), which waived the requirement for individual informed consent because the analysis used de-identified administrative data.

2.1.2 Case Ascertainment

The ascertainment of cardiovascular and cerebrovascular events was conducted in two phases. An initial computerized screening phase automatically extracted all discharge records containing a principal or secondary diagnosis code listed in Supplementary Table 1 (ICD-10 codes I60–I64, I21–I22, I24.1, I46, etc.) from the EMR system. Subsequently, to minimize any misclassification bias, a clinical validation phase was performed, which involved a manual review of digital medical charts and ancillary tests by board-certified specialists. For stroke cases, a neurologist (Y.H.) reviewed brain CT or MRI reports within the Picture Archiving and Communication System (PACS), retaining only those with imaging evidence of acute cerebral infarction, intracerebral hemorrhage, or subarachnoid hemorrhage. For acute coronary syndrome (ACS), a senior cardiologist (L.M.) verified that cardiac troponin I levels exceeded the 99th percentile upper reference limit within 24 h of symptom onset, accompanied by at least one of the following: typical ischemic symptoms, diagnostic ECG changes, or imaging evidence of new wall motion abnormality.

Cases that did not meet these criteria were either reclassified or excluded. This two-phase process identified 81,220 stroke cases, 141,893 cardiovascular disease admissions, and 5,043 acute aortic syndrome cases during the study period. Inter-rater reliability, assessed on a random 10 % sample, was high (κ = 0.91 for stroke and 0.88 for ACS).

2.1.3 Meteorological Data

Gridded meteorological data from the European Centre for Medium-Range Weather Forecasts (ECMWF), specifically the ERA5 dataset with an original spatial resolution of approximately 31 km (0.28125° × 0.28125°)¹⁴⁾. Daily temperature indicators for Zhanjiang (20°13′–21°57′N, 109°40′–110°58′E)¹⁵⁾ were extracted, including the daily mean apparent temperature (aTEP), daily maximum temperature (maxTEP), and daily minimum temperature (minTEP). For subsequent analysis, regional average temperatures were computed by arithmetically averaging the values of all ERA5 grid points whose centroids fell within the specified geographical coordinates of Zhanjiang.

2.2 Statistical Analysis

2.2.1 Descriptive Analysis

First, we conducted descriptive statistical analyses of meteorological data and disease incidence. Time-series line charts were plotted for aTEP, maxTEP, and minTEP using temperature data from 2009 to 2023 in Zhanjiang. These visualizations intuitively demonstrate the temperature variation trends during the study period. For the three disease categories (CVA, ACS, and AAS), the date of disease onset was defined as the date of hospital admission for the corresponding event, which was extracted from EMR. Additionally, monthly incidence counts were plotted separately for the three disease categories to analyze their temporal distribution patterns and seasonal variation characteristics.

2.2.2 Generalized Additive Model

To investigate the nonlinear relationship between temperature and the incidence of cardiovascular and cerebrovascular diseases, we established a generalized additive model (GAM). The model formula is as follows.

log(E[Y_t]) = β₀+s(T_t)+s(time_t)+γ・DOW_t+ϵ_t

where:

•Y_t represents the daily disease incidence count on day t;

•E[Y_t] is the expected value of Y_t;

•β₀ is the intercept term;

•s(T_t) is a nonparametric smoothing function for temperature using cubic regression splines, expressed as s(T_t) = Σβ_jB_j(T_t);

•s(time_t) is a smooth function of the continuous time variable (days since the study start) to simultaneously control for long-term trends and seasonality;

•DOW_t represents the categorical effect of the day of the week;

•ϵ_t is a random error term.

The model adopted a quasi-Poisson distribution to account for potential overdispersion. Using this model, we plotted exposure-response curves for temperature and disease risk, where the y-axis represents the independent effect of temperature on the log-transformed incidence count while controlling for other factors. To investigate potential non-linear associations and threshold effects, daily maximum temperature was stratified into three categories based on its local distribution: low (＜25th percentile, P₂₅), moderate (P₂₅–P₇₅), and high (＞75th percentile, P₇₅). This approach, commonly used in environmental health studies, allows for the examination of risk differences across biologically relevant exposure ranges while maintaining sufficient statistical power within each stratum. The relative risk (RR) values with 95% confidence intervals (CIs) were calculated for each stratum. The optimal smoothing parameter was automatically selected using the generalized cross-validation (GCV) criterion. Model goodness-of-fit was assessed through residual analysis and the overdispersion parameter φ. All analyses were performed using the mgcv package in R 4.2.0, with a significance level of α = 0.05.

2.2.3 Distributed Lag Nonlinear Model (DLNM)

To further explore the lagged effects and nonlinear characteristics of temperature exposure, we constructed a DLNM. The mathematical expression of the model is as follows:

log(E[Y_t]) = β₀+Σf(T_t-l,l)+s(time_t)+γ・DOW_t+ϵ_t

where:

•Y_t is the daily disease incidence count on day t;

•E[Y_t] is the expected value of Y_t;

•β₀ is the intercept term;

•f(T_t-l,l) is the cross-basis function for temperature-lag effects;

•l is the maximum lag period (set to 5 days in this study);

•s(time_t) is a smooth function of the continuous time variable (days since the study started) to simultaneously control for long-term trends and seasonality;

•DOW_t represents the categorical effect of the day of the week;

•ϵ_t is a random error term.

The cross-basis function f(T_t-l,l) is expressed as:

f(T_t-l,l) = ΣΣη_jkϕ_j(T_t-l)ψ_k(l)

where:

ϕ_j(T_t-l) is a natural cubic spline basis function for the temperature dimension (with knots at the 25th, 50th, and 75th percentiles);

ψ_k(l) is a natural cubic spline basis function for the lag dimension with three knots placed on a logarithmic scale;

η_jk is the cross-basis parameter;

T_t-l is the maximum apparent temperature (maxTEP) on day t-l.

The model employed a quasi-Poisson distribution to account for overdispersion. Using this framework, we stratified temperature into quartiles (＜P₂₅, P₂₅–P₇₅, ＞P₇₅) and plotted exposure-response curves for each stratum. Based on the established literature illustrating the distinct temporal patterns of temperature effects on cardiovascular health, lag periods of 1, 3, and 5 days were selected for the primary DLNM analysis, which is commonly used in environmental epidemiology for capturing the immediate physiological stress (e.g., hemodynamic changes within 1 d) and subacute pathological processes (e.g., inflammatory responses peaking around 3-5 days) characteristic of acute heat effects on conditions like stroke and myocardial infarction^{7, 8)}. Concurrently, to assess the prolonged risk associated with cold exposure, a maximum lag of 21 days was applied in sensitivity analyses^{9, 10)}. This dual-lag framework ensures our analysis is sensitive to both the acute risks predominant in our subtropical study setting and the potential for longer-term cumulative burdens. We calculated RRs and 95% CIs using minimum morbidity temperature (MMT) as the reference (RR = 1), and visualized the spatial distribution of temperature-lag-risk associations using 3D surface plots. Model parameters were estimated via maximum likelihood. All analyses were conducted using the dlnm package in R 4.2.0 ¹⁶⁾, with a significance level of α = 0.05.

3. Results

3.1 Descriptive Results

Fig.1 illustrates the time-series trends of the daily mean, maximum, and minimum temperatures in Zhanjiang City from 2009 to 2023. The results demonstrated distinct cyclical variations in temperature, characterized by stable seasonal fluctuations. A slight year-by-year warming trend was observed, with notably higher temperatures recorded during 2020–2022. The trends in the daily maximum and minimum temperatures closely mirrored those of the daily mean temperature.

Fig.1.

Daily ambient temperature and monthly incidence of CVA, ACS, and AAS in Zhanjiang, China, 2009–2023

Fig.1 also displays the monthly incidence counts of CVA, ACS, and AAS from 2009 to 2023. The incidence of CVA exhibited a marked upward trend, with pronounced increases in both CVA and ACS cases after 2014, peaking in 2021. Similarly, the incidence of AAS showed a gradual increase.

A comparative analysis revealed the synchronous temporal patterns between daily mean temperature variations and the incidence of cardiovascular and cerebrovascular diseases. Periods of elevated temperatures appeared to be correlated with disease incidence peaks, and cyclical temperature changes aligned with the seasonal distribution characteristics of disease occurrence. Given Zhanjiang’s subtropical climate and the high similarity across temperature metrics, this study selected maxTEP as the core exposure indicator for subsequent analyses. This choice was based on its heightened sensitivity to extreme heat exposure, existing evidence supporting its association with cardiovascular and cerebrovascular events, and its alignment with the World Health Organization (WHO) guidelines for climate-health research in tropical regions.

3.2 GAM Model Results

3.2.1 CVA

The exposure-response relationship between stroke incidence risk and daily maximum temperature exhibited an N-shaped distribution pattern (Fig.2). The findings demonstrated that under Zhanjiang’s climatic conditions, there exists an optimal temperature range (22.7-30.3℃, corresponding to the 25th-75th percentiles) in which stroke risk was minimized (RR = 1.013). When temperatures exceeded the critical threshold of 30.3℃, the relative risk of stroke incidence increased slightly (RR = 1.017). Notably, in the low temperature range (＜22.7℃), although the risk remained at relatively higher levels (RR = 1.072), it did not reach statistical significance (Table 1).

Fig.2.

Association between the maximum temperature and CVA, ACS, and AAS risks using GAM

Table 1.GAM coefficients of CVA, ACS, and AAS risks by maximum temperature strata

Disease	Group	z	Relative Risk (RR)	P-value
CVA	Low ＜22.7℃	1.335	1.072 (0.968 - 1.186)	0.182
	Medium 22.7 - 30.3℃	0.334	1.013 (0.937 - 1.095)	0.739
	High ＞30.3℃	0.349	1.016 (0.927 - 1.114)	0.727
ACS	Low＜22.8℃	-2.051	0.896 (0.807 - 0.995)	0.040
	Medium 22.8 - 30.3℃	0.110	1.005 (0.925 - 1.091)	0.912
	High ＞30.3℃	-0.074	0.996 (0.907 - 1.095)	0.941
AAS	Low ＜22.7℃	1.325	1.130 (0.943 - 1.353)	0.185
	Medium 22.7 - 30.4℃	0.548	1.034 (0.918 - 1.163)	0.584
	High ＞30.4℃	-1.080	0.927 (0.807 - 1.064)	0.280

3.2.2 ACS

The exposure-response relationship between the acute cardiovascular disease risk and daily maximum temperature exhibited an N-shaped pattern (Fig.2). Specifically, the moderate temperature range of 22.8-30.3℃ (corresponding to the 25th-75th percentiles) showed a disease risk close to the reference level (RR = 1.005). When temperatures fell below 22.8℃, a significantly lower risk was observed (RR = 0.896, p = 0.040). When temperatures exceeded the threshold of 30.3℃, the disease risk decreased slightly (RR = 0.996). This indicates that, compared with the moderate temperature range of 22.8–30.3℃, both lower and higher temperatures were associated with reduced risks of acute cardiovascular disease, with cold exposure showing a stronger protective effect (Table 1).

3.2.3 AAS

The association between acute aortic syndrome risk and daily maximum temperature was not statistically significant (P＞0.05). Temperature-stratified analysis revealed the following patterns: in the low temperature range (＜22.7℃), although a slight 13% risk elevation was observed, it remained statistically non-significant. The moderate temperature range (22.7-30.4℃) showed risk levels close to the reference (RR = 1.034). Interestingly, the high temperature range (＞30.4℃) demonstrated a marginal risk reduction trend (RR = 0.927) (Table 1). These findings suggest that under Zhanjiang’s climatic conditions, the daily maximum temperature exerts a relatively limited influence on the incidence of acute aortic syndrome. Compared to cerebrovascular accidents and acute cardiovascular diseases, acute aortic syndrome risk patterns exhibited clearer fluctuations (Fig.2). This distinction may be attributed to the unique pathophysiological mechanisms and triggering factors specific to aortic diseases, warranting further investigation in future studies.

3.3 DLNM Model Results

This study employed a DLNM to investigate the lagged effects of temperature exposure on three categories of cardiovascular and cerebrovascular diseases. By constructing bivariate cross-basis functions, we systematically evaluated the joint effects of temperature (Var) and lag time (Lag) on the disease risk.

3.3.1 Three-Dimensional Plots

The three-dimensional surfaces in Fig.3 visually depict the non-linear and lag-dependent relationships between daily maximum temperature (X-axis) and the relative risk (RR, Z-axis) of disease incidence across different lag days (Y-axis). The contour lines on the base plane project the complex response onto a two-dimensional view, aiding in the identification of key patterns.

Fig.3.

Temperature-Lag-Response 3D surface for (a) CVA, (b) ACS, and (c) AAS by DLNM

The key findings can be interpreted by focusing on two primary aspects. Temperature-Dependent Risk: For a given lag day (e.g., cutting the surface along the Y-axis), the curve illustrates how the disease risk varies with temperature. The shape indicates whether the exposure-response relationship is J-, U-, or V-shaped and helps identify risk thresholds. Lag-Dependent Progression: For specific temperatures (e.g., cutting the surface along the X-axis), the curve shows how the risk evolves over the lag period. This reveals whether the effect is acute (risk peaks and decays rapidly within a few days) or delayed/protracted (risk accumulates or peaks after a longer period).

Specifically, (a) CVA: The surface suggests that the effect of high temperatures on stroke risk is acute. The RR peak is sharp and concentrated at short lags (likely within 0–3 days), forming a distinct “ridge” in the high-temperature zone that quickly diminishes. (b) ACS: The ACS surface demonstrated a more protracted and cumulative risk pattern. Elevated RR persisted over a wider range of lag days, indicating that the adverse effects of temperature exposure may accumulate over a longer period. (c) AAS: The AAS risk surface shows a crossover interaction: risk peaks at moderate temperatures when lag is very short or long but peaks at extreme temperatures when lag is intermediate (2–4 days). This indicates that the timing of aortic risk depends on the severity of the temperature exposure.

3.3.2 Lag-Specific Effects

Building upon the DLNM, this study further stratified the daily maximum temperatures into three percentile-based intervals: low-temperature range (Low, ＜P25), medium-temperature range (Medium, P25-P75), and high-temperature range (High, ＞P75). Using MMT as the reference level (RR = 1), lag-specific effect models were developed for each of the three disease categories (Fig.4), and detailed statistics are shown in Supplementary Table 2.

Fig.4.

Lag-Response patterns of (a) CVA, (b) ACS, and (c) AAS risks by temperature strata

Supplementary Table 2.DLNM parameter estimates for temperature-stratified lag effects on CVA, ACS, and AAS risks

Disease	Group	Relative Risk (RR)	P-value
CVA	Lower 22.7℃	1.08 (1.00-1.16)	0.037
	Medium 27.4℃	1.00 (0.98-1.03)	0.752
	Higher 30.3℃	1.02 (0.98-1.05)	0.386
ACS	Lower 22.8℃	1.32 (1.05-1.66)	0.017
	Medium 27.4℃	1.32 (1.06-1.63)	0.011
	Higher 30.3℃	1.20 (0.96-1.50)	0.115
AAS	Lower 22.7℃	1.39 (0.81-2.39)	0.227
	Medium 27.4℃	1.40 (0.84-2.35)	0.196
	Higher 30.4℃	1.44 (0.82-2.53)	0.205

(a) CVA

Based on the temperature stratification results, using the minimum morbidity temperature as the reference level (RR = 1.00), the low-temperature range (22.7℃) showed a slight, but statistically significant increase in stroke risk (RR = 1.08, 95% CI: 1.00-1.16, p = 0.037). The medium-temperature range (27.4℃) and high-temperature range (30.3℃) demonstrated non-significant relative risks of 1.00 and 1.02, respectively.

Further analysis of the lag effect patterns revealed significant differences in risk variations over time across different temperature ranges. Under low- and high-temperature conditions, the disease risk showed a fluctuating decreasing trend with increasing lag days, indicating that the impact of extreme temperatures was primarily concentrated in the short term after exposure, whereas the medium-temperature range exhibited a steady pattern of risk around the reference level.

(b) ACS

Using MMT as the reference level, the results showed that the temperature ranges of low (22.8℃) and medium (27.4℃) were significantly associated with acute cardiovascular disease risk (p＜0.05). Specifically, the relative risk was 1.32 (95% CI: 1.05-1.66, p = 0.017) in the low-temperature range and 1.32 (95% CI: 1.06-1.63, p = 0.011) in the medium range.

An analysis of the lag effect patterns revealed distinct dynamic characteristics of risk variation over time across different temperature ranges. The high-temperature range showed decreasing risk trends over the five lag days, suggesting that the acute impact of high temperatures on the cardiovascular system may gradually decline over time. The medium- and low-temperature ranges exhibited similar variations, both showing an N-shaped trend, reaching the highest risk at a 1-day lag, but starting to increase again after day 4.

(c) AAS

Using MMT as the reference level, the study results showed that the disease risk in all temperature ranges remained higher than the reference level, although not statistically significant. The relative risks for the low-, medium-, and high-temperature ranges were 1.39, 1.40, and 1.44, respectively.

Further analysis of the lag effect characteristics revealed similar dynamic patterns of risk variation over time across different temperature ranges. The disease risk reached its lowest point at a 1 d before rebounding until a 3-day lag, suggesting potential delayed effects on aortic diseases. Compared with cerebrovascular accidents and acute cardiovascular diseases, acute aortic diseases showed the strongest risk effect under all temperature conditions, which may be related to the pathophysiological mechanisms specific to aortic diseases.

3.3.3 Temperature-Specific Effects

Based on the DLNM, this study examined the impact of temperature exposure on disease risk at three specific lag periods: lag 1 d (Lag = 1), lag 3 days (Lag = 3), and lag 5 days (Lag = 5), using MMT as the reference level (RR = 1.00) (Fig.5). The corresponding statistics are shown in Supplementary Table 3.

Fig.5.

Temperature-dependent relative risks of (a) CVA, (b) ACS, and (c) AAS by lag strata

Supplementary Table 3.DLNM parameter estimates for lag-stratified CVA, ACS, and AAS risks across the temperature strata

Disease	Group	Relative Risk (RR)	P-value
CVA	Lag = 1	1.01 (1.00–1.02)	0.090
	Lag = 3	1.01 (1.00–1.03)	0.063
	Lag = 5	1.00 (1.00–1.01)	0.506
ACS	Lag = 1	1.19 (1.14–1.24)	＜0.01
	Lag = 3	1.16 (1.11–1.22)	＜0.01
	Lag = 5	1.12 (1.06–1.18)	＜0.01
AAS	Lag = 1	1.10 (0.95–1.28)	0.187
	Lag = 3	1.13 (0.94–1.36)	0.197
	Lag = 5	1.29 (1.03–1.61)	0.027

(a) CVA

The DLNM analysis revealed a nonlinear, N-shaped temperature-response relationship for CVA risk, characterized by an initial increase, subsequent decline, and final marked upswing across the temperature spectrum. Notably, the magnitude of this N-shaped fluctuation intensified during longer lag periods. At lag 1 d, the risk curve was the most attenuated, with modest elevations at median and high temperatures (e.g., RR = 1.01–1.02). By lag 3 days, the curve amplitude increased, showing a more pronounced N-pattern, although the effect estimates remained near the null (RR = 1.01–1.03, p = 0.063). The most dramatic N-shaped oscillation emerged at a lag of 5 days (greatest amplitude), yet paradoxically coincided with point estimates closest to 1.00 (RR = 1.00–1.01, p = 0.506), indicating wider dispersion around the risk continuum.

This temporal amplification of the N-pattern suggests a progressively evolving cerebrovascular response to the stimulus. The early muted response may reflect immediate hemodynamic adjustments, whereas the later exaggerated N-shape could indicate the superposition of delayed pathological processes (e.g., inflammatory activation or coagulation cascade effects) upon the initial thermal stress. The dissociation between pronounced curve morphology and non-significant point estimates at lag 5 days highlights a highly heterogeneous risk distribution, where specific temperature subranges may confer substantial risk that is diluted in aggregate estimates.

(b) ACS

The ACS risk demonstrated a strong, significant, and consistently N-shaped association with temperature across all lag periods, with risk magnitude inversely related to lag length. The most substantial effects were observed at lag 1 d (RR = 1.19, 95% CI: 1.14–1.24, P＜0.01), characterized by a sharp initial rise to a local peak at approximately 25℃, a decline to a trough near 30℃, followed by a second pronounced upswing at extremely high temperatures. This biphasic risk pattern persisted but attenuated at longer lags: lag 3 days (RR = 1.16) and lag 5 days (RR = 1.12), maintaining the characteristic N-shape with a sequentially diminishing amplitude.

The decaying risk magnitude with increasing lag (Lag1 ＞Lag3 ＞Lag5) suggests that the cardiovascular system is most vulnerable to temperature perturbations in the acute phase (within 24 h), likely mediated by rapid hemodynamic stress, platelet activation, and sympathetic surge. The persistence of a significant, albeit declining, N-shaped response at lags of 3 and 5 days implies additional subacute pathways, such as sustained inflammatory cytokine release or endothelial dysfunction. The consistent N-shaped curve across lags indicates a robust, non-monotonic dose-response relationship in which both moderately high and extreme temperatures pose distinct threats to coronary stability.

(c) AAS

In contrast to CVA and ACS, AAS exhibits a delayed and monotonic risk profile. The temperature-response relationship was characterized by a gradual, near-linear decline in risk with increasing temperature, with all curves converging near an RR of 1.0 at higher temperatures. Risk estimates showed a clear temporal progression: a non-significant elevation at lag 1 d (RR = 1.10, p = 0.187) and a similar trend at lag 3 days (RR = 1.13, p = 0.197), culminating in a statistically significant risk peak at lag 5 days (RR = 1.29, 95% CI: 1.03–1.61, p = 0.027).

This distinctly delayed risk escalation highlights the unique pathophysiological timeline of aortic diseases. The monotonic declining curve shape suggests that lower temperatures pose a sustained, progressive threat to aortic integrity, potentially through mechanisms such as chronic vasoconstriction, increased peripheral resistance, and delayed matrix metalloproteinase activation in the aortic wall. The absence of an acute, high-amplitude response underscores that aortic dissection may be less driven by immediate hemodynamic shock and more by cumulative biomechanical stress and wall remodeling over several days. This temporal discordance with cerebrovascular and acute coronary responses necessitates distinct public health and clinical monitoring strategies that focus on medium-term risk windows.

4. Sensitivity Analysis

Based on the sensitivity analysis extending the maximum lag period to 21 days, the overall temperature-risk relationships for cardiovascular and cerebrovascular diseases observed in the main analysis (1–5-day lags) remained consistent, while also revealing important nuances regarding the temporal distribution and cumulative magnitude of the effects (Supplementary Figs.1 and 2).

Supplementary Fig.1.

Lag-Response patterns of (a) CVA, (b) ACS, and (c) AAS risks by temperature strata with 21-day lag

Supplementary Fig.2.

Temperature-dependent relative risks of (a) CVA, (b) ACS, and (c) AAS of 21-day lag

The most notable finding from the extended lag model was the substantial amplification of cumulative risk estimates for all three disease outcomes. For CVA, the cumulative relative risk over 21 days reached 1.57 (95% CI: 1.52–1.62), which was markedly higher than any single lag point in the main analysis. This suggests that the protracted effects of temperature exposure on stroke risk are considerably greater than those predicted by acute-phase models (≤ 5 days). The lag-specific pattern for CVA showed a gradual decline in risk over the extended period before stabilizing near the null, indicating that while the cumulative burden was high, the acute phase still contributed significantly to the overall risk.

Similarly, for ACS, the 21-day cumulative risk was 1.36 (95% CI: 1.32–1.39), representing an aggregation of effects beyond the 5-day period. The lag-specific analysis revealed a continuing downward trend in risk over the extended period, particularly for high temperatures, where the risk diminished from an initial elevation to near baseline levels. This pattern implies that the cardiovascular system experiences prolonged vulnerability to temperature stress, with effects that dissipate gradually rather than abruptly.

AAS exhibited the largest discrepancy between the short-term and extended lag models. While the main analysis showed peak risk at 5 days (RR = 1.29), the 21-day cumulative risk surged to 1.91 (95% CI: 1.68–2.14). This increase underscores that aortic pathology may involve particularly delayed mechanisms that conventional shorter lag periods fail to capture. The lag-specific pattern for AAS, characterized by an initial decline followed by a secondary rise, further supports the existence of multiphasic pathological processes unfolding over weeks rather than days.

A comparison of the disease-specific responses revealed an important hierarchy in temporal vulnerability. Cerebrovascular risk demonstrates an early substantial impact with a significant cumulative burden, coronary risk shows pronounced acute effects that accumulate moderately over time, and aortic risk manifests a minimal immediate response but dramatic delayed accumulation. These differential timelines likely reflect distinct underlying pathophysiological mechanisms: immediate hemodynamic stress for stroke, combined hemodynamic and thrombotic effects for myocardial infarction, and delayed structural changes in the aortic wall for aortic dissection syndromes.

From a methodological perspective, these findings suggest that standard short lag periods (≤ 5 days) systematically underestimate the total burden of temperature-related cardiovascular morbidity, particularly in aortic diseases. This has important implications for both risk assessment models and public health preparedness, indicating that preventive measures should account for extended risk windows, particularly in vulnerable populations and during prolonged temperature extremes.

5. Discussion

This study employed generalized additive models and distributed lag nonlinear models to analyze the acute effects and lagged impacts of daily maximum temperatures on three categories of cardiovascular and cerebrovascular diseases in Zhanjiang from 2009 to 2023, yielding findings with significant clinical and public health implications. The discussion proceeds from three dimensions: epidemiological characteristics, pathophysiological mechanisms, and public health significance.

From an epidemiological perspective, the most notable finding was the distinct disease-specific response patterns to temperature exposure observed across different cardiovascular and cerebrovascular conditions. CVA exhibited a characteristic N-shaped exposure-response relationship. However, this study found that the optimal temperature range in Zhanjiang (23.5-28.7℃) was significantly higher than those reported in temperate regions⁷⁾, potentially reflecting the unique physiological adaptations developed by subtropical populations through long-term heat acclimatization. When temperatures exceeded the critical threshold of 30.3℃, stroke risk increased within 0-3 days of exposure, aligning well with the mechanism of impaired cerebrovascular autoregulation under heat stress¹³⁾. In contrast, ACS demonstrates more complex biphasic effects, showing both immediate heat stress responses and delayed risk elevation after 5 days, possibly representing two distinct pathological processes: early hemorheological changes and subsequent inflammatory activation¹⁷⁾. AAS exhibited the most distinctive risk pattern, with relatively weaker temperature associations and risk elevations only occurring during 0-1 days after temperature drops and 3-5 days following sustained heat, consistent with the biomechanical characteristics of aortic wall stress changes³⁾.

From a pathophysiological standpoint, these findings provide important insights into the biological pathways through which environmental temperatures affect cardiovascular and cerebrovascular health. High temperatures may increase cardiovascular risk through multiple mechanisms: 1) causing hemoconcentration and increased blood viscosity, promoting thrombosis¹⁸⁾; 2) inducing systemic inflammatory responses and endothelial dysfunction¹⁹⁾; and 3) disrupting autonomic balance and increasing arrhythmia risk¹²⁾. The differential temperature effects observed across diseases may be related to the physiological characteristics of the target organs: brain tissue is highly sensitive to ischemia and hypoxia, while the myocardium has a greater metabolic reserve capacity, and the aorta is particularly sensitive to mechanical stress due to its unique structural features. Notably, this study is the first to observe the protective effect of high temperatures on acute aortic diseases in subtropical populations, which is different from the findings in temperate regions²⁰⁾. This discrepancy may reflect the fundamental differences in vascular reactivity between climate zones, requiring further laboratory validation.

This study also yielded innovative findings regarding lag effects. By constructing three-dimensional response surfaces, we systematically characterized the stereoscopic relationship between temperature, lag time, and risk in subtropical regions. The heat effects on stroke exhibited an acute onset pattern, while ACS demonstrated cumulative effect characteristics, consistent with findings from a German cohort study showing that the risk of ischemic stroke peaked immediately during heatwaves²¹⁾. However, our study revealed that the heat effects in Zhanjiang were shorter in duration but more intense, potentially reflecting the differences between subtropical and temperate climates. Particularly noteworthy is the lag pattern of AAS, with its risk peak appearing 5 days after temperature exposure (RR = 1.22). This delayed response may reflect the temporal process of aortic wall matrix metalloproteinase activation and extracellular matrix remodeling, providing new temporal evidence for understanding the environmental triggers of aortic diseases²⁰⁾.

From a public health perspective, these findings have significant practical implications. First, they established temperature-health early warning indicators specifically applicable to the Zhanjiang region. These indicators differ markedly from those in temperate zones, emphasizing that health warning systems must consider regional climatic characteristics. Second, the study identified optimal warning time windows for different diseases: stroke requires attention to risks within 72 h after heat exposure, ACS necessitates vigilance for both immediate and 5-day delayed risks, while AAS demands particular focus on risk changes approximately 5 days after temperature fluctuations.

Several limitations should be considered when interpreting the results of this study. First, the use of single-center data may affect external validity and introduce a selection bias. Specifically, as the hospitalization records were sourced from one major tertiary hospital in Zhanjiang, the study cohort may underrepresent certain populations, such as rural residents, individuals with milder symptoms who sought care at community clinics, and fatal cases occurring before hospital admission. Consequently, the reported effect estimates might be more reflective of severe cases that reach tertiary care, and the generalizability of our findings to the broader population, particularly in primary care settings or for fatal out-of-hospital events, should be interpreted cautiously. Future multi-center studies encompassing different levels of healthcare facilities would help validate and extend these findings. Second, limitations exist in individual-level exposure assessment, as we employed gridded meteorological data rather than personal exposure measurements, potentially causing misclassification of exposure. Third, the study did not account for additional confounding climatic factors, where residual confounding from elements such as air pollution and humidity may persist and influence the results. Furthermore, methodological considerations pertain to the time-series structure of the data. The Ljung-Box test indicated the presence of some residual autocorrelation (p ≈ 0.03) in our models, suggesting that not all short-term temporal dependencies were fully captured by the included smoothers for long-term trends and seasonality. While this is a common challenge in environmental time-series studies and our core model has already been adjusted for key temporal confounders, this residual autocorrelation could potentially lead to an underestimation of standard errors. Future studies could address this by explicitly modeling the autocorrelation structure, for instance, by employing generalized additive mixed models (GAMMs) with an autoregressive term, which may provide more precise variance estimates. Future research could also be improved by 1) conducting multi-center collaborative studies, 2) incorporating wearable devices to obtain individual microenvironment exposure data or considering more complex climate data, and 3) including biomarkers to explore mechanistic pathways.

5. Conclusions

This study systematically analyzed temperature variations and cardiovascular disease data from a hospital in Zhanjiang between 2009-2023, revealing distinct patterns of climate-health relationships in subtropical coastal cities. The main findings can be summarized in three aspects:

First, regarding exposure-response relationships, this study identified disease-specific temperature thresholds applicable to Zhanjiang. While the overall temperature-stroke relationship in GAM models was non-significant, the DLNM analysis indicated a significant elevation in cumulative stroke risk, specifically at lower temperatures (22.7℃, RR = 1.08). For ACS, a strong and consistent risk increase was observed across multiple lag days, with the highest risk at lag 1 d (RR = 1.19). The study also validated, for the first time in tropical regions, the significantly delayed peak risk for AAS at 5 days post exposure (RR = 1.29), providing critical parameters for refining regional health risk assessment models.

Second, to characterize lag effects, temporal risk profiles were developed. Using distributed lag nonlinear models, we quantified the lag-specific burdens: ACS risk was most acute (lag 1-day), whereas AAS risk was most delayed (lag 5-day). The risk of CVA was more modest and distributed across the lag period. These results provide new epidemiological evidence for understanding the disease-specific temporal dynamics of the impact of climate change on cardiovascular health.

Third, for public health applications, this study proposed a differentiated prevention framework. Based on risk stratification, we recommend targeted interventions: enhancing awareness of stroke risk following lower temperature exposure over a multi-day window; establishing “immediate (Lag 1-day) + sustained” monitoring systems for ACS given its strong and prolonged risk elevation; and implementing focused surveillance around 5 days after temperature fluctuations for AAS, corresponding to its delayed risk peak.

Future research should focus on the following: 1) multi-center studies to develop health risk maps across climate zones; 2) integrating genomics and epigenetics to explore the molecular mechanisms of climate adaptation; and 3) developing intelligent early warning platforms that combine real-time meteorological and health data. This work will contribute to more precise climate-resilient public health systems.

These findings provide locally calibrated evidence for Zhanjiang’s climate-health adaptation strategies and offer a transferable research paradigm for other subtropical coastal cities facing climate-related health challenges. As climate change intensifies, evidence-based health risk management and adaptation strategies will become increasingly crucial, requiring continued interdisciplinary collaboration and innovation among the meteorological, medical, and public health sectors.

Acknowledgments

This study was supported by the Unfunded Science and Technology Project in Zhanjiang City (grant number 2022B01029) and the Lingnan Normal University’s 2025 Innovation and Entrepreneurship Training Program (grant number DC [2025]14-93/97).

Conflict of Interest

The authors declare that they have no conflicts of interest associated with this work.

References

• 1)  Vicedo-Cabrera AM, Scovronick N, Sera F, Royé D, Schneider R, Tobias A, Astrom C, Guo Y, Honda Y, Hondula DM, Abrutzky R, Tong S, de Sousa Zanotti Stagliorio Coelho M, Saldiva PHN, Lavigne E, Correa PM, Ortega NV, Kan H, Osorio S, Kyselý J, Urban A, Orru H, Indermitte E, Jaakkola JJK, Ryti N, Pascal M, Schneider A, Katsouyanni K, Samoli E, Mayvaneh F, Entezari A, Goodman P, Zeka A, Michelozzi P, de’Donato F, Hashizume M, Alahmad B, Diaz MH, De La Cruz Valencia C, Overcenco A, Houthuijs D, Ameling C, Rao S, Ruscio FD, Carrasco-Escobar G, Seposo X, Silva S, Madureira J, Holobaca IH, Fratianni S, Acquaotta F, Kim H, Lee W, Iniguez C, Forsberg B, Ragettli MS, Guo YLL, Chen BY, Li S, Armstrong B, Aleman A, Zanobetti A, Schwartz J, Dang TN, Dung DV, Gillett N, Haines A, Mengel M, Huber V, Gasparrini A: The burden of heat-related mortality attributable to recent human-induced climate change. Nat Clim Chang, 2021; 11: 492-500
• 2)  Burkart KG, Brauer M, Aravkin AY, Godwin WW, Hay SI, He J, Iannucci VC, Larson SL, Lim SS, Liu J, Murray CJL, Zheng P, Zhou M, Stanaway JD: Estimating the cause-specific relative risks of non-optimal temperature on daily mortality: a two-part modelling approach applied to the Global Burden of Disease Study. Lancet, 2021; 398: 685-697
• 3)  Yang J, Yin P, Zhou M, Ou CQ, Li M, Li J, Liu X, Gao J, Liu Y, Qin R, Xu L, Huang C, Liu Q: The burden of stroke mortality attributable to cold and hot ambient temperatures: Epidemiological evidence from China. Environ Int, 2016; 92-93: 232-238
• 4)  Ho JY, Shi Y, Lau KKL, Ng EYY, Ren C, Goggins WB: Urban heat island effect-related mortality under extreme heat and non-extreme heat scenarios: A 2010-2019 case study in Hong Kong. Sci Total Environ, 2023; 858: 159791
• 5)  Wang Y, Tian X, Pang B, Ma W, Kou Z, Wen H: Long-term effects of meteorological factors on severe fever with thrombocytopenia syndrome incidence in eastern China from 2014 to 2020: An ecological time-series study. PLoS Negl Trop Dis, 2024; 18: e0012266
• 6)  Wang W, Ma Y, Qin P, Liu Z, Zhao Y, Jiao H: Assessment of mortality risks due to a strong cold spell in 2022 in China. Front Public Health, 2023; 11: 1322019
• 7)  Shan, C., Wang, Z., Wang, Y., Sang, X., Yao, R., & Ai, W: Effect of daily maximum temperature on the number of patients with bronchial and cardiovascular diseases in Nanjing. Journal of the Meteorological Sciences, 2024; 44: 593-602
• 8)  Yao J, Sun J, Kuang D, Qian W, Luo L, Qin F, Zhai Y, Li Y, Huang J, Wang C, Lu R, Gao X: Interaction effects of high temperature and ozone on cardiovascular disease mortality in Chengdu, 2014-2023. Front Public Health, 2025; 13: 1580849
• 9)  Psistaki K, Dokas IM, Paschalidou AK: The Impact of Ambient Temperature on Cardiorespiratory Mortality in Northern Greece. Int J Environ Res Public Health, 2022; 20: 555
• 10)  Ma Y, Li H, Qin P, Cheng B, Feng F, Zhang Y, Jiao H: Extreme temperatures and circulatory mortality in a temperate continental monsoon climate city in Northeast China. Environ Sci Pollut Res Int, 2023; 30: 21661-21670
• 11)  Li X, Zhang Y, Tian Z, Wang J, Zhao J, Lyu Y, Ni Y, Guo Y, Cui Z, Zhang W, Li C: Lag effect of ambient temperature on respiratory emergency department visits in Beijing: a time series and pooled analysis. BMC Public Health, 2024; 24: 1363
• 12)  Peng Z, Wang Q, Kan H, Chen R, Wang W: Effects of ambient temperature on daily hospital admissions for mental disorders in Shanghai, China: A time-series analysis. Sci Total Environ, 2017; 590-591: 281-286
• 13)  Zhai G, Jiang Z, Zhou W: Differences in cardiovascular disease mortality between northern and southern China under exposure to different temperatures: a systematic review. PeerJ, 2024; 12: e18355
• 14)  European Centre for Medium-Range Weather Forecasts. (2023). ERA5 reanalysis (fifth generation). ECMWF
• 15)  Zhanjiang Municipal Bureau of Natural Resources. (2023). Zhanjiang geographic overview. Government of Zhanjiang. http://www.zhanjiang.gov.cn
• 16)  R Core Team. (2022). R: A language and environment for statistical computing (Version 4.2.0). R Foundation for Statistical Computing. https://www.R-project.org/
• 17)  Chan Pin Yin DRPP, Ten Berg JM: Long-term residual cardiovascular risk after acute coronary syndrome: antithrombotic treatment options. Neth Heart J, 2022; 30: 38-46
• 18)  Chen K, Dubrow R, Breitner S, Wolf K, Linseisen J, Schmitz T, Heier M, von Scheidt W, Kuch B, Meisinger C, Peters A; KORA Study Group; Schneider A: Triggering of myocardial infarction by heat exposure is modified by medication intake. Nat Cardiovasc Res, 2022; 1: 727-731
• 19)  Xu J, Wang Q, Anikeeva O, Zhu P, Bi P, Huang C: Effects of extreme heat on physiology, morbidity, and mortality under climate change: mechanisms and clinical implications. BMJ, 2025; 391: e084675
• 20)  Chen J, Gao Y, Jiang Y, Li H, Lv M, Duan W, Lai H, Chen R, Wang C: Low ambient temperature and temperature drop between neighbouring days and acute aortic dissection: a case-crossover study. Eur Heart J, 2022; 43: 228-235
• 21)  He C, Breitner S, Zhang S, Huber V, Naumann M, Traidl-Hoffmann C, Hammel G, Peters A, Ertl M, Schneider A. Nocturnal heat exposure and stroke risk. Eur Heart J, 2024; 45: 2158-2166