Nonoptimal Temperature Is Associated With Higher Atrial Fibrillation Hospitalizations in China

Abstract

Background

Previous studies have evaluated extreme temperatures' impact on cardiovascular health, but few have specifically focused on atrial fibrillation (AF)-related hospitalizations across a wide temperature range.

Objectives

This study aimed to quantify and investigate the association between ambient nonoptimal temperatures and AF hospitalizations.

Methods

A two-stage time-stratified case-crossover study was conducted using a nationwide registry of 1,665,014 AF patients from 251 cities between 2014 and 2023. Conditional quasi-Poisson and distributed lag nonlinear models analyzed associations between nonoptimal temperature and AF hospitalizations. Subgroup and attributable burden analyses identified potentially susceptible subpopulations.

Results

The minimum hospitalization temperature for AF was 24.3 °C (74th percentile). Compared to the minimum hospitalization temperature over a lag of 0 to 14 days, cumulative relative risks for extreme cold and heat (1st and 99th percentiles) were 1.32 (95% CI: 1.24-1.42) and 1.03 (95% CI: 0.99-1.07), respectively. Hospitalization risks related to extreme temperatures were similar across subgroups of age, sex, and baseline diseases. Overall, 14.3% (95% empirical CI [eCI]: 12.2%-14.8%) of AF hospital admissions were attributable to nonoptimal temperatures, with higher burden in northern China (18.4%; 95% eCI: 15.7%-19.4%) than southern China (12.4%; 95% eCI: 9.6%-13.6%).

Conclusions

In this nationwide sample, extreme cold temperatures were associated with a greater risk of AF hospitalization. Excess risk was observed in northern China, where low temperatures prevail. This evidence highlights the importance of effective health care management and early resource allocation in high-risk regions.

Central Illustration

Atrial fibrillation (AF), the most common sustained cardiac arrhythmia, affects approximately 59.7 million individuals worldwide.^1,2 As a chronic, progressive disease, AF is associated with a five-fold risk of stroke, a 3-fold increased risk of heart failure, and a 2-fold increased risk of all-cause mortality.¹ Recurrent prolonged episodes of symptomatic AF can result in (re)hospitalization, accounting for 50% to 70% of the annual direct costs associated with AF.³ Despite significant advances in understanding and managing traditional cardiovascular risk factors in recent decades, the global burden of AF continues to rise, particularly due to population aging and the increasing prevalence of cardiovascular risk factors.² Although extensive research has focused on genetic and lifestyle determinants of AF, the potential impact of environmental exposures on AF occurrence and progression remains to be fully evaluated, especially in the context of ongoing environmental changes such as climate change, urbanization, and environmental pollution.⁴

The Global Burden of Disease Study 2019 indicates that nonoptimal temperatures are among the top ten leading causes of disease burden, accounting for 9.43% of all deaths globally.^5,6 Numerous studies have shown an association between nonoptimal temperature exposure and adverse cardiovascular events.7, 8, 9 However, most published studies to date have focused on mortality rather than morbidity data. As a measure of morbidity, hospitalization data more sensitively reflect the health effects of exposure to environmental factors compared to mortality data. Given that AF is a chronic, recurrent condition characterized by substantial symptom burden rather than immediate mortality risk,^8,10 hospitalization rates serve as a more appropriate indicator of disease burden. Compared to emergency department visits, which may only capture acute episodes, hospitalization records provide more complete clinical documentation and better reflect the sustained impact of environmental exposures. These comprehensive records enable robust analysis of the temporal relationship between environmental exposures and AF exacerbations.¹¹ Given the distinct contribution of AF compared to other types of arrhythmias,⁸ the environmental triggers of AF warrant further attention. Furthermore, the generalizability of these findings to a national level is uncertain because of variations in temperature levels and sociodemographic characteristics across cities of different sizes.12, 13, 14, 15

Given that climate change is expected to increase the frequency with which populations are exposed to extreme temperatures and that developing countries are particularly vulnerable to climate risks,^16,17 examining the potential specific effects of temperature on AF may enhance our current understanding. Such research could provide insights into the differential effects of temperature across various cardiovascular conditions and contribute to the foundation for comparative research. To extend current evidence in this field, this time-stratified case-crossover study was designed to investigate the associations between extreme temperature exposure and hospital admissions for AF events. This large-scale analysis was based on the RWS-CAF (Chinese Atrial Fibrillation Real-World Study database from 866 centers in 251 Chinese cities, ensuring a broad and representative sample for the study.

Methods

Data sources

The data regarding hospital admissions for AF were obtained from the RWS-CAF, an observational cohort study conducted across multiple centers (ChiCTR1900021250) that focuses on prevention, optimizing management, and enhancing surveillance of diagnosed patients.¹⁸ RWS-CAF regularly collects medical records of AF patients from the China Atrial Fibrillation Center (https://www.china-afc.org/), a national registry with 866 centers as of June 1, 2023. Of the 866 initially registered centers, 802 centers from 251 cities across China were included in the final analysis (Figure 1). Sixty-four centers were excluded due to insufficient duration of admission data recording (less than 3 months) to ensure statistical robustness. We included patients who were hospitalized with a primary discharge diagnosis of AF (International Classification of Diseases-10 code I48). Demographic information collected included age, sex, comorbidities, admission date, and city of residence. Full inclusion and exclusion criteria are detailed in the Supplemental Methods.

Figure 1.

Distribution of Participating Centers in China Atrial Fibrillation Program

Map showing locations of the final 802 centers included in the study and the initial 866 registered centers in the China Atrial Fibrillation Center Program.

We obtained daily mean temperature (Tmean, °C) and daily relative humidity (RH) (%) from the Meteorological Data Sharing Service System of People's Republic of China (http://data.cma.cn/). City-level meteorological data were collected from the nearest fixed-site national meteorological monitoring stations to each sample city. Air pollution data were obtained from the National Urban Air Quality Real-time Publishing Platform (http://datacenter.mee.gov.cn/), using daily average concentrations from state-controlled monitoring stations in each patient's city. We used 24-hour arithmetic means for all air pollutants except ozone (O₃), which was expressed as the daily maximum 8-hour moving average.

The Ethics Committee of the First Affiliated Hospital of Dalian Medical University approved the study protocol (number. PJ-KS-KY-2023-295). The need for individual patient consent was waived owing to the retrospective nature of the study, no patient contact, and the use of deidentified data.

Statistical analysis

The daily mean concentrations of meteorological and air pollution factors were summarized using descriptive statistics, including the mean ± SD for continuous variables and count (percentage) for categorical variables. Spearman correlation (r) analysis was performed to evaluate the associations between the meteorological and air pollution factors. We adopted a widely used two-stage analysis strategy to assess the link between ambient temperature and hospital admissions for AF. In the first stage, a time-stratified case-crossover model was established to estimate associations specific to each city across China and over time. In the second stage, a random-effects meta-analysis was used to pool the effect estimates at national and regional levels.

In stage 1, we fitted a conditional quasi-Poisson generalized linear model combined with a distributed-lag nonlinear function to characterize the temperature–AF relationship in each city. The quasi-Poisson specification accommodates overdispersion by assuming only a mean–variance relationship:

$$E\left(Y_t\right)={\mu }_t,Var\left(Y_t\right)=\theta {\mu }_t$$

where $Y_t$ is the daily count of AF admissions, ${\mu }_t$ is the expected count, and $\theta$ is the dispersion parameter estimated from the data. The linear predictor was

$$\mathit{log}\left({\mu }_t\right)=\alpha +\beta DLNM\left(temperature\right)+\gamma NS\left(humidity,3\right)+\delta Holiday$$

$$eliminate=factor\left(stratum\right)$$

The outcome variable $Y_t$ was the daily number of hospitalizations due to AF in each city. Distributed lag nonlinear model (DLNM) (temperature) represents a matrix obtained by applying a cross-basis function in the DLNM. We modeled the exposure-response relationship using a quadratic B-spline with 3 internal knots at the 10th, 50th, and 90th percentiles of the city-specific temperature distributions. A natural cubic spline function with an intercept and 2 internal knots at equally spaced log values of lags with a maximum of 14 days was applied. This 14-day window was selected based on physiological mechanisms of cold-induced cardiovascular stress, empirical evidence on delayed AF hospitalizations, and potential short-term displacement of susceptible cases. A natural cubic spline with degrees of freedom (df = 3) was used to control for potential nonlinear effects of RH on the same day (single-lag exposure). A binary indicator variable for public holidays was included to control for potential bias due to altered hospital access on holidays. α is the model intercept, and β, γ, and δ are the other coefficients. We employed a time-stratified case-crossover design (stratum = same city, year, month, and day-of-week) to assess the association between temperature exposure and AF hospitalizations. By design, this self-matching approach controls for fixed characteristics that do not vary within each stratum—such as sex, age, baseline comorbidities, and socioeconomic status—while accounting for day-of-week patterns, seasonality, and long-term trends. We conditioned on “stratum” using the “eliminate” function in the “gnm” package to implement this time-stratified approach.¹⁹ This design has been widely applied in studies investigating health effects of environmental exposures.

In stage 2, we performed a meta-analysis using a random-effects model to pool the estimated associations between ambient temperature and hospital admissions for AF at regional and national levels.²⁰ The minimum hospitalization temperature (MHT) is defined as the temperature at which hospitalizations for AF are lowest on regional and national levels; this temperature depends on local human adaptability, reflecting the most comfortable or optimal temperature.²¹ To ensure stable reference temperature estimation, we restricted the MHT to be within the 2.5th to 97.5th percentiles of the temperature distribution for each city and at the national level, avoiding imprecise estimates at distribution extremes. We computed relative risks (RRs) with 95% CIs for AF hospitalization at extreme cold and heat temperatures (1st and 99th temperature percentiles of the temperature series, respectively), using the MHT as the reference value. The heterogeneity of effect estimates was assessed using Cochran Q-test and I² statistic.

Given the significant variations in temperature ranges between southern and northern China throughout the year, we geographically classified the cities using the Huai River–Qin Mountains as the demarcation line (Figure 1). Northern China (109 cities) experiences relatively colder temperatures, with central heating available during winter, whereas southern China (142 cities) remains relatively warmer and lacks central heating. Central heating refers to a government-regulated system implemented in regions north of the Qinling–Huaihe River line, typically operating from November 15 to March 15. In addition, based on China's geographical characteristics, we divided the cities into the following 7 regions: North, Northeast, East, Central, South, Southwest, and Northwest China (Supplemental Figure 1).

Moreover, we derived the best linear unbiased prediction of the overall cumulative temperature associations from the fitted meta-analytical model for each city in terms of RRs. The best linear unbiased prediction represented a trade-off between region-specific and region-pooled associations and can provide more accurate estimates, especially in locations with small daily hospitalization counts or short-series areas with small hospital admissions. We determined the MHT for each city using the lowest overall cumulative exposure-response association from the city-specific temperature distribution, with the constraint that MHT must fall within the 2.5th to 97.5th percentiles of each city's temperature distribution to ensure statistical stability of the reference temperature. Population-attributable fractions (PAFs) were calculated using forward perspectives to estimate the burden of AF related to nonoptimal temperatures.²² Given the heterogeneity between the cities, different MHTs were calculated for each city within the specified percentile range. Based on each MHT, the attributable risk was then calculated. We used fractions rather than absolute numbers to ensure comparability across cities. Based on previous literature, we calculated the temperature-attributable risk fractions for AF hospitalizations in 2 temperature ranges: cold (below the MHT for each city) and heat (above the MHT for each city).²² Empirical CIs (eCIs) for PAFs were derived using bootstrap resampling methods with 1,000 iterations. To distinguish the effects of extreme and moderate temperatures, we categorized exposures into extreme cold (<2.5th percentile), moderate cold (2.5th percentile to MHT), moderate heat (MHT to 97.5th percentile), and extreme heat (>97.5th percentile). We calculated attributable risks separately for each range to quantify their relative contributions to the total temperature-related burden.

Subgroup analyses were performed based on age, sex, comorbidities, and estimated stroke risk (CHA₂DS₂-VASc score) to identify potentially vulnerable populations. Each subgroup analysis was conducted independently, with city-specific RRs pooled using random-effects meta-analysis. Results are presented as pooled RRs with 95% CIs. Between-subgroup differences were assessed using z-tests for binary comparisons and Cochran Q-tests for multiple comparisons, with P_interaction values representing the statistical significance of between-subgroup heterogeneity. PAFs were calculated for each subgroup. Empirical CIs for subgroup-specific PAFs were computed using the same bootstrap resampling approach as the main analysis. Detailed information is provided in the Supplemental Appendix.

Several sensitivity analyses were performed to test the robustness of our results. First, we separately added each criterion of air pollutant (O₃, PM_2.5, PM₁₀, CO, NO₂, and SO₂) and all 6 pollutants simultaneously in the same model to control for potential confounders. Second, we analyzed periods before the COVID-19 pandemic aimed to control the impact of prevention and control policies during the period of the COVID-19 pandemic on hospital admissions for AF. Third, we assessed the association between daily ambient temperature and AF admission rates, adjusting for the presence of central heating (a binary variable indicating whether central heating was active on that date) to account for its potential mediating or moderating effect on the exposure-response relationship. Fourth, we used different lag structures and degrees of freedom (df) for temperature (df = 4−7), humidity (df = 3−7), and different maximum lag periods (0, 0-1, 0-2, 0-7, 0-14, 0-21, and 0-28 days) in the DLNM to ensure the robustness of effect estimates. Fifth, we conducted effect modification analyses using both the Köppen–Geiger climate classification system (Dwa, BWh/BSh, Csa, Cwa, and Am, Supplemental Figure 2) and annual mean temperature quartiles (from coldest Q1 to warmest Q4, Supplemental Figure 3) to examine whether temperature-AF associations varied across different climatic contexts.

All statistical analyses were performed using R (version 4.0.4) in RStudio (RStudio). The DLNMs were implemented using the “dlnm” package (version 2.4.7),²³ conditional quasi-Poisson regression was fitted using the “gnm” package (version 1.1.5),¹⁹ and second-stage meta-analysis was conducted using the “mvmeta” package (version 1.0.3).²⁰ PAFs were quantified using the “attrdl” function within the “dlnm” package framework.²² A P value of <0.05 (2-sided) was considered statistically significant. More details are presented in the Supplemental Appendix.

Results

Descriptive statistics

From June 1, 2014, to June 1, 2023, a total of 2,500,578 AF hospitalizations were recorded in the RWS-CAF database. After applying the inclusion and exclusion criteria, 1,665,014 patients from 802 centers in 251 cities were included in the final analysis (Supplemental Figure 4). The mean age of AF patients was 72 ± 12 years; approximately 57% were men, and about three-quarters were aged 65 years or older. The most common comorbidities were congestive heart failure (37.5%), hypertension (33.6%), and diabetes (27.0%). Around 70% of the patients had at least 2 vascular risk factors at baseline, with an average CHA2DS2-VASc score of 3.1 ± 1.8 (Supplemental Tables 1 and 2). From 2014 to 2023, the mean daily ambient temperature (T_mean) was 15.9 °C, and the mean RH was 68.6%. Among environmental covariates, RH (r = 0.19) and O₃ (r = 0.54) exhibited positive correlations with the mean temperature. The remaining air pollutants (PM_2.5, PM₁₀, CO, NO₂, and SO₂) showed low to moderate negative correlations (−0.38 to −0.27).

Temperature–hospitalization association

The cumulative RR illustrates the association between ambient temperature exposure and AF hospitalization over a lag of 0 to 14 days (Figure 2). The relationship between the daily mean temperature and AF hospitalization exhibited an inverted J-shaped curve. The heterogeneity among different cities was moderate but statistically significant, with an I² value of 34.2% and a significant Cochran Q test (Q = 1898, P < 0.001).

Figure 2.

Exposure-Response Curves of Ambient Temperature and AF Admissions in China

(A) Exposure-response curves for daily mean temperature and hospital admissions for atrial fibrillation in mainland China, (B) northern China, and (C) southern China. Solid lines represent cumulative relative risk, with areas indicating 95% CIs. Vertical lines show the MHT, extreme cold (1st percentile), and extreme heat (99th percentile). Temperature distributions are shown at the bottom. AF = atrial fibrillation; MHT = minimum hospitalization temperature.

The minimum AF hospitalization occurred at the 74th percentile of temperature (24.3 °C), which we defined as the MHT at the national level. Cold temperatures exerted a more significant effect on AF hospitalizations (Figure 2). Using the MHT as a reference, the overall cumulative RR for extreme cold was 1.32 (95% CI: 1.24-1.42). However, exposure to extreme heat was not significantly associated with an increased risk of AF hospitalization (RR: 1.03; 95% CI: 0.99-1.07). The minimum risk temperatures for AF-related admissions in northern and southern China were the 78th percentile (21.7 °C) and the 71st percentile (25.6 °C), respectively. The risk of AF hospitalization attributed to extreme cold temperatures was higher in northern China (RR: 1.53; 95% CI: 1.35-1.73, corresponding to −13.9 °C) than in southern China (RR: 1.27; 95% CI: 1.16-1.39, corresponding to 0.7 °C) (P_interaction = 0.019) (Figure 2).

Over lags of 0 to 14 days, extreme cold temperature showed initial protective associations at lag 0 to 1 days (RR <1.0), followed by rapidly intensifying adverse effects, peaking around lag days 2 to 4, and was sustained over a longer period. Unlike the effect of extreme cold, the adverse effects of extreme heat were relatively mild (Supplemental Figure 7).

We found that the cumulative exposure-response and lag-response relationships were similar across subgroups. The cumulative effects of extreme cold and heat temperatures in different subgroups stratified by age, sex, and comorbidities are shown in the forest plot (Figure 3). There were no statistically significant differences between strata (all P_interaction >0.05). The subgroup analysis based on geographical regions suggested that AF patients residing in northern China were more vulnerable to extreme cold temperatures (P_interaction = 0.019) (Supplemental Figure 10). In contrast, the effects of extreme heat on AF hospitalizations showed no significant geographical variation (P_interaction = 0.44), with only the Northwest region demonstrating a statistically significant association (RR: 1.28, 95% CI: 1.06-1.53).

Figure 3.

Relative Risks of AF Admissions With Extreme Cold and Heat Temperatures by Subgroup

The figure shows the relative risks with 95% CIs of daily hospital admissions for atrial fibrillation in association with extreme cold temperature (1st percentile of temperature, left panel) and extreme heat temperature (99th percentile of temperature, right panel), stratified by age, sex, and comorbidities. ^aThe P_interaction values were obtained from Z-tests comparing the differences between the effect estimates of different subgroups. ^bThe risk factors refer to having any item included in the CHA₂DS₂-VASc score beyond sex (congestive heart failure, hypertension, age ≥65 years, diabetes, prior stroke/transient ischemic attack/thromboembolism, vascular disease). RR = relative risks; TIA = transient ischemic attack; other abbreviations as in Figure 2.

Attributable burden related to nonoptimal temperatures

Overall, 14.3% (95% eCI: 12.2%-14.8%) of AF hospitalizations were attributable to ambient temperature exposure during the study period (Table 1). Cold effects accounted for the vast majority (91.4%) of the attributable risk of temperature exposure. When examining the temperature-attributable burden by temperature range, we found that moderate cold temperatures were responsible for the majority of the attributable risk (12.0%; 95% eCI: 10.1%-12.8%). Extreme temperatures were responsible for a small fraction, with extreme cold accounting for 1.1% (95% eCI: 1.0%-1.2%) and extreme heat for 0.2% (95% eCI: 0.1%-0.2%) (Supplemental Figure 11). The attributable risk of AF hospitalization was 14.9% (95% eCI: 12.9%-15.4%) for males and 14.6% (95% eCI: 10.7%-15.4%) for females. In the age-stratified analysis, individuals aged 65 years or older had a lower attributable risk associated with temperature than those younger than 65 years (14.1%; 95% eCI: 12.0%-14.7% vs 15.8%; 95% eCI: 12.0%-16.4%) (Table 1). In addition, AF hospitalizations showed different PAFs among different comorbidity groups. The PAF of nonoptimal temperature was higher in patients with a history of hypertension, stroke/transient ischemic attack, and vascular disease, and lower in those with a history of congestive heart failure. Regional variations in the population-attributable risk were also observed, with higher risk in northern China compared with southern China. Consistently, similar results were observed across the 7 geographical regions (Table 2). This pattern shows moderate cold temperatures contributing substantially more to the attributable burden than heat (Supplemental Figure 11).

Table 1.

Pooled Attributable Fraction of Atrial Fibrillation Burden Related to Nonoptimal Temperature^a by Age, Sex, and Comorbidities

	MHT Percentile	All Nonoptimal (95% eCI)	Cold Related (95% eCI)	Heat Related (95% eCI)
Sex
Male	71st	14.87% (12.92-15.42)	13.56% (11.86-14.15)	1.31% (0.73-1.69)
Female	79th	14.63% (10.65-15.44)	13.12% (9.35-13.94)	1.51% (0.57-2.04)
Age
<65 y	70th	15.77% (12.00-16.35)	13.47% (10.21-14.02)	2.30% (0.97-2.93)
≥65 y	75th	14.14% (11.95-14.69)	13.14% (10.81-13.93)	1.00% (0.55-1.31)
Risk factorb
≤1 Risk factor	73rd	18.88% (14.22-19.04)	15.80% (11.70-16.22)	3.08% (1.46-3.70)
≥2 Risk factors	72nd	14.30% (12.10-14.90)	13.13% (10.97-13.91)	1.17% (0.66-1.53)
Comorbidities
Congestive heart failure
Yes	78th	13.92% (10.79-14.45)	12.36% (9.39-13.07)	1.55% (0.77-1.91)
No	73rd	17.26% (14.23-17.86)	15.47% (12.50-16.16)	1.79% (0.97-2.26)
Hypertension
Yes	76th	18.53% (14.24-19.07)	16.33% (12.26-17.03)	2.21% (1.14-2.62)
No	75th	15.39% (12.94-15.94)	13.88% (11.46-14.66)	1.50% (0.88-1.89)
Diabetes
Yes	74th	15.04% (11.17-15.60)	13.19% (9.67-13.84)	1.86% (0.91-2.22)
No	71st	15.20% (12.96-15.86)	13.63% (11.49-14.30)	1.57% (0.76-2.00)
Stroke/TIA
Yes	93rd	16.74% (10.55-17.01)	13.83% (8.77-14.46)	2.92% (0.85-3.56)
No	73rd	15.14% (12.78-15.76)	13.90% (11.70-14.67)	1.23% (0.63-1.66)
Vascular disease
Yes	88th	18.87% (14.21-19.02)	16.75% (12.43-17.35)	2.12% (1.03-2.53)
No	73rd	14.45% (12.13-15.18)	13.12% (10.86-13.79)	1.34% (0.72-1.70)
Total	74th	14.31% (12.19-14.83)	13.08% (10.96-13.79)	1.23% (0.67-1.58)

eCI = empirical CI; MHT = minimum hospitalization temperature; TIA = transient ischemic attack.

^aAttributable hospitalized burden computed as total and as separate components for cold and heat using the MHT as a referent.

^bThe risk factor refers to having any item included in the CHA₂DS₂-VaSc score beyond sex.

Table 2.

Pooled Attributable Fraction of Atrial Fibrillation Burden Related to Nonoptimal Temperature^a Stratified by Geographical Region

	MHT Percentile	All Nonoptimal (95% eCI)	Cold Related (95% eCI)	Heat Related (95% eCI)
Seven geographical regions
Northeast	94th	31.42% (16.85-38.31)	28.82% (16.19-34.81)	2.60% (−1.07 to 4.59)
North	88th	22.37% (18.06-25.43)	21.23% (17.26-23.65)	1.14% (−0.04 to 2.03)
Northwest	66th	22.45% (11.86-26.30)	17.39% (10.28-20.86)	5.06% (−1.92 to 7.95)
East	75th	12.32% (9.77-14.06)	11.03% (8.56-12.80)	1.29% (0.38-1.99)
Central	95th	14.36% (8.89-16.65)	10.82% (6.48-12.93)	3.54% (0.94-5.26)
Southwest	93rd	14.15% (3.63-19.69)	11.84% (3.48-17.48)	2.31% (−2.12 to 4.76)
South	68th	17.38% (13.79-19.44)	15.16% (11.23-17.54)	2.23% (1.12-2.97)
Regionb
Northern China	78th	18.40% (15.69-19.41)	16.93% (14.55-17.94)	1.47% (0.72-2.04)
Southern China	71st	12.38% (9.64-13.64)	11.18% (8.35-12.67)	1.20% (0.60-1.58)
Total	74th	14.31% (12.19-14.83)	13.08% (10.96-13.79)	1.23% (0.67-1.58)

Abbreviations as in Table 1.

^aAttributable hospitalized burden was computed as total and as separate components for cold and heat using the MHT as a referent.

^bThe 251 cities included in the study were geographically divided into northern (109 cities) and southern (142 cities) regions by the demarcation line of the Huai River–Qin Mountains and divided into 7 geographical regions according to the geographical characteristics of China.

Sensitivity analysis

A series of sensitivity analyses were performed. First, the associations remained robust after adjusting for air pollutants individually or all together (Supplemental Table 5). Second, RRs for extreme cold or heat temperatures were similar during the pre-COVID-19 period compared to the COVID-19 period (Supplemental Table 6). Third, after adjusting for central heating, associations persisted between AF hospitalization and extreme cold (RR: 1.31; 95% CI: 1.23-1.40) (Supplemental Table 7). Fourth, the results were stable in models using different lag structures and degrees of freedom for temperature (Supplemental Table 8) and humidity (Supplemental Table 9). Sensitivity analyses across different maximum lag periods (lag 0, 0-1 d, 0-3 d, 0-7 d, 0-14 d, 0-21 d, and 0-28 d) revealed distinct temporal patterns: heat effects were strongest at immediate exposure (lag 0: RR: 1.09; 95% CI: 1.04-1.13, particularly pronounced in northern China [RR: 1.14; 95% CI: 1.08-1.20]) and diminished rapidly thereafter, becoming borderline significant within 0 to 1 days (RR: 1.04; 95% CI: 1.00-1.08), and approaching the null by lag 0 to 3 days (RR: 1.01; 95% CI: 1.00-1.01). In contrast, cold effects showed delayed patterns, with associations gradually intensifying and peaking at lag 0 to 21 days (RR: 1.67; 95% CI: 1.41-1.98) (Supplemental Table 10). Fifth, effect modification analyses by climate characteristics showed that extreme cold effects did not significantly differ across Köppen–Geiger climate zones (P_interaction = 0.33) but varied significantly across annual temperature quartiles (P_interaction = 0.031). No significant effect modification was observed for extreme heat across either classification system (Supplemental Figures 12 to 14, Supplemental Table 11).

Discussion

To the best of our knowledge, this study is the most extensive nationwide investigation of the association between nonoptimal ambient temperature and hospital admissions for AF morbidity. An inverse J-shaped exposure-response curve was identified, with the lowest incidence of AF hospitalization at the 74th percentile of temperature. Short-term exposure to extreme cold temperatures was associated with increased hospital admissions for AF. In total, 14.3% (95% eCI: 12.2%-14.8%) of AF hospitalizations across 251 Chinese cities were attributable to temperature exposures (Central Illustration). Furthermore, the risk of AF hospitalization associated with extreme cold exposure varied among regions, with a more substantial adverse effect observed in northern China.

Central Illustration.

Ambient Temperature and Hospital Admissions for Atrial Fibrillation in China

The solid lines are the cumulative relative risk of hospital admissions for atrial fibrillation. Vertical lines indicate the minimum hospitalization temperature (MHT), extreme cold, and extreme heat (1st and 99th percentiles of temperature series, respectively). The lower part shows the temperature distributions. AF = atrial fibrillation; RWS-CAF = Chinese Atrial Fibrillation Real-World Study.

Epidemiological evidence consistently supports the notion that cold exposure is an independent risk factor for cardiovascular disease.^{8,9,11,15,24,25} However, most studies have been based on mortality data rather than morbidity data. As a measure of morbidity, hospitalization data are a more sensitive reflection of the health effects of exposure to environmental factors than mortality data, especially in chronic nonfatal conditions.²⁴ Measuring the associations between nonoptimal ambient temperature and AF hospitalizations could play a vital role in guiding prevention strategies, such as issuing extreme temperature warnings to patients. However, the evidence for the links between nonoptimal ambient temperature and the risk of AF-related hospital admissions is limited and may be influenced by regional variations.^8,11 To address this gap, our research found robust increases in daily hospital admissions for AF related to exposure to extreme cold temperatures, even after accounting for other influencing factors.

Cold exposure has been demonstrated to differentially affect hospitalizations across various cardiovascular conditions. A meta-analysis identified a 2.8% increase in overall cardiovascular hospitalizations associated with cold exposure (RR: 1.03; 95% CI: 1.02-1.04).²⁵ In the present study, extreme cold exposure was associated with a cumulative 32% increase in hospitalizations for AF over lag days 0 to 14 compared to the MHT (RR: 1.32; 95% CI: 1.24-1.42). This RR estimate for AF hospitalizations exceeds previous estimates of arrhythmic deaths from multicity studies conducted primarily in developed countries.⁸ Compared to specific cardiovascular conditions in a nationwide study from Japan, the RR for AF hospitalizations is lower than that for heart failure (RR: 1.57; 95% CI: 1.49-1.66) but higher than for total cardiovascular hospitalizations (RR: 1.23; 95% CI: 1.20-1.26), ischemic heart disease (RR: 1.12; 95% CI: 1.04-1.20), and stroke (RR: 1.11; 95% CI: 1.06-1.16).²⁶ The higher RR for AF hospitalizations, relative to both the overall cardiovascular hospitalization risk from the meta-analysis and most disease-specific risks from previous studies, suggests that AF may be particularly susceptible to extreme cold exposure. These findings indicate that different subtypes of cardiovascular diseases exhibit varying levels of sensitivity to extreme cold, underscoring the importance of targeted research to understand these differential impacts.

Unlike cold exposure, the impact of heat on the morbidity and mortality of cardiovascular diseases remains unclear, with contradictory findings reported in various studies.⁷ Some studies have reported no significant effects of extreme heat,¹⁵ whereas others suggest that high temperatures could be either protective²⁷ or detrimental²⁸ to cardiovascular disease. In the present study, the RRs of AF hospitalization were not significantly associated with extreme heat. These findings highlight that heat impacts vary across cardiovascular disease subtypes, underscoring the complexity of underlying physiological mechanisms.^8,26

The present study identified distinct temporal dynamics in the association between ambient temperature and AF hospitalizations: heat effects emerged immediately (lag 0), whereas cold effects appeared delayed, with elevated risks unfolding over longer lag periods. Initially, extreme cold exposure (1st percentile) seemed to confer a transient protective association at lag 0 to 1 days, followed by a rapidly increasing risk thereafter. However, our lag-specific exposure-response curves (Supplemental Figures 15) demonstrate that RR is elevated near the reference temperature (MHT at the 74th percentile) and falls below unity at the extreme cold end—a pattern consistent with a contrast artefact arising from reference point selection rather than true physiological protection. Although brief adaptive responses or short-term hospitalization displacement cannot be excluded, this artefact provides the most parsimonious explanation for the initial dip.²⁹ From lag 2 onwards (especially lag 2-4 days), the adverse cold-related risk becomes apparent and biologically plausible.

Low temperatures adversely affect blood composition and hemodynamics, leading to atrial remodeling.³⁰ These pathophysiological changes promote AF initiation, maintenance, and perpetuation. Research has shown that although initial sympathetic nervous system activation occurs during exposure, biomarkers of cardiovascular strain and myocardial injury (such as Mb, cTnI, and ET-1) remain elevated or even increase further in the days following cold exposure, indicating delayed cardiac cellular damage that cannot be immediately repaired.³¹ This progressive cardiovascular burden might partially account for the observed lag between cold exposure and peak AF hospitalizations.³¹

An estimated 14.3% of the total AF hospitalizations were attributable to nonoptimal temperature. The PAF was higher in patients with comorbidities that are also independent AF risk factors (hypertension, vascular disease), suggesting these high-risk individuals are particularly vulnerable to temperature effects. Regional subgroup analysis revealed that the attributable fractions related to nonoptimal temperatures were higher for individuals in northern China than for those in southern China. The cold effect manifested a faster onset effect and stronger cumulative effects in northern China patients compared to southern regions. Although we used percentile-based definitions to account for local adaptation as is standard in environmental epidemiology, these marked absolute temperature differences may contribute to the higher RRs observed in northern China, as physiological stress likely differs substantially between −13.9 °C and 0.7 °C, despite both representing locally extreme conditions. Future research could explore both percentile-based and absolute temperature thresholds to better disentangle adaptation effects from absolute physiological responses. Although extreme temperatures showed higher RRs, moderate cold temperatures contributed substantially more to the overall AF hospitalization burden, highlighting the importance of public health measures addressing prolonged moderate cold exposure, particularly in northern regions.

In light of the rapidly growing prevalence of AF and global population aging, actions aimed to prevent and control the health implications of climate change are and will be increasingly necessary.¹⁷ Mounting evidence indicates that the effect of cold exposure on the natural history of AF is multifaceted and is contributing not only to pathogenesis, but also to adverse outcomes.³² However, after searching all major AF guidelines from the American College of Cardiology, European Society of Cardiology, and American Heart Association, no coordinated actions are currently recommended to mitigate the effects of nonoptimal temperature.

Recent high-resolution analyses reveal that in northern China, extreme cold events have exhibited nonlinear and regionally intensifying patterns amid global warming. These events display significant asymmetry with respect to extreme heat trends, underscoring a more complex and spatially heterogeneous climate response.³³ Over the past decade, the frequency of extreme heat days across China has increased by approximately 20% to 30%, whereas the number of frost or extreme cold days has declined by over 30%, particularly in regions like the Yangtze River Basin and Eastern China.³⁴ These regional temperature shifts further contextualize the urgency of AF-related public health interventions under evolving climate conditions.

Our findings have implications for clinical practice. Firstly, at the policy and public health levels, the lag effects of cold exposure allow it necessary to address cold-related hazards from the perspective of urban planning early cold spell warnings. Secondly, at the health system level, active screening for AF should be provided at the health facility during the high epidemic season. Lastly, at the individual level, high-risk populations may avoid cold exposure and take protective strategies in cold seasons.

Study strengths and limitations

This study has several notable strengths. Firstly, hospital admission data were sourced from the RWS-CAF database—a standardized, nationwide, disease-specific database focused on AF in China. The China Atrial Fibrillation Centers are directly connected with the General Registry Center of RWS-CAF and provide an excellent representation of the AF burden in China. Secondly, our study investigated the exposure-response relationship between nonoptimal ambient temperature exposure and hospital admissions for AF. It provided information on the hospital admission burden attributable to nonoptimal temperature, which could facilitate more detailed pathogenesis or intervention research on AF. Thirdly, we conducted a subgroup analysis of AF patients with different cardiovascular risk factors to inform targeted public health prevention strategies in different geographical regions and populations.

Our study has several limitations. Firstly, we used a time-stratified case-crossover design to control for confounding variables that varied over time. However, residual confounding is still possible from unmeasured factors linked to hospitalization for AF, such as seasonal flu, acute respiratory infections, and pneumonia. Secondly, we used city-level data from fixed monitoring stations to represent population exposure instead of personal measurements. This approach cannot capture within-city variations such as urban heat island effects or localized air pollution patterns, potentially introducing exposure misclassification. Therefore, we could not observe the individual adaptive responses to ambient temperature exposure, such as staying indoors or using air conditioning during extreme temperature days, which might introduce exposure misclassification bias. Thirdly, our reliance on primary discharge diagnoses from hospital records might underestimate the true incidence of AF in the population. This approach misses individuals with undiagnosed AF, those managed in outpatient settings, and patients with AF as a secondary diagnosis. Furthermore, our data cannot distinguish between silent (asymptomatic) AF and symptomatic AF hospitalizations, which may have different temperature-related triggers. In addition, the RWS-CAF database considers each hospitalization as a separate entry, so our data could not differentiate between elective, emergency hospital admissions, and readmissions, making it difficult to assess the association between temperature exposure and acute symptomatic onset of AF vs planned hospitalizations.

Conclusions

Exposure to nonoptimal ambient temperatures is associated with increased risks of hospitalization related to AF in China. Excess risk was evident in northern China, where cold temperatures dominate. The study also found that patients with a history of hypertension, diabetes, stroke/transient ischemic attack, and vascular disease have a higher PAF. The potential impact of these findings could be substantial, as they provide crucial insights into the management and prevention of AF.

Perspectives.

COMPETENCY IN MEDICAL KNOWLEDGE: Nonoptimal ambient temperatures, particularly extreme cold exposure, are associated with increased hospital admissions for AF across Chinese cities. The risk is more pronounced in northern China and among patients with pre-existing cardiovascular risk factors, indicating the need for targeted preventive approaches.

TRANSLATIONAL OUTLOOK: Future studies should focus on individual-level temperature exposure assessment and adaptive behaviors to better define AF environmental triggers. Research distinguishing between emergency and elective admissions would clarify acute temperature effects on symptomatic AF. Development of targeted interventions for vulnerable populations is critical as climate change intensifies temperature extremes globally.

Funding support and author disclosures

This research was funded by the National Key Research and Development Program of China (Grant No. 2022YFC3601301); the National Key Research and Development Program of China (grant number 2022YFC2405002); the Dalian Science Fund for Distinguished Young Scholars (grant number 2022RJ13); the Medical and Industry Joint Innovation Programs from the First Affiliated Hospital of Dalian Medical University and Dalian Institute of Chemical Physics (DMU-1&DICP) (grant number UN202201); Liao Ning Revitalization Talents Program (grant number XLYC2002096); and the Interdisciplinary Innovative Talents Foundation from Renmin Hospital of Wuhan University (grant number JCRCZN-2022-003), and the Research Impact Fund by the Hong Kong Metropolitan University (RIF/2022/2.2). The authors have reported that they have no relationships relevant to the contents of this paper to disclose.