The interplay of heatwaves, air pollution, and green space on all-cause mortality in older adults with diabetes mellitus: a national cohort study

Abstract

Background

While extreme heat represents an established risk factor for mortality in patients with diabetes mellitus (DM), its synergistic effects with air pollution and green space remain incompletely elucidated. This study aimed to investigate the joint effects of heatwaves, air pollution, and green space on all-cause mortality in older adults aged ≥ 45 years with diabetes mellitus.

Methods

We conducted a longitudinal analysis of data from 1,400 adults aged ≥ 45 years with DM, drawn from the China Health and Retirement Longitudinal Study (CHARLS, 2011–2020). Heatwaves were defined according to six intensity-duration thresholds, based on two temperature thresholds (95th/97.5th percentiles of daily maximum temperature) and three duration criteria (≥ 2/≥3/≥4 consecutive days). Time-varying Cox proportional hazards models were used to estimate hazard ratios (HRs), with adjustments for individual PM₂.₅, NO₂ and greenness exposures as time-varying covariates. Multiplicative and additive interactions were evaluated using product terms and the relative excess risk due to interaction (RERI), complemented by the attributable proportion due to interaction (AP) and the synergy index (S).

Results

Heatwave exposure was significantly associated with increased mortality risk in a graded exposure-response relationship, with HRs increasing from 1.027 (95% CI: 1.013–1.041) to 1.070 (95% CI: 1.048–1.093) as heatwave intensity and duration increased. Additive interaction analyses demonstrated significant synergistic effects between heatwaves and both air pollutants and reduced greenness. These combined effects were more pronounced among older individuals and urban residents.

Conclusion

Heatwaves interact synergistically with air pollution and diminished greenness to elevate mortality risk in diabetic patients. These findings highlight the imperative for integrated public health strategies that concurrently address these coexisting environmental exposures.

Clinical trial registration

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-27606-8.

Background

Diabetes mellitus (DM) poses a substantial and growing global public health burden. According to the International Diabetes Federation (IDF) [1], approximately 589 million adults were living with diabetes in 2024, a number projected to rise to 853 million by 2050. Individuals with diabetes experience elevated all-cause mortality relative to the general population, attributable in part to metabolic dysregulation and a heightened susceptibility to multi-organ complications [2]. Critically, this underlying pathophysiology may also render them particularly vulnerable to environmental stressors, a potentially modifiable risk factor that remains underexplored.

Climate change has increased the frequency and intensity of extreme heat events, including heatwaves, elevating risks to population health [3]. High temperatures are well-established contributors to increased morbidity and mortality from cardiovascular, respiratory, and renal diseases [4–6]. Importantly, individuals with diabetes are at a disproportionately higher risk during heatwaves [7], largely due to impaired thermoregulation, increased susceptibility to dehydration, and exacerbation of pre-existing comorbidities. Accordingly, epidemiological studies have consistently linked heatwave exposure to elevated all-cause mortality in this vulnerable population [7, 8].

Air pollution concurrently constitutes a critical threat to environmental health. Pollutants including fine particulate matter (PM₂.₅) and nitrogen dioxide (NO₂) are well-established contributors to the incidence, progression, and mortality of a range of chronic diseases [9–12]. In diabetic individuals, long-term exposure to elevated pollution levels has been linked to aggravated cardiovascular complications, progressive renal impairment, and elevated mortality risk [13]. Evidence from a major cohort of 174,063 Shanghai residents with newly diagnosed type 2 diabetes confirms that higher residential concentrations of PM₂.₅, coarse particles (PM₂.₅-₁₀), and NO₂ are significantly associated with increased all-cause and cause-specific mortality [14]. Further compounding this risk, the potential synergistic interaction between PM₂.₅ and high temperature may exacerbate adverse health outcomes in diabetic patients via mechanisms such as systemic inflammation, oxidative stress, and endothelial dysfunction [15]. Supporting evidence from animal models indicates that co-exposure to heat and NO₂ markedly accelerates the progression of diabetic kidney disease, manifesting as heightened hyperglycemia, deteriorated renal pathology, and intensified inflammatory and oxidative stress responses [16].

Urban vegetation provides valuable ecosystem services by mitigating urban heat island effects and improving air quality [17, 18]. The mechanisms of evapotranspiration, shading, and pollutant absorption collectively lower local temperatures and reduce ambient pollution levels. A robust body of research confirms that residential greenness confers a range of health benefits, including lower all-cause mortality [19], reduced risk of preterm birth [20], and better mental health [17, 21]. In diabetic individuals, greenness has been linked to both delayed diabetes onset and reduced all-cause mortality [14, 22, 23], relationships partly explained by lower exposure to air pollution. Therefore, green infrastructure emerges as a promising strategic resource to help mitigate the synergistic adverse effects of heatwaves and air pollution in this vulnerable group.

Despite this growing evidence, the combined effect of heatwaves, air pollution, and greenness on mortality risk in diabetic patients remains to be fully elucidated. Given the physiological susceptibility of this population and the escalating threats of climate change and air pollution, understanding these complex interplays is of critical public health importance. This study aims to quantify the association between heatwave exposure and all-cause mortality in Chinese older adults with diabetes and to evaluate the potential effect modifications by air pollution and greenness. We hypothesize that diabetic patients experience not only direct harms from each environmental exposure but also synergistic effects that exceed the sum of individual risks, and that these interactions can be mitigated by greater greenness availability.

Method

Study data

This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative cohort of residents aged ≥ 45 years from approximately 150 county-level districts across 28 provinces in China [24]. CHARLS employs a multistage, stratified probability-proportional-to-size (PPS) sampling design to collect high-quality microdata on the health, economic, and social circumstances of middle-aged and older adults. The baseline survey was conducted in 2011, with follow-ups in 2013, 2015, 2018, and 2020. This analysis adopted a prospective design using data from 2011 to 2020 [25]. Although CHARLS is a dynamic cohort that enrolled limited new participants in later waves, we considered the 2011 wave as baseline and followed participants through 2020. From an initial sample of 25,586 individuals, we excluded those under age 45 and without diagnosed diabetes, resulting in a final analytical sample of 1,400 participants (Figure S1). Therefore, this final sample of 1,400 participants constitutes a subset of individuals with diagnosed diabetes that is drawn from and retains the representativeness of the original, nationally representative CHARLS cohort. The primary outcome was all-cause mortality. Mortality status was ascertained during each follow-up wave (2013, 2015, 2018, and 2020) through interviews with surviving family members or household informants. For deceased participants, the date of death was recorded based on informant reports. Participants were followed from the baseline interview (2011) until the date of death, loss to follow-up, or the end of the study period (December, 2020), whichever occurred first. Participants who remained alive throughout the study period or were lost to follow-up were censored at their last known interview date.

The CHARLS study protocol was approved by the Research Ethics Committee of Peking University (IRB00001052-11015). All participants provided written informed consent. To protect participant confidentiality, individual residential addresses are not publicly available, and geographic identifiers are provided at the city level. Therefore, environmental exposures were assigned at the city level using a six-digit geographic code. These geographic units were derived from the 2010 Population Census of China and community questionnaires [26]. All residents within the same city were assigned identical exposure values.

Air pollution and greenness exposure assessment

Environmental exposures to air pollution and greenness were assessed using high-resolution, contemporaneous datasets. Annual average concentrations of fine particulate matter (PM₂.₅) and nitrogen dioxide (NO₂) at 1 km × 1 km resolution were obtained from the China High Air Pollutants (CHAP) database (available at: https://weijing-rs.github.io/product.html) [27]. Based on existing evidence linking short-term air pollution exposure to adverse health outcomes, we evaluated multiple exposure windows, including 1-, 2-, and 3-year moving averages prior to the event of interest. Preliminary analyses (Table S1) indicated that the 1-year lagged exposure, defined as the mean concentration during the calendar year immediately preceding death, loss to follow-up, or the end of the 2020 study wave, demonstrated the strongest association with all-cause mortality. For exposure assignment, we first calculated annual mean PM₂.₅ and NO₂ concentrations for each city by averaging the 1 km × 1 km grid values within that city. These city-level annual means were then linked to participants using their city geographic codes. Both pollutants were specified as time-varying covariates in the analytical models, with exposure values updated annually.

Greenness exposure was assessed using the Normalized Difference Vegetation Index (NDVI). Annual NDVI data at a 500 m resolution were obtained from NASA’s Moderate-Resolution Imaging Spectroradiometer (MODIS) aboard the Terra satellite [28]. For each prefecture-level city, we calculated the annual mean NDVI by averaging the values of all 500 m grid cells within the city boundaries. Prior to this aggregation, grid cells with NDVI values less than 0, typically representing water bodies or ice/snow cover, were excluded to ensure that the greenness assessment focused solely on vegetated land surfaces. Based on the exposure window analysis, NDVI values were calculated as the 1-year average prior to death, loss to follow-up, or the end of follow-up. Accordingly, in the time-dependent Cox models, NDVI was specified as a time-varying covariate with values updated annually for each participant throughout follow-up. For analytical convenience, original NDVI values (ranging from − 1 to 1) were scaled by a factor of 10,000. This transformation addresses potential numerical precision issues associated with small decimal values and enhances the interpretability of regression coefficients by converting subtle fractional changes into meaningful integer units. Specifically, after scaling, a 1000-unit decrease corresponds to a 0.1 reduction in raw NDVI, allowing effect estimates to be presented as per 1000-unit decrements rather than small decimals, while preserving all underlying statistical relationships. Higher scaled values continue to indicate greater vegetation density.

Heatwave exposure assessment

Daily maximum temperature data (2008–2019) were obtained from over 2,400 meteorological stations across China, sourced from the Resource and Environment Science and Data Center (RESDC) of the Chinese Academy of Sciences [29, 30]. Heatwaves were defined using a relative threshold approach during the warm season (May 1 to September 30) to account for regional climatic adaptation [29, 31]. Specifically, we applied two temperature thresholds (95th and 97.5th percentiles of the local warm-season temperature distribution) across three minimum duration criteria (≥ 2, ≥ 3, or ≥ 4 consecutive days), generating six distinct heatwave definitions (HW1–HW6; Table 1). Individual heatwave exposure was quantified as the annual cumulative number of heatwave days. Our preliminary analysis identified the 1-year lagged exposure window as most strongly associated with mortality (Table S2). Heatwave exposure was modeled as a time-varying covariate, representing the total heatwave days in the calendar year preceding each event.

Table 1.

Definitions of heatwaves based on intensity and duration thresholds

Heatwave definitions	Threshold percentile (°C)	Duration (Day)
HW1 (95th-2D)	95th (36.9)	2
HW2 (95th-3D)	95th (36.9)	3
HW3 (95th-4D)	95th (36.9)	4
HW4 (97.5th-2D)	97.5th (37.8)	2
HW5 (97.5th-3D)	97.5th (37.8)	3
HW6 (97.5th-4D)	97.5th (37.8)	4

Abbreviation: HW Heatwave

A key methodological distinction is that heatwave metrics were constructed using a more spatially refined approach than air pollution or greenness, while remaining aligned with the city-level geographic data available from CHARLS. Specifically, we first calculated annual heatwave metrics at the county/district level based on six definitions (HW1–HW6), ensuring that heatwave identification captured locally defined extreme temperature events at the finest available spatial resolution. These county/district-level metrics were then aggregated to the city level by averaging across all counties/districts within each prefecture, deriving a city-level mean exposure metric that aligns with the geographic resolution of the CHARLS participant data. This averaging process provides an estimate of the average heatwave burden experienced by residents across the city, which is particularly relevant when linking to city-level health outcomes. Finally, participants were linked to these city-level heatwave metrics using their geographic codes. This two-step approach ensures that heatwave events are identified based on local temperature variations that might otherwise be masked by direct city-level aggregation, while the final aggregation aligns with the available participant data.

Covariates

Baseline covariates were selected a priori based on established epidemiological evidence and a directed acyclic graph (DAG, Figure S2) to control for potential confounding [32, 33]. The adjusted covariates encompassed sociodemographic characteristics (age, gender, educational attainment, marital status, residence type, geographic region), socioeconomic status (household total wealth), behavioral factors (smoking, alcohol consumption), anthropometric measures (body mass index), and comorbidity status (history of respiratory disease, hypertension, and cardiovascular or cerebrovascular disease).

Statistical analysis

Sample description

Descriptive statistics summarized baseline characteristics and environmental exposures. Continuous variables are presented as mean ± standard deviation or median with interquartile range (IQR), based on their distribution; categorical variables are expressed as frequencies and percentages. To address missing covariate data (approximately 8.7% of the sample; Table S3), we employed multiple imputation by chained equations (MICE) [34], generating five complete datasets. The imputation model achieved convergence, and the five imputed datasets showed consistent baseline characteristics. One of these datasets was randomly selected for all subsequent analyses. In survival models, time at risk (in days) was calculated from the date of enrollment until the occurrence of death, loss to follow-up, or the end of the study period (2020 wave), whichever came first.

Basic risk models

We employed a time-varying Cox proportional hazards model to assess the association between heatwave exposure and all-cause mortality in the diabetic cohort [35]. To address potential spatial autocorrelation, a random intercept for residential unit (city) was incorporated. Given that global Schoenfeld residual tests indicated violations of the proportional hazards assumption for residence type and educational attainment, indicating that the effects of these two variables on mortality risk varied over time, the models were stratified by these two variables. This approach accounts for their non-proportional hazards by allowing separate baseline hazard functions across strata, while still preserving the accuracy of effect estimates for the primary exposure of interest (e.g., heatwaves) [36]. Furthermore, residence type and educational attainment are not only key demographic characteristics but also factors closely associated with mortality risk in diabetic patients [37–39], underscoring the importance of appropriately accounting for their time-varying effects in the model specification. The model was specified as follows: h(t, X) = h₀(t)exp(β_ix_i + βₘxₘ(t)), where h₀(t) denotes the baseline hazard function, x_i represents time-invariant covariates (e.g., sex), and βₘxₘ(t) corresponds to time-varying exposures (e.g., heatwaves, PM₂.₅, and NO₂).

The analysis proceeded in sequential steps. First, we estimated the hazard ratio (HR) associated with each additional heatwave day, adjusting for baseline covariates. Next, PM₂.₅, NO₂, and greenness were incorporated as time-varying covariates to account for potential confounding. Finally, stratified analyses were conducted by age, gender, residence type, and educational attainment to examine potential effect modification. All results are presented as hazard ratios (HRs) with corresponding 95% confidence intervals (CIs).

Measures of multiplicative interaction

To evaluate multiplicative interactions between heatwaves, air pollution, and greenness on all-cause mortality in diabetic patients, we extended the Cox models by incorporating interaction terms between heatwave exposure (binary, coded as 0/1) and each environmental factor. Specifically, we modeled interactions with a 10 µg/m³ increase in PM₂.₅, a 10 µg/m³ increase in NO₂, and a 1000-unit decrease in greenness (based on the scaled NDVI × 10⁴). The interaction terms were constructed as the product of the binary heatwave indicator and the continuous environmental variables. Results from these interaction models are presented as hazard ratios (HRs) with 95% confidence intervals (CIs), indicating the change in mortality risk associated with each environmental exposure during heatwave periods relative to non-heatwave periods.

Measures of additive interaction

Additive interactions between heatwaves and environmental exposures were assessed using the relative excess risk due to interaction (RERI) to evaluate whether their joint effect exceeded the sum of their individual effects [40]. Heatwave exposure was dichotomized as experienced versus not experienced based on annual cumulative heatwave days. PM₂.₅, NO₂, and NDVI were dichotomized into binary variables using sample-specific medians as cut-off points: values below the median were classified as low, and values at or above the median as high. The RERI was calculated as RERI = HR₁₁ - HR₁₀ - HR₀₁ + 1, where subscripts denote exposure combinations (e.g., HR₁₁ represents the hazard ratio for joint high exposure to both heatwaves and an environmental factor). We further calculated two complementary measures of additive interaction: the attributable proportion due to interaction (AP) and the synergy index (S). AP was defined as AP = RERI / HR₁₁, representing the proportion of the risk in the doubly exposed group attributable to interaction, while S was defined as S = (HR₁₁ − 1) / [(HR₁₀ − 1) + (HR₀₁ − 1)], measuring the extent to which the joint effect exceeds the sum of individual effects. Confidence intervals for all three measures were estimated via the delta method. For RERI and AP, values > 0 indicate a synergistic interaction, values < 0 suggest antagonism, and a 95% confidence interval encompassing zero implies no significant additive interaction. For S, values > 1 indicate synergy, values < 1 suggest antagonism, and a 95% confidence interval encompassing 1 implies no significant additive interaction [41].

Sensitivity analyses

To evaluate the robustness of our primary findings, we conducted multiple sensitivity analyses: (a) adjusting for mean warm-season temperature and its average deviation to address meteorological confounding; (b) excluding participants lost to follow-up to assess potential selection bias; (c) employing heatwave event counts as an alternative exposure metric; (d) performing a complete-case analysis by removing observations with missing covariates; (e) excluding deaths occurring in 2020 to address the potential impact of the COVID-19 pandemic on all-cause mortality; and (f) replacing NDVI with EVI as an alternative greenness metric (with EVI also scaled by 10,000 for consistency with NDVI). All statistical analyses were conducted using R version 4.5.1, with statistical significance defined as a two-sided p-value < 0.05.

Results

Sample characteristics

Table 2 presents the baseline characteristics and environmental exposures of the analytical cohort. The study followed 1,400 individuals over 10,187 person-years, during which 214 deaths were recorded, yielding an incidence rate of 2.1 per 100 person-years. The participants had a mean age of 60.5 years (SD = 9.2) and were predominantly female (56.1%), married (87.8%), and urban residents (53.8%). Over half of the participants reported low household wealth (57.7%) and limited formal education (64.1%). The geographic distribution of participants is illustrated in Figure S3. Environmental exposures, presented as median (IQR) values, were as follows: PM₂.₅ 38.7 (29.5–52.5) µg/m³, NO₂ 26.1 (17.8–35.7) µg/m³, and NDVI 2931.3 (2280.4–3570.3) (based on scaled values ×10⁴). Notably, decedents had experienced significantly more cumulative heatwave days than survivors (p < 0.05).

Table 2.

Baseline characteristics and environmental exposures of the study population

Variable	Overall	Survivor	Death	P-value
	(n = 1400)	(n = 1186)	(n = 214)
Age (years), Mean ± SD	60.5 ± 9.2	59.3 ± 8.5	67.0 ± 9.8	< 0.001
Age group, n (%)
< 60	683 (48.8)	638 (53.8)	45 (21.0)	< 0.001
>=60	717 (51.2)	548 (46.2)	169 (79.0)
Gender, n (%)
Female	785 (56.1)	674 (56.8)	111 (51.9)	0.204
Male	615 (43.9)	512 (43.2)	103 (48.1)
Education attainment, n (%)
Middle school and above	502 (35.9)	451 (38.1)	51 (23.8)	< 0.001
Primary school and below	896 (64.1)	733 (61.9)	163 (76.2)
Marital status, n (%)
Married	1229 (87.8)	1060 (89.5)	169 (79.0)	< 0.001
Single	170 (12.2)	125 (10.5)	45 (21.0)
Residence, n (%)
City/town	753 (53.8)	648 (54.6)	105 (49.1)	0.153
Village	647 (46.2)	538 (45.4)	109 (50.9)
Geographic region, n(%)
East China	522 (37.3)	450 (37.9)	72 (33.6)	0.317
Midland China	539 (38.5)	447 (37.7)	92 (43.0)
West China	339 (24.2)	289 (24.4)	50 (23.4)
Household total wealth, Median (IQR)	65225.0 (15711.2, 186875.0)	73490.0 (17300.0, 202675.0)	32890.0 (9652.5, 124815.2)	0.001
Wealth group, n (%)
High Wealth (> 30,0000)	156 (15.5)	144 (17.1)	12 (7.5)	0.006
Low Wealth (< 10,0000)	579 (57.7)	473 (56.0)	106 (66.2)
Medium Wealth (10,0000–30,0000)	269 (26.8)	227 (26.9)	42 (26.2)
Smoking, n (%)
No	901 (64.6)	771 (65.3)	130 (61.0)	0.264
Yes	493 (35.4)	410 (34.7)	83 (39.0)
Drinking, n (%)
No	884 (63.5)	744 (63.1)	140 (65.7)	0.503
Yes	509 (36.5)	436 (36.9)	73 (34.3)
BMI, n (%)
Normal or below(< 24.9)	533 (49.4)	435 (47.4)	98 (60.9)	0.005
Obesity(25-29.9)	103 (9.6)	94 (10.3)	9 (5.6)
Overweight( > = 30)	442 (41.0)	388 (42.3)	54 (33.5)
Comorbidity, n (%)
No	352 (27.4)	317 (29.3)	35 (17.4)	0.001
Yes	931 (72.6)	765 (70.7)	166 (82.6)
Air pollution (µg/m3), Median (IQR)
PM2.5	38.7(29.5,52.5)	36.0(28.2,49.8)	55.4(41.4,70.5)	< 0.001
NO2	26.1 (17.8, 35.7)	25.1 (16.7, 33.4)	29.2 (22.1, 37.7)	< 0.001
Greenness(NDVI ×10⁴), Median (IQR)
NDVI	2931.3(2280.4, 3570.3)	2951.9(2333.8, 3635.2)	2715.3(2115.5, 3430.6)	< 0.001
Heatwave exposures (days), Median (IQR)
HW1(95 h-2D)	46.0 (29.0, 61.0)	43.0 (28.0, 59.0)	55.0 (37.2, 71.5)	< 0.001
HW2(95 h-3D)	18.0 (3.0, 37.0)	15.0 (3.0, 37.0)	28.0 (14.2, 49.8)	< 0.001
HW3(95th-4D)	0.0 (0.0, 12.0)	0.0 (0.0, 12.0)	4.0 (0.0, 23.8)	< 0.001
HW4(97.5 h-2D)	18.0 (8.0, 26.0)	17.0 (6.0, 25.0)	20.0 (12.0, 30.0)	< 0.001
HW5(97.5 h-3D)	0.0 (0.0, 6.0)	0.0 (0.0, 3.0)	0.0 (0.0, 15.0)	< 0.001
HW6(97.5th-4D)	0.0 (0.0, 0.0)	0.0 (0.0, 0.0)	0.0 (0.0, 0.0)	0.075
Mean temperature in the warm season (◦C), Median (IQR)	24.5 (21.9, 25.6)	24.6 (21.9, 25.6)	24.2 (21.7, 25.3)	0.180
Mean temperature deviation in the warm season (◦C), Median (IQR)	3.6 (3.3, 3.8)	3.6 (3.3, 3.8)	3.6 (3.2, 4.0)	0.754

Abbreviation: PM_2.5 Fine particulate matter with aerodynamic diameter ≤ 2.5 μm, NO₂ Nitrogen dioxide, NDVI Normalized difference vegetation index, IQR Interquartile range, SD Standard deviation, BMI Body mass index, HW Heatwave

Association of heatwaves with all-cause mortality

Table 3 presents the associations between various heatwave definitions and all-cause mortality among diabetic individuals. Analyses revealed a consistent exposure-response relationship, with progressively higher hazard ratios corresponding to greater heatwave intensity and duration. All associations remained statistically significant across sequential adjustment levels. In the fully adjusted model (Model 6), which controlled for sociodemographic factors, lifestyle behaviors, clinical comorbidities, and environmental co-exposures, heatwaves defined as temperatures exceeding the 95th percentile for ≥ 4 consecutive days (HW3) were associated with a 2.7% increase in mortality risk (HR = 1.027, 95% CI: 1.013–1.041). Risk estimates intensified under more stringent definitions using the 97.5th percentile threshold, with HRs reaching 1.052 (95% CI: 1.033–1.072) for HW5 (≥ 3 days) and 1.070 (95% CI: 1.048–1.093) for HW6 (≥ 4 days). The persistence of these graded associations after comprehensive adjustment is consistent with an independent relationship between prolonged, intense heatwaves and elevated mortality risk in this vulnerable population.

Table 3.

Hazard ratios for all-cause mortality associated with heatwave exposure in patients with diabetes mellitus

HW definitions	Model 1			Model 2			Model 3			Model 4			Model 5			Model 6
	HR	95%CI		HR	95%CI		HR	95%CI		HR	95%CI		HR	95%CI		HR	95%CI
HW1 (95th-2D)	1.003	0.995	1.012	1.003	0.995	1.012	1.000	0.992	1.009	1.002	0.994	1.011	1.004	0.996	1.013	1.001	0.992	1.010
HW2 (95th-3D)	1.008	0.997	1.020	1.008	0.997	1.020	1.004	0.993	1.016	1.007	0.995	1.018	1.008	0.996	1.020	1.004	0.993	1.016
HW3(95th-4D)	1.028	1.014	1.042	1.028	1.014	1.041	1.026	1.012	1.039	1.026	1.013	1.040	1.029	1.016	1.043	1.027	1.013	1.041
HW4 (97.5th-2D)	1.021	1.003	1.038	1.020	1.003	1.038	1.015	0.997	1.033	1.018	1.001	1.036	1.022	1.004	1.039	1.018	1.000	1.037
HW5 (97.5th-3D)	1.056	1.037	1.077	1.055	1.035	1.075	1.051	1.032	1.070	1.051	1.032	1.071	1.054	1.034	1.075	1.052	1.033	1.072
HW6 (97.5th-4D)	1.076	1.054	1.099	1.074	1.051	1.096	1.067	1.045	1.087	1.069	1.046	1.092	1.075	1.052	1.098	1.070	1.048	1.093

Bold values indicate statistical significance. Hazard ratios represent the increased mortality risk associated with each additional day of heatwave exposure under the corresponding definition. All models were adjusted for the following covariates:

Model 1: Age, gender, marital status, education, wealth, residence, and geographic regions

Model 2: Model 1 + BMI, smoking, drinking, and comorbidity

Model 3: Model 2 + PM₂.₅

Model 4: Model 2 + NO₂

Model 5: Model 2 + NDVI

Model 6: Model 2 + PM₂.₅, NO₂, and NDVI

Abbreviation: HR Hazard ratio, CI Confidence interval, HW Heatwave

Subgroup analysis

Subgroup analyses assessed potential effect modification by age, gender, educational attainment, and residence type (Fig. 1 and Table S4). The association between heatwave exposure and mortality was generally consistent across most subgroups. However, significant effect modification was observed for age and residence type (p for effect modification < 0.05): adults aged ≥ 60 years and urban residents exhibited significantly elevated mortality risks compared to their younger and rural counterparts, respectively. In contrast, no significant heterogeneity was detected across strata of gender or educational attainment, indicating a consistent heatwave-mortality association regardless of these characteristics.

Fig. 1.

Subgroup analysis of the association between heatwave exposure and all-cause mortality by age, gender, educational attainment, and residence type. All models were adjusted for the following covariates: Model 1: Age, gender, marital status, education, wealth, residence, and geographic regions. Model 2: Model 1 + BMI, smoking, drinking, and comorbidity. Model 3: Model 2 + PM₂.₅. Model 4: Model 2 + NO₂. Model 5: Model 2 + NDVI. Model 6: Model 2 + PM₂.₅, NO₂, and NDVI. PM_2.5, fine particulate matter with aerodynamic diameter ≤ 2.5 μm; NO₂,nitrogen dioxide; NDVI, normalized difference vegetation index; HW, heatwave; HR, hazard ratio; CI: confidence interval

Assessment of multiplicative interaction

We evaluated multiplicative interactions between heatwaves and environmental exposures on mortality risk. As shown in Table 4, the hazard ratios for all interaction terms were consistently close to 1.00. Specifically, the HRs for heatwave–PM₂.₅ interactions ranged from 0.999 to 1.001, for heatwave–NO₂ from 0.998 to 1.001, and for heatwave–NDVI remained at 1.000. Although the 95% confidence intervals for a limited number of interaction terms excluded 1.00, indicating statistical significance, their point estimates were negligible in magnitude and without clinically or epidemiologically meaningful effects. Collectively, these results provide no evidence of a meaningful multiplicative interaction on the relative risk scale.

Table 4.

Multiplicative interactions between heatwaves, PM₂.₅, NO₂, and greenness on all-cause mortality in patients with diabetes mellitus

HW definitions	Multiplicative interaction a			Multiplicative interaction b			Multiplicative interaction c
	HR	95%CI		HR	95%CI		HR	95%CI
HW1 (95th-2D)	0.999937	0.999740	1.000134	1.000639	1.000216	1.001063	0.999996	0.999991	1.000001
HW2 (95th-3D)	0.999716	0.999478	0.999953	1.000275	0.999770	1.000781	1.000002	0.999996	1.00008
HW3 (95th-4D)	0.999587	0.999287	0.999887	1.000283	0.999569	1.000997	1.000008	1.000001	1.000015
HW4 (97.5th-2D)	1.000059	0.999625	1.000493	1.001374	1.000537	1.002211	0.999990	0.999981	0.999999
HW5 (97.5th-3D)	0.999144	0.998496	0.999792	0.998782	0.997478	1.000088	1.000017	1.000004	1.000030
HW6 (97.5th-4D)	1.001199	0.999823	1.002576	0.998394	0.994010	1.002797	1.000066	1.000041	1.000091

Bold values indicate statistical significance. The model was adjusted for age, gender, marital status, education, wealth, residence, geographic regions, BMI, smoking, drinking, and comorbidity

Abbreviation: PM2.5 Fine particulate matter with aerodynamic diameter ≤ 2.5 μm, NO2 Nitrogen dioxide, NDVI Normalized difference vegetation index, HW Heatwave, HR Hazard ratio, CI Confidence interval

^aEach 10 µg/m3 increase in PM_2.5 and increase in heatwave exposure

^bEach 10 µg/m3 increase in NO₂ and increase in heatwave exposure

^cEach 1000 decrease in greenness exposure(NDVI× 10⁴) and increase in heatwave exposure

Assessment of additive interaction

We assessed additive interactions between heatwaves, air pollution, and greenness using the relative excess risk due to interaction (RERI), complemented by the attributable proportion due to interaction (AP) and the synergy index (S) (Table 5). Fully adjusted models revealed several significant synergistic effects. For PM₂.₅, a pronounced interaction was observed between HW5 (97.5th percentile, ≥ 3 days) and high exposure (RERI = 4.40, 95% CI: 0.25–8.54), with consistent support from AP (0.31, 95% CI: 0.10–0.52) and S (1.50, 95% CI: 1.06–2.11), indicating that 31% of the mortality risk in the doubly exposed group was attributable to the interaction. For NO₂, HW3 (95th percentile, ≥ 4 days) combined with high exposure showed a significant positive interaction (RERI = 1.78, 95% CI: 0.51–3.06), with AP (0.44, 95% CI: 0.22–0.66) and S (2.39, 95% CI: 1.17–4.88) providing consistent evidence; 44% of the excess risk in the doubly exposed group was attributable to the interaction. For greenness (NDVI), significant interactions were identified between heatwaves and low greenness for HW2 (95th percentile, ≥ 3 days; RERI = 0.64, 95% CI: 0.23–1.04) and HW6 (97.5th percentile, ≥ 4 days; RERI = 2.54, 95% CI: 0.83–4.25). For HW6, the AP (1.00, 95% CI: 0.80–1.20) indicated that nearly all of the excess mortality risk in the doubly exposed group was attributable to the interaction. Collectively, these findings demonstrate that combined exposure to heatwaves and high air pollution or low greenness produces mortality risks exceeding the sum of their individual effects, with AP and S results largely consistent with the RERI-based conclusions.

Table 5.

Additive interactions between heatwaves, PM₂.₅, NO₂, and greenness on all-cause mortality in patients with diabetes mellitus

HWdefinitions	Category	PM2.5		NO2		NDVI
		HR	95%CI	HR	95%CI	HR	95%CI
HW1 (95th-2D)
No	Low level	[Ref.]	-	[Ref.]	-	[Ref.]	-
Yes	Low level	-	-	-	-	-	-
No	High level	-	-	-	-	-	-
Yes	High level	-	-	-	-	-	-
RERI		-	-	-	-	-	-
AP		-	-	-	-	-	-
S		-	-	-	-	-	-
HW2 (95th-3D)
No	Low level	[Ref.]	-	[Ref.]	-	[Ref.]	-
Yes	Low level	-	-	10.84	1.50-78.24	1.04	0.54–2.02
No	High level	-	-	10.82	1.40-83.43	0.21	0.07–0.61
Yes	High level	-	-	19.59	2.70-142.25	0.89	0.45–1.73
RERI		-	-	-1.07	-9.34-7.20	0.64	0.23–1.04
AP		-	-	-0.05	-0.44-0.33	0.72	-0.06-1.50
S		-	-	0.95	0.64–1.39	NA	-
HW3 (95th-4D)
No	Low level	[Ref.]		[Ref.]		[Ref.]
Yes	Low level	4.08	2.02–8.24	1.64	1.06–2.53	1.66	1.15–2.39
No	High level	11.76	6.09–22.71	1.65	1.05–2.58	0.48	0.31–0.76
Yes	High level	20.17	10.43–38.97	4.07	2.60–6.37	1.30	0.91–1.85
RERI		5.33	-0.83-11.49	1.78	0.51–3.06	0.16	-0.47-0.79
AP		0.26	0.04–0.49	0.44	0.22–0.66	0.12	-0.36-0.60
S		1.39	1.00-1.92	2.39	1.17–4.88	2.15	0.02-202.39
HW4 (97.5th-2D)
No	Low level	[Ref.]	-	[Ref.]	-	[Ref.]	-
Yes	Low level	2.75	0.38–20.12	2.41	0.58–9.92	6.06	2.45–14.98
No	High level	3.04	0.37–24.87	0.74	0.15–3.72	1.09	0.26–4.58
Yes	High level	19.85	2.77–142.00	5.03	1.24–20.46	4.06	1.62–10.18
RERI		15.06	-12.81-42.93	2.88	-0.01-5.76	-2.09	-5.38-1.21
AP		0.76	0.60–0.92	0.57	0.24–0.90	-0.51	-1.08-0.05
S		4.98	1.58–15.67	3.49	0.39–31.36	0.59	0.38–0.92
HW5 (97.5th-3D)
No	Low level	[Ref.]	-	[Ref.]	-	[Ref.]
Yes	Low level	3.05	1.60–5.80	3.07	1.98–4.77	2.45	1.71–3.51
No	High level	7.75	4.48–13.41	2.21	1.44–3.39	0.65	0.44–0.97
Yes	High level	14.20	8.28–24.36	4.43	2.85–6.88	1.65	1.12–2.42
RERI		4.40	0.25–8.54	0.15	-1.46-1.77	-0.46	-1.37-0.45
AP		0.31	0.10–0.52	0.03	-0.32-0.39	-0.28	-0.86-0.30
S		1.50	1.06–2.11	1.05	0.65–1.70	0.59	0.22–1.55
HW6 (97.5th-4D)
No	Low level	[Ref.]	-	[Ref.]	-	[Ref.]	-
Yes	Low level	2.02	0.71–5.79	2.56	1.23–5.33	0.39	0.12–1.23
No	High level	7.01	4.67–10.52	1.91	1.39–2.63	0.61	0.46–0.82
Yes	High level	6.78	3.36–13.69	1.47	0.63–3.46	2.54	1.31–4.91
RERI		-1.25	-5.99- 3.48	-2.00	-4.29-0.29	2.54	0.83–4.25
AP		-0.18	-0.98-0.61	-1.36	-3.69-0.97	1.00	0.80–1.20
S		0.82	0.38–1.79	0.19	0.01–2.86	NA	-

Bold values indicate statistical significance. “-” denotes that the estimate could not be estimated due to the presence of zero death events in the specific exposure combination. “NA” indicates that the synergy index (SI) could not be calculated because the individual exposure effects were protective (OR₁₀ and OR₀₁ < 1), resulting in an undefined denominator in the SI formula. The model was adjusted for age, gender, marital status, education, wealth, residence, geographic regions, BMI, smoking, drinking, and comorbidity

Abbreviation: PM_2.5 Fine particulate matter with aerodynamic diameter ≤ 2.5 μm, NO₂ Nitrogen dioxide, NDVI Normalized difference vegetation index, HW Heatwave, HR Hazard ratio, CI Confidence interval, RERI Relative excess risk due to interaction, AP Attributable proportion, SI Synergy index

Sensitivity analysis

Sensitivity analyses demonstrated the robustness of the primary findings. The association between heatwave exposure and all-cause mortality remained statistically significant for the more intense and prolonged definitions (specifically HW3, HW5, and HW6) across all alternative model specifications (Table S5). Furthermore, synergistic interactions between heatwaves and co-exposures to PM₂.₅, NO₂, and low greenness were consistently observed across multiple sensitivity scenarios, including adjustment for temperature metrics, exclusion of participants lost to follow-up, use of heatwave event counts as an alternative exposure metric, complete-case analysis, exclusion of deaths occurring in 2020 to address the potential impact of the COVID-19 pandemic, and replacing NDVI with EVI as an alternative greenness metric. The close agreement between these sensitivity-derived estimates and those from the primary models (Table S6-S11) strengthens the reliability of the reported interaction effects.

Discussion

This prospective, nationally representative cohort study provides new evidence on how multiple environmental exposures synergistically affect mortality risk in Chinese older adults with diabetes. We identified a robust exposure-response relationship, wherein all-cause mortality risk increased progressively with heatwave intensity and duration. Notably, additive interaction analyses demonstrated that heatwave-related mortality was synergistically exacerbated by co-exposure to elevated PM₂.₅ and NO₂, while greater greenness availability exerted a mitigating effect. These associations remained consistent across extensive sensitivity analyses, underscoring their robustness. Subgroup analyses further identified older adults (≥ 60 years) and urban residents as particularly vulnerable subgroups.

While previous time-series studies have established the acute mortality risk of short-term heatwave events [42, 43], our cohort study provides complementary evidence on the chronic health burden of cumulative heat exposure over an entire warm season. For vulnerable populations such as older adults with diabetes, the physiological strain from repeated or sustained heat may accumulate, leading to a progressive deterioration of health that ultimately increases mortality risk. This finding has distinct environmental implications: it suggests that public health strategies should extend beyond acute heatwave warnings to include long-term adaptation measures, such as increasing urban green coverage and improving housing thermal comfort, aimed at reducing cumulative heat exposure throughout the summer months. Furthermore, the synergistic effects observed with PM₂.₅ and NO₂ underscore the need for integrated policies that concurrently address air pollution and heat mitigation as part of chronic disease prevention.

Building on this evidence of cumulative risk, our study further provides robust evidence of a clear exposure-response relationship between heatwave intensity/duration and mortality risk in a diabetic population, thereby confirming and extending the conclusions of prior meta-analyses on heat-related morbidity and mortality [7]. The underlying physiological mechanisms for this vulnerability, while not fully elucidated, likely involve several diabetes-specific pathways: autonomic neuropathy impairs thermoregulation, increasing susceptibility to heat stress; elevated temperatures disrupt glucose homeostasis, leading to blood glucose dysregulation; and sustained heat exposure promotes dehydration, potentially precipitating acute complications such as diabetic ketoacidosis [2]. A key methodological advancement of our study lies in the application of multiple, precise heatwave definitions, which enabled us to identify particularly hazardous intensity-duration combinations. Specifically, the most intense heatwave conditions (HW5: 97.5th percentile ≥ 3 days; HW6: 97.5th percentile ≥ 4 days) may exacerbate autonomic neuropathy and glucose dysregulation more severely than milder heat events. This finding underscores the synergistic role of these two characteristics and corroborates existing domestic evidence [8]. The substantial health impact is further evidenced by studies linking heat exposure to increased hospitalizations and emergency visits among older diabetics [44]. Therefore, integrating these specific intensity and duration thresholds into public heat-health warning systems and developing targeted interventions are critical steps for mitigating this significant and growing disease burden.

Subgroup analyses revealed markedly elevated vulnerability among two key demographics: older adults (≥ 60 years) and urban residents. For older individuals, this susceptibility stems from a synergistic interplay between age-related physiological decline and diabetes-specific pathologies [2, 45]. Underlying thermoregulatory deficits, including diminished vasodilation, reduced sweating capacity, and impaired cardiac output, severely compromise heat dissipation. These vulnerabilities are frequently compounded by diabetes-related metabolic dysregulation and prevalent comorbidities such as cardiovascular disease, creating a pathophysiological profile of exceptional risk [46]. Meanwhile, urban vulnerability was primarily driven by the urban heat island effect, where impervious surfaces and sparse vegetation elevate ambient temperatures through increased heat absorption and diminished evaporative cooling [17]. This microclimatic amplification disproportionately intensifies heat exposure in urban areas. To effectively address these dual challenges, integrated public health strategies must be implemented. Urban planning interventions should prioritize green infrastructure and high-albedo materials to mitigate ambient heat, while parallel health policies must enhance early warning systems and ensure accessible cooling resources for at-risk communities during extreme heat events.

A critical contribution of this study lies in its distinction between multiplicative and additive interaction scales. While no significant multiplicative interactions were detected, suggesting that the combined effect of heatwaves with air pollution or reduced greenness does not exceed the product of their individual effects on a relative scale, additive interaction analysis revealed a different and more critical public health picture [47]. Significantly positive RERI values indicated that the absolute mortality risk under co-exposure to heatwaves and PM₂.₅, NO₂, or low greenness exceeded the sum of their individual effects, a finding further supported by the attributable proportion (AP) and synergy index (S), which were largely consistent with the RERI-based conclusions. This pattern of synergistic effect on an additive scale is consistent with earlier reports linking heat and PM₂.₅ to exacerbated diabetes burden, preterm birth, and hypertension [20, 48, 49]. The biological plausibility of this synergy is supported by multifactorial mechanisms: co-exposure to heat stress and air pollutants can amplify systemic inflammation and oxidative stress, thereby accelerating cellular damage [50]. Furthermore, this interaction promotes endothelial dysfunction and atherosclerotic instability [48, 51], compounding cardiovascular risk, while pre-existing cardiopulmonary compromise from chronic air pollution exposure diminishes the physiological reserve necessary to withstand additional heat stress. These pathways collectively erode the body’s compensatory capacity, dramatically increasing health risks during concurrent environmental extremes [52].

Our study provides critical evidence that co-exposure to NO₂ and extreme heat significantly increases all-cause mortality among diabetic patients, addressing an important gap in the environmental health literature. This finding is corroborated by converging lines of evidence from both epidemiological and toxicological studies. Population-based research has demonstrated that such co-exposure is associated with elevated risks of cardiovascular and cerebrovascular mortality [53], alongside increased emergency department visits [54, 55]. Complementing these findings, animal models reveal that combined heat and NO₂ exposure can accelerate diabetic nephropathy progression and exacerbate respiratory injury [16, 56]. The underlying mechanisms likely parallel those of PM₂.₅, primarily involving the amplification of systemic inflammation and oxidative stress [57], which collectively impair cardiovascular and respiratory function [53, 57]. Given the intersecting challenges of climate change and air pollution, targeted public health interventions are imperative to mitigate this compounded environmental health threat.

Our findings confirm that reduced greenness amplifies mortality risk during heatwaves in diabetic individuals. This conclusion is strengthened by broader evidence linking the combination of low greenness and heat exposure to an increased incidence of other heat-aggravated conditions, including hypertension, cognitive decline, and preterm birth [20, 30, 48]. Beyond interaction effects, greenness also serves as an independent protective factor, consistent with studies showing a lower incidence of diabetes and improved mental health in older adults with higher exposure [21, 23]. The mechanisms underlying these benefits are multifactorial. Environmentally, vegetation ameliorates microclimatic conditions through pollutant absorption and transpirational cooling. Physiologically, these improvements attenuate systemic inflammation and oxidative stress, thereby supporting cardiopulmonary function [48]. Psychologically, greenness promotes mental restoration and emotional regulation, reducing risks of depression and anxiety. Collectively, this evidence underscores that strategically expanding urban green infrastructure represents a co-beneficial strategy that simultaneously mitigates heat exposure, improves mental well-being, and reduces mortality, offering crucial protection for vulnerable populations such as diabetic patients during extreme heat events.

This study possesses several methodological strengths, including a large, nationally representative cohort that mitigates biases common in smaller or cross-sectional studies. The application of multiple heatwave definitions enabled a nuanced characterization of population responses to extreme heat, effectively capturing local adaptation to events of varying severity. A principal methodological contribution is our integrated assessment of the combined effects of heatwaves, air pollution, and greenness on mortality in diabetic patients, an approach that addresses a significant gap in the existing literature. Notwithstanding these strengths, several limitations warrant consideration. First, exposure assessment was based on participants’ residential locations at the city level due to geographic confidentiality constraints of the CHARLS dataset. This approach does not capture individual mobility or intra-city variation in greenness, which may lead to non-differential exposure misclassification. Second, mortality ascertainment relied on follow-up interviews without precise dates of death, potentially affecting the temporal alignment between exposures and outcomes. Third, although we adjusted for numerous confounders, residual confounding from unmeasured factors, such as specific medication use and quality of glycemic control, cannot be ruled out. Fourth, although our sample was drawn from the nationally representative CHARLS cohort, the application of our specific inclusion criteria (age ≥ 45 years and diagnosed diabetes) resulted in a final analytical sample of 1,400 participants. While this sample is representative of community-dwelling, middle-aged and older adults with diagnosed diabetes in China, caution is warranted when extrapolating these findings to younger diabetic populations, those with undiagnosed diabetes, or institutionalized older adults. Finally, the generalizability of our findings to populations outside of China, with differing climatic and socioeconomic contexts, requires further investigation. Future studies should integrate higher-resolution, activity-aware exposure assessments with detailed individual-level clinical data to better elucidate the complex interplay between environmental factors and mortality risk in diabetic populations.

Conclusions

This study identified heatwave exposure as a significant risk factor for all-cause mortality in diabetic patients, with older adults and urban residents showing heightened vulnerability. Moreover, we demonstrated synergistic effects on an additive scale between heatwaves and co-exposure to air pollution and low greenness. This collective evidence underscores the need for integrated public health strategies that simultaneously address extreme heat, air quality, and urban greening. Future research should prioritize further validation of these associations across diverse populations and elucidation of the underlying biological mechanisms to support evidence-based public health interventions.

Abbreviations

BMI

Body mass index

CHAP

China High Air Pollutants database

CHARLS

China Health and Retirement Longitudinal Study

CIs

Confidence intervals

DAG

Directed acyclic graph

DM

Diabetes mellitus

EVI

Enhanced vegetation index

HRs

Hazard ratios

HW

Heatwave

IDF

International Diabetes Federation

IQR

Interquartile range

MICE

Multiple imputation by chained equations

MODIS

Moderate-Resolution Imaging Spectroradiometer

NDVI

Normalized Difference Vegetation Index

NIA

National Institute on Aging

NO₂

Nitrogen dioxide

PM2.5

Fine particulate matter with aerodynamic diameter ≤ 2.5 μm

PPS

Probability-proportional-to-size

RERI

Relative excess risk due to interaction

RESDC

Resource and Environment Science and Data Center

SD

Standard deviation

Authors’ contributions

**Jie Gao: ** Conceptualization, Methodology, Data curation and analysis, Writing-original draft. **Pei Sun: ** Methodology, Software. **Mei Huang: ** Methodology, Validation. **Shi-yang Chen: ** Data Curation, Writing-original draft. **Xiao Zhang: ** Methodology. **Xiao-peng Yan: ** Methodology. **Xiao Liang: ** Methodology. **Xin Zhang: ** Methodology. **Chao Zhang: ** Conceptualization, Technical support, Supervision. **Chun-ping Ni: ** Project administration, Supervision, Funding acquisition.

Funding

This study received financial support from the “Clinical Medicine + X” Research Center at the Air Force Medical University (Grant No. LHJJ24HL02) and the National Natural Science Foundation of China (Grant No. 72574232). The funding bodies had no involvement in the research’s development.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The CHARLS study protocol was approved by the Research Ethics Committee of Peking University (IRB00001052-11015). All participants provided written informed consent. All procedures performed in this study involving human participants were in accordance with the ethical standards of the national research committee and with the 1975 Helsinki Declaration and its later amendments.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.