Heat stress exposure and incidence of chronic kidney disease

Summary

Long-term repeated heat stress may contribute to chronic kidney disease (CKD). In this ecological study, we examined country-level associations between heat stress exposure and CKD incidence using data from 174 countries during 1990–2021. Annual numbers of days with four heat stress levels (moderate, strong, very strong, and extreme) were assessed using the Universal Thermal Climate Index. A weighted heat stress score was generated from these four levels. Linear mixed-effects models were used to evaluate the associations. Each additional day of moderate, strong, very strong, and extreme heat stress was associated with increases of 0.088 (95% confidence interval [CI]: 0.035–0.140), 0.095 (0.047–0.142), 0.097 (0.036–0.159), and 0.782 (0.643–0.921) incident cases per 100,000 population, respectively. The weighted heat stress score was consistently associated with an increased incidence of CKD. These findings may imply that attention should be paid to the potential disease burden of heat-related CKD.

Graphical abstract

Highlights

• •Four heat stress levels were positively associated with CKD incidence
• •The weighted heat stress score was consistently associated with higher CKD incidence
• •Associations were stronger among female populations and older adults
• •Heat-related CKD associations were stronger at higher levels of NO₂ and O₃ exposure

Biomedical discipline; chemistry; environmental science; molecular biology; neuroscience; cell biology

Introduction

Under ongoing climate warming, the frequency and intensity of hot conditions are increasing worldwide.¹ Heat exposure imposes a substantial public health burden, including excess morbidity and mortality.^2,3 Chronic kidney disease (CKD) is a progressive disorder defined by structural or functional renal abnormalities persisting for ≥3 months.⁴ While diabetes and hypertension are the principal causes, in many low- and middle-income settings, glomerulonephritis and CKD of unknown etiology (CKDu) are more prevalent.^5,6 CKDu is a major concern among agricultural workers in tropical-subtropical regions.⁷ Epidemiological evidence links heat exposure to kidney-related outpatient visits and hospital admissions.^8,9,10 CKDu in agricultural populations is likewise thought to be related to repeated heat stress; however, causal pathways remain under investigation.^11,12 More direct evidence comes from a post hoc analysis of randomized clinical trials: across 21 countries, higher environmental heat exposure indices were associated with faster declines in estimated glomerular filtration rate (eGFR) among patients with CKD.¹³ Given the increasing frequency of hot days and heat waves, understanding the relation between heat stress and CKD has become increasingly important.

The Universal Thermal Climate Index (UTCI) is a heat exposure metric. Its central premise is to represent the integrated thermal strain of real environments as an equivalent temperature for any combination of air temperature, wind, radiation, and humidity. UTCI is defined as the air temperature in a standardized outdoor reference condition that elicits the same response from a human thermophysiological model as that of the actual environment.¹⁴ UTCI couples a multi-node thermophysiological model (UTCI-Fiala) with an advanced clothing model to translate meteorological inputs into physiological responses.¹⁴ Comparative assessments indicate that UTCI outperforms legacy heat indices in characterizing biothermal conditions.^15,16

To date, large-scale epidemiological investigations of the association between heat stress and CKD incidence are limited. Existing evidence has mainly come from regional studies (e.g., CKDu endemic regions) or from high-risk populations (e.g., tropical agricultural workers). Thus, the association between heat stress and CKD in the general population remains insufficiently understood. Previous studies have primarily examined the short-term effects of temperature fluctuations on kidney-related health outcomes (e.g., emergency department visits), whereas research on the long-term association between heat stress and CKD remains scarce. Moreover, human heat exposure occurs across a range of stress intensities, yet many existing studies have focused primarily on extreme heat events, such as heatwaves, and few have simultaneously examined multiple heat stress levels in relation to kidney disease. In light of these gaps, we linked UTCI data with CKD incidence estimates for 174 countries/regions from 1990 to 2021. We examined four UTCI-based heat stress categories (i.e., “moderate,” “strong,” “very strong,” and “extreme”), as well as a heat stress score, to characterize country/region-level heat exposure and its association with CKD incidence. We hypothesized that greater heat stress would be associated with higher CKD incidence.

Results

Study countries/regions characteristics

From 1990 to 2021, the global mean (SD) estimates of moderate heat stress, strong heat stress, very strong heat stress, and extreme heat stress were 80.7 (63.0) days, 108.9 (88.9) days, 43.4 (55.7) days, and 6.8 (20.8) days, respectively (online Table S6). The global mean (SD) estimates of CKD incidence and CKD caused by type 1 diabetes mellitus, type 2 diabetes mellitus, glomerulonephritis, hypertension, and other/unspecified causes per 100,000 population, were 211.5 (85.8), 1.3 (0.3), 21.1 (9.4), 4.8 (1.7), 13.1 (5.6), and 171.2 (69.7), respectively (online Table S7).

Heat stress days occurred in a latitudinal gradient: tropical and subtropical regions (e.g., Central America, sub-Saharan Africa, Southeast Asia) experienced more heat stress days per year, whereas temperate or high-latitude countries had fewer heat stress days (Figure 1A). With regard to CKD incidence, high-rate areas were clustered in Central America, North Africa and the Middle East, and Southeast Asia (Figure 1A). Overall, a greater number of heat stress days coincided with higher CKD incidence, suggesting similar geographic clustering.

Figure 1.

Distribution and trends of heat stress and CKD incidence from 1990 to 2021

The term “heat stress” includes moderate heat stress, strong heat stress, very strong heat stress, and extreme heat stress.

(A) The global map shows the local average number of heat stress days for 1990–2021, divided into five intervals. The color-coded dots represent the intensity of the local mean incidence rate for each country over the 1990–2021 period.

(B) The change in the number of heat stress days from 1990 to 2021.

(C) The change in the global CKD incidence from 1990 to 2021.

CKD, chronic kidney disease; LOESS, locally estimated scatterplot smoothing.

Figures 1B and 1C further illustrate upward trends in heat stress days and CKD incidence. From 1990 to 2021, the annual number of heat stress days increased from approximately 232 days–248 days, whereas CKD incidence rose from 170 cases to 250 cases per 100,000 population.

Heat stress and CKD incidence

Greater numbers of heat stress days were associated with a higher CKD incidence. In particular, each additional day of moderate, strong, very strong, and extreme heat stress was associated with an increase in CKD incidence of 0.088 (95% confidence interval [CI]: 0.035–0.140), 0.095 (0.047–0.142), 0.097 (0.036–0.159), and 0.782 (0.643–0.921) cases per 100,000 population, respectively (Figure 2). Similar associations were observed for CKD attributable to type 1 diabetes mellitus, type 2 diabetes mellitus, glomerulonephritis, hypertension, and other/unspecified causes (Figure 2).

Figure 2.

Multivariable-adjusted β for the associations of daily increase in heat stress and incidence of CKD

Adjustments were made for SDI, PM, NO₂, and O₃, with year included as a categorical fixed effect.

CKD, chronic kidney disease; SDI, sociodemographic index; PM, particulate matter; NO₂, nitrogen dioxide; O₃, ozone.

A higher heat stress score was positively associated with CKD incidence: for each standard deviation (SD) increase in the score, CKD incidence increased by 0.890 cases per 100,000 population (95% CI: 0.717–1.063) (Table 1). For cause-specific outcomes, the heat stress score was associated with CKD caused by type 1 diabetes mellitus, type 2 diabetes mellitus, glomerulonephritis, hypertension, and other/unspecified causes; the adjusted β (95% CI) values were 0.002 (0.001–0.003), 0.109 (0.090–0.128), 0.013 (0.009–0.017), 0.061 (0.050–0.072), and 0.711 (0.567–0.854) cases per 100,000 population, respectively (Table 1).

Table 1.

Multivariable-adjusted β and 95% confidence interval for the associations of per standard deviation increase in the heat stress score and CKD incidence

Heat stress score	Crude β (95% CI)	Crude p	Adjusted β (95% CI)	Adjusted p
Chronic kidney disease	3.382 (3.091, 3.668)	<0.001	0.890 (0.717, 1.063)	<0.001
Chronic kidney disease caused by diabetes mellitus type 1	0.014 (0.012, 0.015)	<0.001	0.002 (0.001, 0.003)	0.001
Chronic kidney disease caused by diabetes mellitus type 2	0.358 (0.329, 0.387)	<0.001	0.109 (0.090, 0.128)	<0.001
Chronic kidney disease caused by glomerulonephritis	0.049 (0.045, 0.054)	<0.001	0.013 (0.009, 0.017)	<0.001
Chronic kidney disease caused by hypertension	0.224 (0.205, 0.242)	<0.001	0.061 (0.050, 0.072)	<0.001
Chronic kidney disease caused by other and unspecified causes	2.742 (2.505, 2.975)	<0.001	0.711 (0.567, 0.854)	<0.001

Adjustments were made for SDI, PM, NO₂ and O₃, with year included as a categorical fixed effect.

SDI, sociodemographic index; PM, particulate matter; NO₂ nitrogen dioxide; O₃, ozone.

Subgroup analyses suggested effect modification by socio-demographic index (SDI) (p for interaction <0.001); however, the estimates did not show a steady decrease from low to high SDI (Figure 3). The observed associations between heat stress score and CKD incidence were stronger among female populations and older age groups (Figure 3). In particular, for the stratified SDI groups, the adjusted β (95% CI) from the low to high SDI levels were −0.05 (−0.25–0.15), 1.06 (0.74–1.37), 0.81 (0.42–1.18), 1.30 (0.93–1.65), and 0.82 (0.41–1.19) per 100,000 population, respectively. The estimated association was stronger among female than male populations, with the adjusted β (95% CI) of 1.01 (0.83–1.19) and 0.84 (0.66–1.01) per 100,000 population, respectively. With regard to the stratified age groups, the estimated associations between heat stress score and CKD generally increased with advancing age, but this pattern was not observed in the ≥80-year group.

Figure 3.

Subgroup analysis findings on the association of heat stress score and incidence of CKD

Adjustments were made for SDI, PM, NO₂, and O₃, with year included as a categorical fixed-effect Models for subgroup analyses across categorical variables did not adjust for the stratification factor itself.

SDI, sociodemographic index; PM, particulate matter; NO₂, nitrogen dioxide; O₃, ozone.

The estimated association between the heat stress score and CKD incidence was stronger at higher quantiles of NO₂ and O₃ (both p for interaction <0.05), whereas no significant interaction was observed for PM (p for interaction = 0.074) (Figure 4).

Figure 4.

Effect of air pollution on the association of heat stress score and CKD incidence

Adjustments were made for SDI, PM, NO₂, and O₃, with year included as a categorical fixed effect. The p value indicates the statistical significance of the interaction terms of heat stress score and air pollution exposure.

SDI, sociodemographic index; PM, particulate matter; NO₂, nitrogen dioxide; O₃, ozone.

Shapley additive explanation (SHAP) analysis ranked the relative importance of the heat stress score, ambient air pollution exposure, and socioeconomic level. Overall, socioeconomic level exerted the greatest influence on CKD incidence (Figure 5). Among environmental determinants, heat stress had more influence than ambient air pollution (Figure 5).

Figure 5.

SHAP summary plots

Variables are ranked by importance, based on the mean absolute SHAP value, from top to bottom. Each dot represents one observation; the x axis shows the SHAP value. Positive values increase the model-predicted CKD incidence, whereas negative values decrease it. Dot color indicates the feature value for each observation.

CKD, chronic kidney disease; O₃, ozone; NO₂, nitrogen dioxide; PM, particulate matter; SHAP, shapley additive explanations.

Sensitivity analyses

The findings were generally consistent across all sensitivity analyses. The association between the heat stress score and CKD incidence remained in the same direction after log-transforming CKD incidence, applying two-way fixed-effects models with Driscoll-Kraay standard errors, further adjusting for life expectancy, diabetes prevalence, hypertension prevalence, and urbanization rate in 2000–2019, and excluding observations with extreme heat stress score values. These analyses supported the robustness of the main findings (online Tables S8–S11).

Discussion

In this study, we evaluated the associations between heat stress across four intensity levels and a heat stress score with CKD incidence. We found that, across individual levels, a greater number of heat stress days was positively associated with CKD incidence. The score was similarly associated with higher CKD incidence. Associations were generally stronger among women and older adults, and estimates varied across SDI strata; however, they did not show a steady decrease from low to high SDI. Moreover, the observed association between heat stress and CKD incidence was stronger at higher levels of NO₂ and O₃ exposure.

Large scale epidemiological studies on heat stress and CKD incidence remain limited. To our knowledge, only one post hoc analysis of a clinical trial has reported relevant findings. Using data from 4,017 patients with established CKD across 21 countries, Zhang et al. found that each additional 30 days with a heat index above 30°C was associated with a −0.6% change in eGFR (95% CI: −0.9% to −0.3%).¹³ However, Zhang et al. addressed disease progression in patients with pre-existing CKD, whereas our study extends the evidence base to CKD incidence in the general population. Previous regional studies have also focused mainly on extreme heat events. For example, a national time series study of 261 major Chinese cities showed that heatwave exposure was associated with a higher risk of CKD hospitalization than non-heatwave days (relative risk [RR] = 1.116; 95% CI: 1.069–1.166).¹⁷ By contrast, we found that even moderate heat stress may be associated with CKD incidence, suggesting that heat-related kidney disease may not be limited to extreme events alone.

Although our analyses do not identify CKDu cases directly, we observed a comparatively stronger association between heat stress and CKD due to other and unspecified causes. This finding is compatible with the broader epidemiologic literature describing CKD of unknown/non-traditional cause, as occurring disproportionately in hot settings and often among populations engaged in physically demanding outdoor work.^18,19,20 Importantly, the GBD category “other and unspecified causes” is a heterogeneous residual group rather than a CKDu-specific diagnosis. Therefore, our results should be interpreted as country/region-level correlations that may partially capture CKDu in endemic settings. Future work using occupational heat exposure metrics and clinical CKDu registries is needed to validate the association.

We observed that the association between heat stress and CKD incidence was more pronounced among women and in older adults, broadly aligning with the findings of previous studies. For example, Wen et al.²¹ found in a nationwide Brazilian case-crossover study that each 1°C increase in the daily mean temperature was associated with a 0.6% increase in kidney-related hospitalizations among men (RR = 1.006; 95% CI: 1.004–1.008) versus 1.1% among women (RR = 1.011; 95% CI: 1.009–1.012). In addition to biological differences in thermoregulation and dehydration responses,^22,23 women in some settings may also experience substantial heat and dehydration exposure through gendered activity patterns, including unpaid domestic work and water collection.^24,25 Numerous studies have also reported greater heat-related renal risks among older adults.^26,27,28 These findings may help identify population groups in which the observed associations were stronger, particularly female populations and older adults.

We also observed that the association differed across SDI strata, but the estimates did not display a monotonic trend. The lack of a monotonic pattern in our study may reflect several factors. First, dividing SDI into categories reduces the effective sample size within each subgroup, limiting statistical power to detect a clear trend across SDI levels. Second, cross-country differences in CKD detection may distort differences by SDI, particularly in the lowest SDI settings. Additionally, residual confounding may further mask any underlying monotonic relationship.

Air pollution may modify the association between heat stress and CKD incidence. In our interaction analyses, the association strengthened across increasing quantiles of NO₂ and O₃, whereas we did not observe clear evidence of interaction for PM. Previous studies^29,30,31 have linked air pollutants to kidney disease risk and to declines in renal function. For example, in a synthesis of 14 studies, Ye et al.³¹ estimated an odds ratio of 1.15 (95% CI: 1.07–1.24) per 10 μg/m³ increase in PM₂.₅ and an odds ratio of 1.10 (95% CI: 1.03–1.17) per 10 ppb increase in NO₂. More direct evidence comes from a case-crossover analysis of renal care records from 28 U.S. counties by Remigio et al.,³² which revealed interdependent joint effects of extreme heat with PM and O₃ among patients receiving maintenance dialysis (i.e., end-stage renal disease), with higher risks of hospitalization and death than either exposure alone. Our findings support potential joint effects of heat stress with NO₂ and O₃ on kidney health, while the evidence for PM as an effect modifier was less clear in our data.

The mechanisms by which heat stress affects renal health remain incompletely understood. Recurrent heat stress may contribute to CKD progression through several pathways. First, heat exposure commonly induces profuse sweating and dehydration, which reduces effective circulating volume and renal perfusion, thereby increasing (often subclinical) episodes of acute kidney injury (AKI).^33,34 Second, recurrent tubular injury can lead to incomplete repair, promoting tubulointerstitial inflammation and fibrosis.³⁵ Third, dehydration-related hyperosmolar and metabolic stress can amplify oxidative stress and inflammatory signaling, further exacerbating tubular damage and fibrogenesis.^36,37 Finally, sustained activation of neurohormonal responses (e.g., renin-angiotensin-aldosterone system) helps preserve fluid homeostasis acutely but may, over time, increase renal workload and reinforce inflammation-related injury pathways.^38,39 Whether heat stress is the principal driver of CKD epidemics in certain regions remains contested. Nevertheless, recent integrative assessments increasingly support a central role for heat stress.^40,41 Accordingly, preventive strategies should focus on reducing heat strain and ensuring hydration, particularly during sustained heat exposure.⁴²

In summary, our findings suggest that heat stress may be associated with higher CKD incidence at the country/region level, and that this association may not be limited to extreme heat events. These results indicate that reducing sustained heat stress exposure should be considered in kidney health protection strategies under global warming, particularly in settings with prolonged heat exposure and among populations in which the observed associations were stronger. They also suggest that future research should consider the cumulative implications of moderate but frequent heat stress.

Limitations of the study

Some limitations must be acknowledged. One, this ecological study measured both heat stress and CKD incidence at the country/region level; therefore, the findings should not be interpreted as causal effects on individual risk. Two, we could not incorporate many potentially important country-level determinants because harmonized, annually resolved data were unavailable for all 174 countries/regions across the full study period. Residual confounding is likely, as key drivers of CKD operate at the individual level and are not captured by national aggregates. Three, our exposure metric reflects national average heat stress and cannot capture within-country variability. Such heterogeneity may lead to exposure misclassification, likely attenuating associations. Four, our analysis relied on globally harmonized, model-based datasets. Although this enabled consistent long-term global coverage, the estimates may be affected by model-based uncertainty and country-specific measurement error. Therefore, our findings should be interpreted as country/region-level evidence derived from harmonized global data, and their generalizability may vary across national settings.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Jianxiong Ma (daxiong1990@zcmu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •This paper analyzes existing, publicly available data, accessible at Mendeley Data: https://doi.org/10.17632/xtnm57syhh.1.
• •All original code has been deposited and is publicly available at Mendeley Data: https://doi.org/10.17632/xtnm57syhh.1.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This study was supported by the Zhejiang Provincial Natural Science Foundation of China, Hangzhou, China [grant number: ZCLMS25H2706], China Postdoctoral Science Foundation (CPSF) Beijing, China (grant no.: 2023M743146 and 2025T181077), Postdoctoral Fellowship Program of CPSF, Beijing, China (grant no.: GZC20232373], and the Zhejiang Chinese Medical University Scientific Research Project for Talent, Hangzhou, China (grant number: 2023RCZXZK47).

Author contributions

Conceptualization, J.M. and C.M.; methodology, J.Q., J.M., and C.M.; investigation, J.Q., Y.D., and Y.C.; writing – original draft, J.Q., R.W., and D.W.; writing—review and editing, Y.S., L.C., W.L., and Q.W.; funding acquisition, J.M.; resources, J.M. and C.M.; supervision, J.M. and C.M.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Environmental data; incidence data; analysis code	Mendeley Data	Mendeley Data: https://doi.org/10.17632/xtnm57syhh.1.
Software and algorithms
R software	R Core Team. R Foundation for Statistical Computing, Vienna, Austria	4.3.1
ArcGIS	Environmental Systems Research Institute, Inc., Redlands, CA, USA	10.8

Experimental model and study participant details

Study approval and ethical statement

Not applicable.

Method details

Sources of data

The Global Burden of Disease (GBD) study,⁴³ led by the Institute for Health Metrics and Evaluation with a global network of collaborators, systematically quantifies health loss attributable to diseases and risk factors. The GBD 2021 provides country-level estimates of the incidence, prevalence, and mortality of 371 conditions from 1990 to 2021 under a unified modeling framework. We used 1990 to 2021 data on CKD incidence from GBD 2021. We also obtained country-level estimates of ambient air pollution exposure (i.e., PM, NO₂ [nitrogen dioxide], O₃ [ozone]) and the Sociodemographic Index (SDI).

ERA5-HEAT (Human thErmAl comforT) is the first global historical gridded dataset of human biometeorological indices, produced by the European Center for Medium-Range Weather Forecasts and distributed through the Copernicus Climate Data Store.⁴⁴ ERA5-HEAT provides hourly UTCI at a 0.25 ° × 0.25 ° spatial resolution, covering land and ocean areas worldwide from 1979 to the present [18]. For this study, we downloaded global UTCI rasters for 1990–2021. Shapefiles of administrative areas were available for 174 GBD countries/territories, our ecological analysis was restricted to this set (see online Table S1 for the full list).

Assessment of chronic kidney disease incidence

In GBD 2021, CKD was defined as an eGFR <60 mL/min/1.73 m² or albuminuria (i.e., the urinary albumin-to-creatinine ratio [uACR]) ≥3 mg/mmol.⁴⁵ eGFR was calculated using the CKD-EPI equation (online Table S2). The country-level CKD incidence was estimated by using DisMod-MR2.1. The GBD 2021 also reported cause-specific incidence attributable to type 1 diabetes mellitus, type 2 diabetes mellitus, glomerulonephritis, hypertension, and other unspecified causes. For each individual, primary kidney disease was mapped to an etiology, based on the International Classification of Diseases (ICD)-9/ICD-10 codes; individuals with CKD who lacked a primary kidney disease ICD code were classified as “CKD of unknown cause”⁴⁵ (online Table S3). The outcome for this study was the mean estimate of the age-standardized incidence rate of CKD, calculated using the GBD standard population.

Assessment of UTCI

UTCI expresses heat exposure as an equivalent temperature, defined as the air temperature in a reference setting that would elicit the same physiological response as that of the actual environment. The reference activity is walking at 4 km/h (≈1.1 m s⁻¹) with a metabolic rate of 2.3 metabolic equivalent of task (MET) (≈135 W m⁻²). The reference environment specifies a 10-m wind speed of 0.5 m s⁻¹, a mean radiant temperature equal to air temperature, and a relative humidity of 50% (for air temperature >29°C, the reference vapor pressure is fixed at 20 hPa).¹⁴ The computational form of UTCI can be written as:

$$UTCI=Ta+Offset\left(Ta;Tmrt;va;RH\right)$$

An automated routine approximates the UTCI-Fiala thermophysiological model by computing the equivalent temperature offset from a sixth-order polynomial with main and interaction terms, given four meteorological inputs: 2-m air temperature (Ta), mean radiant temperature (Tmrt), 10-m wind speed (va), and relative humidity (RH). The routine yields an average root-mean-square error of approximately 1.1°C, relative to the original model.¹⁴ UTCI values were classified using the standard UTCI thermal stress scale: moderate (26°C–32°C), strong (32°C–38°C), very strong (38°C–46°C), and extreme (>46°C) heat stress (see online Table S4 for the full classification).¹⁴ In the present study, we calculated the annual number of days on which the daily maximum UTCI reached each category as indicators of heat stress at different intensities, then aggregated to the country-level by masking with shapefiles of GBD countries’ administrative areas. The distribution and temporal trend of these metrics over the study period are shown in online Figures S1 and S2.

Assessment of the covariates

Two country-level covariates were included: the Sociodemographic Index (SDI) and ambient air pollution exposure (i.e., PM, NO₂, O₃). The SDI is calculated as the geometric mean of three standardized components—income per capita, years of schooling among persons aged ≥15 years, and the fertility rate among women <25 years—scaled from 0 to 1 to reflect socioeconomic development relevant to health.⁴⁶ Methods for estimating air pollution exposure have been detailed previously.⁴⁷

Quantification and statistical analysis

We applied linear mixed-effects models to assess the association between annual number of days increase in the various levels of heat stress and CKD incidence. In this model, country/region codes were included as random intercept, whereas adjustments were made for SDI, PM, NO₂, and O₃. To control for long-term secular trends shared across countries/regions, we further included calendar year as a fixed effect. β and 95% confidence intervals (CIs) were estimated by using a crude model (i.e., without any adjustment) and a multivariable-adjusted model.

To capture overall heat exposure across the four heat stress categories, we constructed a weighted heat stress score, with higher values indicating greater overall heat stress exposure. The score was calculated using the following equation: (β [moderate heat stress] × moderate heat stress days + β [strong heat stress] × strong heat stress days + β [very strong heat stress] × very strong heat stress days + β [extreme heat stress] × extreme heat stress days), where the multivariable-adjusted β coefficients for the composite CKD outcome, derived from the linear mixed-effects model described above, were used as weights. To reduce overfitting of the weights during the calculation process, we applied a 5-fold cross-fitting procedure. Countries and regions were randomly divided into five groups; in each round, four groups were used to estimate the weights and the remaining group was used to calculate the score. This process was repeated until each group had served once as the validation group, and the resulting scores were then combined. Before weighting, the heat stress variables were standardized within the training data to place them on a comparable scale. The final heat stress score ranged from −33.0 to 103.6 (online Table S5). We then used linear mixed-effects models to estimate the change in CKD incidence per standard deviation (SD) increase in this score.

For the subgroup analysis, we examined the associations between per SD increase in the heat stress score and CKD incidence, stratified by SDI, sex, and age. SDI was categorized into five levels (“low,” “low-middle,” “middle,” “high-middle,” and “high”), based on GBD definitions.⁴⁷

Interaction models were used to investigate the modifying effects of air pollution exposure on the association between heat stress score and CKD incidence. In the interaction model, multiplicative cross-product terms between the heat stress score and each pollutant (i.e., PM, NO₂, O₃) were included. Conditional effects of heat stress score on CKD incidence were then calculated across quantiles of air pollution exposure.

To further quantify the contribution of heat stress to CKD incidence relative to social and environmental determinants, we trained an XGBoost model (a machine learning algorithm) and applied SHapley Additive exPlanations (SHAP) to rank the importance of the heat stress score, ambient air pollution exposure, and socioeconomic level. A machine learning approach avoids parametric assumptions about the exposure–response relation and accommodates correlation and nonlinearity. SHAP, grounded in Shapley values, assigns an additive contribution to each factor, thereby enabling an interpretable ranking of predictor importance.

We conducted several sensitivity analyses to assess the robustness of the association between the heat stress score and CKD incidence: (1) log-transforming CKD incidence to better approximate normality; (2) re-estimating models using two-way fixed effects (country and year) with Driscoll–Kraay standard errors to account for serial correlation and cross-sectional dependence; (3) additionally adjusting for life expectancy, diabetes prevalence, hypertension prevalence, and urbanization rate, restricting this analysis to 2000–2019 due to data availability; and (4) refitting the models after excluding observations in the lowest and highest 5% of the heat stress score distribution to evaluate the influence of extreme values. All analyses were performed in R (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria) and ArcGIS (version 10.8; Environmental Systems Research Institute, Inc., Redlands, CA, USA). A two-sided p value <0.05 was statistically significant.