Spatiotemporal Trends in Chronic Kidney Disease Mortality and Its Association with Low Temperature in the United States (1999–2023)

Abstract

Purpose

Chronic kidney disease (CKD) is a major public health concern in the United States, with persistently rising incidence and mortality. This study aimed to quantify geographic and demographic disparities in CKD mortality across the United States and to examine the association between CKD mortality and ambient temperature variation.

Patients and Methods

CKD deaths and age-adjusted mortality rates (AAMRs) for 1999–2023 were obtained from CDC WONDER and stratified by census region, state, sex, race, age group, and urbanization level. Mortality trends were quantified using joinpoint regression. Primary temperature–mortality analyses used CDC WONDER-linked NLDAS temperature data for 1999–2011, while supplementary descriptive analyses used independently retrieved NLDAS temperature estimates for 2012–2023. Temperature–mortality associations were evaluated using Spearman rank correlations, quasi-Poisson regression models, and distributed lag nonlinear models (DLNMs).

Results

From 1999 to 2023, 628,937 CKD deaths occurred, with national AAMR increasing steadily. Mortality rates were consistently higher in men and nonmetropolitan areas. AAPC rose fastest in the West and Midwest, though recent AAMR declines were noted in the Northeast and South. Winter mortality exceeded summer across regions. In the primary DLNM analysis, cold exposure at the region-specific 5th percentile was associated with higher cumulative CKD mortality risk, with risk ration (RR) ranging from 1.065 in the Northeast to 1.505 in the Midwest; the association reached statistical significance only in the Midwest (RR 1.505, 95% CI: 1.091–2.075), while hot-exposure estimates at the 95th percentile were generally imprecise and not statistically significant.

Conclusion

CKD mortality in the United States increased from 1999 to 2023 with marked geographic and demographic disparities. Cold air temperature was associated with higher short-term mortality risk.

Introduction

Chronic kidney disease (CKD) is a global public health problem that affects more than 750 million people worldwide.1,2 In 2023, CKD accounted for 1.5 million deaths.3 By 2040, this number is projected to rise to 2.2–4.0 million, making CKD the fifth leading cause of death globally.4 In the United States alone, approximately 14% of the population has CKD.5,6 CKD is defined as a group of clinical syndromes characterized by structural and/or functional abnormalities of the kidney persisting for ≥3 months due to various etiologies.7 Patients with CKD frequently have comorbid or secondary chronic conditions, including cardiovascular disease, chronic respiratory disease, diabetes, and osteoarthritis.8,9

The Lancet has reported that climate change poses a profound threat to global health in the 21st century.10 Increases in extreme heat and cold spells have been linked to a higher risk of CKD.11 The incidence and mortality associated with heat-related illness are projected to rise substantially in the future.12,13 As the global mean temperature has increased by 0.8–0.9 °C, the frequency of extreme heat and cold events has grown, contributing to higher rates of heat stroke and exacerbations of pre-existing chronic diseases, thereby imposing a substantial disease burden.14 Prior studies indicate that long-term exposure to high temperatures is a risk factor for CKD.15 Potential mechanisms include dehydration, reduced renal perfusion, and tubular injury resulting from recurrent heat stress.14 Equally critical, though less frequently highlighted, is the role of cold exposure as a potential contributor for adverse renal outcome. Epidemiological evidence has increasingly linked cold temperature exposure to elevated CKD mortality.16,17 A large time-stratified case-crossover study conducted across 47 prefectures in Japan, analyzing nearly one million renal disease deaths over four decades, found a reversed J-shaped temperature–mortality relationship, with lower temperatures associated with increased mortality across all renal disease categories, with a cumulative relative risk of 1.34 (95% CI: 1.29–1.40) at the 2.5th temperature percentile.17 In terms of mechanism, cold temperatures induce peripheral vasoconstriction18,19 and activate both the sympathetic nervous system and the renin-angiotensin-aldosterone system (RAAS).20,21 This cascade leads to elevated systemic blood pressure, increased cardiovascular strain, and heightened hemodynamic stress on the kidneys, which can exacerbate renal impairment in susceptible individuals.

Although prior studies using the United States Centers for Disease Control and Prevention Wide-ranging Online Data for Epidemiologic Research (CDC WONDER) database have reported CKD-related mortality trends in the United States population, they were limited to periods before 2020.6,22–24 With comprehensive national coverage, standardized mortality coding, and sufficient regional sample size, CDC WONDER provides a robust platform for examining population-level CKD mortality patterns across diverse geographic and climate regions. In this study, we conducted an up-to-date analysis of CKD mortality trends in the United States from 1999 to 2023, examined the association between ambient temperature and CKD mortality, and projected future CKD mortality trajectories over the next 20 years to inform healthcare resource planning.

It is important to note that the present study focuses on short-term associations between non-optimal temperatures and CKD mortality, rather than the long-term effects of chronic temperature exposure on CKD incidence. The distributed lag non-linear model (DLNM) framework is particularly well-suited for capturing such short-term, lagged relationships between daily temperature fluctuations and daily mortality counts, allowing us to simultaneously characterize both the non-linear and delayed effects of temporally proximate temperature exposure.25

However, it is crucial to recognize that CKD mortality is multifaceted. The interpretation of these trends requires consideration of potential confounders, including the prevalence of baseline comorbidities including diabetes and hypertension, and disparities in healthcare infrastructure, such as access to timely dialysis care and specialized nephrology services.26 These findings are critical for elucidating the spatiotemporal variation in the CKD burden across the United States and for clarifying the potential impact of temperature exposure on CKD mortality.

Materials and Methods

Mortality Data Source and Collection

We investigated national trends in deaths involving CKD in the United States from 1999 to 2023. CKD was identified using the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) code N18: N18.0 (End-stage renal disease); N18.1 (Chronic kidney disease, stage 1); N18.2 (Chronic kidney disease, stage 2); N18.3 (Chronic kidney disease, stage 3); N18.4 (Chronic kidney disease, stage 4); N18.5 (Chronic kidney disease, stage 5); N18.8 (Other chronic renal failure); N18.9 (Chronic renal failure, unspecified). Mortality data were provided by the National Center for Health Statistics (NCHS) and accessed via the CDC WONDER database, which compiles death certificate records from all 50 states and the District of Columbia. The analysis included death certificates for individuals aged ≥25 years in which CKD was recorded. This dataset was integrated by matching unique spatial identifiers (Census region and state codes) and temporal variables (year and month) to ensure precise alignment of exposure and outcome data. We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.27

We extracted data on census region, state, demographic characteristics, and the geographic area of residence. Demographics included sex, age group (10-year bands), and race as defined by the United States Census Bureau: Hispanic or Latino; non-Hispanic Black or African American (NH-Black); non-Hispanic White (NH-White); and non-Hispanic Other (NH-Other). Urbanization was classified using the National Center for Health Statistics Urban–Rural Classification Scheme for Counties, and census regions were assigned as Northeast, Midwest, South, and West.28

According to CDC WONDER confidentiality rules, monthly death counts fewer than 10 in a given state or region were suppressed; these suppressed observations were treated as missing, and complete-case analysis was applied when constructing the linked mortality–temperature analytic dataset.

Temperature Data Source and Collection

Temperature data for the four United States census regions were derived from two sources covering different study periods. Regional temperature records for 1999–2011 were extracted through the CDC WONDER North America Land Data Assimilation System Daily Air Temperatures and Heat Index database. For 2012–2023, matching temperature estimates were independently retrieved from the NLDAS archive. Both datasets were based on the NLDAS Primary Forcing Data L4 Monthly 0.125° × 0.125° V2.0 product (NLDAS_FORA0125_M).

For meteorologic exposure, monthly temperature metrics included the average daily maximum temperature (Max Temp), average daily minimum temperature (Min Temp), monthly mean temperature (Mean Temp), the within-month temperature range (Monthly range), and the month-to-month temperature range (Month-to-month range). All temperatures were expressed in degrees Fahrenheit (°F).

Mortality Analysis

All seasonal and temperature-related analyses were conducted separately at the United States census-region level, using monthly aggregated mortality counts and region-specific temperature metrics for the four regions (Northeast, Midwest, South, and West). This region-stratified approach accounts for the marked climatic heterogeneity across the contiguous United States and avoids ecological confounding from between-region differences in temperature distributions and baseline mortality. We directly standardized to the year 2000 United States population to compute age adjusted mortality rates (AAMR), which reduces bias from temporal and demographic shifts.29 From AAMRs we estimated the average annual percent change (AAPC) with 95% confidence intervals (CI). To assess trends, we applied joinpoint regression using the Joinpoint Regression Program version 5.0.2 (National Cancer Institute), fitting log-linear models to identify changes in slope over time. The software’s model selection procedure determined the optimal number of joinpoints, yielding period-specific annual percent changes (APC) with 95% CIs.

For monthly crude mortality rates and count regression models, annual region-specific population denominators were obtained from the population field included in the CDC WONDER mortality extracts. The annual population for each region-year was assigned to all monthly observations within that year, as monthly population estimates were unavailable. The logarithm of the corresponding region- and year-specific annual population was included as an offset in all count regression models.

We constructed annual time series using the AAMR. The minimum required order of differencing was determined via the KPSS test, and variance was stabilized using Box–Cox transformation. Three candidate models were fitted: non-seasonal ARIMA, ETS exponential smoothing (including damped trend forms), and TBATS. Each model captures a distinct structure of temporal dependency. ARIMA models autoregressive and moving-average dynamics in differenced series.30 ETS models adaptively smoothed level and trend components suited to series with decelerating trends.31 TBATS accommodates complex seasonal patterns and trigonometric error structures.32 Because different subgroups (sex, census region) exhibit heterogeneous trend profiles, no single model is universally optimal; therefore, model selection was based on rolling-origin cross-validation with a 3-year forecast horizon, using root mean squared error (RMSE) as the criterion. The best-performing model was then refitted on the full dataset to generate 20-year forecasts, with 80% and 95% prediction intervals reported. In cases where the automatically selected model produced nearly flat extrapolations, a damped-trend ETS model was used as a conservative alternative.

Temperature–Mortality Association Analysis

To examine the association between CKD mortality and season, we stratified by the United States census region and by season. Seasonal differences were evaluated with one way analysis of variance, followed by Tukey’s honestly significant difference test for post hoc comparisons. For each census region, Spearman rank correlation coefficients were computed between monthly crude mortality rates and temperature metrics, with two-sided P values reported.

Region-specific generalized linear models were used to examine the lagged associations between monthly mean temperature and CKD mortality. Monthly death counts were modelled using quasi-Poisson regression with a log link, and the logarithm of the corresponding regional population was included as an offset. Monthly mean temperature was centred at the region-specific mean and divided by 10 so that the resulting relative risks represented the change in CKD mortality associated with each 10°F increase in temperature. Temperature variables at lags of 0, 1, 2, and 3 months were generated separately within each United States census region after observations had been ordered chronologically, thereby preventing lagged values from being carried across regions.

Two complementary model specifications were fitted. First, separate single-lag models were used to estimate the association between temperature and CKD mortality at each lag independently. Second, temperature terms for lags 0–3 months were entered simultaneously into an all-lags model to obtain mutually adjusted lag-specific estimates. The cumulative association across lags 0–3 months was calculated by summing the four regression coefficients on the log-relative-risk scale. The corresponding standard error was derived from the complete variance–covariance matrix using the delta method, thereby accounting for covariance among the lag-specific coefficients.

All models included calendar month as a categorical variable to control for seasonality and a natural cubic spline of calendar time to account for long-term and secular trends. The time spline was specified using 3 degrees of freedom per year, with the total number of degrees of freedom restricted to no more than approximately one-tenth of the available monthly observations to reduce potential overfitting. Overdispersion was assessed in corresponding conventional Poisson models using the Pearson dispersion statistic, calculated as the sum of squared Pearson residuals divided by the residual degrees of freedom. The dispersion statistics ranged from 1.151 to 1.455, with a median of 1.229, indicating mild-to-moderate overdispersion. Therefore, the final analyses used quasi-Poisson regression, which scales the standard errors and 95% confidence intervals according to the model-specific dispersion parameter.

Distributed Lag Non-Linear Model Analysis

Distributed lag non-linear models (DLNM) were used to characterize the potentially non-linear and delayed associations between monthly temperature indicators and CKD mortality. Analyses were conducted separately for each United States census region. Monthly mortality counts were modelled using quasi-Poisson regression with a log link to account for overdispersion, with the logarithm of the corresponding regional population included as an offset.

The exposure–lag relationship was represented by a cross-basis function. In the primary model, the exposure dimension was specified using a natural cubic spline with three internal knots placed at the 25th, 50th, and 75th percentiles of the region-specific temperature distribution and boundary knots at the observed minimum and maximum values. This specification yielded 4 degrees of freedom for the exposure basis. The lag dimension was modelled using a natural cubic spline over lags of 0–3 months, with one internal knot placed at the midpoint of the lag interval and boundary knots at 0 and 3 months. An intercept was included in the lag basis, resulting in 3 degrees of freedom.

Long-term and secular temporal trends were controlled using a natural cubic spline of calendar time with 3 degrees of freedom per year. To prevent overfitting in shorter time series, the total degrees of freedom for the time spline were restricted to no more than approximately one-tenth of the number of monthly observations. Calendar month was additionally included as a categorical variable to control for residual seasonality. The median value of each temperature indicator within each region was used as the reference. Cold and hot temperatures were defined as the 5th and 95th percentiles, respectively. We estimated the overall cumulative relative risks across the lag period, the complete cumulative exposure–response curves, and lag-specific relative risks at cold and hot temperatures, together with their 95% confidence intervals.

The robustness of the findings was examined through a series of one-at-a-time sensitivity analyses. First, the maximum lag period was changed from 3 months to 2, 4, and 6 months. Second, alternative knot placements were applied to the exposure dimension, including knots at the 10th, 50th, and 90th percentiles and at the 33rd and 67th percentiles. Third, the lag structure was alternatively specified using a natural cubic spline with two equally spaced internal knots at one-third and two-thirds of the maximum lag interval, or as a linear lag function. All other model components were held constant when each alternative specification was evaluated.

Statistical Analysis

Two sided P values <0.05 were considered statistically significant. All analyses and visualizations were conducted in R version 4.4.0. As this is an ecological study using aggregate data at the state and regional levels, we acknowledge the potential for ecological bias; thus, the observed associations should be interpreted as population-level patterns rather than individual-level risks. Calendar month was included as a categorical variable to control for recurring within-year seasonality, whereas a natural cubic spline of calendar time was included to account for gradual, non-linear long-term and secular trends.

Results

Characteristics of Chronic Kidney Disease Mortality in the United States

In 2023, there were 37,401 deaths from CKD among the United States adults aged ≥25 years, with an AAMR of 13.65 (13.51, 13.79) (Table 1). Among the four census regions (Table 1 and Supplementary Table S1), the South recorded the largest number of deaths at 15,251 with an AAMR of 14.68 (14.44, 14.91). The Midwest reported 8657 deaths with an AAMR of 15.01 (14.69, 15.33), the West reported 6970 deaths with an AAMR of 11.33 (11.06, 11.60), and the Northeast reported 6523 deaths with an AAMR of 12.77 (12.46, 13.08). At the state level (Figure 1a–c and Supplementary Figure S1a, b), California had the highest number of CKD deaths at 4190, followed by Texas at 3071. The highest AAMR was observed in West Virginia at 21.60 (19.32, 23.87), followed by Louisiana at 21.56 (20.01, 23.10). Washington exhibitedthe lowest AAMR at 5.64 (5.03, 6.24) (Figure 1d).

Table 1.

Counts and Age-Adjusted Mortality Rates (per 100,000) from Chronic Kidney Disease Among ≥ 25 Years Adults, Stratified by Sex, Race, Census Region, and Urbanization Level

	Deaths 2023	Population 2023	Crude Rate 2023 (95% CI)	AAMR 1999 (95% CI)	AAMR 2023 (95% CI)	AAPC (95% CI)
Overall	37401	231529762	16.15 (15.99, 16.32)	7.88 (7.75, 8.02)	13.65 (13.51, 13.79)	2.29 (2.17, 2.40)
Census Region
Northeast	6523	40402747	16.14 (15.75, 16.54)	8.72 (8.42, 9.02)	12.77 (12.46, 13.08)	1.47 (1.29, 1.64)
Midwest	8657	47312396	18.30 (17.91, 18.68)	7.79 (7.53, 8.06)	15.01 (14.69, 15.33)	2.96 (2.68, 3.35)
South	15251	89219544	17.09 (16.82, 17.37)	8.40 (8.17, 8.63)	14.68 (14.44, 14.91)	2.34 (2.17, 2.61)
West	6970	54595075	12.77 (12.47, 13.07)	6.20 (5.94, 6.46)	11.33 (11.06, 11.60)	3.01 (2.60, 3.54)
Gender
Female	17974	118635845	15.15 (14.93, 15.37)	6.76 (6.60, 6.92)	11.64 (11.47, 11.81)	2.43 (2.26, 2.65)
Male	19427	112893917	17.21 (16.97, 17.45)	9.89 (9.65, 10.13)	16.35 (16.11, 16.58)	2.19 (2.00, 2.52)
Race
Hispanic or Latino	3992	38599156	10.34 (10.02, 10.66)	9.38 (8.71, 10.05)	14.03 (13.58, 14.48)	1.89 (1.63, 2.26)
NH Black or African American	7835	28042483	27.94 (27.32, 28.56)	23.00 (22.22, 23.77)	29.41 (28.74, 30.07)	1.08 (0.89, 1.32)
NH White	23401	144415042	16.20 (16.00, 16.41)	6.14 (6.01, 6.27)	11.58 (11.43, 11.73)	2.73 (2.54, 2.96)
NH Other	2093	20473081	10.22 (9.79, 10.66)	8.98 (8.06, 9.89)	10.95 (10.48, 11.43)	1.26 (0.88, 1.69)
Urbanization	Deaths 2020	Population 2020	Crude Rate 2020 (95% CI)	AAMR 1999 (95% CI)	AAMR 2020 (95% CI)
Metropolitan	30261	194600110	15.55 (15.38, 15.73)	7.94 (7.79, 8.09)	13.50 (13.34, 13.65)	2.60 (2.42, 2.80)
Nonmetropolitan	6565	32028405	20.50 (20.00, 20.99)	7.65 (7.36, 7.95)	15.17 (14.80, 15.55)	3.37 (3.16, 3.72)

Figure 1.

Trends of mortality in patients with chronic kidney disease (CKD) in the United States from 1999 to 2023. (a) Line graph of age adjusted mortality rate (AAMR) in CKD patients stratified by state; (b) Heapmap of deaths in CKD patients stratified by state; (c) Scatter diagram of AAMR and deaths in CKD patients stratified by state in 2023; (d) Ranking dynamics of top 10 states with highest AAMR in America from 1999 to 2023.

In sex-specific analyses for 2023 (Supplementary Table S1), women accounted for 17,974 CKD deaths with an AAMR of 11.64 (11.47, 11.81), whereas men accounted for 19,427 deaths with an AAMR of 16.35 (16.11, 16.58). By race (Supplementary Table S1), the AAMR was highest among non-Hispanic Black or African American individuals at 29.41 (28.74, 30.07), followed by Hispanic or Latino individuals at 14.03 (13.58, 14.48). In urbanization-stratified analyses for 2020 (Table 1), the Metropolitan group had an AAMR of 13.50 (13.34, 13.65), which was lower than the nonmetropolitan group at 15.17 (14.80, 15.55).

A pronounced gradient was observed in the age-stratified results (Supplementary Table S1). In 2023, those aged 85 years or older had the most deaths at 11,260 and the highest AAMR at 181.76 (178.40, 185.12). Adults aged 25 to 34 years had the fewest deaths at 201 and the lowest AAMR at 0.44 (0.38, 0.50).

Trends in Chronic Kidney Disease Mortality in the United States

From 1999 to 2023, there were 628,937 deaths from CKD among adults aged ≥25 years in the United States, and both the number of deaths and the AAMR increased (Supplementary Table S2). Across census regions (Figure 2a and Supplementary Table S3), the West showed the fastest rise (AAPC 3.01%, 95% CI 2.60 to 3.54), followed by the Midwest (2.96%, 2.68 to 3.35). In most states the AAMR increased (Figure 1a, b and Supplementary Figure S1a, b); Utah had the steepest increase (AAPC 5.01%, 3.63 to 6.23), followed by Iowa (5.00%, 4.09 to 5.88). Nevada was an exception, with a decline (AAPC −1.05%, −1.70 to −0.33).

Figure 2.

Subgroup analysis of chronic kidney disease (CKD) mortality trends in the United States from 1999 to 2023. (a) Mortality trends by Census region; (b) Mortality trends by sex; (c) Mortality trends by urbanization level; (d) Mortality trends by age group. Abbreviations: AAMR, age adjusted mortality rate. Joinpoint analysis of CKD mortality trends in the United States from 1999 to 2023. (e) Mortality trends by Census region; (f) Mortality trends by sex; (g) Mortality trends by urbanization level; (h) Mortality trends by age group. *Indicates that the APC is significantly different from zero at the alpha = 0.05 level.

Abbreviations: AAMR, age adjusted mortality rate; APC, annual percentage change.

In sex specific analyses (Figure 2b and Supplementary Table S3), women had a faster rise in AAMR than men (AAPC 2.43%, 2.26 to 2.65 vs 2.19%, 2.00 to 2.52). By race (Supplementary Figure S2a), non-Hispanic White individuals had the fastest increase (AAPC 2.73%, 2.54 to 2.96), while non-Hispanic Black or African American individuals had the lowest AAPC (1.08%, 0.89 to 1.32). By urbanization (Figure 2c), from 1999 to 2020 the nonmetropolitan group rose faster than the metropolitan group (AAPC 3.37%, 3.16 to 3.72 vs 2.60%, 2.42 to 2.80).

In age specific analyses (Figure 2d and Supplementary Table S3), the ≥85 years group had the fastest increase (AAPC 3.20%, 2.82 to 3.59), followed by those aged 45 to 54 years (2.81%, 2.43 to 3.35). The 25 to 34 years group rose the slowest (AAPC 1.41%, 0.68 to 2.14).

Joinpoint Analysis of Chronic Kidney Disease Mortality in the United States

Using joinpoint regression, we characterized temporal shifts in CKD age adjusted mortality in the United States from 1999 to 2023. By census region (Figure 2e and Supplementary Table S4), the Midwest and the West showed sustained increases during 2009 to 2023, the Northeast declined during 2019 to 2023 (APC −0.43), and the South declined during 2021 to 2023 (APC −2.46). In sex specific analyses (Figure 2f and Supplementary Table S4), women increased during 2009 to 2023, whereas men declined during 2021 to 2023 (APC −2.26). In race analyses (Supplementary Figure S2b and Supplementary Table S4), non-Hispanic Black or African American individuals declined during 2020 to 2023 (APC −1.22), and non-Hispanic White individuals declined during 2021 to 2023 (APC −1.90). By urbanization level (Figure 2g and Supplementary Table S4), the nonmetropolitan group showed a more pronounced upward trend than the metropolitan group.

In age specific analyses (Figure 2h and Supplementary Table S4), AAMR declined during 2021 to 2023 for adults aged 45 to 84 years, including ages 75 to 84 years (APC −3.34), ages 65 to 74 years during 2019 to 2023 (APC −0.44), ages 55 to 64 years during 2020 to 2023 (APC −0.21), and ages 45 to 54 years during 2021 to 2023 (APC −1.28); other age groups predominantly showed increasing trends.

Exploratory Forecasting of AAMR by Sex and Census Region Over the Next 20 Years

Figure 3a–f depicts the historical and forecasted trajectories of AAMR across sex and the United States census regions. For both females and males (Figure 3a and b), AAMR showed a sustained rise from the early 2000s to 2019, followed by a transient plateau during the COVID-19 period (2019–2021). Forecasts based on the best-fit ARIMA models suggested that mortality rates were projected to continue an upward trend over the next two decades, with males consistently exhibiting higher absolute rates than females.

Figure 3.

Forecasts of age-adjusted mortality rates (AAMR) over time by subgroup. (a–f) show the observed AAMR (solid lines) and model-based forecasts (dashed lines) with 95% prediction intervals (shaded areas). The vertical shaded bar denotes the COVID-19 period (2019–2021). Subgroups are presented as: (a) females, (b) males, (c) Midwest, (d) Northeast, (e) South, and (f) West. The best-fitting model (ARIMA or ETS) for each subgroup is indicated above the corresponding panel.

Regional analyses revealed marked geographic variation. The Midwest and Northeast experienced the steepest historical increases (Figure 3c and d), and projections suggested a continued upward trajectory in these regions. The South demonstrated persistently elevated rates throughout the study period; ETS modeling projected stabilization after 2025, albeit with wide confidence intervals (Figure 3e). In contrast, the West maintained comparatively lower AAMR levels but was still projected to rise steadily over time (Figure 3f). Across all regions, the widening 95% prediction intervals underscored increasing uncertainty in long-term forecasts. Taken together, these exploratory estimates may provide forward-looking information that could help inform, rather than directly guide, future nephrology workforce planning, regional healthcare resource allocation, and the prioritization of preventive strategies. However, these projections should be interpreted cautiously given the widening prediction intervals and substantial uncertainty inherent in long-term forecasting.

Seasonal Variation in Chronic Kidney Disease Mortality in the United States

In time series analyses (Figure 4a), CKD mortality showed a potential association with monthly mean temperature during 1999 to 2021. We then calculated mean monthly mortality by season (Figure 4b). In the Northeast, the mean monthly mortality rate was 0.90 (95% CI 0.87 to 0.92) in winter and 0.79 (95% CI 0.76 to 0.81) in summer, with P<0.001. Other regions showed similar patterns, with higher CKD mortality observed in winter.

Figure 4.

Association between temperature/seasonality and chronic kidney disease (CKD) mortality in the United States from 1999 to 2011. (a) Mortality and temperature trends by census region; (b) Seasonal mortality variation across census regions; (c) Spearman correlation between mortality and temperature metrics.

We next performed Spearman’s rank correlations between monthly average maximum temperature, average minimum temperature, mean temperature, within month temperature range, and month to month temperature range, and monthly CKD mortality (Figure 4c and Supplementary Table S5). Overall, across the four census regions, higher temperatures were associated with lower CKD mortality. In analyses of month to month temperature variability versus month to month changes in deaths (Supplementary Figure S3), larger temperature increases were likewise associated with reduced mortality. Notably, in the Midwest and the South, the temperature–mortality relationship displayed a U shaped pattern at higher temperature levels, higher mortality rates observed as temperatures became extreme.

As a sensitivity analysis, we repeated the temperature–mortality analyses for 2012–2023 using temperature estimates obtained from the NLDAS monthly forcing dataset. The monthly CKD crude mortality rate continued to display a clear seasonal pattern, generally increasing during colder months and declining during warmer months (Supplementary Figure S4a). The mean monthly crude mortality rate was highest in winter 1.32 (1.27–1.38) and lowest in summer 1.16 (1.12–1.20), with a significant difference between the two seasons (P = 0.000488) (Supplementary Figure S4b and c). An inverse association between temperature and CKD mortality was consistently observed across all four United States census regions, with Spearman correlation coefficients ranging from −0.35 in the West to −0.43 in the Northeast (all P < 0.001) (Supplementary Figure S4d). In the within-season analysis, month-to-month increases in temperature were significantly associated with reductions in CKD crude mortality during winter (r = −0.57, β = −0.0162, P < 0.001), whereas no significant associations were observed in spring, summer, or fall (Supplementary Figure S4e). These findings were consistent with the main analysis and supported the robustness of the observed inverse relationship between ambient temperature and CKD mortality despite the use of an alternative temperature data source.

Low Air Temperature and Rapid Temperature Changes are Associated with Increased Chronic Kidney Disease Mortality Risk

The Pearson dispersion statistics ranged from 1.151 to 1.455, indicating mild-to-moderate overdispersion; therefore, quasi-Poisson regression was used for the final analyses (Figure 5 and Supplementary Table S6). After adjustment for calendar month and long-term temporal trends, higher monthly temperature was associated with lower contemporaneous CKD mortality in the Northeast and West, while a delayed inverse association at lag 3 months was observed in the Midwest. In the South, temperature showed opposite associations at lag 0 and lag 2 months. However, the cumulative associations across lags 0–3 months were not statistically significant in any region, suggesting regional and lag-specific heterogeneity rather than a consistent overall cumulative effect.

Figure 5.

Lagged associations between monthly mean temperature and chronic kidney disease (CKD) mortality across the United States census regions. Points and error bars show relative risks (RRs) and 95% confidence intervals per 10°F increase in temperature. Upper panels present single-lag models, and lower panels present mutually adjusted lag-specific and cumulative effects over lags 0–3 months. All estimates were obtained from quasi-Poisson models adjusted for calendar month and long-term temporal trends. The vertical dashed line indicates RR = 1.0.

The distributed lag non-linear models revealed regional heterogeneity in the cumulative exposure–response relationship between monthly mean temperature and CKD mortality (Supplementary Figure S5a–d). In the Northeast and Midwest, cumulative mortality risk generally decreased as temperature increased above the region-specific median, whereas lower temperatures tended to be associated with elevated risk. This inverse pattern was most pronounced in the Midwest, although the confidence intervals widened substantially at the extremes of the temperature distribution. In the South, the cumulative exposure–response curve was comparatively flat around the median temperature, with only modest increases in risk at lower temperatures. By contrast, the West exhibited a dominant cold-related pattern, with the lowest estimated risk near the median temperature and higher risks at both lower and higher temperatures. Nevertheless, estimates at extreme temperatures were imprecise, as indicated by the wide 95% confidence intervals.

Lag-specific analyses suggested that the association with cold temperature was strongest in the same month and generally attenuated over subsequent months (Supplementary Figure S6a–d and Supplementary Table S7). In the Northeast, cold exposure at lag 0 was associated with an estimated relative risk of approximately 1.25 compared with the median temperature, after which the association declined toward or below the null. Similar, although less pronounced, immediate increases in risk were observed in the Midwest, South, and West. The estimated effects of hot temperature were generally weaker and fluctuated around the null across lags 0–3 months. No consistent delayed pattern for hot temperature was identified across regions, and most estimates were accompanied by confidence intervals that included unity.

The sensitivity analyses supported the overall robustness of these findings (Supplementary Figure S7a–d). Altering the maximum lag period from 2 to 6 months, changing the placement of exposure knots, or using alternative lag structures produced broadly comparable estimates for cold and hot temperatures. Cold-related cumulative risks generally remained above those associated with hot temperature in the Northeast and Midwest, while estimates in the South remained close to the null. In the West, the direction of the estimates was largely stable, although the model using a 6-month maximum lag generated a less precise estimate for hot temperature. Overall, the principal exposure–lag patterns were not materially altered by alternative model specifications, although uncertainty remained substantial at temperature extremes and under longer lag structures.

Discussion

This study leveraged the CDC WONDER database to comprehensively analyze CKD-related mortality in the United States from 1999 to 2023. We observed a sustained increase in CKD mortality, a pattern evident across geography, sex, race, and age groups. Moreover, increases in CKD mortality risk were significantly associated with declines in ambient temperature. These findings underscore the need for sustained, targeted interventions and for equitable allocation of healthcare resources.

Indeed, CKD is a chronic condition whose mortality partly reflects underlying socioeconomic circumstances.33,34 Over time, AAMRs for CKD and their rates of increase have been consistently higher in nonmetropolitan (rural) than in metropolitan (urban) areas, highlighting a substantial disparity in CKD risk between urban and rural populations in the United States. Numerous studies likewise report greater prevalence and mortality from chronic diseases in nonmetropolitan regions.33 Contributing factors likely include lower socioeconomic status, delayed or limited access to medical services, transportation barriers, and reduced uptake of preventive care in rural communities.34 Notably, the density of primary care providers have risen much more rapidly in urban than in rural areas, suggesting a widening access gap.35 These observations support policies that strengthen rural primary care capacity for CKD surveillance and prevention, an approach with clear potential to reduce CKD prevalence and mortality in nonmetropolitan populations.

In this study, from 1999 to 2023, male consistently exhibited higher AAMR from CKD. We hypothesize that this pattern is partly reflecting the potential influence of estrogen-mediated protection of the vascular endothelium. The renoprotective effects of estrogen are well documented; estradiol significantly antagonizes apoptosis of podocytes and mesangial cells36 and reduces the formation of reactive oxygen species.37 Additional sex differences likely reflect a combination of biological and social-behavioral factors. Epidemiologic studies indicate that male are more frequently exposed to CKD risk factors such as smoking, alcohol consumption, and diets high in salt and protein, and they demonstrate lower care-seeking and adherence, which is often associated with delayed diagnosis.38,39

Interestingly, this study shows that in recent years the AAMR for CKD among female has begun to increase, whereas the AAMR among men has declined. In addition, the South and Northwest began to exhibit decreases in AAMR after 2019. This timing suggests that COVID-19 may have been associated with shifts in CKD management.40 For example, the pandemic heightened attention to CKD patients with impaired innate immunity and multiple comorbidities, which may have facilitated earlier diagnosis and more effective treatment. The observed sex difference remains difficult to explain and is probably multifactorial. Notably, after the COVID-19 pandemic, some US populations experienced increases in alcohol and psychoactive substance misuse and in smoking, with higher prevalence reported among women.41,42

The present study found that lower ambient temperature was consistently associated with higher CKD mortality risk across all four United States census regions, representing the dominant temperature-related signal in our data. In the primary DLNM analysis, the cumulative RRs at the 5th percentile of monthly mean temperature were 1.065, 1.505, 1.178, and 1.309 in the Northeast, Midwest, South, and West, respectively; the corresponding RRs at the 95th percentile were 0.644, 0.808, 0.983, and 1.226. Because most 95% CIs included unity, these findings should be interpreted as exploratory regional patterns rather than definitive evidence of a consistent U-shaped association.

These findings are directly corroborated by existing evidence for cold-temperature effects on CKD.43–47 The Korean cohort reported individual-level hazard ratios of 1.049 per unit decrease in perceived temperature under routine winter conditions, rising to 1.837 during cold surge events—effects estimated via time-varying Cox regression in clinically confirmed CKD patients.48 By contrast, our analysis yielded a cumulative RR of 0.955 per 10°F increase (equivalent to approximately 1.047 per 10°F decrease) in the Northeast across a 0–3 month lag window, a magnitude broadly comparable in direction but not directly scalable given differences in exposure metric, temporal resolution, and outcome definition. A global analysis from the Global Burden of Disease study showed that CKD mortality associated with low temperature exceeds that associated with high temperature, largely because cold spells generally last longer than heat waves.43 The Brazilian study used hospitalization as its endpoint, whereas our outcome is mortality, suggesting that cold-related renal stress may extend from acute exacerbation to fatal outcomes.49 Notably, while these studies converge on the direction of cold-related risk, important quantitative differences exist across settings that merit discussion. The biological plausibility of cold-mediated renal harm is well established: cold exposure triggers peripheral vasoconstriction and redistribution of cardiac output, increases blood viscosity and systemic vascular resistance, and activates the renin–angiotensin–aldosterone system, collectively imposing acute haemodynamic stress on already-compromised renal vasculature in CKD patients.18,19

While the DLNM cumulative exposure–response curves suggested a modest upturn in risk at extreme high temperatures in the Midwest and South, which is consistent with a dominant cold-related pattern. This finding should be interpreted cautiously. The heat-effect signal was not the primary or pre-specified focus of this analysis, was not formally tested for statistical significance as a departure from monotonicity, and was based on monthly aggregated data that may over-smooth within-month temperature extremes. The existing literature on heat and kidney disease predominantly documents associations with hospitalization, emergency visits, or renal disease incidence rather than CKD mortality.44–47 Future studies with daily temperature resolution would be better positioned to formally characterize the heat-effect component of the temperature–CKD mortality relationship.

From a clinical perspective, these findings highlight the necessity of integrating environmental risk factors into CKD management. Nephrologists and primary care providers should proactively counsel vulnerable patients, particularly the elderly and those in rural areas, about the risks associated with extreme temperature exposure. Practical strategies may include utilizing early weather warning systems, optimizing indoor temperature control, and carefully monitoring hydration status and medication regimens including adjusting diuretics or antihypertensives during extreme heat or cold to prevent acute exacerbations of kidney function.

This analysis has limitations. First, the absence of individual-level data on baseline comorbidities precluded statistical adjustment for potential confounders, and we could not differentiate CKD stages or severity, and substantial heterogeneity likely exists across these patient groups. Consequently, this study is subject to potential ecological bias, meaning the correlations identified at the population level may not directly reflect individual-level pathophysiological processes. Accordingly, future research should aim to utilize individual-level health records and personal exposure data to perform more granular stratified analyses, which would help clarify the direct impact of temperature on specific patient subgroups and reduce the influence of ecological bias. Second, our reliance on monthly aggregated data from CDC WONDER precludes the analysis of daily temperature fluctuations and short-term lag windows.50 Future studies should aim to integrate more recent meteorological cohorts as updated databases become accessible. Third, other atmospheric environmental variables, such as atmospheric pressure, humidity, and wind speed, were not included in the study.48,51 Future analyses incorporating matched humidity data as an additional covariate in the DLNM linear predictor would strengthen causal inference.

Conclusion

In summary, this study provides a comprehensive assessment of CKD-related mortality from 1999 to 2023. CKD mortality increased over the study period and showed marked disparities by sex, race, age, geographic region, and urbanization level. In the temperature-related analyses, colder temperatures were directionally associated with higher cumulative CKD mortality risk across the United States census regions; however, the association reached statistical significance only in the Midwest, while estimates in the other regions were not statistically significant. These exploratory findings suggest a potential role of cold exposure in CKD-related mortality, particularly in the Midwest, but should be interpreted cautiously and validated in future studies using more granular exposure and individual-level data. Overall, the observed demographic and geographic disparities may help inform future CKD surveillance and population-level prevention strategies.

Funding Statement

This work was supported by the National Health and Medical Research Council (NHMRC) of Australia (Grant No. 2027965).

Data Sharing Statement

The datasets analyzed during the current study are available in the public database, https://wonder.cdc.gov/. The data used during the current study are available from the corresponding author on reasonable request.

Ethical Approval

This study used publicly available, de-identified, and aggregated mortality data from the CDC WONDER database. No individual-level or identifiable information was accessed, and no attempt was made to re-identify individuals. In accordance with the National Statement on Ethical Conduct in Human Research (Australia, 2023), Section 5.1.17(a) and (d), this research is eligible for exemption from ethics review given its exclusive use of publicly available, de-identified aggregate data; this was confirmed with the Ethics Committee of the Western Sydney Local Health District. Compliance with the Declaration of Helsinki is maintained through adherence to principles of data integrity, privacy protection, and non-maleficence applicable to population-level research.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.