Influence of Weather and Air Quality on Clinical Outcomes in End-Stage Kidney Disease: A Retrospective Hospital-Based Study

Abstract

Environmental exposures are recognized as determinants of health, particularly in vulnerable populations such as hemodialysis patients, whose impaired kidney function may heighten susceptibility. This study assesses the association between meteorological and air-pollution factors and the risk of hospital admission and mortality in end-stage kidney disease patients, considering both acute peaks and sustained exposure through weekly and monthly averages. A retrospective cohort of 336 hemodialysis patients treated between 2016 and 2024 at Hospital Universitario Príncipe de Asturias (Madrid, Spain) was analyzed, including 563 admissions and 90 deaths. Time-to-event analyses were conducted separately for mortality and admissions. Cox proportional hazards and parametric survival models (Weibull, log-normal, log-logistic) were applied for mortality, while recurrent-event models (Andersen–Gill, Prentice–Williams–Peterson, frailty) used for admissions. Sulfur dioxide (SO₂) was consistently associated with increased risk in both acute and cumulative exposure models for mortality and admissions. Nitrogen dioxide (NO₂) predicted mortality only in long-term analyses, suggesting a cumulative effect. Solar radiation was linked to accelerated mortality in non-linear models, whereas atmospheric pressure appeared protective for admissions. These findings highlight the influence of environmental exposures, particularly SO₂, on adverse outcomes in hemodialysis patients and support incorporating environmental indicators into clinical risk prediction and public health strategies.

Plain Language Summary

Bad Weather, Dirty Air: How Pollution and Climate Increase the Risk of Hospitalization and Death for Kidney Failure Patients

Environmental exposures are increasingly recognized as serious health factors, especially for high-risk groups like hemodialysis patients. These individuals have severely compromised kidney function, meaning their bodies struggle to detoxify, potentially making them highly susceptible to environmental toxins and climate stressors. This study aimed to determine if changes in local air quality and weather patterns increase the risk of hospital admission and mortality for patients with end-stage kidney disease. We analyzed data from 336 hemodialysis patients treated in Madrid, Spain, between 2016 and 2024, examining 563 hospital admissions and 90 deaths. We looked at two types of exposure: acute (short-term pollution peaks) and sustained (weekly and monthly averages of environmental data). Our statistical analysis showed strong links between environmental factors and adverse health outcomes. Sulfur dioxide (SO₂) pollution was the most consistent threat, significantly increasing the risk of both hospitalization and death under both acute and sustained exposure scenarios. Nitrogen dioxide (NO₂) was linked to a higher risk of mortality only when patients were exposed to sustained long-term averages, suggesting a cumulative impact. Regarding weather, solar radiation was associated with accelerated mortality in certain models, while atmospheric pressure appeared to offer a protective effect against hospital admissions. These findings emphasize that environmental factors, particularly SO₂, are critical and preventable drivers of adverse clinical outcomes in the hemodialysis population. The study supports integrating environmental data and warning systems into clinical risk assessment and public health strategies to better protect this highly vulnerable group.

Keywords

chronic kidney disease health deterioration hemodialysis meteorology pollution survival models

Introduction

Chronic kidney disease (CKD) affects overall health through increased mortality, long-term morbidity and impaired quality of life. In this way, it affects functional capacity, social interaction, psychological well-being and life satisfaction, which in turn are influenced by access to treatment, socioeconomic factors and coexisting diseases.¹

At present, CKD is estimated to affect approximately 9.5% of the global population.² A significant proportion of these individuals eventually progress to end-stage kidney disease (ESKD), a condition that requires renal replacement therapy to sustain life. Among the available options, hemodialysis remains the most commonly used worldwide.³ Patients undergoing hemodialysis represent a particularly vulnerable subgroup, with a markedly increased risk of cardiovascular events, infections, hospitalizations, and premature mortality.⁴ The healthcare expenditures associated with ESKD are disproportionately high, largely due to the complexity of care and the need for continuous, resource-intensive treatment. Although the median global spending per CKD patient stands at $353, costs rise steeply in the dialysis-dependent population.⁵

Epidemiological studies have explored how climatic conditions may influence the general well-being of the population. The World Meteorological Organization (WMO) emphasizes that changes in weather patterns can affect people’s daily lives, impacting both their physical health and mental well-being. Integrating health surveillance systems with meteorological data has been suggested as a proactive approach to more effectively identifying at-risk populations.⁶

On the other hand, air pollution is now 1 of the most pressing public health challenges. These environmental exposures can cause considerable harm, contributing to respiratory infections, strokes, cardiovascular disease, and cancer. According to the World Health Organization (WHO), air pollution is among the most significant preventable causes of illness and death globally.⁷

Given the current high prevalence and clinical relevance of ESKD, along with the growing recognition of the health effects of environmental factors on medically complex patients, this study hypothesizes that specific environmental variables contribute to health deterioration in patients undergoing hemodialysis. In particular, we analyze the associations between air pollutant content and meteorology and increased rates of hospital admissions and mortality in patients at the Hospital Universitario Príncipe de Asturias in Madrid (Spain), considering both acute exposure to short-term peaks and sustained exposure through weekly and monthly averages of environmental variables.

Related Works

A growing body of evidence has investigated the environmental determinants of morbidity and mortality among patients with ESKD undergoing hemodialysis. Air pollution and climate-related exposures such as extreme heat and humidity have emerged as significant contributors to adverse outcomes in this particularly vulnerable population.

Numerous studies have demonstrated associations between fine particulate matters (PM) and adverse health outcomes in this population. In 2022, Xi et al reported that short-term elevations in PM_2.5 were associated with increases in cardiovascular disease (CVD) incidence, CVD-related mortality, and all-cause mortality among U.S. hemodialysis patients, with hazard ratios ranging from 1.03 to 1.05 per 10 μg/m³ increment.⁸ Xi et al also found that long-term PM_2.5 exposure was significantly associated with cardiovascular risk in this population.⁹ Further, in 2023, Xi et al. observed that short-term exposure to PM_2.5 correlated with abnormalities in inflammatory biomarkers, supporting a mechanistic link through systemic inflammation.¹⁰

In a nationwide Chinese case-control study, Lou et al found that each 10 μg/m³ increase in PM_2.5 and PM₁₀ was associated with a relative increase in daily mortality among dialysis patients.¹¹ Similarly, Wyatt et al showed that short-term PM_2.5 exposure increased the risk of cardiovascular-related hospital admissions and 30-day readmissions following discharge.¹²

Other pollutants have also shown harmful associations. In 2020, Jung et al found that long-term SO₂ exposure increased mortality risk in dialysis patients,¹³ while Huh et al reported in 2022 that elevated ambient carbon monoxide (CO) was linked to increased all-cause mortality, especially in older high-risk patients.¹⁴ Additionally, Hamroun et al conducted a nationwide French registry-based study and found that exposure to multiple air pollutants significantly raised all-cause mortality among over 90 000 dialysis patients.¹⁵

Climate-related stressors also contribute substantially to adverse outcomes. Remigio et al demonstrated that extreme heat events (EHEs) were associated with higher risks of hospital admission and death in hemodialysis patients.¹⁶ Xi et al observed that elevated ambient temperatures increased cardiovascular risk and healthcare utilization.¹⁷ In 2024, Blum et al identified that extreme humid-heat exposure raised mortality risks among urban-dwelling dialysis patients in the U.S.¹⁸ Remigio et al highlighted that inclement weather—such as heavy rain, snow, and hurricanes—increased the likelihood of missed dialysis appointments, potentially leading to life-threatening complications.¹⁹ Similarly, Blum et al found that hurricane exposure significantly elevated 30-day mortality risk.²⁰

Recent studies suggest that environmental exposures may have synergistic effects. In 2022, Remigio et al found that concurrent exposure to EHEs and elevated air pollution significantly increased all-cause mortality in ESKD patients in the Northeastern United States.²¹ Conversely, some environmental exposures may offer protective benefits. Yoon et al showed that increased sunlight exposure was inversely associated with all-cause mortality in dialysis patients.²²

Methodology

Hemodialysis Patient and Environmental Databases

This retrospective observational study was conducted using anonymized data from a cohort of patients undergoing chronic hemodialysis at the Hospital Universitario Príncipe de Asturias, located in Alcalá de Henares, Madrid, Spain. The study covers an 8-year period, from January 5, 2016, to May 31, 2024. The participant selection process is illustrated in the STROBE flow diagram (Figure 1). From an initial cohort of 405 patients with ESKD on hemodialysis screened at the Hospital Universitario Príncipe de Asturias, 69 individuals were excluded. Specifically, 47 patients were excluded due to incomplete medical records, and 22 were excluded because their residence fell outside the coverage area of the environmental sensors. Consequently, a final analytic sample of 336 patients with complete clinical and environmental data was included in the complete-case analysis, accounting for 90 deaths and 563 independent hospital admissions analyzed after the filtering process.

Figure 1.

STROBE flow diagram of participant selection.

The analysis integrates 3 distinct databases, as shown in Figure 2: 1 containing clinical events and outcomes, other compiling meteorological data, and the third 1 recording levels of air pollution. Clinical data were sourced directly from hospital records and digitized for analysis.

Figure 2.

Data pipeline for research: from source integration to analytical insights.

Meteorological variables were retrieved from the Visual Crossing Weather Data Platform,²³ which utilizes a global weather engine processing billions of hourly and sub-hourly observations from more than 100,000 worldwide observation stations, including satellite and maritime sources. This engine analyzes every report for errors, missing information, and other anomalies to ensure the accuracy of the data. For this study, the platform’s engine aggregated and interpolated records from multiple local weather stations—including data from leading meteorological services such as NOAA—to produce a single, coherent weather report specifically for the hospital’s catchment area.

Air quality data were obtained from the Air Quality Historical Data Platform.²⁴ This platform compiles official, real-time pollutant measurements from the local monitoring network of the Comunidad de Madrid (Spain). The system utilizes machine learning algorithms to perform real-time data consistency checks, comparing values between neighboring stations to identify and exclude potentially defective sensors. For this study, environmental data were spatially matched to the hospital’s catchment area (Alcalá de Henares and surrounding municipalities), ensuring the exposure assessment reflects the specific urban and climatic conditions of the study population.

All data were anonymized in accordance with applicable data protection regulations and ethical standards. The geographic scope of environmental data corresponds to the hospital’s catchment area in the Madrid region, where all patients most likely live.

Data Processing and Variables Considered in the Study

Following the initial data extraction, a total of 1360 clinical events were identified. To ensure that only independent and clinically distinct events were analyzed, a filtering process was applied to exclude any admissions that occurred within 7 days of a previous hospital admission. This step aimed to minimize the inclusion of readmissions related to the same clinical episode and better isolate discrete events. In addition, each event was checked to ensure it occurred within the patient’s active dialysis treatment period. Events that fell outside the documented dialysis timeframe were excluded to avoid misattribution and to maintain consistency in the population under study.

Data from multiple sources—including clinical records, meteorological databases, and air pollution registries—were integrated and consolidated into a single dataset. During preprocessing, particular attention was paid to identifying and resolving inconsistencies across sources. A detailed review of each variable was conducted to detect quality issues such as missing, implausible, or outlier values. A conservative complete-case analysis approach was adopted for the environmental datasets. While multiple imputation was considered, the authors opted for a complete-case analysis given that the proportion of missing data for most key variables was low—for instance, 3.5% for PM_2.5 and 3.7% for NO₂ in the mortality dataset, and 3.8% and 2.5% respectively in the admissions dataset. Although SO₂ presented a higher missingness rate (up to 18.6%), preliminary comparisons of baseline clinical characteristics between included and excluded cases showed no significant differences, suggesting a minimal risk of selection bias. This approach ensures that the reported associations are based strictly on recorded environmental measurements. Despite the reduction in the total sample size, the findings remained statistically significant, reinforcing the robustness of the observed environmental signals.

In addition to assigning environmental variables according to the exact date of each clinical event (acute exposure), we also constructed datasets in which weekly and monthly averages of all meteorological and air pollution variables were calculated for each patient. These aggregated datasets allowed us to investigate the influence of sustained exposure on clinical outcomes, complementing the analyses based on short-term peaks.

The variables included in the study were structured as follows:

- Dependent variables:

• Hospital admission (based on date of admission)

• Mortality (based on date of death)

- Independent variables:

• Air pollution indicators: PM₂.₅, PM₁₀, nitrogen dioxide (NO₂), sulfur dioxide (SO₂), and ozone (O₃).

• Meteorological variables: maximum, minimum and mean temperature, relative humidity, precipitation, atmospheric pressure, wind speed, solar radiation, and UV index

Environmental variables were assigned according to the date of each clinical event (admission or death), or to the date of censoring in the absence of an event, and additionally through the patient’s cumulative exposure reflected in weekly and monthly averages of the same variables. All numerical variables were normalized prior to analysis in order to facilitate the performance of the statistical models and to ensure comparability across different scales.

Survival Analysis

The steps taken to analyze patient survival and to assess the influence of environmental factors on clinical outcomes are detailed below. Two separate analyses were conducted, 1 focusing on the risk of death and the other on the risk of hospital admission among hemodialysis patients. The statistical approach relied on survival models selected and adapted according to the characteristics of each event type and data structure.

In this context, a multi-model approach was adopted to ensure the robustness of the results and to minimize potential algorithmic bias. Given the uncertainty regarding the functional form of the relationship between environmental stressors and ESKD outcomes, we compared linear proportional hazards models (Cox Proportional Hazards Model and Penalized Cox with LASSO regularization) with non-linear parametric models (Weibull, Log-logistic, Log-normal). This diversity allows for the detection of both constant hazard ratios and accelerated failure times. For recurrent hospitalizations, models were selected to account for different aspects of event history: the Andersen–Gill (AG) model for overall intensity, the Prentice–Williams–Peterson (PWP) model for the effect of preceding events, and the shared frailty model to control for unobserved individual-level heterogeneity. Findings that demonstrated consistency across these different mathematical frameworks were considered the most reliable indicators of environmental influence.

Each analysis was performed twice: first using environmental variables at the exact day of the clinical event (acute exposure), and then using the aggregated datasets with weekly and monthly averages (sustained exposure). This dual approach enabled comparison between short-term and long-term environmental effects.

Death Survival Analysis

To evaluate the impact on mortality, the aforementioned linear and parametric models were implemented using Python. Prior to model selection, the proportional hazards assumption was tested using Schoenfeld residuals, to assess whether the covariate effects remained constant over time. These models were fitted separately for acute exposures and for sustained exposures (weekly and monthly averages).

Admission Survival Analysis

For the analysis of hospital admissions, which involve recurrent events, the analysis was conducted using R. Model selection among the recurrent event frameworks (AG, PWP, and Frailty) was guided by multiple criteria: the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC), visual inspection of Martingale residuals to assess model fit, formal goodness-of-fit tests, and verification of the proportional hazards’ assumption using the Schoenfeld residuals test. After comparing the different approaches, the final analysis was based on the results obtained from the shared frailty and PWP models, which showed the best performance and interpretability in the context of this study.

As with mortality, both acute exposure and sustained exposure (weekly and monthly averages) were analyzed to capture potential differences between short-term peaks and cumulative effects of environmental factors.

Results

Cohort Description

The study cohort consisted of 336 patients with a mean age of 68.1 years, with a gender distribution of 64.9% male and 35.1% female. The clinical profile was characterized by a high prevalence of comorbidities, with 86.9% of the population suffering from hypertension and 41.5% diagnosed with diabetes mellitus (37.3% Type 2 and 4.2% Type 1). Chronic Obstructive Pulmonary Disease (COPD) was present in 8.6% of the patients. The median dialysis vintage was 2.6 years, with individual histories ranging from recent initiation to over 40 years of treatment. Data on lifestyle factors such as smoking and alcohol consumption were partially limited in the retrospective records, with 50.3% and 80.1% of entries being unspecified, respectively. Among those with recorded data, 50.9% were identified as ex-smokers and 25.1% as active smokers.

The final dataset comprised 336 patients, with a total of 563 hospital admissions and 90 deaths. Tables 1 and 2 show a summary of the environmental variables whose influence on deaths and admissions of renal patients was analyzed. No extreme outliers were found to distort the distribution of the co-variables. Figure 3 displays the age distribution at hospital admission (right panel) and at death (left panel). Admissions were most frequent between ages 60 and 85, with a peak around 70, while deaths were concentrated in older age groups, suggesting that fatal events occurred predominantly among the elderly population. A comparison of the above distributions with the age distribution of the population studied, shown in Figure 4, reveals that the drop in the number of deaths and admissions that occurs at around 75 years of age corresponds to a lower number of patients of that age.

Table 1.

Summary of Environmental Variables in the Mortality Analysis Dataset.

Variable	Mean (SD)	Median (IQR)	Min – Max	Missing data (n, %)
PM₂.₅	63.81 (26.50)	60.00 (43.00-72.00)	21.00-222.00	12 (3.5%)
PM₁₀	33.47 (23.13)	28.00 (19.00-34.00)	6.00-92.00	12 (3.5%)
O₃	32.44 (11.88)	30.00 (25.00-40.00)	3.00-65.00	13 (3.7%)
NO₂	18.57 (8.28)	16.00 (14.00-23.00)	2.00-56.00	13 (3.7%)
SO₂	2.41 (1.56)	2.00 (1.00-3.00)	1.00-10.00	53 (15.3%)
Maximum temperature	21.97 (7.53)	20.90 (16.90-26.90)	5.50-39.90	0 (0.0%)
Minimum temperature	9.81 (5.77)	9.40 (6.45-13.10)	−3.90 to 25.20	0 (0.0%)
Mean temperature	15.78 (6.80)	15.10 (11.00-19.55)	1.30-32.00	0 (0.0%)
Humidity	63.68 (18.85)	72.00 (49.55-74.70)	20.80-96.60	0 (0.0%)
Precipitation	3.19 (6.72)	0.00 (0.00-1.31)	0.00-29.93	0 (0.0%)
Wind speed	40.06 (22.43)	33.80 (25.20-46.60)	10.80-146.50	0 (0.0%)
Atmospheric pressure	1018.06 (5.96)	1017.40 (1014.90-1020.00)	996.20-1035.70	0 (0.0%)
Solar radiation	184.44 (95.15)	161.30 (124.30-268.75)	24.30-374.30	0 (0.0%)
UV index	6.85 (2.44)	6.00 (6.00-9.00)	1.00-10.00	0 (0.0%)

Table 2.

Summary of Environmental Variables in the Admissions Analysis Dataset.

Variable	Mean (SD)	Median (IQR)	Min – Max	Missing data (n, %)
PM₂.₅	55.52 (22.11)	53.00 (40.00-65.00)	18.00-222.00	31 (3.8%)
PM₁₀	25.54 (18.13)	23.00 (15.00-32.00)	5.00-344.00	20 (2.5%)
O₃	33.77 (13.38)	33.00 (26.00-42.00)	1.00-109.00	20 (2.5%)
NO₂	20.32 (9.62)	19.00 (13.00-26.00)	2.00-62.00	20 (2.5%)
SO₂	2.63 (1.83)	2.00 (1.00-4.00)	1.00-10.00	150 (18.6%)
Maximum temperature	21.52 (9.07)	20.80 (13.90-28.80)	1.30-41.00	0 (0.0%)
Minimum temperature	9.01 (7.03)	8.40 (3.40-14.50)	−5.70 to −25.20	0 (0.0%)
Mean temperature	15.15 (8.11)	13.70 (8.45-21.70)	−0.60 to 32.40	0 (0.0%)
Humidity	60.21 (20.22)	60.90 (42.85-77.15)	20.00-98.20	0 (0.0%)
Precipitation	1.26 (4.81)	0.00 (0.00-0.30)	0.00-69.21	0 (0.0%)
Wind speed	38.25 (19.58)	34.60 (24.80-43.90)	10.80-120.60	0 (0.0%)
Atmospheric pressure	1018.08 (6.38)	1017.70 (1013.75-1021.90)	993.70-1038.20	0 (0.0%)
Solar radiation	195.06 (103.39)	204.80 (103.25-292.50)	6.30-377.20	0 (0.0%)
UV index	6.90 (2.75)	7.00 (5.00-9.00)	0.00-10.00	0 (0.0%)

Figure 3.

Comparison of Age Distributions for Hospital Admissions and Death Events.

Figure 4.

Population Age Distribution.

All patients were residents of the Alcalá de Henares health area (primarily within postal codes 28801 to 28807, 28810, 28816, and 28880), ensuring a homogeneous socioeconomic and geographic background. The cohort exhibits high geographical stability due to their mandatory clinical routine, as patients must undergo hemodialysis sessions at the Hospital Universitario Príncipe de Asturias every 48 to 72 hours. This frequent and consistent presence within the same environmental monitoring zone, combined with the lack of long-distance commuting in this population, significantly reduces the potential for exposure misclassification and justifies the use of regional meteorological and pollutant data as a representative proxy for their daily exposure.

Environmental Influence on Deaths

Following the exclusion of missing data, the analysis comprised 292 observations, with 66 related to deaths.

Two complementary approaches were performed: an analysis of acute exposure to daily environmental conditions at the time of each clinical event, and an analysis of sustained exposure through weekly and monthly averages.

Acute Exposure Analysis

The proportional hazards assumption was confirmed for all covariates through Schoenfeld residuals (p-value > .05 for each), endorsing the application of proportional hazards models and their variations.

The standard Cox Proportional Hazards model found that SO₂ is a major risk factor for mortality (hazard ratio (HR) = 1.62, 95% confidence interval (CI): 1.30–2.02, p-value < .005), yielding similar outcomes in the penalized Cox model utilizing LASSO regularization (HR = 1.28, 95% CI: 1.03–1.60, p-value = .02). Conversely, parametric models that account for non-linear relationships (log-logistic, log-normal, and Weibull) showed that both SO₂ and solar radiation had a significant impact on mortality. In particular, SO₂ reliably appeared as a risk factor in these parametric models (time ratio < 1 because of reversed coding, p-value < .005), whereas solar radiation correlated with a hastened time to death (p-value < .03). These results, shown in more detail in Table 3, emphasize SO₂ as a strong environmental indicator of mortality in individuals with CKD, and indicate that solar radiation might also influence survival duration in this group. The rest of pollution values have not shown significance in these analyses.

Table 3.

Statistically Significant Results Obtained After Analysis of the Influence of Environmental Variables on Patient Survival.

Model	Covariate	Coefficient (β)	Effect estimate (e^β)	95% confidence interval (CI)	p-value
Cox	SO₂	0.48	1.62	1.30-2.02	<.005
Cox + LASSO	SO₂	0.25	1.28	1.03-1.60	.02
Log-logistic	SO₂	−2.48	0.08	0.02-0.34	<.005
Log-logistic	Solar radiation	−4.43	0.01	0.00-0.44	.02
Log-normal	SO₂	−2.36	0.09	0.02-0.42	<.005
Log-normal	Solar radiation	−5.91	0.00	0.00-0.17	<.005
Weibull	SO₂	−1.99	0.14	0.05-0.38	<.005
Weibull	Solar radiation	−3.43	0.03	0.00-0.74	.03

In the Cox model and the Cox model with LASSO regularization, effect estimates are expressed as hazard ratios (HRs): values greater than 1 indicate an increased risk of death, while values less than 1 suggest a protective effect. In contrast, the coefficients (β) for the log-logistic, log-normal, and Weibull models are derived from Accelerated Failure Time (AFT) models, where effect estimates represent acceleration factors (time ratios). In this context, values less than 1 indicate a shorter time to event (ie, higher risk), and values greater than 1 indicate a delayed event (ie, lower risk).

Due to the consistent significance of SO₂ in all models, initial survival comparisons were conducted using Kaplan-Meier curves stratified by SO₂ exposure levels (⩽2 ppb vs >2 ppb). These unadjusted curves demonstrated a marked difference in survival probabilities between groups, with higher SO₂ exposure linked to reduced survival times (log-rank test, p-value < .005). (See Figure 5, left panel).

Figure 5.

Left: Kaplan–Meier Survival Curves Stratified by SO₂ Exposure Levels (⩽2 ppb vs >2 ppb). Right: Adjusted Survival Curves Derived from the Cox Proportional Hazards Model, Stratified by SO₂ Exposure Levels.

To further explore this association while controlling for potential confounders, a Cox proportional hazards model was fitted adjusting for the environmental and meteorological covariates. Adjusted survival curves derived from this model—representing predicted survival probabilities at average covariate values—confirmed the increased mortality risk associated with elevated SO₂ exposure (HR = 1.62, 95% CI: 1.30-2.02, p-value < .005) (See Figure 5, right panel).

Although several models were evaluated (standard Cox, penalized Cox, and parametric models), adjusted survival curves are presented only for the standard Cox model for clarity and interpretability.

Sustained Exposure Analysis (Weekly and Monthly Averages)

For the monthly dataset, proportional hazards assumptions were satisfied across variables. In contrast, in the weekly dataset PM₁₀, maximum temperature, and minimum temperature violated proportionality, and their interactions with time were included in models.

No covariates showed significant linear associations with mortality in Cox models, including penalized versions. However, parametric non-linear models (Weibull, log-normal, log-logistic) consistently identified NO₂ and SO₂ as significant predictors. For monthly averages, NO₂ (p-value < .005, time ratio = –2.82) and SO₂ (p-value = .02, time ratio = –1.10) emerged as risk factors. Weekly averages yielded similar findings, with NO₂ (p-value < .005, time ratio ≈ –2.3) and SO₂ (p-value = 0.01-0.02, time ratio ≈ –2.0-–2.5) repeatedly significant in parametric models.

Taken together, the sustained exposure analyses reinforce the role of SO₂ as a robust predictor of mortality risk and indicate that NO₂ may exert additional long-term effects not fully captured by daily exposure models.

Environmental Influence on Admissions

Acute Exposure Analysis

Model comparison based on the AIC and BIC revealed that the PWP model achieved the lowest values in both criteria (AIC = 2992.573, BIC = 3052.473), indicating the best balance between goodness-of-fit and model complexity. The frailty model showed the second-best AIC (4762.975), suggesting a reasonable fit; however, its BIC was substantially higher (5455.156), reflecting the penalty for added complexity. The AG demonstrated poorer fits (AIC = 5018.078; BIC = 5077.977) (See Table 4).

Table 4.

AIC and BIC Comparison of Recurrent Event Models: AG, PWP and Frailty.

Criterion	AG	PWP	Frailty
Akaike information criterion (AIC)	5018.078	2992.573	4762.975
Bayesian information criterion (BIC)	5077.977	3052.473	5455.156

Visual inspection of Martingale residuals shown in Figure 6 supported these findings. The PWP model exhibited more concentrated and symmetrically distributed residuals, whereas the AG model showed greater dispersion and more extreme negative values, indicative of potential lack of fit. Additionally, the frailty model demonstrated a statistically significant improvement over the standard Cox model without random effects (p-value < .001), confirming the presence of unobserved heterogeneity among patients.

Figure 6.

Comparison of Martingale Residuals Between AG and PWP Models.

Proportional hazards assumptions were assessed using Schoenfeld residual tests. Most covariates met the assumption satisfactorily, with the exception of O₃, which showed a marginal p-value (.050). The global test was not significant (p-value = .058), indicating that the overall model does not violate the proportionality assumption.

Given the combination of statistical criteria, residual diagnostics, and proportionality tests, further analyses focused on the PWP and frailty models. These models capture complementary aspects of recurrent event data: the ordering of events and patient-level heterogeneity, respectively.

After identifying the 2 survival models that best captured the data structure, PWP and the frailty models, the following results were obtained including the environmental influence on hospital admissions in CKD patients. After removing records with missing data, 631 observations were retained, including 533 events.

The PWP model demonstrated moderate predictive ability, with a concordance index of 0.547. Model fit was supported by the likelihood ratio test (χ² = 24.99, df = 14, p-value = .03), Wald test (χ² = 23.66, df = 14, p-value = .03), and the logrank score test (χ² = 25.3, df = 14, p-value = .03), suggesting a significant overall effect of the included covariates. However, the robust score test (χ² = 17.34, p-value = .2) advised caution in interpreting these results due to potential within-cluster correlations.

Analysis of individual covariate effects in the PWP model indicated that exposure to SO₂ was associated with an increased risk of admission (HR = 1.067, 95% CI = 1.007-1.131, p-value = .028). Conversely, atmospheric pressure showed a protective association (HR = 0.975, 95% CI = 0.951-0.999, p-value = .031).

The frailty model, which incorporates random effects to account for patient-level variability, confirmed the presence of significant unobserved heterogeneity (frailty variance = 1.28; SD = 1.13). This suggests that there are individual differences in risk beyond those explained by measured environmental factors. In line with the PWP model, the frailty model also found a significant association for SO₂ (HR = 1.14, 95% CI = 1.01-1.30, p-value = .02) and pressure (HR = 0.79, 95% CI = 0.65-0.96, p-value = .02).

Sustained Exposure Analysis (Weekly and Monthly Averages)

In contrast to the acute exposure analysis, the evaluation of sustained exposure did not reveal consistent associations between environmental variables and hospital admissions. For both weekly and monthly averages, no pollutant or meteorological variable showed significance across survival models (AG, PWP and frailty). These negative findings suggest that short-term peaks may be more relevant than long-term cumulative exposure in triggering hospital admissions among hemodialysis patients.

Discussion

This study demonstrates the substantial influence of environmental factors on clinical outcomes in patients with ESKD undergoing dialysis. Among pollutants, SO₂ consistently emerged as a strong predictor of adverse outcomes, showing associations with both hospital admissions and mortality across acute (daily) and sustained (weekly, monthly) exposure windows. This relationship remained robust across multiple statistical approaches, including Cox proportional hazards models, LASSO penalized models, and nonlinear parametric models. For long-term exposures, NO₂ was identified as a significant predictor of mortality, though no associations were seen in acute analyses. This suggests a cumulative rather than immediate effect, contrasting with SO₂, which exerted both short- and long-term influence. Additionally, solar radiation was strongly linked to mortality, indicating that greater exposure accelerates time to death in this vulnerable population. In contrast, higher atmospheric pressure appeared protective, being inversely associated with hospital admission risk. These findings were consistent in both the PWP and frailty models, with the latter also revealing substantial patient heterogeneity, indicating that unobserved individual-level factors significantly contribute to readmission risk.

Regarding particulate matter, a notable finding in our study is the lack of a statistically significant association between PM_2.5 and adverse clinical outcomes, which stands in contrast to previous literature. For instance, studies by Xi et al⁸ in the U.S. and Lou et al¹¹ in China have identified PM_2.5 as a major driver of cardiovascular mortality and hospitalizations in hemodialysis patients. The discrepancy between our findings and these large-scale studies may be attributed to several factors. First, the specific demographic and clinical profile of our Mediterranean cohort—such as age distribution and primary CKD etiologies—might differ from those in North American or Asian populations. Second, the concentration levels and chemical composition of particulate matter in the Madrid region may vary significantly from those in more heavily industrialized or densely populated global regions. Our results suggest that in certain urban environments, other pollutants like SO₂ may act as more sensitive triggers for acute decompensation in ESKD patients than PM_2.5, emphasizing the need for locally tailored environmental health assessments.

Importantly, no consistent associations were found between sustained environmental exposures and admissions, suggesting that admissions are more sensitive to short-term peaks of pollution or meteorological stressors, while mortality risk reflects both acute and cumulative exposures.

Despite methodological strengths, including robust statistical modeling and consistent findings, some limitations warrant consideration. The relatively small sample size (336 patients) restricts generalizability; larger multi-center cohorts could enhance statistical power and uncover additional associations. Furthermore, missing data, particularly for SO₂, may have reduced model precision. Implementing dedicated air quality sensors near the hospital would enable more continuous and accurate pollutant monitoring in future studies. Similarly, the reliance on ambient monitoring data as a proxy for personal exposure may introduce misclassification, as indoor environments or individual mobility were not tracked. However, this risk is mitigated by the high geographical stability and mandatory clinical routine of the hemodialysis cohort, who undergo treatment at the same reference center every 48 to 72 hours. Additionally, the dual focus on acute versus sustained exposures, while informative, introduces methodological complexity. Furthermore, the survival models focused primarily on environmental predictors; while the cohort is clinically well-defined, the lack of adjustment for individual-level clinical confounders such as diabetes or smoking status should be noted. Finally, while this study identifies significant associations with all-cause mortality and total hospitalizations, the retrospective nature of the data did not allow for a systematic classification of events by specific clinical etiology (eg, cardiovascular or respiratory).

Future research will address these limitations by expanding the cohort across multiple centers and utilizing personal monitoring to refine exposure assessments. Improved pollutant monitoring infrastructure would reduce missing data and improve exposure assessment. More detailed classification of causes of death and hospital admissions could facilitate more precise correlations with environmental factors. Further, explicitly distinguishing between acute and cumulative effects of pollutants could clarify whether they act primarily as short-term triggers of decompensation (eg, admissions) or exert long-term cumulative damage (eg, mortality). Lastly, investigating the acute effects of extreme weather events, such as heatwaves or storms, on the renal health of chronically ill patients is warranted, especially given the increasing impact of climate change on chronic disease epidemiology.

Conclusion

This study provides strong evidence that environmental exposures significantly influence outcomes in patients with ESKD undergoing dialysis. SO₂ emerged as the most consistent predictor, associated with both mortality and hospital admissions across acute and sustained exposure windows. In contrast, NO₂ was linked only to long-term mortality, suggesting a cumulative effect. Beyond pollutants, solar radiation and atmospheric pressure also modulated outcomes, underscoring the multifactorial nature of environmental impacts on renal health.

The application of advanced survival and recurrent event statistical models allowed for a thorough and more accurate assessment of these relationships, accounting for both time-dependent covariates and recurrent hospital admissions. While acute exposures showed clearer associations with admissions, long-term exposures were primarily linked to mortality outcomes, underscoring the need to jointly assess short-term triggers and cumulative effects. Despite these strengths, the study acknowledges limitations related to sample size and environmental data granularity, underscoring the need for further investigations with larger cohorts and more refined exposure assessment to confirm and expand upon these findings.

Overall, these findings emphasize the need for stricter air quality surveillance in areas hosting vulnerable populations, such as those with ESKD, and advocate for the integration of environmental data into clinical risk assessment protocols. Incorporating environmental considerations into public health policies could contribute to improved patient outcomes, better resource allocation in healthcare systems, and more precise preventive strategies for chronically ill populations.

Footnotes

Acknowledgements

This work is part of the project “Prevention of serious pathological events in hemodialysis patients by non-invasive continuous monitoring of vital signs and analysis of circular biomarkers (ALLPREVENT),” PMPTA23/00033, which has been funded by the Instituto de Salud Carlos III (ISCIII) within the Program of R&D projects linked to personalized medicine and advanced therapies.

ORCID iDs

María Santamera-Lastras

Melisa Granda

Alberto Garcés-Jiménez

Francisco Javier Bueno-Guillén

Patricia Martínez-Miguel

Diego María Rodríguez-Puyol

José Manuel Gómez-Pulido

Ethical Considerations

This study was conducted in accordance with the principles of the Declaration of Helsinki and current Spanish legislation on biomedical research. The project was reviewed and approved by the Research Ethics Committee for Medicines (Comité de Ética de la Investigación con Medicamentos, CEIm) of Hospital Universitario Príncipe de Asturias (Madrid, Spain) with approval code OE 29/2025.

Consent to Participate

This study was granted a waiver of informed consent by the Research Ethics Committee for Medicines (Comité de Ética de la Investigación con Medicamentos, CEIm) of Hospital Universitario Príncipe de Asturias (Madrid, Spain). The project involves a retrospective observational analysis using previously collected clinical data that were pseudo-anonymized prior to analysis. No intervention or contact with patients was required. Given the large study period (2016–2024), the high number of eligible patients, and the likelihood that many were deceased due to their underlying condition, obtaining individual consent was not feasible. As the study poses minimal risk and has potential social benefit, the Ethics Committee approved the exemption from informed consent.

Author Contributions

MSL: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Supervision, Visualization, Writing–original draft, Writing–review & editing. MG: Validation, Writing–review & editing. AGJ: Conceptualization, Investigation, Supervision, Validation, Writing – review & editing. FJBG: Resources, Validation. PMM: Supervision, Validation, Writing–review & editing. DMRP: Conceptualization, Funding acquisition, Project administration, Supervision, Validation, Writing–review & editing. JMGP: Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Validation, Writing–review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work forms part of the project “Prevention of serious pathological events in hemodialysis patients by non-invasive continuous monitoring of vital signs and analysis of circular biomarkers (ALLPREVENT)” (PMPTA23/00033), funded by the Instituto de Salud Carlos III (ISCIII) under the Program of R&D projects linked to personalized medicine and advanced therapies.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data presented in this study are not publicly available due to restrictions imposed by the Research Ethics Committee for Medicines (Comité de Ética de la Investigación con Medicamentos, CEIm) of Hospital Universitario Príncipe de Asturias (Madrid, Spain). The approved study protocol does not authorize the public sharing of raw patient data, as this could compromise participant confidentiality.

References

Jacob

Kannan

Raveendran

Wilson

Quality of life in chronic kidney disease: a community perspective using world health organization quality of life: BREF questionnaire. Int J Community Med Public Health. 2020;7(10):4105.

Bello

Okpechi

Levin

, et al. An update on the global disparities in kidney disease burden and care across world countries and regions. Lancet Glob Health. 2024;12(3): e382-e395.

Liyanage

Ninomiya

Jha

, et al. Worldwide access to treatment for end-stage kidney disease: a systematic review. Lancet. 2015;385(9981):1975-1982.

United States Renal Data System. Epidemiology of Kidney Disease in the United States. Published online 2023. https://adr.usrds.org

Honeycutt

Segel

Zhuo

Hoerger

Imai

Williams

Medical costs of CKD in the Medicare population. J Am Soc Nephrol. 2013;24(9):1478-1483.

World Meteorological Organization. World Meteorological Organization Home. WMO. Accessed May 21, 2025. https://wmo.int

World Health Organization. World Health Organization Home. WHO. Accessed May 21, 2025. https://www.who.int

Richardson

Kshirsagar

, et al. Effects of short-term ambient PM2.5 exposure on cardiovascular disease incidence and mortality among U.S. Hemodialysis patients: a retrospective cohort study. Environ Health. 2022;21(1):33.

Richardson

Kshirsagar

, et al. Association between long-term ambient PM2.5 exposure and cardiovascular outcomes among US hemodialysis patients. Am J Kidney Dis. 2022;80(5):648-657.e1.

10.

Richardson

Kshirsagar

, et al. Associations of air pollution and serum biomarker abnormalities in individuals with hemodialysis-dependent kidney failure. Kidney360. 2023;4(1):63-68.

11.

Lou

Zhang

Shi

, et al. Associations between short-term exposure of ambient particulate matter and hemodialysis patients death: a nationwide, longitudinal case-control study in China. Sci Total Environ. 2022;852:158215.

12.

Wyatt

Kshirsagar

, et al. Association of short-term exposure to ambient PM_2.5 with hospital admissions and 30-day readmissions in end-stage renal disease patients: population-based retrospective cohort study. BMJ Open. 2020; 10(12):e041177.

13.

Jung

Park

Kim

, et al. Long-term effects of air pollutants on mortality risk in patients with end-stage renal disease. Int J Environ Res Public Health. 2020;17(2):546.

14.

Huh

Kim

Yoon

, et al. Ambient carbon monoxide correlates with mortality risk of hemodialysis patients: comparing results of control selection in the case-crossover designs. Kidney Res Clin Pract. 2022;41(5):601-610.

15.

Hamroun

Génin

Glowacki

, et al. Multiple air pollutant exposure is associated with higher risk of all-cause mortality in dialysis patients: a French registry-based nationwide study. Front Public Health. 2024;12:1390999.

16.

Remigio

Jiang

Raimann

, et al. Association of extreme heat events with hospital admission or mortality among patients with end-stage renal disease. JAMA Netw Open. 2019;2(8):e198904.

17.

Wettstein

Kshirsagar

, et al. Elevated ambient temperature associated with increased cardiovascular disease-risk among patients on hemodialysis. Kidney Int Rep. 2024;9(10):2946-2955.

18.

Blum

Feng

Tuholske

, et al. Extreme humid-heat exposure and mortality among patients receiving dialysis. Am J Kidney Dis. 2024;84(5):582-592.e1.

19.

Remigio

Song

Raimann

, et al. Inclement weather and risk of missing scheduled hemodialysis appointments among patients with kidney failure. Clin J Am Soc Nephrol. 2023;18(7):904-912.

20.

Blum

Feng

Anderson

Segev

McAdams-DeMarco

Grams

ME.

Hurricanes and mortality among patients receiving dialysis. J Am Soc Nephrol. 2022;33(9): 1757-1766.

21.

Remigio

Raimann

, et al. Combined effects of air pollution and extreme heat events among ESKD patients within the Northeastern United States. Sci Total Environ. 2022;812:152481.

22.

Yoon

Kim

Lee

, et al. The impact of sunlight exposure on mortality of patients with end stage renal disease. Sci Rep. 2019;9(1):2230.

23.

Visual Crossing Weather Data. Visual Crossing. Accessed January 24, 2025. https://www.visualcrossing.com/

24.

Air Quality Historical Data Platform. Air Quality Historical Data Platform. Accessed January 24, 2025. https://www.iqair.com