Associations of air pollutant concentrations with longitudinal kidney function changes in patients with chronic kidney disease

This longitudinal cohort study investigated the associations of air pollutant exposures, including CO, NO, NO2, NOx, O3, PM10, PM2.5, and SO2, with long-term kidney function changes in patients with chronic kidney disease (CKD). We enrolled 447 CKD patients who took part in a universal hospital pre-ESRD care program during 2011–2015. The daily average air pollutant exposures and temperature were estimated for each patient, with different levels of air pollutant concentrations defined by 5-knot and restricted cubic spline function. Predicted annual estimated glomerular filtration (eGFR) slope values by one mixed model were considered as the study outcome. The average age of the study population was 77.1 ± 12.6 years, and the median annual eGFR decreased by 2.1 ml/min/1.73 m2 per year from 30 ml/min/1.73 m2 at baseline during a mean follow-up time of 3.4 years. The univariable and multivariable analyses revealed no significant linear and non-linear associations between 5-knot air pollutant concentrations and annual eGFR slope. In addition, the visualized spline effect plots show insignificant variation patterns in annual eGFR slope values with increased air pollutant concentrations. These results encourage more extensive studies to clarify the causal relationships and mechanisms of long-term specific air pollutant exposures and longitudinal kidney function change, especially in CKD populations.


Results
Patient and ambient air pollutant characteristics. After excluding patients who had fewer than three estimated glomerular filtration rate (eGFR) measurements in their first year after enrollment (n = 75), as well as those under the age of 20 (n = 1), those who lived outside of Pingtung city (n = 6), and those who lived in highaltitude areas (n = 11), a total of 447 patients with CKD were enrolled in this study (Fig. 1). The patients' mean age was 77.2; three-fourths of them had hypertension, and nearly half of them had diabetes mellitus. Over half of the eGFR measurements at enrollment were below 30.0 ml/min/1.73 m 2 , and most (over 80%) were accompanied by proteinuria ( Table 1). The air pollutant average concentrations of coarse particulate matter (particulate matter < 10 μm in aerodynamic diameter [PM 10 ] (64.31 ± 1.13) and PM 2.5 (35.75 ± 0.90) in the participants were much higher than the 2021 annual air quality standard set by the WHO 22 , which was 15 μg/m 3 and 5 μg/m 3 , respectively. The patient characteristics by quartile of air pollutants are provided in Supplementary Tables S1-S8). In general, those who were old with a high proportion of high educational years, low prevalence of hypertension, and a low proportion of high rank of urine protein creatinine ratio (UPCR) were significantly more likely to be exposed to the highest quartile of most air pollutants (CO, NO 2 , nitrogen oxide [NO x ], PM 10 , and PM 2.5 ) concentrations. However, the highest quartile group of sulfur dioxide (SO 2 ) concentration had a significantly lower proportion of the high educational years, more prevalent angiotensin-converting enzyme inhibitors or angiotensin II receptor blockers (ACEI/ARB) use, and a higher proportion of high-rank UPCR. Only a significant difference in hypertension prevalence was found between quartiles of nitrogen monoxide (NO) concentration, and only an educational level difference was found between quartiles of ozone (O 3 ) concentration.
Associations of air pollutant concentration with annual eGFR slope. During the observation period, 5,811 eGFR measurements were recorded showing that the median annual eGFR decreased by 2.1 ml/ min/1.73 m 2 per year with the interquartile range: 0.9, 3.4 reduction per year (Table 1, Fig. 2, and Supplementary  Fig. S1). The cubic spline plots imply non-linear relationships between long-term eGFR change and the quartile of air pollutant concentration ( Supplementary Fig. S2). In the modelings, we first determined 5-knot of air pollutants for all models because they produced smaller Akaike information criterion (AIC) values. There were no significant differences in the 5-knot groups of any air pollutant concentrations associated with annual eGFR slope after putting a group of air pollutant concentrations in the model (Table 2). Even adjusting for age, sex, educational level, diabetes mellitus, hypertension, cerebrovascular accident, congestive heart failure, ischemic heart disease, gout, ACEI/ARB, temperature, and log (UPCR), no selected air pollutant concentrations significantly associated with annual eGFR slope (The ranges of Type III sum of squares P-values: 0.05-0.26). After transforming the air pollutant concentrations by a 5-knot restricted cubic spline function, model one's results in Table 3 neither support significant non-linear relationships between air pollutants concentrations and longterm kidney function changes. Similar findings were demonstrated in the model, further adjusted by the selected covariates. The visualized spline effect plots show insignificant variation patterns in predicted annual eGFR slope values with increased air pollutants' concentrations ( Fig. 3).

Discussion
In this retrospective cohort study, we traced 447 CKD patients who regularly received care in one hospital. After a mean follow-up of 3.4 years, each patient's annual eGFR slope was estimated from 5811 consequent eGFR measurements. We surprisedly found that 5-knot linear and 5-knot restricted cubic splines models findings did not support that exposure to different CO, NO, NO 2 , NO X , O 3 , PM 10 , PM 2.5 , and SO 2 concentrations were significantly associated with varying annual eGFR slope among patients with advanced CKD. There is a shortage of evidence regarding the types of ambient air pollutants and which doses of those pollutants can lead to kidney function change, especially in the CKD population. Most previous studies have mainly focused on deterministic kidney outcomes. Lin et al. applied claim data to explore the effects of air pollution exposure on the incidence of CKD or ESKD in the general population, showing that high concentrations of air pollutants (NO, NO 2 , NOx, PM 2.5, and SO 2 ) increased CKD and ESKD intensities 18 . Another study used a clinical database with a more comprehensive confounder collection and demonstrated that PM 2.5 concentration Table 1. Patient characteristics. ppb, parts per million; ppm, parts per billion; μg/m 3 , micrograms per cubic meter; NSAIDs, non-steroidal anti-inflammatory drugs; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; eGFR, estimated glomerular filtration rate; CO, carbon monoxide; NO, nitrogen monoxide; NO 2 , nitrogen dioxide; NOx, nitrogen oxide; O 3 , ozone; PM 10 , ambient coarse particulate matter ( ≤ 10 μm); PM 2.5 , ambient coarse particulate matter ( ≤ 2.5 μm); SO 2 , sulfur dioxide. a The subjects' annual eGFR slope was estimated by one mixed model with random intercept and slope. www.nature.com/scientificreports/ significantly increased (by nearly 20%) kidney replacement therapy incidence in patients with advanced CKD 21 . Moreover, a recent study defined two consequent eGFR measurements which declined by more than 25% from the baseline within one year as kidney function deterioration. Almost all air pollutants, except O 3 , were significantly associated with kidney function deterioration in the CKD cohort 20 . In addition, another study observed that PM 2.5 and NO 2 exposure with a mean 2.3 years follow-up was associated with an increased incidence of first hitting eGFR reduction by more than 5 ml/min/1.73 m 2 /year 19 . Compared with the above studies, the present study recorded more eGFR measurements with a much longer duration, thus making it more appropriate for evaluating the associations between air pollution and long-term kidney function change. However, our results did not indicate that any air pollutant concentration was significantly associated with long-term kidney function deterioration. Indeed, our findings do not argue against the impacts of air pollutants on kidney health, but instead emphasize the importance of precisely arranging care resources in response to the influences of air pollution on regular kidney function measurement between various areas. Certain factors may explain the current findings, which were different from the observations forwarded by the previous cohort studies involving advanced CKD patients. First, oxidative stress induced by PM, NO, NO 2 , and O 3 , could lead to inflammatory reactions, DNA damage, and organ dysfunction. The degree of oxidative reactions is a critical factor affecting pathogenic mechanisms in CKD progression 23 . In addition, the different compositions of PM may affect the degree of oxidative stress. The concentration of transition metals and organic carbon compounds within PM may be different in various areas, which can induce different degrees of oxidative reactions 24 , thus leading to various observations on CKD progression. Second, and similar to results from animal studies 25,26 , most previous studies observed the association using short-term rather than long-term kidney function declines. Indeed, employing this approach may bring into the picture rapid kidney function decline caused by other acute conditions, such as trauma and the use of nephrotoxin drugs, which occur more often in urban than rural areas; this would result in a higher likelihood of the above-mentioned association being falsely confirmed. Thirdly, it is not impossible that our study population was too homogeneous to observe the impacts of high air pollutant concentration on kidney function progression. The lowest air pollutant concentrations in the reference groups were higher than those in previous studies [19][20][21] , which may have weakened the associations. Finally, air pollutant-induced cardiovascular dysfunction in the study might be modified by regular care. Through this, it is then possible to offset the impact of air pollutants on kidney function decline. More research is needed to understand the mechanisms of air pollutants and their effect on long-term kidney function progression.
Our study observed potential negative associations between NO x concentration and kidney function decline ( Table 2). The mechanisms were unclear. Endogenous NO is a common intercellular messenger in all vertebrates and has physiological functions in blood flow regulation, inhibiting platelet adhesion and neuron activity 27 . In the kidney, NO modulates hemodynamics, medullary perfusion, pressure natriuresis, tubuloglomerular feedback, tubular sodium reabsorption, and the kidney sympathetic nerve 28 . Therefore, we speculate that ambient air pollutants, including PM, NO, and O 3 , as potent oxidants could generate superoxides that interact with endogenous NO to form peroxynitrite, thus resulting in lipid peroxidation, protein oxidation, inactivation of enzymes, worsening of NO insufficiency in cells, and impaired vascular relaxation, finally leading to cardiovascular dysfunction. Cardiorenal syndrome is a concept used to connect cardiovascular dysfunction to kidney function deterioration 29 . Our observation offers one possible explanation for this connection.
This study has several strengths. First, a much longer follow-up time allowed us to observe more long-term kidney function change than was possible in other studies. In addition, the patients were regularly cared for by the CKD care team, which ensured that our findings were not affected, or only affected to a limited degree by care quality. Third, we evaluated kidney function change almost per 3 months for each patient, which could have decreased the influences of acute kidney injury on eGFR slope estimation. Fourth, the laboratory data were  www.nature.com/scientificreports/ examined in the same laboratory, thereby reducing experimental error. Finally, the high density of air quality stations within the study area increased the accuracy of individual air pollutant exposure estimations. Nonetheless, there are limitations in this study. First, there may be some potential unmeasured confounding factors (either time-fixed or time-vary) associated with kidney function deterioration, such as detailed outdoor activities, income, and convenience of access to healthcare, which may have partially influenced our estimations. Second, the residential addresses may not have comprehensively reflected individual air pollutant exposure due to migration. Since work is the main reason for migration, we expected that these influences would be minor because most participants were elderly and retired. Third, the study averaged air pollutant values rather than directly using hourly raw data, so some information regarding short-term air pollutant effects on kidney function changes may be lost. Finally, our findings were based primarily on elderly patients with advanced CKD, so the results may not be generalizable to the whole CKD population.
In conclusion, this longitudinal study revealed that ambient air pollutant concentrations were not significantly associated with long-term kidney function decline in patients with advanced CKD. Although our findings did not support the modern mainstream contention, the findings encourage more extensive studies to clarify the causal relationships and mechanisms between long-term specific air pollutant exposures and longitudinal kidney function change, especially in vulnerable populations.

Materials and methods
Study design and population. This retrospective cohort study enrolled stage 3b to 5 CKD patients who were new participants in a universal national pre-ESRD care program at the Pingtung Hospital from 2011 to 2015. The above-mentioned regional hospital and another 16 hospitals offer acute and chronic medical services and hospital admission for nearly 50,400 residents in southern Taiwan. The participants all lived at a longitude between 120°42′ E and 120°85′ E and a latitude between 22°02′ N and 23°86′ N. The care program reimbursed nephrologists who organized care teams to deliver multidisciplinary care for patients with uncontrolled proteinuria or eGFR less than 45 ml/min/1.73 m 2,30,31 . The care teams regularly evaluated the patient's blood pressure and kidney functions, as well as proteinuria, and offered diet and kidney care knowledge, all of which allowed us to study long-term kidney function change. The Institutional Review Board of Kaohsiung Medical University Hospital reviewed and approved the study protocol (KMUHIRB-E(I)-20,210,306). The Institutional Review Board of Kaohsiung Medical University Hospital agreed to our request that informed consent be waived because of the use of secondary data analysis with anonymous personal identification numbers. All the study procedures were conducted according to the principles of the Declaration of Helsinki.
Air pollutants. The study obtained air pollutant concentrations from 2011 to 2015, including CO, NO, NO 2 , NO x , O 3 , PM 2.5 , PM 10 , and SO 2 from five central (Pingtung, Daliao, Linyuan, Meinong, and Chaouzhou) and one local (YanZhou) air quality monitoring stations near Pingtung Hospital (Fig. 4). In addition, the ambient temperatures were collected as a covariate. The Environmental Protection Administrative of Taiwan and the Environmental Protection Bureau, Pingtung County regularly maintain the stations. The hourly air pollutant data recorded by the central stations were publicly available on websites, while the local station offered daily data through average hourly data 32 .
Kidney function measurements. Serum creatinine, urine creatinine, and urine protein were measured using the Jaff method with a Beckman Coulter DxC 700 AU Clinical Chemistry analyzer. The patients' serum creatinine values were obtained from care program enrollment to the end of June 2021 or before withdrawal from the program due to death, dialysis, kidney transplantation, or lost follow-ups. Kidney function was represented by an eGFR that was calculated through the age, sex, and serum creatinine values using a CKD-EPI equation 33 . Since every study subject had several eGFR measurements after the care program enrollment, the eGFR measurements over time were repeated and correlated.   www.nature.com/scientificreports/ Covariates. Several baseline covariates, including demographic features (age, sex, educational level, smoking, and alcohol consumption), comorbid conditions (diabetes mellitus, hypertension, cerebrovascular accident, congestive heart failure, ischemic heart disease, gout, and cancer), laboratory data (blood creatinine, urine protein, and urine creatinine), as well as medications with refillable prescriptions for more than 3 months (nonsteroidal anti-inflammatory drugs (NSAIDs) and ACEI/ARB) were also recorded. Patients' demographic characteristics and comorbid conditions were obtained from the records for reimbursement of the pre-ESRD care program. Educational level was classified into three categories (0, 1-12, and > 12 years). Comorbidities were selfreported and confirmed by clinical diagnosis by a trained nurse educator. The medication history was obtained by reviewing electronic medical records, whereby the user was defined by a cumulative prescription of over 3 months during the observed period. In addition, the ambient temperature was used as a covariate because it may confound the effect of air pollutants on health.

P-value
Statistical analysis. The distribution of the patient characteristics was described by mean and standard deviation for continuous variables with approximately normal distribution and median and interquartile range for those with no normal distribution. Differences between air pollutant concentration quartile groups were assessed by one-way ANOVA or the Kruskal-Wallis test. In addition, we applied counts and percentages for the categorical variables of patient characteristics and tested the differences in distributions between groups by Chi-square test or Fisher exact test. The raw air pollutants and ambient temperature data at the central air quality monitoring stations were represented by an hour on each measured date. We first averaged the hourly data, which served as daily concentrations for the central stations. Subsequently, we averaged the daily concentrations of each station during the observed period to represent long-term air pollutant concentrations and ambient temperature for each air quality monitoring station. Distances were calculated by longitude and latitude to understand the distance between patient residences and air quality monitoring stations. Each patient's long-term air pollutant concentration was estimated by employing inverse distance weighting with a distance power of 1 approach and the subjects were grouped by different percentiles of each selected air pollutant concentration. First, we examined the potential confounding factors by inspecting the differences in characteristic distributions between each air pollutant concentration quartile. Then, to explore relationships between long-term eGFR changes at different quartiles of air pollutant concentrations, longitudinal spline curves of eGFR change were constructed by cubic splines by setting 50 smoothness. The subjects' annual eGFR slope was estimated by one mixed model with random intercept and slope. In modelings, patients' air pollution concentration was grouped by 3 (10th, 50th, and 90th percentile) and 5 knots (5th, 27.5th, 50th, 72.5th, and 95th percentile) as suggested by previous literature 34 to explore the linear and non-linear relationships of air pollutant concentration with annual eGFR slope.
The univariable and multivariable general linear models were applied to explore the associations between each selected air pollutant concentration and annual eGFR slope. Linear and non-linear relationships were evaluated by putting knot grouping air pollution concentrations and air pollution concentrations with a restricted cubic spline function into models. The optimal fitted number of knots for modeling was evaluated based on the AIC (smaller is better). Log transformation was performed for UPCR due to a highly right-skewed distribution. We first developed model 1 by only putting a group of air pollutant concentrations. Then, model 2 further added age, sex, educational level, diabetes mellitus, hypertension, cerebrovascular accident, congestive heart failure, ischemic heart disease, gout, angiotensin-converting enzyme inhibitor/ angiotensin II receptor blocker, temperature, and log (urine protein creatinine ratio). Finally, the univariable and multivariable analyses were further Model 1  www.nature.com/scientificreports/  www.nature.com/scientificreports/