Introduction

Long-term exposure to nitrogen oxides (NOx) and nitrogen dioxide (NO2) have continued to pose a substantial threat to global public health, even in regions with low air pollution concentrations1. NOx consisting mainly of nitric oxide (NO) and NO2 serves as a critical precursor in photochemical reactions leading to ozone (O3) formation and secondary inorganic aerosols2,3, posing significant threats to human health. High NOx emissions from light-duty diesel vehicles in Europe in 2013 were responsible for up to 10,000 premature deaths4, supporting that reducing NOx emissions could have substantial health implications. The control of NOx emissions might also contribute to the mitigation of population exposure to fine particulate matter (PM2.5) and O35. Nevertheless, research on the health effects of NOx remains limited and the broad disease spectrum that might be affected by NOx has not been fully characterized. An extensive literature review conducted in the preliminary stage of our study indicated that long-term exposure to NOx and NO2 might be associated with mortality from certain diseases, mainly cardiovascular diseases, chronic respiratory conditions, and pulmonary carcinoma6,7 (Supplementary Fig. S1). However, the distinct impacts of NOx on cause-specific mortality, particularly regarding diseases beyond cardiorespiratory conditions, remain inadequately explored. Therefore, there is a critical gap in understanding the full spectrum of health effects associated with NOx exposure, necessitating a comprehensive examination to elucidate their broader implications and explore potential exposure-response relationships.

Here, we analyze a cohort data comprising over 500,000 participants in the UK Biobank (UKB) from 2006 to 2022. Time-varying covariates Cox proportional hazards models are employed to examine the impact of atmospheric NOx exposure on deaths from all-cause, non-accidental cause, accidental cause, 15 main causes and 103 specified subcategories diseases. Furthermore, we explore the potential exposure-response relationships and conduct stratification analyses by age, sex and regional characteristics. We find that long-term exposure to NOx is associated with increased mortality risks of all-cause, non-accidental diseases, and 20 specific diseases. Associations between NOx and cause-specific mortality are generally linear. Sex and residential area seem to be the effect modifiers of NOx-mortality associations. The findings may inform the development of stricter air quality regulations to control atmospheric NOx levels, and provide evidence for developing adaptive strategies to mitigate the social inequalities in public health.

Results

Descriptive data

Baseline characteristics of the participants and air pollutants exposure are shown in Table 1. A total of 502,040 participants with an average age of 56.5 [standard deviation (SD): 8.1] years at the baseline were included in the final analyses, with 54.4% being female. During a median follow-up of 13.7 years, 44,090 individuals died from all-causes, and 1297 participants were lost to follow-up. Compared to the study subjects who survived until the end of follow-up, deceased participants were more likely to be older, male, white, and urban residents, with a higher proportion of them possessing lower household income, having lower educational attainment, and belonging to retired demographics. The prevalence of smoking, obesity, and low physical activity levels was also higher in the deceased group.

Table 1 Baseline characteristics of participants and air pollutants exposure in UK Biobank cohort
Full size table

Temporal and spatial trends in nitrogen oxides

During 2006–2022, the average concentrations (SD) of NOx and NO2 were 24.5 (11.8) μg/m3 and 16.6 (6.6) μg/m3, respectively. The concentrations of NOx exhibited an overall declining trend but manifested large spatial variations (Supplementary Figs. S2–S4). NOx exhibited the highest yearly average exposure in 2006 (38.9 μg/m3). Supplementary Fig. S5 illustrates the Spearman’s correlation coefficients (rs) among air pollutants. NOx exhibited positive correlations with PM2.5, particulate matter with aerodynamic diameter ≤ 10 μm (PM10) and sulfur dioxide (SO2) (rs = 0.434–0.777).

Effects of nitrogen oxides on mortality from major categories of diseases

Table 2 shows effects of NOx on mortality from 18 broad categories of diseases. Hazard ratios (HRs) for all-cause mortality associated with per 10 μg/m3, SD (11.75 μg/m3) and interquartile range (IQR) (14.16 μg/m3) increases in NOx were 1.036 [95% confidence interval (CI): 1.024, 1.049], 1.043 (95% CI: 1.028, 1.058), and 1.052 (95% CI: 1.034, 1.070), respectively. Regarding non-accidental diseases, the HR was 1.032 (95% CI: 1.019, 1.045) with each 10 μg/m3 increase in NOx. Across the 15 major cause-specific categories, elevated hazards per 10 μg/m3 increase in NOx levels were observed for respiratory diseases (HR: 1.139; 95% CI: 1.090, 1.190), mental and behavioral disorders (HR: 1.096; 95% CI: 1.002, 1.199), and circulatory diseases (HR: 1.043; 95% CI: 1.015, 1.072). Moreover, the notably increased risks associated with NOx were observed for those who died from other symptoms, signs and abnormal findings, as well as external causes. Supplementary Table S1 shows effect estimates of NO2 on mortality from major disease categories. We found significant associations between NO2 and mortality from aforementioned diseases, except for other symptoms, signs and abnormal findings.

Table 2 Hazard ratios (95%CIs) for mortality from major categories of diseases linked to per 10 μg/m3, SD, and IQR increases in NOx
Full size table

Effects of nitrogen oxides on mortality from specific subcategories of diseases

Among 103 subcategories of cause-specific mortality, NOx was linked to mortality risk of 15 types of disease (Fig. 1 and Supplementary Table S2). The NOx-related diseases comprised other peripheral vascular diseases (ICD-10: I73), hypertensive heart disease (I11), phlebitis and thrombophlebitis (I80), accidental poisoning by and exposure to noxious substances (accidental poisoning) (X40-X49), event of undetermined intent (Y10-Y34), fibrosis and cirrhosis of liver (K74), other chronic obstructive pulmonary disease (COPD) (J44), intentional self-harm (suicide) (X60-X84), COVID-19 (U07.1), malignant neoplasm of lip, oral cavity and pharynx (C00-C14), malignant neoplasm of bladder (C67), unspecified dementia (F03), pneumonia (J12-J18), malignant neoplasm of esophagus (C15), and chronic ischemic heart disease (I25), with HRs ranging from 1.075 to 1.342 for each 10 μg/m3 increase in NOx. For NO2, we observed that mortality from 14 types of disease were significantly associated with this pollutant, with HRs ranging from 1.137 to 1.790 (Supplementary Table S2).

Fig. 1: Hazard ratios (95% CIs) for mortality by cause-specific diseases, linked to per 10 μg/m3 increase in NOx.
figure 1

Time-varying Cox regression was used to assess the associations between NOx and mortality, with the adjustment of age, sex, ethnicity, body mass index, education, occupation, smoking status, alcohol consumption, physical activity, residential area, and household income. Dots represent the estimated hazards ratios for mortality risk associated with per 10 μg/m3 increase in NOx, and horizontal lines denote corresponding 95% confidence intervals. Two-sided P values less than 0.05 were considered as statistically significant. Asterisk (*) denotes P < 0.05. Source data for Fig. 1 are provided as Source Data 1.

Full size image

Effects of nitrogen oxides on mortality from common subcategories of diseases

Supplementary Table S3 shows effect estimates for several common causes of death, which are often examined in previous studies on health impact of air pollution. Statistically significant associations were identified for mortality from hypertension (I10-I15), COPD (J40-J44), chronic lower airway disease (J40-J47), and ischemic heart disease (I20-I25). The HRs ranged from 1.041 to 1.250 for each 10 μg/m3 increase in NOx, and from 1.071 to 1.554 for each 10 μg/m3 increase in NO2. We found no association of NOx or NO2 with mortality from lung cancer (C33-C34), cardiac arrest, heart failure (I30-I51), cerebrovascular disease (I60-I69), or asthma (J45-J46).

Exposure-response relationships between nitrogen oxides and mortality

Figure 2 depicts exposure-response curves between NOx and all-cause and cause-specific mortality. A linear and monotonical increase associated with NOx was found for mortality from all-cause, non-accidental, circulatory diseases, respiratory diseases, esophageal cancer, chronic ischemic heart disease, phlebitis and thrombophlebitis, pneumonia, and other COPD, without discernible thresholds (Supplementary Tables S4–S5). Conversely, non-linear associations between NOx and mortality were observed for outcomes of mental and behavioral disorders, hypertensive heart disease, COVID-19, intentional self-harm, and other symptoms, signs, and abnormal findings. Supplementary Fig. S6 indicates similar exposure-response curves between NO2 and cause-specific mortality as NOx.

Fig. 2: Exposure-response relationships between NOx and mortality.
figure 2

Associations were assessed utilizing a restricted cubic spline function with 3 knots in time-varying Cox regression, with the adjustment of age, sex, ethnicity, body mass index, education, occupation, smoking status, alcohol consumption, physical activity, residential area, and household income. The minimum concentration of NOx was 0.68 μg/m3, which served as the reference. The lines represent the point estimates, and shaded areas indicate corresponding 95% confidence intervals. The dashed vertical lines are 25th and 75th percentiles of NOx (16.27 μg/m3 and 30.42 μg/m3). Source data for Fig. 2 are provided as Source Data 2.

Full size image

Subgroup-specific effect estimates of nitrogen oxides on mortality

Figure 3 compares the subgroup-specific risk estimates of all-cause and cause-specific mortality with long-term exposure to NOx, stratified by age, sex, and regional characteristics. A greater effect estimate of NOx on mortality from mental and behavioral disorders was observed in the females, with HR of 1.229 (95% CI: 1.074, 1.407) per 10 μg/m3 increase, compared to the males (P value of 0.037 for the difference test). For residential area, we observed a notably heightened risk of NOx on mortality from fibrosis and cirrhosis of liver in urban areas, but greater risks for pneumonia and accidental poisoning in rural settings. Additionally, we observed that the impact of NO2 on mortality from accidental poisoning was more pronounced in the youth (≤ 65 years) compared to those aged > 65 years (Supplementary Tables S6–S8).

Fig. 3: Hazard ratios (95% CIs) for mortality associated with per 10 μg/m3 increase in NOx, stratified by age, sex and residential area.
figure 3

a Models (time-varying Cox regression) adjusted for sex, ethnicity, body mass index, education, occupation, smoking status, alcohol consumption, physical activity, residential area, and household income. b Models adjusted for age, ethnicity, body mass index, education, occupation, smoking status, alcohol consumption, physical activity, residential area, and household income. c Models adjusted for age, sex, ethnicity, body mass index, education, occupation, smoking status, alcohol consumption, physical activity, and household income. Dots represent the estimated hazards ratios for mortality risk associated with per 10 μg/m3 increase in NOx, and vertical lines denote corresponding 95% confidence intervals. The P values are the difference tests between subgroups by the Z-test. Two-sided P values less than 0.05 were considered as statistically significant. Source data for Fig. 3 are provided as Source Data 3.

Full size image

Sensitivity analyses

Supplementary Tables S9–S18 show the results of sensitivity analyses. Effect estimates remained generally similar when implementing double-pollutant model, integrating additional covariates related to history of chronic diseases, excluding certain individual-level covariates, excluding participants with follow-up durations of less than two years, applying time-varying exposures on extended scale of 5-year average for NOx, limiting to a maximum of ten year follow-up period, employing multiple imputation, excluding participants who exhibited residential mobility, and excluding outliers of NOx concentrations. After excluding the COVID-19 period, effect estimates of NOx on deaths from mental and behavioural disorders (including unspecified dementia), circulatory diseases (including other peripheral vascular diseases), lip, oral cavity and pharynx cancer, bladder cancer, and fibrosis and cirrhosis of liver became statistically insignificant. After false discovery rate (FDR) correction, positive associations remained between NOx exposure and hypertension, hypertensive heart disease, ischemic heart disease, phlebitis and thrombophlebitis, COPD, chronic lower airway disease, COVID-19, and intentional self-harm (all with q values < 0.1, Supplementary Tables S2–S3).

Discussion

This is a large-scale study to conduct a pioneering and comprehensive assessment of the impact of long-term exposure to ambient NOx on mortality, encompassing both all-cause and a wide range of specific causes, thereby expanding our knowledge on the spectrum of mortality associated with nitrogen oxides and uncovering additional potential disease mortality links. Overall, long-term exposure to NOx exhibited statistically significant associations with increased mortality risk of all-cause, non-accidental diseases, and 20 cause-specific diseases. The exposure-response associations between NOx and mortality were generally linear. Elevated mortality risks were observed for all-cause diseases, mental and behavioural disorders, respiratory diseases, unspecified dementia, hypertensive heart disease, COVID-19, and intentional self-harm, even at low NOx concentrations.

Consistent with previous epidemiological studies, our findings substantiated positive associations between long-term exposure to NOx and mortality risks across several common diseases, including all-cause diseases, non-accidental diseases, circulatory diseases, respiratory diseases, ischemic heart disease, and COPD. For example, a previous study in the UK demonstrated positive associations of long-term exposure to NOx with all-cause mortality, cardiovascular mortality, and respiratory mortality, reporting HRs of 1.05 (95% CI: 1.04, 1.06), 1.06 (95% CI: 1.03, 1.08), and 1.10 (95% CI: 1.06, 1.15) per IQR increase in NOx, respectively6. Furthermore, a cohort of Norwegian men found that each 10 μg/m3 increase in NOx was linked to an elevated risk of ischemic heart disease mortality (HR: 1.08; 95% CI: 1.03, 1.12), but not for cerebrovascular diseases8. Another study from China found NOx to be an independent risk factor for mortality from respiratory diseases and COPD9. Regarding associations between NOx and cancer, current evidence was inconsistent. For instance, in line with results of other cohort studies conducted in Israel and China10,11, we found null effect estimates of NOx on overall cancer or lung cancer. However, positive associations were reported in another UK study6. The discrepancies observed in these studies may be due to differences in the characteristics of the study populations, exposure assessment, methodologies, and adjustment for covariates.

Meanwhile, we found significant positive associations of NOx and NO2 with mortality from psychiatric disorders, unspecific dementia, pneumonia, COVID-19, and intentional self-harm, which were consistent with previous studies. For instance, a Danish nationwide study showed that per 10 µg/m3 increase in NO2 was associated with higher mortality from dementia (HR: 1.05; 95% CI: 1.03, 1.07), and psychiatric disorders (HR: 1.23; 95% CI: 1.19, 1.28)12. And compared to our findings, a nationwide cohort study in the UK reported a lower estimated effect (HR: 1.09; 95% CI: 1.03, 1.16) for pneumonia mortality for each IQR increase in NO213. A sub-regional study in England estimated that each 1 µg/m3 increment in NOx was associated with a 1.5% increase in COVID-19 deaths (odds ratio: 1.015; 95% CI: 1.011, 1.019)14. An IQR increase in NO2 was associated with an HR of 1.18 (95% CI: 1.03, 1.34) for COVID-19 death in Denmark15, while an HR of 1.33 (95% CI: 1.09, 1.64) was observed for intentional self-harm mortality in South Korea16.

In a more granular exploration of diseases, we uncovered several underexplored conditions intricately associated with long-term exposure to NOx and NO2, such as lip, oral cavity and pharynx cancer, esophageal cancer, bladder cancer, hypertensive heart disease, other peripheral vascular diseases, phlebitis and thrombophlebitis, hepatic fibrosis and cirrhosis, and accidental poisoning. Although these common but underexplored conditions represent only a small fraction of total mortality, our study offers early evidence supporting their potential role in disease progression. Previous studies have demonstrated that long-term exposure to NOx and NO2 elevated mortality risks of hypertension, ischemic heart disease and heart failure17,18,19. Notably, hypertensive heart disease is intrinsically linked to hypertension and frequently co-occurs with ischemic heart disease or progresses to heart failure20,21. For phlebitis and thrombophlebitis, its subcategory superficial venous thrombosis may increase the risk of deep vein thrombosis, which can potentially lead to life-threatening pulmonary embolism22. Likewise, deep vein thrombosis is associated with adverse outcomes, including stroke, acute coronary syndrome, and increased mortality23. A retrospective case-control study found that long-term exposure to PM2.5, PM10, and O3 may increase the risk of lower extremity deep vein thrombosis23. Another national cohort study conducted in China observed that 62% of the mortality attributable to air pollution occurred within three months following ischemic stroke, elucidating the potential association with acute complications of stroke (e.g., respiratory infection and venous thromboembolism), which may contribute to decreased early survival rate24. As for the liver cirrhosis, a cohort study conducted in UK found that NOx and NO2 increased the risk of non-alcoholic fatty liver disease25, which typically manifests as an early precursor to liver fibrosis and cirrhosis. Overall, further investigation is needed to verify these associations and to understand the underlying mechanisms.

We revealed diverse features in the shapes of exposure-response curves across diseases, suggesting that NOx and NO2 may affect people with various diseases differently. The large European cohorts within the ELAPSE project similarly reported a non-linear association between NO2 exposure and all-cause mortality26. In a previous study conducted in Canada, a supra-linear exposure-response relationship was found for non-accidental mortality27. Eum et al. documented a linear association between NO2 and non-accidental mortality, but non-linear relationships for cardiovascular and respiratory mortality28. Nevertheless, our study provides compelling evidence that there is no safe level of low-dose NOx exposure for mortality from all-cause, mental and behavioural disorders, respiratory diseases, unspecified dementia, hypertensive heart disease, COVID-19, and intentional self-harm. Elevated risks of mortality from these diseases were observed despite concentrations falling below the 25th percentiles of NOx (16.27 μg/m3). Consequently, it is imperative to implement more stringent air pollution control standards or regulations, as mitigating exposure has the potential to reduce death burden.

Stratified analyses revealed that age, sex, and residential area may modify the association between exposure and cause-specific diseases mortality. In age-stratified analyses, we observed that the association between NO2 and accidental poisoning was more pronounced in the younger group (≤ 65 years). An analysis on accidental injury mortality in the US indicated that, compared to those aged 15–24 years, the relative risk of accidental poisoning was strongest among individuals aged 35–54, followed by those aged 25–34 and 55–7429. Age and NO2 may act synergistically on the risk of dying from accidental poisoning.

Our sex-stratified findings indicated a heightened susceptibility among females to the effects of NOx and NO2 concerning mental and behavioral disorders. A study conducted in South Korea found that adverse effects of NO2 on cognitive function were significantly higher among females residing in urban areas compared to males30. Song et al. observed a positive correlation between NO2 and the levels of biomarkers indicating neurological damage, with the estimated impact being notably elevated in females relative to males31. The underlying explanation may be related to estrogen fluctuation among females and difference in hypothalamic-pituitary-adrenal (HPA) axis functioning between males and females, genetic susceptibility, and socio-psychological factors between males and females32.

The area-stratified analysis revealed regional variations in the mortality risks of NOx and NO2, particularly for patients with pneumonia, fibrosis and cirrhosis of liver, and accidental poisoning. We observed a stronger risk of air pollutants for those with fibrosis and cirrhosis of liver among urban dwellers, possibly due to differing lifestyles between urban and rural areas. Previous studies reported that particulate matters could increase the risk of advanced fibrosis in patients with metabolic dysfunction-associated fatty liver disease, which is closely linked to urban lifestyles (e.g., sedentary behavior, urbanization, and Westernized dietary patterns)33,34. We speculated that NOx and NO2 may act similarly to particulate matter by worsening the outcome of patients with pre-existing liver conditions. For pneumonia and accidental poisoning, previous studies identified an elevated risk of Klebsiella pneumoniae oropharyngeal carriage in rural areas35. A study investigating urban-rural disparities in accidental injuries found that the hospitalization rates resulting from accidental poisoning were significantly higher among rural populations36. This may be related to disparities in access to healthcare service. However, these studies did not assess the potential associations between NOx and NO2 and mortality due to pneumonia, fibrosis and cirrhosis of liver, and accidental poisoning. To date, research on pollutant exposure and mortality remains notably limited in examining population characteristics and is marked by considerable heterogeneity. Considering the paucity of epidemiological evidence on susceptible populations and the yet incompletely elucidated etiology, future endeavors necessitate further comprehensive investigations.

It is presently hypothesized that oxidative stress and inflammation are the predominant biological pathways connecting air pollution with health outcomes. The main manifestation involves inducing systemic inflammation and targeting multiple systems or organs, resulting in functional impairment and endocrine disorder, thus contributing to long-term sub-clinical effects and elevating the risks of various disease-related mortality. As there is a lack of evidence on the underlying biological mechanism for NOx, we mainly discussed the potential mechanism of its primary components (NO2 and NO). Animal-model analyses conducted by Poynter et al. revealed that NO2 exposure enhanced and prolonged inflammatory responses and airway hyperresponsiveness, characterized by significantly augmented eosinophilic inflammation and enhanced terminal bronchiolar lesions, which extended markedly into the alveoli37. The COPD mouse model further elucidated that long-term exposure to NO2 may foster the accumulation of inflammatory cells via oxidative stress, signifying a progression from acute to chronic pulmonary injury38. Additionally, exposure to NO2 could disrupt calcium ion equilibrium via apoptotic signaling pathways, actin cytoskeletal reorganization, and negatively impact myocardial contractility and vascular dilation, ultimately resulting in apoptosis of myocardial cells39. Throughout the course of central nervous system (CNS) and psychiatric disorders, the accumulation of abnormal proteins (Aβ42 and α-synuclein), heightened levels of inflammatory cytokines, and elevated concentrations of reactive oxygen species served as mediators for the adverse impacts of air pollution on the CNS, ultimately leading to structural brain impairment and marking the early pathological stages of neurodegenerative diseases40,41. Moreover, systemic inflammation, oxidative stress, and structural damage to the brain caused by air pollution can contribute to the onset of psychological disorders and heighten the risk of accidental injuries16.

In contrast to the harmful effects of NO2 on human health, NO plays an unexpectedly significant role in physiological processes. NO is a critical signaling molecule involved in numerous physiological processes, such as blood pressure regulation, vasodilation, neurotransmission, and immunity42. Despite its role as an atmospheric pollutant, carefully controlled inhaled NO therapy produces selective pulmonary vasodilation, effectively increasing arterial oxygen levels, and providing an effective treatment for pulmonary hypertension in both adults and children43,44. However, dysregulated NO signaling is a key characteristic in the pathogenesis of numerous major diseases, including cardiovascular disease, diabetes, and cancer42. Although NO rapidly reacts with atmospheric oxygen to form NO2, its impact on human health remains significant and should not be overlooked. A multidisciplinary approach will be crucial in the future to comprehensively investigate the role of NOx within both environmental and biological frameworks.

Our study has several strengths, including a large sample size, longitudinal study design, long-term follow-up, and access to high-quality data covering a wide range of diseases and potential covariates. As a result, it provides robust epidemiological evidence on associations of NOx with the full spectrum of diseases, uncovers novel mortality risk linkages, and advances the existing knowledge on NOx-related health effects—an effort that might be challenging in smaller cohorts. Additionally, the atmospheric pollution exposure assessment method used in this study is highly reliable, as it accounts for individuals’ residential mobility throughout the follow-up period and incorporates a long temporal scope, allowing for a more accurate evaluation of individual exposure levels.

This study is subject to certain limitations. Firstly, UKB is based on volunteer participants rather than random selection, which introduces the potential for volunteer bias. Nevertheless, owing to the large sample size advantage of the UKB, the impact of the healthy volunteer effect on the validity of exposure-outcome associations is likely minimal45. Secondly, regarding generalizability and external validity of the results, caution is needed when interpreting our findings beyond the UKB cohort due to differences in populations, geographic regions, demographic characteristics, and exposure levels. Thirdly, we exclusively utilized residential geographical coordinates to evaluate individual exposure to ambient air pollution, without accounting for indoor air pollution, commuting patterns, workplace exposures, and other relevant factors. Further research into exposure to pollutants from various sources is needed, given the continuous evidence of a significant disease burden linked to indoor air pollution46,47. Fourthly, observational study designs inherently struggle to eliminate all potential confounding factors or reverse causality. Nonetheless, this study employed a prospective cohort design and sensitivity analyses that enhanced the ability to effectively assess relationships between nitrogen oxides exposure and cause-specific mortality. Lastly, although we examined the impact of NOx and NO2 on various diseases, an in-depth analysis of disease progression, such as through multi-state modeling, was not conducted. We propose an in-depth examination of disease progression in future endeavors to support earlier intervention strategies and, consequently, reduce the air pollution-related disease burden.

Methods

Study design and participants

This study used data from the UKB, with ethical approval granted by the Northwest Multiple Center Research Ethics Committee, and all participants provided written informed consent. This large-scale national prospective cohort study recruited over 500,000 individuals aged 37–73 encompassing 22 centers across England, Wales, and Scotland during 2006–2010. The UKB dataset provides a rich resource of information on its participants, including demographics from questionnaire surveys, anthropometric measurements, biochemical samples, genomic research data, and longitudinal tracking of participant health outcomes48,49. For this study, we began with 502,356 participants from the UK Biobank. We excluded the participants with missing residential addresses (n = 13) and mortality information (n = 303). A total of 502,040 participants were included in the final analysis (Supplementary Fig. S7).

Air pollution assessment

The annual average concentrations of atmospheric pollutants were obtained from the UK’s Department for Environment Food & Rural Affairs (DEFRA, https://uk-air.defra.gov.uk/data/pcm-data), which functions as a high-resolution data platform for near-surface atmospheric pollution. DEFRA primarily utilizes multi-source estimates of emissions based on the National Atmospheric Emissions Inventory (NAEI), combined with meteorological data and calibrated oxidant distribution modeling processes, to establish an Atmospheric Dispersion Modeling System (ADMS), which evaluates the annual average concentrations of major air pollutants through simulation within a grid with a horizontal resolution of 1 × 1 km50. Subsequently, calibration was performed using data obtained from background monitoring network sites. DEFRA performs comprehensive comparisons between the modelled and measured annual average concentrations of air pollutants to validate the reliability of the models, showing good model performance (R-squared over 0.80)50,51. The National administrative political boundary Shapefile was obtained from the Office for National Statistics licensed under the Open Government Licence v.3.052.

The study employed previously proposed methods to calculate average exposure concentrations of NOx, NO2, PM2.5, PM10 and SO2 during the follow-up period of participants53,54. To simultaneously account for changes in participants’ residential addresses and the long-term exposure status, geospatial information on participants’ residential addresses along with the duration of residence at each address were collected from UKB. Geographic grid cell in the annual concentration map were then aggregated to the residential address history code51. We employed linear interpolation to calculate the time-weighted annual average air pollutant exposure for each participant from enrollment until the occurrence of any outcomes53.

Follow-up and outcome

All participants were followed up prospectively until death, loss to follow-up, or being censored (Nov 30, 2022), whichever occurred first. The all-cause and cause-specific mortality events for this study were sourced from the Office for National Statistics mortality registry55. The specific mortality outcomes were defined according to the International Classification of Diseases Tenth Revision (ICD-10). We conducted a thorough analysis on not only deaths from all causes (A00-Z99), non-accidental (A00-R99), accidental (V01-X59, Y85-Y86), and the 15 major disease causes, but also all subcategories of diseases with mortality rates ≥ 1‰ to ensure statistical precision. The detailed information of outcomes is shown in Supplementary Tables S1–S2.

Covariates

To control for potential confounding biases, we included covariates such as demographic, geographic, and socio-economic status data collected during the baseline assessment period. Specifically, we adjusted for age, sex, ethnicity, body mass index (BMI), education, occupation, smoking status, alcohol consumption, physical activity, residential area, and household income. Details on the covariates are presented in Supplementary Table S19. The proportions of missing values were low for most variables (Supplementary Table S20), except for physical activity, which was dealt with by creating dummy variable56. When analyzing sex-specific mortality outcomes (e.g., malignant tumor of male or female reproductive system), we did not include sex as a covariate in the models.

Statistical analyses

Associations between long-term exposure to NOx and mortality from all-cause, non-accidental, accidental, and cause-specific diseases, namely 15 major disease causes (encompassing 103 subcategories), were investigated using Cox proportional hazards models with time-varying covariates. Considering that NO2 is the major component of NOx, we also assessed the health impact of NO2 as an add-on analysis.

Since all-cause mortality is a competing risk for cause-specific mortality, the models censored participants at death from any cause (or loss to follow-up, or the censoring date of the study) to estimate cause-specific HRs, which compared event rates in participants who were alive and free of the event of interest57. The modeling is detailed as shown below.

$$ {{{\rm{Survival}}}}\,({tstart},{tstop},{status}) \sim {Air\; pollutant}+{Age}+{Sex}+{Ethnicity}\\ +{Body\; mass\; index}+{Education}+{Occupation}+{Smoking\; status}\\ +{Alcohol\; status}+{Physical\; activity}+{Residential\; area}\\ +{Household\; income}+{strata}\,({Year})$$
(1)

The adjustment of time-varying covariates was considered due to the violation of the proportional hazard assumption inherent in traditional Cox proportional hazards models, as well as the consideration that concentrations of air pollutants change over time58,59,60. Therefore, we incorporated ambient pollutants and age as external time-varying variables on a 1-year temporal scale into the Cox proportional hazards model, while adjusting for the aforementioned covariates to control for potential confounding effects. In addition, calendar year was included as a indicatory variable to account for potential temporal trends61. The effect estimates of this study are presented as HRs with 95% CIs for each 10 μg/m3, SD, and IQR increases in the concentrations of NOx and NO2.

A restricted cubic spline (RCS) function was used to smooth exposure-response relationships between long-term air pollutant exposure and all-cause or cause-specific mortality risk. Knots (3–6) were selected based on the Akaike and Bayesian information criteria (AIC/BIC) to optimize model fit by minimizing AIC and BIC values. Likelihood ratio test was further used to test the potential nonlinearity by comparing likelihood values between models with smoothed term of air pollutants and those with linear function62.

We investigated effect modification through stratification by age groups (≤ 65 versus > 65 years), sex (male versus female), and residential-level (rural versus urban) to discern disparities among various subgroups. Significance testing was conducted utilizing two-sample Z-tests63,64, with the specific formula provided below:

$$\,Z=\frac{{E}_{1}-{E}_{2}}{\sqrt{{SE}{\left({E}_{1}\right)}^{2}+{SE}{\left({E}_{2}\right)}^{2}}}$$
(2)

E1 and E2 are the natural logarithmically transformed values of HRs, \({SE}\left({E}_{1}\right)\) and \({SE}\left({E}_{2}\right)\) are the corresponding standard errors.

Sensitivity analyses

We conducted a series of sensitivity analyses to examine the robustness of our findings: (1) incorporating PM2.5, PM10 and SO2 to establish double-pollutant models; (2) additionally adjusting for baseline chronic conditions (e.g., cancer, hypertension, and diabetes) to examine whether the observed associations were influenced by pre-existing health status; (3) adjusting for only a portion of individual-level covariates (including age, sex, BMI, ethnicity, household income, and residential area); (4) excluding participants with less than 2 years of follow up to minimize the potential for reverse causality; (5) applying time-varying exposures on extended scales of 5-year averages for NOx; (6) setting the follow-up duration to a maximum of ten years from recruitment; (7) censoring follow-up data by the end of 2019 to eliminate the potential influence of the COVID-19 pandemic; (8) employing the multiple imputation for covariates with missing values; (9) excluding participants whose geographic codes changed during follow-up to account for potential measurement bias induced by residential mobility; (10) excluding the outliers of NOx concentrations (values below the 1st percentile or above the 99th percentile) to test the possible influence of outliers; (11) to control for multiple testing, we employed FDR-adjusted q-values < 0.1 to assess significance. When P < 0.05 but q ≥ 0.1, the relationship between air pollutant and mortality was considered suggestive65.

All statistical analyses were conducted using R 4.2.3 software. The ‘survival’ package and the ‘tmerge’ function were utilized for constructing time-varying covariates Cox proportional hazards models. The ‘rms’ package was employed to construct RCS. This study employed two-sided tests with a significance level of α = 0.05, where P < 0.05 indicated statistical significance.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.