Air pollution and human health: a phenome-wide association study

Abstract

Objectives

To explore the associations of long-term exposure to air pollution with onset of all human health conditions.

Design

Prospective phenome-wide association study.

Setting

Denmark.

Participants

All Danish residents aged ≥30 years on 1 January 2000 were included (N=3 323 612). After exclusion of individuals with missing geocoded residential addresses, 3 111 988 participants were available for the statistical analyses.

Main outcome measure

First registered diagnosis of every health condition according to the International Classification of Diseases, 10th revision, from 2000 to 2017.

Results

Long-term exposure to fine particulate matter (PM_2.5) and nitrogen dioxide (NO₂) were both positively associated with the onset of more than 700 health conditions (ie, >80% of the registered health conditions) after correction for multiple testing, while the remaining associations were inverse or insignificant. As regards the most common health conditions, PM_2.5 and NO₂ were strongest positively associated with chronic obstructive pulmonary disease (PM_2.5: HR 1.06 (95% CI 1.05 to 1.07) per 1 IQR increase in exposure level; NO₂: 1.14 (95% CI 1.12 to 1.15)), type 2 diabetes (PM_2.5: 1.06 (95% CI 1.05 to 1.06); NO₂: 1.12 (95% CI 1.10 to 1.13)) and ischaemic heart disease (PM_2.5: 1.05 (95% CI 1.04 to 1.05); NO₂: 1.11 (95% CI 1.09 to 1.12)). Furthermore, PM_2.5 and NO₂ were both positively associated with so far unexplored, but highly prevalent outcomes relevant to public health, including senile cataract, hearing loss and urinary tract infection.

Conclusions

The findings of this study suggest that air pollution has a more extensive impact on human health than previously known. However, as this study is the first of its kind to investigate the associations of long-term exposure to air pollution with onset of all human health conditions, further research is needed to replicate the study findings.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The phenome-wide association study approach to investigate the influence of air pollution on human health makes it possible to generate hypotheses about the associations of particulate matter and nitrogen dioxide with all health conditions, including both those associations that are well-established and those that so far have not been investigated.

• The study includes the entire Danish population aged ≥30 followed for up to 18 years (N=3 111 988).

• The use of individual-level data from administrative registers guarantees a continuous and complete follow-up, as well as reliable and valid information on numerous health conditions and possible confounders.

• Residual confounding is likely as the included covariates in some cases may be incomplete, and stratification for baseline residential municipality may not only have removed confounding due to geographical location but also reduced the variance in exposure.

• Replications of the study findings in other samples are necessary to confirm the significant associations with air pollution exposure.

Introduction

Air pollution is one of the leading global risk factors for human health, responsible for more than 3% of disability-adjusted life years (DALYs) and 2.92 million (11.3%) female and 3.75 million (12.2%) male deaths in 2019.¹ It has well-established associations with a wide range of health conditions affecting various organ systems,^2–10 but many health conditions remain unexplored. To improve our understanding of air pollution’s impact on human health, it has recently been proposed to embrace the use of phenome-wide association studies (PheWAS), which would make it possible to study the associations of the most important air pollutants with a vast number of health outcomes.⁷ A PheWAS is essentially like a traditional association study of one or a few air pollutants as exposures but is repeated across all possible health outcomes. In this context, the PheWAS has the advantage that it can provide a comprehensive account of the total impact of air pollution on health outcomes. Thus, a PheWAS can confirm existing associations and generate new hypotheses about air pollution and human health by evaluating the associations of an air pollutant with all possible health conditions, including both those that are well-established and those that so far have not been investigated, and it can improve our understanding of specific air pollutants and related biological pathways by making it possible to examine patterns of health conditions.⁷

However, up to now, only a single PheWAS has been conducted focusing on the associations of short-term exposure to fine particulate matter (PM_2.5) with all registered causes of hospital admission among Medicare fee-for-service beneficiaries aged ≥65 hospitalised in the USA from 2000 to 2012.⁸ This study found that short-term exposure to PM_2.5 was associated with several prevalent, but rarely studied disease groups, such as septicaemia, fluid and electrolyte disorders, and acute and unspecified renal failure, but also with previously identified disease groups.⁸ Yet studies of short-term exposure can only demonstrate the acute impact, and therefore, underestimate the influence of air pollution on human health. To gain an insight into the true magnitude of the influence of air pollution on human health, prospective PheWAS of long-term exposure to various air pollutants are needed.

Therefore, in a nationwide cohort of more than three million Danish adult residents, the aims of this study were to investigate the prospective associations of long-term exposure to two of the most prevalent and harmful air pollutants—PM_2.5 and nitrogen dioxide (NO₂)—with the onset of all health conditions from 2000 to 2017, as well as to estimate the population attributable fractions for the most prevalent health conditions.

Methods

Study design and participants

This study was based on a Danish nationwide sample originally conceived and designed for inclusion in the Effects of Low-Level Air Pollution: A Study in Europe project.^{2 9 10} In brief, this sample included all Danish residents who were ≥30 years old on 1 January 2000 and had resided in Denmark for at least 1 year (N=3 323 612). The sample’s residential addresses were geocoded at baseline to be able to link the individuals and their place of residence to the air pollution level in that specific location. By law, individuals moving from one residential address to another have to register this, and individuals moving abroad have to register their emigration although it is possible to obtain permission to still be registered as living in the country if the stay abroad lasts for less than 6 months and the individual has full disposal of his or her home during the period. From the original sample, we excluded 211 623 due to missing data in geocoding and one due to missing data in exposure, resulting in an analytical sample of a maximum of 3 111 988 individuals. For each association of interest, the analytical sample consisted of individuals who had not been registered with the specific outcome at any time before baseline. These individuals were followed from 1 January 2000 to the date of the first occurrence of the outcome, emigration, death from other causes or 31 December 2017, whichever came first.

Procedures

Health outcomes were measured using information on all health conditions registered in the Danish National Patient Register and all underlying causes of death in the Danish Register of Causes of Death. The National Patient Register contains individual-level data on inpatients since 1977, as well as individual-level data on both inpatients and outpatients since 1995. For each observation, the contact date and primary and secondary diagnoses are noted. The Register of Causes of Death contains individual-level data on all individuals dying in Denmark since 1970. Both registers have classified diseases according to the International Classification of Diseases, 8th revision (ICD-8) until 1993 and the ICD, 10th revision (ICD-10) since then. In the current study, the registered health conditions in the two registers were combined and linked to the individuals in the study population. This was done to capture the individuals’ onset of all health conditions, including health conditions that are not recorded before a fatal outcome occurs. We note that for more than 90% of the registered health conditions, the fatal outcomes constituted less than 5% of the incident events. In the study, the ICD codes were classified with three characters’ precision, resulting in a total of 1390 registered health conditions. This choice was based on the balancing of benefits and drawbacks since a more precise classification (eg, the aggregation of ICD codes into disease-relevant groupings) may be more informative, but this would come at the expense of the number of new hypotheses that could be tested. Only the 1044 health conditions with a prevalence of ≥2:10 000 (ie, non-rare conditions according to the definition of the Danish Health Authority) were included. With regard to the validity and reliability of the registered health conditions in the two registers, a review of the National Patient Register reported that a study had found an overall positive predictive value of 88% more than 30 years ago, but the positive predictive values varied wildly between diagnosis codes and increased continuously over time.¹¹ To our knowledge, no review of the data quality of the Register of Causes of Death exists, but it is often pointed out that the registration of cause(s) of death may be limited by the country’s relatively low and declining autopsy rate.¹²

Annual average exposure to PM_2.5 and NO₂ for the year 2010 was estimated from European-wide hybrid land-use regression (LUR) models at a 100×100 m resolution at the individuals’ baseline geocoded residential address and extrapolated back in time to reflect the individuals’ long-term exposure to PM_2.5 and NO₂, respectively, at baseline. Figure 1 shows the population’s annual average exposure to PM_2.5 and NO₂ according to municipality. Methods for the development and evaluation of the exposure prediction models were previously described.¹³ In brief, the European-wide hybrid LUR models for PM_2.5 and NO₂ were developed based on the European Environment Agency AirBase routine monitoring data for 2010. The LUR models included variables reflecting land use and traffic data, air pollution data from satellites and dispersion model estimates. Ordinary kriging was applied to the residuals of spatial variation from the LUR models. The R² indicating explained spatial variations in the measured concentration by the LUR models were 72% for PM_2.5 and 59% for NO₂.

Figure 1.

Annual average exposure to PM_2.5 and NO₂ according to municipality. NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

Potential confounders included age, sex (male, female), country of origin (Denmark, Western country, non-Western country), cohabitation status (cohabiting, living alone), educational level (low: primary and lower secondary education; medium: upper secondary education; high: higher education), occupational status (in employment, not in employment), the family’s household size equivalised disposable income and residential municipality. Information on the individual-level covariates (obtained in 1999) was obtained via Statistics Denmark. Area-level covariates were not explicitly modelled, but the influence of within-municipality differences in long-term exposure to PM_2.5 and NO₂ was taken into account by stratifying for municipality. Denmark has 98 municipalities; the municipalities had on average 32 000 individuals from the study population residing at baseline, ranging from around 1000 individuals in the smallest to around 256 000 individuals in the largest municipality. Half of the municipalities had between 17 000 and 35 000 individuals from the study population residing.

Statistical analysis

First, the characteristics of the study population were summarised using descriptive statistics.

Second, the associations of PM_2.5 and NO₂ with all registered health conditions were estimated by use of separate Cox regression models including age (the underlying time axis), country of origin, cohabitation status, educational level, occupational status and income (log-transformed) as covariates. Moreover, the models were stratified for sex and municipality to allow for non-proportional hazards between groups defined by sex and geographical location. Individuals with a given health condition registered before study entrance were excluded from the statistical analyses of this specific ICD-10 diagnosis code. The remaining individuals were followed from 1 January 2000 to the first occurrence of the health condition under consideration, emigration, death from causes other than the aforementioned health condition or 31 December 2017 (end of follow-up), whichever came first. To ensure statistical power, only health conditions with ≥250 incident events within each sex (or ≥250 incident events in total for the sex-specific conditions) were considered. Due to the large number of investigated associations, the probability of a type I error would be one with a traditional α level of 0.05. Therefore, multiple testing was handled by evaluating the null hypotheses using false discovery rate q values. These q values are analogous to p values, but the α level controls the proportion of associations that are falsely declared positive rather than the chance of observing data (or something more extreme) given the null hypothesis of no association.¹⁴ The final estimated associations were presented in what we have chosen to term a Copenhagen plot; this is a plot of a single exposure with a large number of phenotypes (health conditions) based on a phenome-wide association analysis.

Third, population attributable fractions were estimated for the most common health conditions due to PM_2.5 and NO₂ exposure levels exceeding WHO’s recommended air quality guideline levels, that is, 5 µg/m³ for PM_2.5 and 10 µg/m³ for NO₂.

In sensitivity analyses, we investigated the impact of using the PheCode system’s clinically meaningful disease groups¹⁵ as outcomes instead of the registered ICD codes. In the PheCode system, one or more related ICD codes have been translated into clinical diseases with the help of clinical experts in the relevant domains. For instance, the ICD-10 code K40 (inguinal hernia) is translated into the PheCode 550.1 (inguinal hernia), while ICD-10 codes E11 (type 2 diabetes mellitus) and E13 (other specified diabetes mellitus) both are translated into the PheCode 250.2 (type 2 diabetes). Furthermore, we investigated the impact of modelling the associations as possibly non-linear. More specifically, the investigated associations were reanalysed using penalised spline bases for PM_2.5 and NO₂ with an optimal number of df chosen based on the AIC. The concordance statistic from the survival package in R was used to compare the goodness-of-fit of corresponding models of linear and non-linear relationships, respectively. Since our statistical analyses were based on more than a thousand models and a large number of covariates, we did not test the proportional hazards assumption and test for influential observations using graphical diagnostics. However, it is important to keep in mind that even if the proportional hazards assumption should not hold for some of the investigated associations, the estimated HRs just present the average influence of long-term exposure to PM_2.5 or NO₂ on the investigated health outcome across the observed time to event. Moreover, given the large sample sizes, individual observations are unlikely to exert any substantial influence on the findings although it is possible that the influence of long-term exposure to PM_2.5 and NO₂ differs in particular subsets of the population, such as in certain municipalities.

All statistical analyses were carried out using R 4.1.0.

Patient and public involvement

As this is a register-based study, patients and the public were not involved in the design, conduct or reporting of this research. The study did not have the funding or researchers and patient training necessary to include patients or members of the public in the design of the study.

Results

Characteristics of the study population

We included 3 111 988 individuals aged ≥30 years who lived in Denmark on 1 January 2000. Table 1 summarises the sample’s characteristics at the beginning of the study period. The sample consisted of 48.3% men and 51.7% women with a median age of 51.4 years (ranging from 30.0 to 110.7 years) at baseline. The vast majority were of Danish origin and lived together with a partner. Among the men, the major part had completed an upper secondary education, while the women generally had lower educational levels. However, among both men and women, the majority of the sample was in employment although a considerable proportion, particularly of the women, were not so. This category also comprised retired individuals.

Table 1.

Characteristics of the study population in 2000

	Men (N=1 502 795)	Women (N=1 609 193)
PM2.5 (µg/m3)
Median (IQR)	12.3 (11.3–13.3)	12.4 (11.4–13.3)
Minimum-maximum	5.7–19.6	5.7–19.6
NO2 (µg/m3)
Median (IQR)	18.5 (14.2–24.5)	18.9 (14.8–25.0)
Minimum-maximum	3.7–72.2	3.7–72.2
Age at study entrance (years)
Median (IQR)	50.5 (39.8–61.6)	52.3 (40.8–65.4)
Minimum-maximum	30.0–104.9	30.0–110.7
Country of origin (%)
Denmark	94.2	94.5
Western country	3.5	2.9
Non-western country	2.4	2.7
Missing	0.0	0.0
Cohabitation status (%)
Cohabiting	75.0	66.8
Living alone	25.0	33.2
Missing	0.0	0.0
Educational level (%)
Low	35.3	45.9
Medium	43.7	33.1
High	21.0	21.0
Missing	0.0	0.0
Occupational status (%)
In employment	66.9	54.4
Not in employment	33.1	45.6
Missing	0.0	0.0
Income (€1000)
Median (IQR)	28.7 (22.1–36.4)	27.1 (20.5–35.0)
Minimum-maximum	9.6–80.1	9.6–80.1

Educational level is categorised as ‘low’ (primary and lower secondary education), ‘medium’ (upper secondary education) and ‘high’ (higher education). For occupational status, the category of not in employment includes both retired individuals and others outside of the labour force. Income is adjusted to 2020 prices and converted from DKK to EUR (DKK7.4467=€1).

EUR, Euros; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

Associations of PM_2.5 and NO₂ with 922 health conditions

In this sample, we identified 922 non-rare health conditions with at least 250 incident cases during the study period. Figure 2 shows the HR for each of these per 1 IQR increase in long-term exposure to PM_2.5 (IQR: 1.96 µg/m³) and NO₂ (IQR: 10.23 µg/m³), respectively, after adjustment for confounders. Long-term exposure to PM_2.5 was positively associated with 745 (81%) health conditions, of which 340 (37%) associations were statistically significant after correction for multiple testing. The remaining 177 (19%) associations were inverse, of which 27 (3%) associations were statistically significant after correction for multiple testing. Long-term exposure to NO₂ was positively associated with 813 (88%) health conditions, of which 589 (64%) associations were statistically significant after correction for multiple testing. The remaining 109 (12%) associations were inverse, of which 27 (3%) associations were statistically significant after correction for multiple testing. All the investigated associations are summarised in online supplemental material 1. The sensitivity analyses showed that the findings were essentially the same irrespective of whether the health conditions were based on registered diagnosis codes or clinically meaningful disease groups as defined by the PheCode system (online supplemental material 2). For instance, when comparing the findings for ICD-10 codes E11 (type 2 diabetes mellitus) and E13 (other specified diabetes mellitus) that were both translated into the PheCode 250.2 (type 2 diabetes) with the findings for the PheCode itself, the HRs were similar and the CIs were overlapping. Moreover, the sensitivity analyses showed that although the vast majority of the associations fitted the data better when modelled as non-linear, the concordance statistic was on average only 0.001 higher for the non-linear models compared with the linear models, ranging from −0.002 to 0.025 for PM_2.5 and from −0.002 to 0.027 for NO₂ (online supplemental material 3). As an example, the linear and non-linear associations of PM_2.5 and NO₂ with other chronic obstructive pulmonary disease (COPD) are illustrated in online supplemental material 4. Considering 95% of the sample has an exposure level to PM_2.5 between 4.89 and 8.05 per 1 IQR (corresponding to 9.59–15.77 µg/m³) and an exposure level to NO₂ between 0.83 and 3.66 per 1 IQR (corresponding to 8.54–37.42 µg/m³), it is clear that both air pollutants mainly have a linear influence on the hazard of other COPD. Therefore, the linear models were presented as the main results according to the principle of parsimony.

Figure 2.

Copenhagen plot. The x-axis indicates the ICD-10 chapter and the y-axis is the HR for all the registered health conditions per 1 IQR increase in PM_2.5 (IQR 1.96) and NO₂ (IQR 10.23). The sizes of the circles are proportional to the number of cases. Filled circles indicate statistically significant associations, while empty circles indicate insignificant associations after correcting for multiple testing. The ICD-10 chapters refer to (1) certain infectious and parasitic diseases, (2) neoplasms, (3) diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism, (4) endocrine, nutritional and metabolic diseases, (5) mental and behavioural disorders, (6) diseases of the nervous system, (7) diseases of the eye and adnexa, (8) diseases of the ear and mastoid process, (9) diseases of the circulatory system, (10) diseases of the respiratory system, (11) diseases of the digestive system, (12) diseases of the skin and subcutaneous tissue, (13) diseases of the musculoskeletal system and connective tissue, (14) diseases of the genitourinary system, (15) pregnancy, childbirth and the puerperium, (16) certain conditions originating in the perinatal period, (17) congenital malformations, deformations and chromosomal abnormalities and (18) symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified. ICD-10, International Classification of Diseases, 10th revision; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

Regarding the most common health conditions, figure 3 shows the HRs for each of the 25 most prevalent conditions per 1 IQR increase in long-term exposure to PM_2.5 and NO₂, respectively, after adjustment for confounders. Long-term exposure to PM_2.5 was positively associated with 21 of the health conditions, of which 15 associations were significant after correction for multiple testing. The remaining four associations were inverse, of which three were significant after correction for multiple testing. Moreover, long-term exposure to NO₂ was positively associated with 22 of the health conditions, of which 19 associations were significant after correction for multiple testing. The remaining three associations were inverse, of which two were significant after correction for multiple testing. Long-term exposure to PM_2.5 and NO₂ were both strongest positively associated with ‘other COPD’ (HR_PM2.5=1.06 (95% CI 1.05 to 1.07), HR_NO2=1.14 (95% CI 1.12 to 1.15)), ’type 2 diabetes mellitus’ (HR_PM2.5=1.06 (95% CI 1.05 to 1.06), HR_NO2=1.12 (95% CI 1.10 to 1.13)) and ‘chronic ischaemic heart disease’ (HR_PM2.5=1.05 (95% CI 1.04 to 1.05), HR_NO2=1.11 (95% CI 1.09 to 1.12)) and strongest inversely associated with ‘inguinal hernia’ (HR_PM2.5=0.96 (95% CI 0.95 to 0.97), HR_NO2=0.93 (95% CI 0.91 to 0.94)). Baseline hazard plots of ‘other COPD’, ‘type 2 diabetes mellitus’ and ‘chronic ischaemic heart disease’ can be found for selected individuals in online supplemental material 5.

Figure 3.

Associations of long-term exposure to PM_2.5 and NO₂ with the top 25 prevalent conditions. The x-axis indicates the HR per 1 IQR increase in PM_2.5 (IQR 1.96) and NO₂ (IQR 10.23), respectively, and the y-axis is the top 25 prevalent health conditions. Filled circles indicate statistically significant associations, while empty circles indicate insignificant associations after correcting for multiple testing. Error bars represent 95% CIs. NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

Population attributable fractions

Figure 4 shows the population attributable fraction for each of the 25 most prevalent conditions due to PM_2.5 and NO₂ exposure levels exceeding WHO’s recommended air quality guideline levels. The population attributable fraction of ‘other COPD’, ‘type 2 diabetes mellitus’ and ‘chronic ischaemic heart disease’ attributable to PM_2.5 exposure were 0.06, 0.05 and 0.04, respectively, corresponding to 6%, 5% and 4% of cases. Furthermore, the population attributable fraction of these three health conditions attributable to NO₂ exposure were 0.11, 0.10 and 0.09, respectively, corresponding to 11%, 10% and 9% of cases.

Figure 4.

Population attributable fractions for the most common health conditions. The x-axis indicates the population attributable fraction due to PM_2.5 and NO₂ exposure levels exceeding WHO’s recommended air quality guideline levels, and the y-axis is the top 25 prevalent health conditions. Filled bars indicate significant associations, while empty bars indicate insignificant associations after correcting for multiple testing. NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

Discussion

In this nationwide study of more than three million Danish residents, we found that long-term exposure to PM_2.5 and NO₂, respectively, was positively associated with the onset of more than 700 health conditions (ie, >80% of the registered conditions). With regard to the most common health conditions, PM_2.5 and NO₂ were both strongest positively associated with COPD, type 2 diabetes and ischaemic heart disease. Moreover, PM_2.5 and NO₂ were both positively associated with so far unexplored, but highly prevalent outcomes relevant to public health, including senile cataract, hearing loss and urinary tract infection.

Strengths and limitations

The major strength of this study is its inclusion of the entire Danish population aged ≥30 years followed for up to 18 years. Moreover, the use of individual-level data from administrative registers has guaranteed a complete and continuous follow-up, reliable and valid information on numerous health conditions and possible confounders, and a limited amount of missing information. Thus, socioeconomic characteristics could be accounted for in detail even though these are not particularly linked to air pollution exposure in Denmark.¹⁶ Furthermore, long-term average exposure at the residential address was assessed by a well-validated European LUR model. However, we only included information about long-term exposure at the residential address and did not take occupational exposure into account. Furthermore, the large number of investigated associations made it impossible to identify confounders for each association separately and the list of included confounders might, therefore, in some cases be incomplete. Moreover, the stratification for baseline residential municipality does not only remove confounding due to geographical location but also reduces the variance in exposure. Thus, the findings may provide conservative estimates of the true influences on human health. In any case, the study findings need replication to ensure that statistically significant associations were not spurious and to determine their generalisability to other populations, times and places. Furthermore, studies of the possible underlying mechanisms of the observed associations are needed to assess the evidence for causality.

Comparison with the existing literature

The finding that long-term exposure to PM_2.5 and NO₂ has an extensive impact on human health is in line with the existing literature, reporting associations with a wide range of health conditions targeting multiple organ systems.^2–10 This includes respiratory and cardiovascular disease morbidity and mortality,^2–6 as well as adverse influences on diabetes, neurological outcomes, and pregnancy and developmental outcomes.^{4 17–20} Yet many aspects of human health have not been explored until now.

In our study, six circulatory system diagnoses were among the most common health conditions. Essential hypertension, angina pectoris, acute myocardial infarction, chronic ischaemic heart disease and heart failure showed increased hazards associated with both PM_2.5 and NO₂ in accordance with the consensus of causal associations with PM_2.5 by, for example, the American Heart Association.²¹ Atrial fibrillation was not related to PM_2.5 and the hazard was only marginally elevated due to NO₂ in accordance with inconclusive evidence for such associations related to long-term exposure.²²

With regard to the most common respiratory conditions, we found increased hazards for COPD and pneumonia for both PM_2.5 and NO₂, whereas ‘abnormalities of breathing’ was only positively associated with NO₂. This is in keeping with meta-analyses showing an increased risk of COPD incidence associated with long-term exposure to both pollutants.²³ Regarding pneumonia in adults, a large pooled European cohort study showed a marginally increased risk of mortality associated with the black carbon fraction of PM_2.5 and NO₂,²⁴ whereas cohort studies of older American adults showed increased mortality and hospitalisation associated with PM_2.5 and NO₂.^{25 26} ‘Abnormalities of breathing’ includes dyspnoea, stridor and wheezing likely to be related to air pollution exposure as they are symptoms of airway disease, especially COPD and asthma. For asthma, a number of cohort studies point to increased risk for adult-onset associated with long-term exposure to PM_2.5 and NO₂.^{27 28}

For the most common health conditions from the ICD chapter on endocrine, nutritional and metabolic diseases, our finding of an increased hazard of type 2 diabetes mellitus associated with both PM_2.5 and NO₂ is consistent with several recent meta-analyses of cohort studies.^{29 30} Similarly, meta-analyses of long-term exposure to PM_2.5 and NO₂ indicate associations with increased levels of triglycerides in the blood, whereas individual cohort studies further support associations with increased LDL, total cholesterol and/or dyslipidaemia.^31–33

The most common health conditions from the sensory systems, ‘senile cataract’ and ‘other hearing loss’ (including presbycusis), were positively associated with PM_2.5 and NO₂, representing population attributable fractions of 3%–6% for senile cataract and 4%–7% for other hearing loss. In consistence, three cohort studies have found an increased risk of cataract surgery associated with long-term exposure to PM_2.5 in Canada³⁴ and NO₂ in Korea.³⁵ Similarly, hearing loss was associated with long-term exposure to PM₁₀ and NO₂ in a Korean cohort study.³⁶

We found the hazard of cystitis to be associated with both PM_2.5 and NO₂ although there appear to be no published studies addressing long-term exposure to air pollution and this outcome. Nevertheless, a PheWAS-like study of hospital admissions related to short-term exposure levels found urinary tract infections to be the fifth-highest increased hazard associated with PM_2.5.⁸ The other common condition from the genitourinary system chapter in our study was ‘excessive, frequent and irregular menstruation’ showing null associations with PM_2.5 and NO₂. There is very limited evidence published with only one time-series analysis showing short-term associations between outpatient visits for menstrual disorders and PM₁₀ and NO₂.³⁷

We found the hazard of ‘abdominal and pelvic pain’ to be associated with exposure to both PM_2.5 and NO₂ although there appear to be no published studies addressing long-term exposure to air pollution and this outcome. Nevertheless, short-term studies suggest associations between high ambient levels of PM_2.5 and NO₂ and increased emergency room visits for abdominal pain.^{38 39}

Six musculoskeletal conditions were among the most common health conditions in our study. As expected, ‘shoulder lesions’, ‘internal derangement of the knee’ which is mainly traumatically induced, and ‘gonarthrosis (arthrosis of knee)’ showed no positive associations with PM_2.5 or NO₂. However, ‘dorsalgia’ (ie, back pain) showed positive associations with both PM_2.5 and NO₂, whereas ‘other intervertebral disc disorders’ was positively associated with only NO₂. Plausible mechanisms for adverse effects of air pollution in this respect are not intuitive and there appear to be no published studies except a PheWAS-like analysis of short-term associations between PM_2.5 and hospital admissions finding positive associations with Clinical Classification Software code 205 (spondylosis, intervertebral disc disorders and other back problems), which covers both dorsalgia and other intervertebral disc disorders.⁴⁰ ‘Other soft tissue disorders, not elsewhere classified’ covers a range of pain from connective tissues, and the opposing associations with PM_2.5 (protective) and NO₂ (harmful) make causal inference unlikely as this would also be unexpected.

Two pregnancy outcomes were among the most common conditions in our study. The associations between single spontaneous delivery and both PM_2.5 and NO₂ suggest that pregnant women are in general more exposed to air pollution than are otherwise comparable women as there is no reason to assume that air pollution per se increases the probability of pregnancy. For instance, women of younger age are typically residing in areas with higher levels of air pollution and also have a higher fertility rate. However, our large number of included covariates (age, sex, country of origin, cohabitation status, educational level, occupational status and the family’s household size equivalised disposable income) should be able to take this kind of confounding into account, and since our statistical models are also stratified for residential municipality, the influence of within-municipality differences in long-term exposure to PM_2.5 and NO₂ is likewise taken into account. Yet we cannot exclude the possibility of residual confounding. But if our findings for single spontaneous delivery can be replicated, it is a large public health problem as many birth outcomes are affected by air pollution.⁴¹ However, our findings suggest that adherence to WHO’s recommended air quality guideline levels for PM_2.5 and NO₂ may reduce the percentage of exposed pregnancies by 4% and 5%, respectively.

Finally, the hazard of inguinal hernia was inversely associated with PM_2.5 and NO₂. This has no plausible explanation as the external risk factors are mainly related to high intra-abdominal pressure from, for example, daily lifting and standing/walking and are only relevant for lateral hernias.⁴²

Overall, our study findings suggest that all organ systems may be adversely affected by exposure to air pollution. Adverse outcome pathways for effects of air pollution are well described for respiratory, cardiovascular, endocrine, immune, neuronal, digestive and reproductive systems.⁴³ Well-documented common mechanisms of action encompass oxidative stress, inflammation, autophagy and apoptosis with more recent evidence also including pyroptosis, ferroptosis and epigenetic modifications.⁴³

In conclusion, this study suggests that air pollution has a more extensive impact on human health than previously known. However, as this study is the first of its kind to investigate the associations of long-term exposure to air pollution with onset of all human health conditions, further research is needed to replicate the study findings.

Data availability statement

Data may be obtained from a third party and are not publicly available. The data that support the findings of this study are available from the Danish Health Data Authority and Statistics Denmark but restrictions apply to the availability of these data, which were used under licence for the current study and so are not publicly available. Data are, however, available from the authors on reasonable request and with permission of the Danish Health Data Authority and Statistics Denmark. The analytical code will be shared immediately following publication with anyone who wishes to access this document for any purpose. Requests should be directed to ehe@dst.dk.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

According to Danish legislation, no ethics approval is needed for register-based studies. The study is subject to and conducted according to the rules and regulations of the Danish Data Protection Agency. The results of our study are planned to be disseminated to relevant patient and public communities.