Air pollution and the risk of bronchiolitis-related hospital admission in a birth cohort of over 400,000 children born in London

Authors: Amal Rammah, Bianca De Stavola, Samantha Hajna, Jonathon Taylor, Glory Atilola, Joana Cruz, Steve Cunningham, Alison Macfarlane, Ai Milojevic, Pia Hardelid

Abstract

Air pollution exposure in early life may be associated with an increased risk of bronchiolitis in children, but evidence for long-term exposures (over weeks or months) is limited. We estimated associations between fine particulate matter (PM2.5) and nitrogen dioxide (NO2) exposure during pregnancy and in the first year of life and bronchiolitis-related hospital admissions in a London birth cohort. We used a national birth cohort to identify London-resident mothers whose children were born in London from 2010 to 2013 and extracted information from birth and death registrations and maternal and child longitudinal Hospital Episode Statistics. We linked modeled PM2.5 and NO2 data to residential postcode histories during pregnancy and infancy. We applied a landmark approach with Cox proportional hazard models, adjusted for sociodemographic characteristics and housing energy efficiency, to estimate associations between time-varying monthly PM2.5 or NO2 exposure and first bronchiolitis admission. Among 415,311 children, we found inconclusive evidence overall, with suggestive signals of increased risk associated with pregnancy and exposures in the final month of infancy to PM2.5 and NO2 and time to first bronchiolitis-related hospital admission. There were modest increases in risk in the first month after birth, corresponding to prenatal exposures, for PM2.5 (adjusted Hazard ratio [HRa] = 1.07, 95% confidence interval [CI]: 0.70, 1.61 per 5 ug/m^3^) and NO2 (HRa = 1.15, 95% CI: 0.98, 1.35 per 10 ug/m^3^). We found a similar estimated increased risk in the last month of follow-up (PM2.5 HRa = 1.12, 95% CI: 0.94, 1.34; NO2 HRa = 1.31, 95% CI: 1.00, 1.71), corresponding to late infancy exposures. These findings highlight the uncertainty about critical windows of susceptibility during infancy. Future studies should examine the association between air pollution and bronchiolitis in emergency departments and primary care settings.

Lower respiratory tract infections (LRTIs) remain a leading cause of morbidity and mortality worldwide, particularly among children.^1^ Bronchiolitis, an LRTI marked by airway inflammation and obstruction of the lower respiratory tract bronchioles, is a major driver of infant hospital admissions globally,^2^ including the United Kingdom.^3^ Beyond the immediate clinical impact, bronchiolitis may have later consequences for respiratory health, including susceptibility to recurrent LRTIs and recurrent wheezing/asthma in childhood and adolescence,^4,5^ and is associated with lung function impairment.^6^ Bronchiolitis incidence is highest in the first year of life,^7^ but risk of bronchiolitis-related hospital admission changes rapidly over this period (with a peak at 1–2 months),^8^ and lung development and immune function continue to mature during infancy.^9,10^ In addition, susceptibility to environmental exposures, including air pollution, may vary across prenatal and postnatal periods.^11,12^ While several systematic reviews have investigated air pollution in relation to respiratory health in childhood, particularly bronchiolitis and other LRTIs, most have focused on short-term air pollution exposures (i.e., daily changes in exposure).^13–16^ In contrast, the evidence on prenatal exposures^17^ and long-term exposure (i.e., exposure over weeks and months) during infancy and bronchiolitis remains limited and inconclusive.

Exposure to air pollution during pregnancy and infancy can disrupt lung development and immune function in several ways, which may increase a child’s vulnerability to respiratory infections, including bronchiolitis. For particulate matter with a diameter of 2.5 micrometres or less (PM2.5), although no specific chemical components have been identified as posing the greatest risk, oxidative stress is generally understood to be the central mechanism behind the biological effects of toxic species in PM2.5.^18^ Oxidative stress may disrupt lung development and repair mechanisms and trigger inflammation in the placenta and the fetus, as well as promote pulmonary and systemic inflammation in childhood.^18^ Furthermore, evidence from mechanistic studies suggests that exposure to air pollutants, including PM2.5 and nitrogen dioxide (NO2), may impact DNA methylation and epigenetic alterations of genes involved in oxidative stress response as well as inflammation,^18^ which can impact susceptibility to respiratory infections in the long term.^19^

Over the past 2 decades, epidemiological investigations of the relationship between long-term exposure to PM2.5 and NO2 and the risk of bronchiolitis in early childhood have employed a range of study designs, data types (e.g., recruited cohorts and routinely collected administrative data), and exposure assessment methods, yet the evidence is equivocal. In three case-control studies conducted between 2006 and 2009, Karr and colleagues^20–22^ examined monthly and lifetime exposure to air pollution and bronchiolitis hospital admissions in the first year of life among children in the United States and Canada, using air pollution data from local monitoring networks assigned to the address at birth. They reported null to modest associations for PM2.5, as well as modest estimates for NO2. More recent investigations have additionally evaluated satellite-derived and model-estimated air pollution In 2017, a case-control study of infant bronchiolitis in Massachusetts, United States, found no association between lifetime exposure to satellite-derived PM2.5 estimates and bronchiolitis in the first year of life.^23^ Conversely, a retrospective cohort study in Georgia, United States, found elevated risks for bronchiolitis admission associated with modeled annual PM2.5 and NO2 exposure estimates among children under 2 years of age.^24^ A 2018 ecological study employing regional PM2.5 estimates derived from monitoring networks in Chile, also reported statistically significant associations between annual mean PM2.5 and bronchiolitis hospital admissions in the first year of life.^25^ More recent evidence from two cohort studies conducted in urban centers in Italy, which also relied on daily ground-level air pollution estimates, suggests that children’s increased exposure to PM2.5 and NO2 in the medium (i.e., 4-week) term may be associated with elevated risk of bronchiolitis admission before their first birthday.^26,27^ Finally, in another cohort study in Jakarta, Indonesia, increases in annual average PM2.5 and NO2 exposures (derived using land use regression models) were associated with increased risk of LRTI hospital admissions in children under 6 months of age.^28^

Despite this evolving body of evidence on long-term air pollution exposure and bronchiolitis, several important limitations persist. First, these investigations have relied on exposure estimates of limited spatial or temporal resolution, employing either regional estimates or annual means that might not capture variability in individual exposures or the variability in exposures over time. Second, none have accounted for adverse housing conditions, such as energy efficiency, which contributes to cold and damp conditions and is associated with poor respiratory health in children.^29,30^ Finally, and crucially, no studies have interrogated whether exposure during specific developmental stages in infancy represents heightened vulnerability to air pollution with respect to the risk of developing bronchiolitis, despite the rapid changes in susceptibility that characterize this period.^8^ Using a birth cohort based on linked administrative data from all children born in London between 2010 and 2014, as part of the PICNIC study,^31^ we examined the relationship between time-varying early life (pregnancy and infancy) exposures to PM2.5 and NO2 and the risk of bronchiolitis-related hospital admission in the first year of life.

Methods

Data sources

We used data from a large national mother-child cohort for England originally developed by researchers at City, University of London.^31^ The spine consisted of birth registrations from the Office for National Statistics linked to National Health Service (NHS) birth notifications for all children born in England between 2005 and 2014, from which we obtained information on sociodemographic characteristics. Birth registration is mandatory in England and Wales; all children in England have to be registered at a Register Office within 42 days of their birth.^32^ NHS birth notifications are created by staff in maternity units to register children with the NHS and generate an NHS number, the unique identifier in the NHS.^33^ The spine was linked to Hospital Episode Statistics Admitted Patient Care (HES APC), which includes information on all inpatient hospital admissions to the English NHS for both mothers and children. It excludes emergency department (ED) and primary care visits. Linkage was carried out by NHS Digital using a deterministic algorithm based on NHS number, date of birth, sex, and postcode.^34^

Maternal and child residential postcode histories were obtained via linkage to the Personal Demographic Service (PDS), a national address register within the English NHS.^35^ Linkage to PDS was carried out by NHS Digital using similar methods to link birth records to HES APC. Smoking expenditure data was available from CACI Ltd., a UK commercial data provider.^36^

Air pollution information for the Greater London Area was obtained from Cambridge Environmental Research Consultants.^37^ We obtained information on housing characteristics from the Energy Performance Certificates (EPC) database, which contains dwelling-level information on energy efficiency and building characteristics for all buildings constructed, let or sold since 2007.^29^

Study population and follow-up

The PICNIC London birth cohort included all singleton children born in Greater London (i.e., roughly within the borders of the M25 Motorway, the main ring road encircling the majority of the Greater London Authority, and covering inner and outer London, based on their birth registration residential postcode) who were conceived and born at gestational ages between 24 and 45 weeks, inclusive, between January 1, 2010, and December 31, 2014, with linked HES APC data. We estimated conception dates by subtracting the gestational age (in weeks) from the date of birth and adding 2 weeks to account for the time from the likely last menstrual period to conception. We additionally excluded children born after December 31, 2013, to allow for 1 year of follow-up for all children, and those conceived less than 45 weeks before that date to account for fixed cohort bias.^38^ We followed children from their date of birth until the earliest of the following first admission for bronchiolitis, first birthday, death, relocation outside of London, or the end of the study period. We defined follow-up over the first year of life to capture the period of highest incidence of bronchiolitis during childhood.^7^

Bronchiolitis-related admissions

The basic unit of HES APC is a finished consultant episode, a continuous period of care under one consultant. We linked finished consultant episodes into continuous admissions using methods described previously.^39^ We defined a bronchiolitis-related admission within the HES APC data as an emergency admission where the primary diagnosis, that is, the primary diagnosis of the first episode of the admission, was bronchiolitis (J21), as defined in the International Classification of Diseases version 10. In sensitivity analysis, we also considered emergency and nonemergency admissions with any bronchiolitis diagnosis in any episode.

Air pollution exposure

We obtained daily modeled estimates of PM2.5 (μg/m^3^) and NO2 (μg/m^3^) concentrations from 2010 to 2014 at 100 × 100 m grids for the Greater London Area, based on the ADMS Urban model, an atmospheric dispersion model that captured both traffic and area sources of pollution.^37,40^ Figure 1 shows a map of PM2.5 and NO2 distributions across the study area for the year 2012 (mid-year of the study period). We used maternal and child residential postcode histories from the PDS to assign weekly air pollution exposure estimates during pregnancy and infancy. UK postcodes provide fine spatial resolution for exposure assignment, as they are defined by groups of delivery addresses (with an average population of 40 individuals and 20 households^41^) rather than by geographic areas, and are regularly updated. We used a single, consistent snapshot of postcode units (Code-Point with Polygons, October 2019 version)^42^ to ensure temporal consistency in the linkage with air pollution data. Within the Greater London Administration study area (defined using administrative boundaries rather than the M25), this corresponded to 149,598 postcode units. Postcode histories were cleaned and validated using a structured algorithm developed in our previous work,^43^ which identified and resolved duplicate records, overlapping address periods, implausible date sequences, gaps in residential histories, and relocation outside of London to ensure consistency in residential records during the study period. This allowed for complete ascertainment of residential mobility of mother and child during pregnancy and infancy, respectively. We assigned weekly air pollution concentrations to each residential episode based on the grid cell nearest to the postcode centroid. Birth registration postcodes were used to assign air pollution exposure averages over the pregnancy period for children (14.2%) for whom no maternal postcodes were available, either due to linkage error or because mothers had opted out of sharing their NHS data for research. For each pollutant, we calculated an overall pregnancy exposure average as the mean of weekly exposure estimates assigned from conception to birth. Postnatal exposures were summarized as monthly (28-day period) averages from birth through the end of follow-up for each child.

Sociodemographic and housing characteristics

We used Office for National Statistics birth registration and NHS birth notification data to obtain information on the following child’s sex; child’s ethnicity (categorized as White, Black, Asian and Other); year and month of birth; mother’s age at delivery (categorized as <20, 20–24, 25–29, 30–34, 35–39, and ≥40 years); country of birth (categorized as UK-born or non-UK-born) and marital status (categorized as joint-married or civil partnership; sole registration; joint-cohabiting, and joint-not cohabiting); and quintiles of the Index of Multiple Deprivation (IMD) 2010,^44^ a small-area measure of deprivation based on Census, tax, benefit, and other local data at the lower super output area level, covering populations of around 1000–3000 residents.^45^ As an indicator of secondhand smoking exposure, we used information on weekly smoking expenditure (GBP) for 2012 at the Output Area level, which represents areas with an average population of 300 residents.^45^ EPC data were filtered to obtain the most recent certificate for each dwelling, aggregated to calculate the median energy efficiency by postcode, and linked to the cohort using the residential postcode recorded at birth registration. Comparison of EPC counts with the number of households from the 2011 census indicated that postcodes had, on average, 67% of dwellings with EPCs. We categorized postcode-level median energy efficiency as poor (F–G), low (E–D), or good (A–C).

Statistical analysis

We described the distribution of air pollution exposures and each of the key covariates of interest within the birth cohort. We also calculated bronchiolitis admission rates per 1000 child-months, including all repeat admissions rather than only the first. We modeled the association between PM2.5 and NO2 and time to first bronchiolitis admission by fitting Cox proportional hazard models within a landmarking approach to account for time-varying exposures^46,47^ and rapidly changing risk of bronchiolitis admissions.^8^ For this analysis, we used only the first hospital admission for each child to define time-to-event as time to admission. Separate single-pollutant models were fitted, and observations with missing data were excluded. Landmarks were selected at the end of each month (or 28-day period) during the year of follow-up starting from the date of birth. These time points defined a series of consecutive landmark periods, starting from birth for each child, where the first landmark period covered birth through day 28, the second landmark period covered days 29 through 56, and so on. This approach allowed us to estimate the association separately across successive age-defined intervals and capture how the risk varies over early life. For each landmark period, we separately estimated hazard ratios (HR) and 95% confidence intervals (CI) for the risk of hospital admission associated with a 5 μg/m^3^ and 10 μg/m^3^ increase in PM2.5 and NO2 exposure, respectively, in the preceding 28-day period (or during pregnancy for the first landmark period). In the first month of follow-up, we evaluated admissions associated with average pollutant exposure during pregnancy only to assess the impact of prenatal exposures during a critical window of fetal development, which may shape susceptibility in the first month of life. Models were adjusted for year and month of birth, child’s ethnicity, maternal age, marital status, IMD quintiles, and median energy efficiency of postcode dwellings, based on a DAG which conceptualized our reasoning regarding confounders (Figure S1; https://links.lww.com/EE/A430). We additionally adjusted for prenatal exposure and average pollution exposures from birth up until the previous landmark. Finally, we examined tobacco expenditure at output area level as a confounder, but this did not change the effect estimate by more than 10% and we excluded it from the models.^48^

We conducted two kinds of sensitivity analyses. In one analysis, a more sensitive definition of bronchiolitis was used, that is, hospital admissions that have bronchiolitis as a diagnosis on any episodes in all emergency and nonemergency admissions. In another sensitivity analysis, we restricted the analysis to children whose maternal postcode history was not missing (i.e., whose air pollution exposures during pregnancy were not assigned based on their birth registration postcode). Statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, North Carolina).

Results

There were 565,734 children born alive between 24 and 45 weeks of gestation in the Greater London area between 2010 and 2014. After exclusions, 415,311 children were included in our study, contributing 176,271,610 months of follow-up (Figure 2). There were 15,347 bronchiolitis-related admissions during the follow-up period, of which 13,555 (88%) had bronchiolitis as the primary diagnosis in the first episode of an emergency admission. The overall bronchiolitis admission rate in the cohort was 2.81 per 100 child-years. The majority (11,351; 86.8%) of children had a single bronchiolitis-related admission. We observed higher emergency admission rates (in which bronchiolitis was the primary diagnosis in the first episode) among children who were male, born preterm, and those whose mothers were under the age of 20, UK-born, and registered as sole parents or as joint but not cohabiting parents (Table 1). Children born in areas of higher deprivation levels also had higher rates than those born in more privileged areas of London (IMD Quintile 3.1 admissions per 100 child years vs. IMD Quintile 2.1 admissions per 100 child years).

There was limited variability in prenatal exposures (PM2.5 IQR = 2.7 μg/m^3^, NO2 IQR = 9 μg/m^3^) compared with broader variability in postnatal exposures, with IQRs ranging from 6.6 to 7.6 μg/m^3^ for PM2.5 and 12.0 to 12.9 μg/m^3^ for NO2 (detailed distribution summaries are available in Table S4; https://links.lww.com/EE/A430).

Across most landmark periods, there was no consistent association between exposure to either pollutant and bronchiolitis-related admission in the subsequent month (Figure 3). HRs for both PM2.5 and NO2 were generally close to the null and value of 1 with wide 95% CIs across most of the landmark periods. In the first landmark period, where we evaluated pregnancy exposures alone, we observed a 7% increase in the rate of bronchiolitis that was associated with a 5 μg/m^3^ increase in PM2.5 exposure (HRa = 1.07, 95% CI: 0.70, 1.61), as well as a 15% increase in rate associated with a 10 μg/m^3^ increase in NO2 exposure (HRa 1.15, 95% CI: 0.98, 1.35), although both of these effect estimates were inconclusive. We observed associations of stronger magnitude in the final landmark a 12% increase in rate associated with a 5 μg/m^3^ increase in PM2.5 exposure (HRa = 1.12, 95% CI: 0.94, 1.34), and a 31% increase in rate associated with a 10 μg/m^3^ increase in NO2 exposure (HRa = 1.31, 95% CI: 1.00, 1.71) in the last month.

In sensitivity analyses, results were broadly similar when we defined a bronchiolitis admission using emergency and nonemergency admissions with bronchiolitis as a diagnosis on any episode (Table S2; https://links.lww.com/EE/A430); however, the suggestive elevated rate in the first landmark period associated with PM2.5 pregnancy exposures was attenuated. Results were also similar when we restricted the analyses to children with available maternal postcode histories (Table S3; https://links.lww.com/EE/A430).

Discussion

In a population-based birth cohort based on linked administrative data for London, we applied a landmarking approach to model time-varying exposures across defined periods during the first year of life, including prenatal exposures. Our analyses yielded inconclusive, mixed results, with associations between PM2.5 or NO2 exposure and the risk of bronchiolitis-related admission varying in direction and magnitude across landmark periods. We observed modest increases in risk in the earliest and latest landmark periods, with wide 95% CIs, corresponding to prenatal exposure and exposure near the end of the first year of infancy. Effect estimates were generally close to the null and imprecise across most landmark periods. Findings were consistent in sensitivity analyses using a broader definition of bronchiolitis admission, as well as in analyses restricted to children with complete residential history during pregnancy.

Comparison to previously published work

Our findings highlight the uncertainty around the timing of potential vulnerability to air pollution in early life and its impact on respiratory health in children. The point estimate of the HRs for bronchiolitis-related admission in the first landmark period (where we considered prenatal exposure only) was above 1 for both pollutants, but with wide CIs. This is consistent with suggestive evidence of a positive association for PM2.5 from a recent meta-analysis of 12 studies investigating prenatal exposures and childhood LRTIs, although these studies assessed outcomes over longer follow-up periods (ranging from 1 to 14 years of age), rather than with the narrowly defined 28-day risk window used in our analysis.^17^ For prenatal NO2 exposures, evidence from the same review was mixed and inconsistent in direction across studies and exposure windows, with few statistically significant associations reported, which is also reflected in the imprecision of our estimates.^17^

For postnatal exposures, a limited number of studies have employed exposure windows comparable to ours, such as monthly exposure averages during infancy. We observed variation in effect estimates across landmark periods in the first year of life, with HRs generally close to the null and imprecise, and without a consistent pattern of association. In some landmark periods, we observed inverse associations for PM2.5 and bronchiolitis-related admission risk. These findings were not anticipated and may reflect random variation or residual confounding (see below). The point estimate of the HRs for bronchiolitis-related admission in the last landmark period was above 1, with wide 95% CIs. This finding of a positive association between PM2.5 and bronchiolitis admission risk was inconclusive, but consistent with studies that have evaluated monthly exposures. In 2006, Karr et al.^49^ reported a 9% increase in risk of bronchiolitis admission per 10 μg/m^3^ increase in PM2.5 exposure among children in California, United States. And while they found no association for all bronchiolitis admissions in a subsequent case-control study in Washington, United States, they reported a 14% increase in risk, though imprecise, for bronchiolitis caused by Respiratory Syncytial Virus (RSV-bronchiolitis) per 10 μg/m^3^ increase in exposure.^22^ Dondi et al.^26^ evaluated PM2.5 exposures over a similar period and reported a 6% increase in the odds of bronchiolitis-related admissions in the first year of life that was associated with increased exposure to PM2.5 in the preceding 4 weeks among children in an urban center in Italy. We found a stronger association for NO2 and bronchiolitis in the month preceding the first birthday. In contrast, in the only other investigation of monthly NO2 exposures, Karr et al.^22^ reported null associations with both overall bronchiolitis admissions as well as those specifically attributed to RSV infections.

Our study therefore adds to previous literature, primarily from North America, by examining temporally resolved exposure windows within the first year of life in a large population-based cohort in London and providing insight into how associations between air pollution and bronchiolitis may vary across infancy. In this context, the associations observed in the first and final landmark periods may point to periods warranting further investigation, although the inconsistency of associations across the landmark periods calls for cautious interpretation of these findings.

Strengths

Key strengths of this study include the use of a birth cohort based on linkage between civil registration data and hospital admission data for all children born in Greater London, a large and diverse urban area, but with substantial variation in exposure to both PM2.5 and NO2. Our use of linked administrative data minimizes selection bias, particularly as all children ordinarily resident in England can access secondary care services within the NHS free at the point of need.^50^ We applied a landmarking approach that allowed us to examine age-specific associations across successive periods during infancy, providing insight into potential windows of susceptibility in early life. This is particularly important for bronchiolitis, for which the risk of hospital admission varies greatly during infancy.^8^ Furthermore, we used a spatio-temporally fine measure of air pollution, derived from a well-established atmospheric transport model verified using measured concentrations from continuous monitoring sites throughout the Greater London Area. We were able to make use, for the first time in an English setting, of linked, longitudinal NHS address registers and capture residential mobility over time for mothers during pregnancy and children during infancy and assign air pollution exposures at each residential postcode. This allowed us to derive longitudinal measures of air pollution exposure where mothers and children resided, rather than solely using the address at birth registration, as in previous studies.^51^ We accessed information from birth registration and notification records to control for several sociodemographic characteristics for nearly half a million children, both at the individual child level (such as maternal age and registration type; both strongly associated with poverty^51,52^), and the IMD, a small-area level indicator of deprivation. This is the first study to incorporate information on home energy efficiency at the postcode level, a key housing-related determinant of respiratory health,^53–55^ independent of socioeconomic status. Energy efficiency may influence exposure to outdoor air pollution, with building airtightness and ventilation systems affecting the infiltration of outdoor pollutants, the removal of indoor-generated pollutants, and potential mold risks.^56^ Accounting for it may help reduce potential confounding by housing-related factors.

Limitations

Several limitations should be considered. We obtained information on bronchiolitis admission from the HES APC dataset, which only includes admissions to hospital and not bronchiolitis diagnosed and treated in primary care or in emergency departments (ED). Therefore, while we were able to assess the impact of air pollution exposure during pregnancy and infancy on NHS inpatient admissions due to bronchiolitis, we could not determine the association between air pollution exposure and bronchiolitis that does not require hospital admission. Of note is that ED visitation rates among infants in London are generally higher, and emergency admission rates lower, than in other parts of England,^57^ which may partially explain why we did not observe an association between either PM2.5 or NO2 exposure and bronchiolitis admissions. We were not able to carry out a national analysis for England as we only had air pollution data with such high temporal and spatial resolution available for London.

Furthermore, ED data for England for our study period did not cover all EDs, and diagnostic coding in the HES dataset for ED attendances was generally considered to be poor.^58^ A new ED dataset for England, the Emergency Care Dataset, was fully rolled out across England in 2020, with an improved diagnostic coding system. Linkage between environmental data and the Emergency Care Dataset will allow research into the impact of air pollution on bronchiolitis seen in EDs. Primary care data can currently be linked to environmental data for research in Wales,^59^ but not England. Linkage between environmental and primary care data for the large and diverse English population would be a major step forward, allowing the association between air pollution and less severe bronchiolitis symptoms treated in primary care to be investigated. Further, routinely collected data from hospital laboratories on test results for specific viral pathogens such as RSV, the main cause of bronchiolitis,^60^ were not widely available for linkage for research studies in England. Although we adjusted for month and year of birth, residual confounding by seasonal variation in circulating viruses cannot be ruled out.

Data on exposure to tobacco smoke, either during pregnancy or in the home during infancy, is not available in England unlike in Scotland.^61^ The Maternity Services Dataset was introduced in April 2015 across England and collects data on maternity service use in the NHS, including on smoking and other risk factors during pregnancy.^62^ Future studies in England should include further linkage to the Maternity Services Dataset to account for smoking exposure during pregnancy. Our measures of both air pollution concentrations and energy efficiency were at the postcode level and not assigned to individual addresses, which could introduce non-differential exposure misclassification, although UK postcodes represent small groups of addresses and provide relatively fine spatial resolution. Furthermore, this exposure assessment approach may not fully capture time-activity patterns or exposures in other environments, such as schools. A new national birth cohort for England, the Kids’ Environment and Health Cohort, will allow linkage of air pollution, housing quality, and other environmental exposures to health data for children in England at the individual address level over time.^63^ We also did not account for exposure to other air pollutants, which could potentially confound the observed estimates, such as coarse particulate matter (PM10), ozone (O3), and sulphur dioxide (SO2), or for indoor air pollution, despite the substantial amount of time that mothers and young children spend indoors. Future analyses should consider other household characteristics, such as damp/mold, overcrowding, and fuel poverty, that may contribute to poor respiratory health in children, to examine the joint impact of poor housing conditions and outdoor air pollution exposures. Further, our analysis included only the first admission for each child, and we did not assess recurrent cases. Going forward, research could examine the impact of air pollution on repeat admissions for bronchiolitis and other lower respiratory infections in children.

Conclusions

Using a population-based administrative data birth cohort, we found limited evidence of associations between PM2.5 or NO2 exposure and the risk of bronchiolitis-related admission, although suggestive signals were observed for both pollutants in relation to prenatal exposures and exposure in the final month of infancy. Future studies should look beyond hospital admissions to include bronchiolitis episodes in EDs and primary care, to interrogate the relationship between air pollution and bronchiolitis and identify potential windows of vulnerability.

Conflicts of interest statement

The authors declare that they have no conflicts of interest with regard to the content of this report.

ACKNOWLEDGMENTS

We thank Tom Clemens, Chris Dibben and Linda Wijlaars for their valuable support and contributions to this project.

Supplementary Material